Progress in Advanced Computing and Intelligent Engineering: Proceedings of ICACIE 2019, Volume 1 [1st ed.] 9789811565830, 9789811565847

Citation preview

Advances in Intelligent Systems and Computing 1198

Chhabi Rani Panigrahi · Bibudhendu Pati · Prasant Mohapatra · Rajkumar Buyya · Kuan-Ching Li   Editors

Progress in Advanced Computing and Intelligent Engineering Proceedings of ICACIE 2019, Volume 1

Advances in Intelligent Systems and Computing Volume 1198

Series Editor Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Advisory Editors Nikhil R. Pal, Indian Statistical Institute, Kolkata, India Rafael Bello Perez, Faculty of Mathematics, Physics and Computing, Universidad Central de Las Villas, Santa Clara, Cuba Emilio S. Corchado, University of Salamanca, Salamanca, Spain Hani Hagras, School of Computer Science and Electronic Engineering, University of Essex, Colchester, UK László T. Kóczy, Department of Automation, Széchenyi István University, Gyor, Hungary Vladik Kreinovich, Department of Computer Science, University of Texas at El Paso, El Paso, TX, USA Chin-Teng Lin, Department of Electrical Engineering, National Chiao Tung University, Hsinchu, Taiwan Jie Lu, Faculty of Engineering and Information Technology, University of Technology Sydney, Sydney, NSW, Australia Patricia Melin, Graduate Program of Computer Science, Tijuana Institute of Technology, Tijuana, Mexico Nadia Nedjah, Department of Electronics Engineering, University of Rio de Janeiro, Rio de Janeiro, Brazil Ngoc Thanh Nguyen , Faculty of Computer Science and Management, Wrocław University of Technology, Wrocław, Poland Jun Wang, Department of Mechanical and Automation Engineering, The Chinese University of Hong Kong, Shatin, Hong Kong

The series “Advances in Intelligent Systems and Computing” contains publications on theory, applications, and design methods of Intelligent Systems and Intelligent Computing. Virtually all disciplines such as engineering, natural sciences, computer and information science, ICT, economics, business, e-commerce, environment, healthcare, life science are covered. The list of topics spans all the areas of modern intelligent systems and computing such as: computational intelligence, soft computing including neural networks, fuzzy systems, evolutionary computing and the fusion of these paradigms, social intelligence, ambient intelligence, computational neuroscience, artificial life, virtual worlds and society, cognitive science and systems, Perception and Vision, DNA and immune based systems, self-organizing and adaptive systems, e-Learning and teaching, human-centered and human-centric computing, recommender systems, intelligent control, robotics and mechatronics including human-machine teaming, knowledge-based paradigms, learning paradigms, machine ethics, intelligent data analysis, knowledge management, intelligent agents, intelligent decision making and support, intelligent network security, trust management, interactive entertainment, Web intelligence and multimedia. The publications within “Advances in Intelligent Systems and Computing” are primarily proceedings of important conferences, symposia and congresses. They cover significant recent developments in the field, both of a foundational and applicable character. An important characteristic feature of the series is the short publication time and world-wide distribution. This permits a rapid and broad dissemination of research results. ** Indexing: The books of this series are submitted to ISI Proceedings, EI-Compendex, DBLP, SCOPUS, Google Scholar and Springerlink **

More information about this series at http://www.springer.com/series/11156

Chhabi Rani Panigrahi Bibudhendu Pati Prasant Mohapatra Rajkumar Buyya Kuan-Ching Li •

•

•

•

Editors

Progress in Advanced Computing and Intelligent Engineering Proceedings of ICACIE 2019, Volume 1

123

Editors Chhabi Rani Panigrahi Department of Computer Science Rama Devi Women’s University Bhubaneswar, Odisha, India Prasant Mohapatra Department of Computer Science University of California Davis, CA, USA Kuan-Ching Li Department of Computer Science and Information Engineering Providence University Taichung, Taiwan

Bibudhendu Pati Department of Computer Science Rama Devi Women’s University Bhubaneswar, Odisha, India Rajkumar Buyya Cloud Computing and Distributed Systems (CLOUDS) Laboratory, School of Computing and Information Systems University of Melbourne Melbourne, VIC, Australia

ISSN 2194-5357 ISSN 2194-5365 (electronic) Advances in Intelligent Systems and Computing ISBN 978-981-15-6583-0 ISBN 978-981-15-6584-7 (eBook) https://doi.org/10.1007/978-981-15-6584-7 © Springer Nature Singapore Pte Ltd. 2021 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

Preface

This volume contains the papers presented at the 4th International Conference on Advanced Computing and Intelligent Engineering (ICACIE) 2019: The 4th International Conference on Advanced Computing and Intelligent Engineering (www.icacie.com) was held during December 21–23, 2019, at Rama Devi Women’s University, Bhubaneswar, India. There were 287 submissions, and each qualified submission was reviewed by a minimum of two Technical Program Committee members using the criteria of relevance, originality, technical quality, and presentation. The committee accepted 86 full papers for oral presentation at the conference, and the overall acceptance rate is 29%. ICACIE 2019 was an initiative taken by the organizers which focuses on research and applications on topics of advanced computing and intelligent engineering. The focus was also to present the state-of-the-art scientific results, to disseminate modern technologies, and to promote collaborative research in the field of advanced computing and intelligent engineering. Researchers presented their work in the conference and had an excellent opportunity to interact with eminent professors, scientists, and scholars in their area of research. All participants were benefitted from discussions that facilitated the emergence of innovative ideas and approaches. Many distinguished professors, well-known scholars, industry leaders, and young researchers were participated in making ICACIE 2019 an immense success. We had also an industry panel discussion, and we invited people from software industries like TCS, Infosys, and Cognizant and entrepreneurs. We thank all the Technical Program Committee members and all reviewers/sub-reviewers for their timely and thorough participation during the review process. We express our sincere gratitude to Prof. Padmaja Mishra, Honorable Vice Chancellor and Chief Patron of ICACIE 2019, to allow us to organize ICACIE 2019 on the campus and for her unending timely support toward organization of this conference. We would like to extend our sincere thanks to Prof. Bibudhendu Pati and Dr. Hemant Kumar Rath, General chairs of ICACIE 2019, for their valuable guidance during review of papers as well as other aspects of the conference. We appreciate the time and efforts put in by the members of the local organizing team at Rama Devi Women’s University, Bhubaneswar, India, v

vi

Preface

especially the faculty members of the Department of Computer Science, student volunteers, and administrative staff, who dedicated their time and efforts to make ICACIE 2019 successful. We would like to extend our thanks to Er. Subhashis Das Mohapatra for designing and maintaining ICACIE 2019 Website. We are very grateful to all our sponsors, especially Department of Science and Technology (DST), Government of India, under Consolidation of University Research for Innovation and Excellence in women universities (CURIE) project for its generous support toward ICACIE 2019. Bhubaneswar, India Bhubaneswar, India Davis, USA Melbourne, Australia Taichung, Taiwan

Chhabi Rani Panigrahi Bibudhendu Pati Prasant Mohapatra Rajkumar Buyya Kuan-Ching Li

Contents

Advanced Image and Video Processing Applications Crack Detection on Inner Tunnel Surface Using Image Processing . . . . Debanshu Biswas, Ipsit Nayak, Shaibal Choudhury, Trishaani Acharjee, Sidhant, and Mayank Mishra Novel Approach for Resolution Enhancement of Satellite Images Using Wavelet Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mansing Rathod, Jayashree Khanapuri, and Dilendra Hiran Vehicle Number Plate Detection: An Edge Image Based Approach . . . . Kalyan Kumar Jena, Soumya Ranjan Nayak, Sasmita Mishra, and Sarojananda Mishra A High-Precision Pixel Mapping Method for Image-Sensitive Areas Based on SVR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Huang Jing, Amit Yadav, Asif Khan, and Dakshina Yadav Nucleus Segmentation from Microscopic Bone Marrow Image . . . . . . . Shilpa, Rajesh Gopakumar, and Vasundhara Acharya Developing a Framework for Acquisition and Analysis of Speeches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Md. Billal Hossain, Mohammad Shamsul Arefin, and Mohammad Ashfak Habib

3

13 24

35 44

51

Gait-Based Person Identification, Gender Classification, and Age Estimation: A Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Rupali Patua, Tripti Muchhal, and Saikat Basu

62

Efficient Watermarking in Color Video Using DWT-DFT and SVD Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mangal Patil, Preeti Bamane, and Supriya Hasarmani

75

vii

viii

Contents

Motif Discovery and Anomaly Detection in an ECG Using Matrix Profile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Rutuja Wankhedkar and Sanjay Kumar Jain ISS: Intelligent Security System Using Facial Recognition . . . . . . . . . . . Rajesh Kumar Verma, Praveen Singh, Chhabi Rani Panigrahi, and Bibudhendu Pati

88 96

G.V Black Classification of Dental Caries Using CNN . . . . . . . . . . . . . . 102 Prerna Singh and Priti Sehgal Electronics and Electrical Applications Pre-emptive Spectrum Access in Cognitive Radio for Better QoS . . . . . 115 Avirup Das, Sandip Karar, Nabanita Das, and Sasthi C. Ghosh Design of Smart Antenna Arrays for WiMAX Application Using LMS Algorithm Under Fading Channels . . . . . . . . . . . . . . . . . . . 127 Anupama Senapati, Pratyushna Singh, and Jibendu Sekhar Roy Performance Comparison of Variants of Hybrid FLANN-DE for Intelligent Nonlinear Dynamic System Identification . . . . . . . . . . . . 137 Swati Swayamsiddha Load Cell and FSR-Based Hand-Assistive Device . . . . . . . . . . . . . . . . . . 148 Acharya K. Aneesha, Somashekara Bhat, and M. Kanthi Power Quality Analysis of a Distributed Generation System Using Unified Power Quality Conditioner . . . . . . . . . . . . . . . . . . . . . . . 157 Sarita Samal, Akansha Hota, Prakash Kumar Hota, and Prasanta Kumar Barik A Hybrid: Biogeography-Based Optimization-Differential Evolution Algorithm Based Transient Stability Analysis . . . . . . . . . . . . . . . . . . . . 170 P. K. Dhal Restraining Voltage Fluctuations in Distribution System with Jaya Algorithm-Optimized Electric Spring . . . . . . . . . . . . . . . . . . . . . . . . . . . 180 K. Keerthi Deepika, J. Vijayakumar, and Gattu Kesava Rao An Experimental Setup to Study the Effects of Switcing Transients for Low Voltage Underground Cable . . . . . . . . . . . . . . . . . . . . . . . . . . . 191 Sanhita Mishra, A. Routray, and S. C. Swain Effect of Distributed Generator on Over Current Relay Behaviour . . . . 199 Tapaswini Biswal

Contents

ix

Advanced Network Applications Detection and Prevention from DDoS Attack Using Software-Defined Security . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207 Sumit Badotra, Surya Narayan Panda, and Priyanka Datta Dynamic Resource Aware Scheduling Schemes for IEEE 802.16 Broadband Wireless Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218 M. Deva Priya, A. Christy Jeba Malar, S. Sam Peter, G. Sandhya, L. R. Vishnu Varthan, and R. Vignesh Evaluation of the Applicability and Advantages of Application of Artificial Neural Network Based Scanning System for Grid Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231 Shubhranshu Kumar Tiwary, Jagadish Pal, and Chandan Kumar Chanda A Realistic Sensing Model for Event Area Estimation in Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244 Srabani Kundu, Nabanita Das, and Dibakar Saha Generic Framework for Privacy Preservation in Cyber-Physical Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257 Rashmi Agarwal and Muzzammil Hussain A Novel Authentication Scheme for VANET with Anonymity . . . . . . . . 267 Harshita Pal and Bhawna Narwal A Survey on Handover Algorithms in Heterogeneous Wireless Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277 Mithun B Patil and Rekha Patil A Non-stationary Analysis of Erlang Loss Model . . . . . . . . . . . . . . . . . . 286 Amit Kumar Singh, Dilip Senapati, Sujit Bebortta, and Nikhil Kumar Rajput A Novel Authentication Scheme for Wireless Body Area Networks with Anonymity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295 Upasna Singh and Bhawna Narwal Preserving Privacy of Data in Distributed Systems Using Homomorphic Encryption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306 P. Kalyani, M. Masooda, and P. Namrata Advanced Algorithms and Soft Computing Applications Performance Evaluation of Composite Fading Channels Using q-Weibull Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317 Tanmay Mukherjee, Bibudhendu Pati, and Dilip Senapati

x

Contents

Benchmarking Performance of Erasure Codes for Linux Filesystem EXT4, XFS and BTRFS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325 Shreya Bokare and Sanjay S. Pawar Exploration of Cognition Impact: An Experiment with Cover Song Retrieval Through Indexing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335 D. Khasim Vali and Nagappa U. Bhajantri An Effective Hybrid Approach for Solving Prioritized Cube Selection Problem Using Particle Swarm Optimization and Tabu Search . . . . . . . 347 Anjana Gosain and Heena Madaan BOSCA—A Hybrid Butterfly Optimization Algorithm Modified with Sine Cosine Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360 Sushmita Sharma and Apu Kumar Saha Adaptive Applications of Maximum Entropy Principle . . . . . . . . . . . . . 373 Amit Kumar Singh, Dilip Senapati, Tanmay Mukherjee, and Nikhil Kumar Rajput Identifying Challenges in the Adoption of Industry 4.0 in the Indian Construction Industry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380 Arpit Singh and Subhas Chandra Misra Human Action Recognition Using STIP Evaluation Techniques . . . . . . . 399 H. S. Mohana and U. Mahanthesha Fuzzy Cognitive Map-Based Genetic Algorithm for Community Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412 K. Haritha and M. V. Judy Evaluation of Digital Forensic Tools in MongoDB Database Forensics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427 Rupali Chopade and Vinod Pachghare

About the Editors

Dr. Chhabi Rani Panigrahi is Assistant Professor in the P.G. Department of Computer Science at Rama Devi Women’s University, Bhubaneswar, India. She completed her Ph.D. from Department of Computer Science and Engineering, Indian Institute of Technology Kharagpur, India. Her research interest areas include Software Testing and Mobile Cloud Computing. She holds 19 years of teaching and research experience. She has published several international journals and conference papers. She is a Life Member of Indian Society of Technical Education (ISTE) and member of IEEE and Computer Society of India (CSI). Dr. Bibudhendu Pati is Associate Professor and Head of the P.G. Department of Computer Science at Rama Devi Women’s University, Bhubaneswar, India. He completed his Ph.D. from IIT Kharagpur. Dr. Pati has 21 years of experience in teaching, research. His interest areas include Wireless Sensor Networks, Cloud Computing, Big Data, Internet of Things, and Network Virtualization. He has got several papers published in journals, conference proceedings and books of international repute. He is a Life Member of Indian Society of Technical Education (ISTE), Life Member of Computer Society of India, and Senior Member of IEEE. Prof. Prasant Mohapatra is serving as the Vice Chancellor for Research at University of California, Davis. He is also a Professor in the Department of Computer Science and served as the Dean and Vice-Provost of Graduate Studies during 2016-18. He was the Department Chair of Computer Science during 2007-13. In the past, Dr. Mohapatra has also held Visiting Scientist positions at Intel Corporation, Panasonic Technologies, Institute of Infocomm Research (I2R), Singapore, and National ICT Australia (NICTA). Dr. Mohapatra received his doctoral degree from Penn State University in 1993, and received an Outstanding Engineering Alumni Award in 2008. He is also the recipient of Distinguished Alumnus Award from the National Institute of Technology, Rourkela, India. Dr. Mohapatra received an Outstanding Research Faculty Award from the College of Engineering at the University of California, Davis. He received the HP Labs Innovation awards in 2011, 2012, and 2013. He is a Fellow of the IEEE and a xi

xii

About the Editors

Fellow of AAAS. Dr. Mohapatra’s research interests are in the areas of wireless networks, mobile communications, cyber security, and Internet protocols. He has published more than 350 papers in reputed conferences and journals on these topics. Dr. Mohapatra’s research has been funded through grants from the National Science Foundation, US Department of Defense, US Army Research Labs, Intel Corporation, Siemens, Panasonic Technologies, Hewlett Packard, Raytheon, and EMC Corporation. Prof. Rajkumar Buyya is a Redmond Barry Distinguished Professor and Director of the Cloud Computing and Distributed Systems (CLOUDS) Laboratory at the University of Melbourne, Australia. He is also serving as the founding CEO of Manjrasoft, a spin-off company of the University, commercializing its innovations in Cloud Computing. He has authored over 650 publications and seven text books including “Mastering Cloud Computing” published by McGraw Hill, China Machine Press, and Morgan Kaufmann for Indian, Chinese and international markets respectively. Dr. Buyya is one of the highly cited authors in computer science and software engineering worldwide (h-index=120, g-index=255, 76,800+ citations). He is named in the recent Clarivate Analytics’ (formerly Thomson Reuters) Highly Cited Researchers and “World’s Most Influential Scientific Minds” for three consecutive years since 2016. Dr. Buyya is recognized as Scopus Researcher of the Year 2017 with Excellence in Innovative Research Award by Elsevier for his outstanding contributions to Cloud computing. He served as founding Editor-in-Chief of the IEEE Transactions on Cloud Computing. He is currently serving as Editor-in-Chief of Software: Practice and Experience, a long standing journal in the field established *50 years ago. Prof. Kuan-Ching Li is currently a Professor in the Department of Computer Science and Information Engineering at the Providence University, Taiwan. He was the Vice-Dean for Office of International and Cross-Strait Affairs (OIA) in this same university since 2014. Prof. Li is recipient of awards from Nvidia, Ministry of Education (MOE)/Taiwan and Ministry of Science and Technology (MOST)/ Taiwan, as also guest professorship from different universities in China. He got his PhD from University of Sao Paulo, Sao Paulo, Brazil in 2001. His areas of research are networked and GPU computing, parallel software design, and performance evaluation and benchmarking. He has edited 2 books: Cloud Computing and Digital Media and Big Data published by CRC Press. He is a Fellow of the IET, senior member of the IEEE and a member of TACC.

Advanced Image and Video Processing Applications

Crack Detection on Inner Tunnel Surface Using Image Processing Debanshu Biswas(B) , Ipsit Nayak, Shaibal Choudhury, Trishaani Acharjee, Sidhant, and Mayank Mishra School of Electronics Engineering, Kalinga Institute of Industrial Technology (KIIT) Deemed to be University, Bhubaneswar, India [email protected], [email protected], [email protected], [email protected], [email protected], [email protected]

Abstract. Cracks in the concrete structures such as cracks in the inner surface of tunnels are minor fault, however, can cause major damage or loss of lives if not checked frequently. The current method of detecting cracked surface is manual inspection by hand measuring tools and drawing sheets which may not be feasible as the tunnel needs to be blocked for a limited time period, till the inspection is in progress. By image processing, we can analyze the digital images captured from inside the tunnel for localization of cracks. The algorithm proposed in this paper can be applied to an image of the cracked surface of a tunnel for detecting the crack. Moreover, the length of the crack can also be measured in pixels. Keywords: Crack detection · Tunnel faults · Surface crack · Image processing · Automatic detection

1 Introduction Cracks may seem very minor when their formation starts but, it is one of the most fatal defects in concrete structures. The crack increases the stress at a point and as a result, causes damage in the structure. Initially, the cracks may be very difficult to spot with naked eye as it is very thin and hardly visible. Early detection of the crack can prevent a lot of damages from occurring. Some widely used techniques to examine the cracks include Scanning Electron Microscopy (SEM), Infrared (IR) Spectroscopy, and Ultrasonography [1, 2]. However, these methods may not be feasible in various situations because of the complexity of these methods. Underground tunnels are one of the main ways to decrease traffic congestion in urban cities. However, due to aging, temperature difference, topographic change, tunnels suffer from internal cracks and defects which need to be monitored for health status of the tunnel. Currently, in India, many major cities like Delhi have major metro line connectivity under the ground and require regular inspection and maintenances. Conventional methods of determining cracks are performed manually by hand tools such as measuring © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_1

4

D. Biswas et al.

tools and drawing sheets [3]. These methods have less reliability and it is not feasible as the tunnel needs to be blocked for some limited period for inspection before the tunnel can again function normally. Moreover, quality of inspection can vary from person to person and their work experiences [4]. A person with a good amount of work experience will focus on the elements which require inspection but another person may not be able to do so. Thus, by removing person to person view of crack and improve the crack finding method, we need to create a fully automated crack detection system for more dependable techniques, using image processing [5, 6]. For making the maintenance work faster, it is quite essential to develop some systems that can automatically detect the cracks from the images captured. The width, length and type of a crack can give a lot of information about the crack and can also determine the duration up to which the concrete can take up the given amount of load at the site of the crack. By using commercial purpose cameras, many images of concrete surface inside a tunnel can be taken efficiently in short period of time. For practical application setups, these cameras with the help of MATLAB can be used for analysis of the cracks. Image-based crack detection can also face a lot of difficulties. These difficulties mostly arise due to deformed shapes and diverse sizes of the crack, poor lighting in the area and concrete spawl in the image taken. These may result in the detection of various other deformities that might not be present but still comes in the foreground as a part of the crack, reducing the accuracy of the output. Generally, the time needed for detecting the crack using image processing depends on the size of the image. Since the digital cameras have high resolutions such as 20 megapixels and beyond, these can increase the details of the captured image but subsequently reduce the speed of image processing. Therefore, it is important and required to reduce the hardware cost for detecting cracks efficiently. Even after overcoming all the possible difficulties, enhancing the image and sheerly highlighting the crack, for better understanding, we could convert the two dimensional (2D) image to a three dimensional (3D) image. It would help us to concentrate on the crack in a very detailed manner, on an additional Z plane which intensifies the crack from the image giving the exact shape and range of the area which need immediate attention. In this paper, Sect. 2 gives a brief review of related work; Sect. 3 presents the algorithm proposed which would clear up the captured image using various functions; the experimental results with images are given in Sect. 4; conclusion and future work related to this research are given in Sect. 5.

2 Related Work Crack detection through image processing based method facilitates quick check-up on a large surface area of a structure. Moreover, automated process enables stable and accurate inspection. Currently, crack detection using image processing for inspection of big structures like bridges, pavements and buildings is trending and is under high focus. Image processing based crack inspection techniques have been a point of attention for many researchers. Cracks are fragile areas of a concrete surface, which provides important information for segmentation. Because of uneven lighting, crooked surface

Crack Detection on Inner Tunnel Surface

5

and variation of crack types, image segmentation became a compulsory footstep during extraction of crack. Hutchinson et al. [7] suggested a method to distinguish the crack from the images of concrete by combining wavelet and cranny transform. These image segmentation algorithms can completely extract the edge of a crack under uniform illumination. An adaptive threshold algorithm to solve the image segmentation was also suggested by Navon et al. [8]. A crack identification system was devised by Yu et al. [9] by applying the Dijkstra algorithm to locate the cracks from the pavement. However, the recognition precision was low. Machine vision technology was applied by Sinha et al. [10] to recognize and distinguish the cracks inside a concrete pipe.

3 Methodology In this section, the key means for detecting cracks is provided. As the traditional methods were not feasible for analyzing the cracks of the tunnel, image processing based analysis is used which provides more accurate results and takes comparatively less time than traditional manual methods [11]. The proposed algorithm is shown in Fig. 1. First, collect the image of the tunnel surface which will work as the prime subject to crack detection. After image acquisition, the images are pre-processed using few filters and techniques such as median filtering. Finally, the dark areas with potential crack defects are segmented out from background of the gray-scale image by using morphological image processing operations. The following steps show the general architecture of the proposed algorithm. 3.1 Image Acquisition The images are acquired from inside the tunnel for crack analysis. The images collected are stored for later or real-time observation of the cracks. General or commercial purpose cameras can be used for capturing the images. An automated system can be used to capture the images. 3.2 Preprocessing Analysis Image taken from the tunnel is converted to gray-scale image; then it is processed to enhance various aspects of an image such as contrast, hue, saturation by different functions as shown in Fig. 2. The resolution of the image also plays an important part in finding a crack in an image, if the resolution of the image taken from the tunnel is increased without increasing the quality, the amount of unnecessary data is increased which creates unwanted noise resulting in poor accuracy. Inside a tunnel, there may be uneven illumination during image capturing, so a low light image enhancement technique is implemented for improving the brightness of the image as per our requirement.

6

D. Biswas et al.

Fig. 1. Flow chart of the algorithm

3.3 Median Filtering Median filtering is a nonlinear technique used to remove noise from image. It is a very efficient way of removing the noise while preserving the edges. This method works by moving through the image pixels and replacing the target value of the pixel with the median value of the neighboring pixel. First all the pixel values are sorted from the window in numerical order and then changing the pixel value with the median pixel value. Moreover, the median value should be the value from one of the adjacent pixel, so the values created after this filtering is not unrealistic. 3.4 Enhancement by Bottom–Hat Function The proposed algorithm uses bottom-hat function which enhances dark spots in a light background. It subtracts the morphological close (joins the circles in an image together

Crack Detection on Inner Tunnel Surface

7

Fig. 2. Flowchart of pre-processing

by filling in the gaps between them) of the image, fills the holes and joins nearby objects. Bottom-hat is an operation performed in order to highlight all the unlighted regions in different images. Bottom-hat efficiently flips high-frequency regions. In image processing, bottom-hat transforms are extensively used for performing several tasks such as feature extension, image enhancement, background equalization, etc. Using Bottom-hat transform in an image, it is possible to obtain details such as the edge, surface. This process allows extracting the dark features. 3.5 Binarization Binarization is used for separating background from crack as the gray value of the background is higher than the crack. There are two major problems that arise from this method: 1. Some gray value in the background can be similar to the crack making some objects difficult to detect from the crack. 2. Many unwanted objects such as wires, pipes and other objects may be present which causes interference in the image. Therefore, it is important to solve these problems. For this, we calculated the average value of T using the N × N adjacent pixel value surrounded by pixels and set a threshold A.  0, T (a, b) < T −  S(a, b) = (1) 255 T (a, b) ≥ T −  S(a, b) stands for the processed image; T (a, b) stands for the original image; and  is the threshold.

8

D. Biswas et al.

3.6 Noise Removal After the process of binarization, noise can be present in the images in different forms which also need to be removed such as: 1. Uneven illumination or background structures in the tunnel may have higher gray value than that of crack. 2. The gray value of pipes, wires and other structure is low as compared to the background and they take up much greater portion than the crack. Noise removal is followed to overcome the above interferences, i.e. looking for connected black pixels. As gray value of some facilities such as pipes, wires inside a tunnel are low and concentrated. According to this feature, the noise is removed by converting the gray value of the other disturbances to zero. Moreover, the width of any crack does not exceed a fixed amount of threshold. If the adjacent pixels are located in dark region having a width greater than the fixed threshold, then that region will be removed. 3.7 Skelton Function This process is further used for calculating the length of the crack in terms of pixels. By using pixels in an image, estimation of crack length can be done if the value of pixels denoting millimeter (mm) or meter (m) is known in real life. This process is further used for calculating the length of the crack in terms of pixels. By using pixels in an image, estimation of crack length can be done if the value of pixels denoting mm or m is known in real life. 3.8 Pixel Length By using measurement function, length of a crack in a particular image can be interpreted in terms of pixels. If the number of pixels representing real-life mm or m is known for a particular image, then an approximate length can be measured for the respective crack. 3.9 3D Representation For having a better understanding of a crack, we convert the two-dimensional (2D) image into three-dimensional (3D) structure. The 3D view can project the crack in an additional plane (z), in this, only the pixels denoting the cracks are intensified and is shown escalated in an extra plane.

4 Results and Discussion In this section, the results of the experiment are presented. The proposed algorithm was implemented in MATLAB 2014a and tested in windows 10. An algorithm for detecting tunnel cracks was implemented which includes a demonstration of cracks while

Crack Detection on Inner Tunnel Surface

9

removing all the noise from the background. The nature of the crack depends on the composition of the material, geometry and structural integrity. These features are not consistent thus the cracking behavior of concrete tunnel elements are more complicated than it seems. The main advantage of this algorithm is getting an accurate and noise-free image of the crack. In order to make our crack distinct, we use the median filtering, bottom-hat technique, binarization to get rid of the surrounding objects. Median filter is used as it efficiently removes the noises and preserves the edges as shown in Fig. 3b.

Fig. 3. a Original gray image and b median filter

We also use low light image enhancement technique to improve the brightness and remove the shadows in the image as per our requirement as shown in Fig. 4a. Bottomhat function is used to enhances and separate all the dark spots in a light background as shown in Fig. 4b. Binarization is used for separating background from crack as the gray value of the background is higher than the crack as shown in Fig. 4c. The resultant image is not always noise-free and it may give normal background objects as faults in concrete, as some gray values of the background may look similar to the crack, and thus making it difficult to distinguish the crack from the background. Noise removal is followed to overcome the interferences, which is looking for pixels which less intensity than the crack as shown in Fig. 4d. The length of the crack can be calculated in terms of pixels, as shown in Fig. 4e for a better understanding of a crack while detection. Moreover, the 2D image is converted into 3D structure. In this, only the pixels representing the cracks are elevated in additional planes as shown in Fig. 5. This may help in analyzing the characteristics of crack just by looking at the 3D view of the graph.

5 Conclusions In this paper, an efficient approach for crack detection on inner tunnel surface using image processing is proposed. This algorithm’s main contribution is highlighting cracks, by an automated crack detection system, which will discard all decision making steps made in traditional methods.

10

D. Biswas et al.

Fig. 4. Illustrations of the procedure. a Low lighting enhancement image. b Image after bottomhat process. c Binary image. d Image after noise removal. e Skeleton function and length estimation. Pixel count 1498.58

While simplifying the images by preprocessing, the algorithm makes use of median filter, low-light image enhancement function, bottom-hat function, binarization function and noise removal process for effectively isolating the crack features from other objects or background. Moreover, this paper provides an automated system for measuring the length of the crack in terms of pixel and 3D view of the crack.

Crack Detection on Inner Tunnel Surface

11

Fig. 5. 3D view of the crack

Proposed algorithm can be easily integrated into many crack detection models as it is relatively straightforward and as demonstrated in experimental results, is capable of delivering precise crack detection. Therefore, this algorithm can potentially be applied by the maintenance department of a tunnel for crack detection and appraisal. A limitation of this approach is that the users have to adjust some parameters such as threshold in binarization and noise removal for some different types of background features. Our future work will be based on how to remove this limitation. Acknowledgements. The authors are grateful to the School of Electronics Engineering, Kalinga Institute of Industrial Technology (KIIT) Deemed to be University, Bhubaneswar, India for providing help and all support to carry out this work.

References 1. Talaba AMA, Huanga Z, Xia F, HaiMinga L (2016) Detection crack in image using Otsu method and multiple filtering in image processing techniques. Optik 127:1030–1033 2. Mohan A, Poobal S (2018) Crack detection using image processing: A critical review and analysis. Alexandria Eng J 57:787–798 3. Sohn HG, Lim YM, Yun KH, Kim GH (2005) Monitoring crack changes in concrete structures. Comput-Aid Civil Infrastr Eng 20:52–61 4. Rimkus A, Podviezko A, Gribniak V (2015) Processing digital images for crack localization in reinforced concrete members. Proc Eng 122:239–243 5. Yokoyama S, Matsumoto T (2017) Development of an automatic detector of cracks in concrete using machine learning. Proc Eng 177:1250–1255 6. Hutchinson MTC, Chen ZQ (2006) Improved image analysis for evaluating concrete damage. J Comput Civ Eng 20:210–216 7. Navon E, Miller O, Averbuch A (2005) Color image segmentation based on adaptive local thresholds. Image Vis Comput 23:69–85

12

D. Biswas et al.

8. Yu N, Jang JH, Han CS (2007) Auto inspection system using a mobile robot for detecting concrete cracks in a tunnel. Autom Constr 16:255–261 9. Sinha SK, Fieguth PW (2006) Automated detection of cracks in buried concrete pipe images. Autom Constr 15:58–72 10. Wang P, Huang H (2010) Comparison analysis on present image-based crack detection methods in concrete structures. In: Proceedings of 3rd International Congress on Image and Signal Processing (CISP2010), 2530–2533, https://doi.org/10.1109/CISP.2010.5647496 11. Qi D, Liu Y, Wu, X, Zhang Z (2014) An algorithm to detect the crack in the tunnel based on the image processing. In: Proceedings of Tenth International Conference on Intelligent Information Hiding and Multimedia Signal Processing, https://doi.org/10.1109/IIH-MSP.201 4.217

Novel Approach for Resolution Enhancement of Satellite Images Using Wavelet Techniques Mansing Rathod1,2(B) , Jayashree Khanapuri3 , and Dilendra Hiran2 1 Information Technology Department, K.J.S.I.E/IT, Mumbai 400022, India

[email protected]

2 Pacific University, Udaipur 313003, India

[email protected]

3 Electronics and Telecommunication Department, K.J.S.I.E/IT, Mumbai 400022, India

[email protected]

Abstract. Today, many researchers are working on satellite images, to solve resolution problems. Some of the techniques are used to enhance the resolutions that are Stationary Wavelet (SWT), Discrete Wavelet (DWT) and Integer Wavelet Transforms (IWT). These wavelet transforms are considered for improving the quality in terms of resolution enhancement. Low resolution (LR) image is used for processing and decomposed by mentioned transforms. Interpolation techniques are applied to manipulate the output of transform images. Some estimated images determined through the interpolation factor 2 to make equal sizes of images. All the images are integrated with inverse wavelet transform to generate high resolution (HR). In this work, we have compared wavelet transform. LANDSAT5 satellite images are considered for verification of implemented techniques. In this paper, three types of results are presented that are enhanced the image with size 128 × 128 to 512 × 512, 256 × 256 to 512 × 512 and iterative resolution. Some of the quality parameters are applied, i.e. PSNR, RMSE, MAE and MSE for verification for the performance of implemented techniques. Keywords: Stationary wavelet · Discrete wavelet and integer wavelet transforms · Resolution enhancement

1 Introduction Today, many of the fields need satellite images for example geographical information systems, building construction, GPS system and geosciences. But the resolution is the main issue [1]. Resolution Enhancement (RE) is an essential tool to break the mentioned problem [2]. In this paper, we have focused on wavelet transform techniques. Because frequency components can be easily altered in image and help in global processing of image. Computational cost be will reduced due to global processing. Here, we have taken the high resolution (HR) image with size 515 × 512, converted into low resolution (LR) image with size 128 × 128. LR image is processed by enhancing techniques, i.e. SWT&DWT, DWT&IWT and proposed technique. All the techniques © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_2

14

M. Rathod et al.

are compared with evaluation factors such as RMSE, PSNR, MAE and TIME. These techniques decomposed LR image, providing four types of sub-band images that are low-low (LL), low-high (LH), high-low (HL) and high-high (HH). DWT and IWT give downsampled images. But it is not in of case SWT, because it gives same sized sub-band images. Therefore, interpolation factor 2 is required only in DWT and IWT to resize the downsampled images, but not in SWT [4–11]. Estimated images are measured after applying interpolation factor 2. All the images are integrated by applying an inverse wavelet transform to provide HR image [21]. The result obtained from the proposed technique is better than the other two techniques. Computational time for proposed techniques is more as compared to others. Motivation and contribution is to integrate the characteristics of different wavelet transform techniques such as DWT, SWT and IWT. It gives a good resolution in terms of PSNR as compared to other techniques. But need more time for execution. The organization of the paper is as follows. In section one brief introduction and contribution are explained, Sect. 2 many literature survey have studied and summarized. In Sect. 3 various wavelet domain techniques have been explained with their block diagrams. Sections 4 and 5 provide the evaluation techniques and results discussion for implemented methods. Conclusion and future work have explained in Sect. 6 and at the end of the paper some of the references are listed.

2 Literature Survey There are many literature surveys that have studied about these techniques. It provides details about past and current status of resolution enhancement for the mentioned image. Paper [1] has proposed DWT technique. The authors have obtained sharper images and taken satellite image to verify the performance of proposed technique. It is also compared with other traditional techniques. In paper [2], the authors have implemented SWT&DWT techniques. It has integrated characteristics of both transform. They have compared the performance with interpolation and some of the wavelet transforms. It is concluded that the proposed technique gauge much better in terms of PSNR than others. In paper [3, 4], the authors have proposed enhancement of edges with the help of SWT and considered normal images, i.e. baboon and Lena. The results are compared with bilinear and bi-cubic interpolation and images are enhanced from 256 × 256 to 512 × 512. In paper [4, 5], the comparison has been done between DWT and SWT. Authors have achieved better resolution with sharpened image. Paper [5] has proposed only DWT and authors have compared with bi-cubic interpolation. Some evaluation parameters are considered for performance measures such as PSNR, RMSE and MSE. Paper [6] has discussed a novel technique by using interpolation and DWT. They have considered three images, i.e. Coloba-Mumbai, NE cost and Himalayas. The proposed technique is compared with interpolation techniques. Authors have proposed DWT&IWT technique in paper [7] and the results are compared with only DWT. It is concluded that proposed technique provides good results as compared to DWT. Authors have proposed the use of DWT&SWT and singular value decomposition (SVD) techniques to increase the resolution and brightness of satellite images [8]. Paper [9] has discussed DWT, LANDSAT8 satellite images are used for testing the mentioned technique. Authors have compared the

Novel Approach for Resolution Enhancement

15

proposed technique with conventional techniques. Paper [10] has discussed review and given brief explanation about DWT and SWT. Paper [11] has presented and described the transform techniques. They have added error back projection in DWT-SWT and also calculated correlation coefficient for measuring the performance of discussed techniques. Resolution and contrast are enhanced in paper [12]. Singular Value Decomposition is combined in DWT. They have compared with general and local histogram techniques. The authors have calculated number of pixels in blurred, noisy and clarified images and also determined the total objects, missing object and missing rate [13] in images. Paper [14] has studied denoising and resolution enhancement and authors have used retinex technique for the mentioned purpose. The quality is measured in-term of PSNR. Paper [15–18] discussed and compared transform domain techniques. The wavelet tool is used for implementation. Ikonos satellite images are used for verification of implemented techniques. Paper [19] has compared SWT, DWT and IWT. LANDSAT8, LANDSAT7 and IKONOS satellite images are considered to determine the performance of implemented techniques. High-quality image is achieved for LANDSAT8 in-terms of PSNR. Mathematical equations are discussed for 1D, 2D and 3D wavelet transforms [20]. In this section, we have studied literature related to resolution enhancement algorithms such as DWT, SWT and IWT for satellite images. Some of the papers are used different satellite images that are LANDSAT8, 7, 5 and IKONOS. Some of the authors enhanced image 256 × 256 to 512 × 512 and compared other size of, i.e. 128 × 128. It is concluded that greater input size provides better results [21].

3 Frequency Domain Methods 3.1 SWT and DWT This technique uses the feature of both the transforms. But the difference is SWT does not provide down sampled image as in DWT [21]. LR image is broken down into four subband images and bi-cubic interpolation factor 2 is applied to modify the down sampled images of DWT. But this is not necessary for SWT and estimated images are measured by integrating the high-frequency components of both the transforms. In this, α/2 factor is applied on LR and estimated images and all the images combined using Inverse (DWT) for high-resolution image [1–10] (Fig. 1). 2D-Discrete Wavelet Transform is scaled and translated by the following components [20]. LL = φ(x, y) = φ(x)φ(y)

(1)

LH =  H (x, y) = (x)φ(y)

(2)

HL =  V (x, y) = φ(x) (y)

(3)

HH =  D (x, y) = (x) (y)

(4)

16

M. Rathod et al.

Fig. 1. Integrated SWT and DWT technique

H, V and D are the super scripts for direction decomposition of wavelet. 2D wavelet is applied for image manipulation.  is wavelet function and φ is the scaling function derived from . Power of 2 is used for scaling and shifting of the mother wavelet in DWT. where k is the shift, j is the scale parameters and (t) is mother wavelet.   t − k2j 1 (5) j,k (t) = √  2j 2j

3.2 DWT and IWT This technique integrates two wavelet transforms, i.e. DWT and IWT. It is similar to above-described technique, but the only difference is interpolation factor 2 is applied to the output of both the transforms because the images are down sampled after decomposition of LR image. Integer coefficients from IWT and floating point coefficients from DWT are merged for better resolution. Estimated images are generated by combining high-frequency sub-band images. All the images are merged by Inverse DWT to enlarge the small into bigger image [7–21] (Fig. 2).

Novel Approach for Resolution Enhancement

17

Fig. 2. Integrated DWT and IWT technique

3.3 Proposed Technique Proposed technique is an integration of the characteristics of three wavelet transforms and decomposed LR image by using all the transforms. Thus, it provides four sub-band images. The high-frequency components are selected from the sub-band images. The down sampled images are resized by applying interpolation factor 2. Similarly, estimated images are determined by merging the high-frequency components of all transforms and getting more detailed image. Inverse lifting wavelet transform is applied to combine all the images for enhancing LR image. It provides good results, but computational time is more as compared to other techniques [15–21] (Fig. 3).

4 Performance Techniques Some of the various techniques for performance, i.e. MSE, RMSE, MAE, PSNR and TIME are studied and used to verify the implemented techniques [21]. MSE: It is one of the parameters for evaluating the performance of implemented technique. It is measuring the MSE between the input image (I) and the original image (O).

18

M. Rathod et al.

Fig. 3. Proposed technique

The size of the images is MXN [21].  MSE =

i,j (I in (i, j) − I org (i, j))

MXN

(6)

RMSE: It is another parameter to test quality of the mentioned technique. It measures the pixels in reference image and enhanced high-resolution image (H r ) [21].  1 M N (7) RMSE = (Hr (i, j) − H (i, j)) i=1 j=1 MN MAE: This parameter determines the difference between reference image and enhanced high-resolution image (H r ) [21]. MAE =

M N 1  |Hr (i, j) − H (i, j)| MN i=1 j=1

(8)

Novel Approach for Resolution Enhancement

19

PSNR: It does to determine the ratio between original image and reconstructed image. Input image fluctuation is R [21].  2  R PSNR = 10 log10 (9) MSE TIME (second): It determines the computational time for the mentioned techniques.

5 Result and Discussion Some of the resolution enhancement techniques are implemented, compared with evaluation parameters are calculated. In this, work is carried on LANDSAT5 satellite image of different size for testing the techniques, i.e. LR image with size 128 × 128, 256 × 256 and output of first iteration are taken as LR input image with size 128 × 128 and all these images are converted into 512 × 512 sizes. It is concluded that iterated image provides very good resolution as compared to other sizes of image. The tables and graphs show the supremacy of the implemented techniques. 5.1 Resolution Enhancement from 128 × 128 to 512 × 512 In this scenario 128 × 128 size LR image and enhanced 512 × 512 with α = 4. But the result obtained is so decreased as compared to other image sizes. Table 1, Figs. 4 and 5 declare the result of this section. Table 1. The result obtained for LANDSAT5 satellite image and enhanced from 128 × 128 to 512 × 512 with α = 4 Techniques/parameters RMSE PSNR MAE TIME (s) SWT and DWT

28.80

43.05 19.40 12.09

DWT and IWT

27.06

43.40 18.70 12.30

Proposed technique

24.60

44.41 16.20 14.34

5.2 Resolution Enhancement from 256 × 256 to 512 × 512 In this case, we have considered 256 × 256 LR image and enlarged with size 512 × 512. It is concluded that the result obtained from this LR size is improved as compared to above size of image. Table 2, Figs. 6 and 7 show the status of the result. 5.3 Iterative Resolution Enhancement from 128 × 128 to 512 × 512 In this section re-processed image i.e. it means first processed image is taken as input image. The result has significantly improved as compared to other sizes of images. Table 3, Figs. 8 and 9 show the betterment [21].

20

M. Rathod et al.

Fig. 4. a LR input image. b SWT&DWT enhanced image. c DWT&IWT enhanced image. d Proposed method

Fig. 5. Graph of LANDSAT5 image with size 128 × 128 to 512 × 512 Table 2. The result obtained for LANDSAT5 satellite image and enhanced from 256 × 256 to 512 × 512 with α = 4 Techniques/parameters RMSE PSNR MAE TIME(s) SWT and DWT

23.40

44.85 14.90 10.76

DWT and IWT

22.60

45.15 14.56 10.50

Proposed technique

20.81

45.90 13.11 12.50

6 Conclusion Various transform domain techniques are implemented and compared. The results are evaluated with LANDSAT5 satellite image with different sizes of LR image. Evaluation parameters such as RMSE, PSNR, MAE and TIME are measured to determine the performance of the enhancement techniques. In this work, 128 × 128, 256 × 256 and

Novel Approach for Resolution Enhancement

21

Fig. 6. a LR input image. b SWT&DWT enhanced image. c DWT&IWT enhanced image. d Proposed method

Fig. 7. Graph of LANDSAT5 image with size 256 × 256 to 512 × 512 Table 3. The result obtained for LANDSAT5 satellite image and iteratively enhanced from 128 × 128 to 512 × 512 with α = 4 Techniques/Parameters RMSE PSNR MAE TIME(s) SWT and DWT

22.07

DWT and IWT

18.50

Proposed technique

5.70

45.36 14.66 10.90 46.90 12.77 11.13 57.16

3.90 12.33

iterative resolution enhancement LR images are considered to bring the high-resolution image. It is concluded that the result obtained from iterative LR image is better than other LR images. The result obtained from the proposed technique is better than others. Present work is on spatial resolution, but future work we can try on temporal resolution, i.e. often images are taken for the same area and verify the resolution for two different

22

M. Rathod et al.

Fig. 8. a LR input image. b SWT&DWT enhanced image. c DWT&IWT enhanced image. d Proposed method

Fig. 9. Graph of LANDSAT5 image for iterative resolution with size 128 × 128 to 512 × 512

images for same location and minimize execution time for proposed method and work for better resolution.

References 1. Demirel H, Anbarjafari G (2011) Discrete wavelet transform-based satellite image resolution enhancement. IEEE Trans. Geosci. Remote Sens. 49:1997–2004 2. Hasan D, Gholamreza Anbarjafari F (2011) Image resolution enhancement by using discrete and stationary wavelet decomposition. IEEE Trans Image Process 20(5):1458–1460 3. Xiaobing B (2011) A wavelet-based Image resolution enhancement techniques. In: IEEE international conference on electronics and optoelectronics (ICEOE), IEEE-978-1-61284276-9, pp. 62–64 4. Ahir R, Patil VS (2013) Overview of satellite image resolution enhancement techniques. In: IEEE conference IEEE-978-1-4673-5999-3/13, pp 62–64 5. Karunakar P, Praveen V, Ravi Kumar O (2013) Discrete wavelet transform based satellite image resolution enhancement. Adv Electric Electr Eng 3(4):405–412. ISSN: 2231-1297

Novel Approach for Resolution Enhancement

23

6. Venkataramana S, Narayana Reddy S (2014) A novel method to improve resolution of satellite image using DWT and interpolation. Int J Adv Res Electric Electr Instrum Eng 3(1):6530– 6536. ISSN: 2320-3765 7. Sagar E, Kumar M, Ramakrishnaiah T (2015) Satellite image resolution enhancement technique using DWT and IWT. Int J Comput Appl Technol Res 4(1):70–76. ISSN: 2319-8656 8. Shamna KS (2014) Satellite image resolution and brightness enhancement using discrete & stationary wavelet and singular value decomposition. In: International conference on power signals controls and computation (EPSCICON), IEEE: 978-1-4799-3612-0, pp 01–04 9. Jadhav BD, Patil PM (2015) An effective method for satellite image enhancement. In: IEEE international conference on computing communication and automation (ICCCA), pp 1171– 1175. ISBN: 978-1-4799-8890-7 10. Anuradha, Aashish, Hiradhar (2015) Image resolution enhancement: a review. Int J Res Advent Technol 3(5):204–207. ISSN: 2321-9637 11. Shashi Vardhan N, Akanksha G, Hussein Y, Singh D (2015) A comparison of wavelet based techniques for resolution enhancement of moderate resolution satellite images. In: National Conference on Recent Advances in Electronics & Computer Engineering (RAECE), IIT Roorkee, India, pp. 263–266 (2015) 12. Sharma A, Khunteta A (2016) Satellite image contrast and resolution enhancement using discrete wavelet transform and singular value decomposition. In: IEEE international conference on emerging trends in electrical electronics and sustainable energy system (ICETEESES), 978-1-5090-2118-5, pp 01–05 13. Mazhar A, Hassan F, Anjum MR, Saher M, Muhammad AS (2016) High resolution image processing for remote sensing application. In: IEEE International conference on innovative computing technology (INTECH), 978-1-5090-2000-3, pp 302–305 14. Sontakke MD, Kulkarni MS (2016) Multistage combined image enhancement technique. In: IEEE international conference on recent trends in electronics information communication technology, 978-1-5090-2118-5, pp 212–215 15. Rathod M, Jayashree K (2017) Satellite image resolution enhancement using SWT and DWT with SWT. In: IEEE international conference on nascent technologies in engineering (ICNTE), ISBN: 978-1-5090-2794-1, pp 01–05 16. Rathod M, Jayashree K (2017) Comparative study of transform domain methods for image resolution enhancement of satellite image. In: IEEE international conference on intelligent systems and control (ISCO), pp 287–291. ISBN:978-1-5090-2717 17. Rathod M, Jayashree K (2017) Resolution enhancement of satellite image using DCT and DWT. Asian J Converg Technol (AJCT) 3(3):67–71. ISSN: 2350-1146 18. Rathod M, Jayashree K (2018) Performance evaluation of transform domain methods for satellite image resolution enhancement. In: Springer international conference on wireless communication (ICWiCom), eBook: lecture notes on data engineering and communications technologies, vol 19, pp 227–236. ISBN: 978-981-10-8338-9, Series ISSN: 2367-4512 19. Rathod M, Jayashree K (2018) Comparative study of wavelet methods for resolution enhancement of satellite images. Int J Adv Comput Sci Technol (IJACST) 11(1):28–33. ISSN: 0973-6107 20. Ravichandran D, Nimmatoori R, Ahamad MG (2017) Mathematical representation of 1D, 2D and 3D wavelet transform for image coding. Int J Adv Comput Theor Eng (IJACTE) 5(3):20–27. ISSN:2319-2526 21. Rathod M, Jayashree K, Dilendra H (2018) Thesis on satellite image resolution enhancement using frequency domain, submitted to Pacific University, pp 41–44

Vehicle Number Plate Detection: An Edge Image Based Approach Kalyan Kumar Jena1,3,4 , Soumya Ranjan Nayak2(B) , Sasmita Mishra3 , and Sarojananda Mishra3 1 Utkal University, Bhubaneswar, India

[email protected]

2 Amity School of Engineering and Technology, Amity University Noida, Noida, Uttar Pradesh,

India [email protected] 3 Department of Computer Science Engineering and Applications, Indira Gandhi Institute of Technology, Sarang, India [email protected], [email protected] 4 Department of Computer Science and Engineering, Parala Maharaja Engineering College, Berhampur, India

Abstract. Transportation system (vehicle communication) plays a major role in today’s scenario. Detection of vehicle number plate exactly in blurry conditions was the most challenging issue found in the last three decades. Although many intensive studies were undertaken, none addressed this problem exhaustively. Various methods are introduced by several researchers for detecting the vehicle number from the vehicle number plate images. The purpose of this study was to investigate this current issue by implementing an edge-based approach on the basis of quantitative combination of Canny, Morphological and Sobel methods for the accurate detection of vehicle number in blurry conditions. The experimental results demonstrated that the proposed scheme outperforms its counterparts in terms of Sobel, Prewitt, Roberts, Laplacian of Gaussian (LoG), Morphological and Canny methods in all aspects with higher peak signal to noise ratio (PSNR) and signal to noise ratio (SNR) values. Hence, the proposed hybrid scheme is better and robust and results in accurate estimation of vehicle number from the blurry vehicle number plate (BVNP) images for the given datasets. Keywords: Transportation system · BVNP image · Canny · Morphological · Sobel · Prewitt · Roberts · LoG · PSNR · SNR

1 Introduction Extracting features from blurry images is a very difficult as well as challenging work in the field of digital image processing. Different methods are provided by several researchers to extract features from blurry images. However, no method works in a better way for extracting features from blurry images. The performance of each method © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_3

Vehicle Number Plate Detection: An Edge Image Based Approach

25

varies from image to image in case of extracting features from blurry images under several conditions. Transportation system takes an important role in the day to day life. It is very difficult to survive without this system. This system possesses both merits and demerits. From a merit point of view, one can reach the destination place at almost right time under normal situations. However, from a demerit point of view, it may take the life of human beings within a fraction of second. The life risk factor of human beings increases gradually in the current scenario of the transportation system due to the carelessness driving of drivers. So, it is very much essential to track each and every vehicle in order to take action against the drivers those are responsible for the occurrence of any kind of unpleasant situation such as road accident, damage of vehicle due to higher speed, etc. so, vehicle number plate tracking system is very much essential in today’s ear. In this paper, the processing of several BVNP images is focused in order to identify the vehicle number of the victim vehicle. Several methods [1–15] are provided by different researchers to detect the vehicle number from the BVNP images. However, no method is able to detect the vehicle number from all the BVNP images in a better way. Several existing standard edge detection operators [16–27] such as Sobel, Prewitt, Roberts, LoG, Canny, and Morphological operators or image processing approaches [28–40] can be used to detect the vehicle number from BVNP images. This paper focuses on an edge-based approach to detect the vehicle number from BVNP images. As transportation system plays a vital role in day to day life, so it is very much essential to track the number plate of each and every vehicle. Tracking the number plate of vehicles will help in analyzing any kind of unpleasant circumstances that arises due to some specific vehicles and further actions can be taken accordingly. However, it is not so easy to track the number from BVNP images for further actions. So, we are motivated to detect the number plate of several vehicles under blurry condition. The main contributions in this work are stated as follows: • A quantitative combination of Canny, Morphological and Sobel edge detectors is urged to accurately track the vehicle number from BVNP images. • The proposed method is accomplished using MATLAB R2015b and evaluated using PSNR and SNR values. • The output results show that the proposed method acts better in identifying the vehicle number from BVNP images along with higher PSNR and SNR values as compared to Sobel, Prewitt, Roberts, Laplacian of Gaussian (LOG), Morphological and Canny methods. The organization of rest of the paper is stated as follows: Sects. 2, 3, 4, and 5 describe the related works, proposed methodology, results and discussion, and conclusion of the work, respectively.

2 Related Works Several methods [1–15] are provided by different researchers to detect the vehicle number from vehicle number plate images. In this section, some of the methods are discussed. Jagadamba et al. [1] focus on the management of campus vehicle with the help of an

26

K. K. Jena et al.

image processing mechanism. This method deals with the reorganization of each character in the license plate (registered) and the main objective of this method is to gain high accuracy by considering the optimization of several parameters whose rate of recognition is higher than the existing techniques. Tejas et al. [2] focus on an Internet of Things (IoT) based method to recognize the license plate of vehicle. This method deals with the processing of image, identifying the region of license plate region, applying segmentation mechanism and recognizing the character. Shemarry et al. [4] urged an efficient texture descriptor, local binary pattern (multi-level extended) to detect the license plates under several image situations such as distorted, foggy, dusty as well as low or high contract situations. Angara et al. [6] urged an automatic license plate recognition system for capturing and recognizing the license plate of a vehicle. This method deals with the creation of a dataset for the simulation of license plate image (captured), then application of multiple binarization mechanisms for the segmentation of characters from the state, plate as well as from each other and at least this dataset is used to train a convolutional neural network. Babbar et al. [9] urged a system based on machine learning and image processing mechanism to meliorate the identification of number plate of vehicle in overexposure as well as low light conditions. The captured vehicle image is preprocessed by the help of grayscale, binarization mechanisms, and then the image is used for the plate localization in order to extract the number plate by the help of connected component analysis and ratio analysis. Several filters are used to de-noise the number plate. From the above discussion, it is depicted that, several researchers have presented several methods or approaches for vehicle number plate detection from vehicle number plate images. However, no method performs better in all situations.

3 Proposed Methodology The proposed method focuses on the quantitative combination of Canny [16–20] Morphological [21] and Sobel [22, 23, 26, 27] edge detection methods to detect the vehicle number from BVNP images as mentioned in Fig. 1. The Canny edge detector is considered an optimal method and it deals with multi-stage processing mechanisms. The morphological operators such as erosion, dilation, opening and closing can be used in edge detection process. The Sobel edge detector is a gradient-based edge detection method. Figure 1 describes that initially the original BVNP image (IO ) is read by using MATLAB R2015b. Afterward, IO is processed with the help of Canny method to generate the image IC . Subsequently, IC is processed using Morphological method to generate image ICM. Then, ICM is processed by Sobel method to generate the resultant image ICMS . ICMS is compared with the output of Sobel, Prewitt, Roberts, Laplacian of Gaussian (LOG), Morphological and Canny methods along with PSNR and SNR values. 3.1 Steps for Proposed Methodology The proposed method deals with the following steps: Step I: Input original BVNP image (IO ) Step II: Apply Canny method on IO to generate the image IC

Vehicle Number Plate Detection: An Edge Image Based Approach

27

Input Original BVNP Image (IO)

Application of Canny Edge Detection Approach on the Image IO to generate the image IC

Application of Morphological Operation on the Image IC to generate the image ICM

Application of Sobel Edge Detection Approach on the image ICM to generate the resultant image ICMS and compare with the following methods along with PSNR and SNR values Comparison of ICMS with the output of Sobel Method

Comparison of ICMS with the output of Prewitt Method

Comparison of ICMS with the output of Canny Method

Comparison of ICMS with the output of Roberts Method

Comparison of ICMS with the output of LoG Method

Comparison of ICMS with the output of Morphological Operation

Fig. 1. Descriptive representation of proposed methodology

Step III: Apply Morphological operation on IC to generate the image ICM Step IV: Apply Sobel method on ICM to generate the image ICMS (resultant image of urged method) Step V: The image ICMS is generated using MATLAB R2015b and compared with the output of Sobel, Prewitt, Roberts, LoG, Morphological and Canny methods along with PSNR and SNR values.

4 Results and Discussion In this paper, several BVNP images with different size as mentioned in Figs. 2a, 3a, 4a, 5a and 6a are taken from Google image source and processed using MATLAB R2015b. The urged method is focused on the comparison of six standard existing methods in detecting the vehicle number from BVNP images. The upshots of Sobel, Prewitt, Roberts, LoG, Morphological, Canny and the urged method are mentioned in Figs. 2, 3, 4, 5 and 6 and PSNR as well as SNR values of each method is mentioned in Table 1. The quality of output image enhances if the PSNR or SNR value increases. From Figs. 2 and 3, it is realized that the urged method detects the vehicle number in a better way than other methods. However, other methods except Canny are not able to identify the vehicle number and the performance of Canny method is not so good as

28

K. K. Jena et al.

Fig. 2. Processing of BVNP-1 image. a Original Number Plate (1280 × 720). b Upshot using Sobel method. c Upshot using Prewitt method. d Upshot using Roberts method. e Upshot using LoG method. f Upshot using Morphological method. g Upshot using Canny method. h Upshot using proposed method

compared to urged method. From Fig. 4, it is observed that the urged method acts better as compared to other methods. However, LoG, Morphological and Canny methods try to detect the vehicle number but their performance is not good enough as compared to urged method. Similarly, From Figs. 5 and 6, it is observed that the urged method also acts better as compared to other methods and Canny method also tries to detect the vehicle number and its performance is not good enough as compared to the urged method. So, from the analysis of Figs. 2, 3, 4, 5, 6 and 7 and Table 1, it is concluded that the urged method detects the vehicle number from BVNP images efficiently with higher PSNR and SNR values than Sobel, Prewitt, Roberts, LoG, Morphological and Canny methods.

5 Conclusion This paper focuses on an edge-based approach on the basis of quantitative combination of Canny, Morphological and Sobel methods for detecting the vehicle number from the BVNP images. Several BVNP images are processed using the proposed approach to outperform better with all counterparts. The proposed scheme is implemented in a simulated platform of MATLAB R2015b and finally equaled with Sobel, Prewitt, Roberts, LoG, Morphological and Canny methods. From the analysis, we have observed that the proposed scheme provides better estimation in all aspect of experimental analysis in a better way with higher PSNR and SNR values than the existing methods such as Sobel, Prewitt, Roberts, LoG, Morphological and Canny methods. Hence, the proposed

Vehicle Number Plate Detection: An Edge Image Based Approach

29

Fig. 3. Processing of BVNP-2 image. a Original Number Plate (307 × 164). b Upshot using Sobel method. c Upshot using Prewitt method. d Upshot using Roberts method. e Upshot using LoG method. f Upshot using Morphological method. g Upshot using Canny method. h Upshot using proposed method

method is better and robust and results in accurate estimation from the given datasets. The presented work can be extended to detect the vehicle number from vehicle number plate images under distorted, foggy, dusty, low and high contract conditions by using an improved method and can be compared with other existing methods.

30

K. K. Jena et al.

Fig. 4. Processing of BVNP-3 image. a Original Number Plate (500 × 500). b Upshot using Sobel method. c Upshot using Prewitt method. d Upshot using Roberts method. e Upshot using LoG method. f Upshot using Morphological method. g Upshot using Canny method. h Upshot using proposed method

Fig. 5. Processing of BVNP-4 image. a Original Number Plate (615 × 447). b Upshot using Sobel method. c Upshot using Prewitt method. d Upshot using Roberts method. e Upshot using LoG method. f Upshot using Morphological method. g Upshot using Canny method. h Upshot using proposed method

Vehicle Number Plate Detection: An Edge Image Based Approach

31

Fig. 6. Processing of BVNP-5 image. a Original Number Plate (660 × 452). b Upshot using Sobel method. c Upshot using Prewitt method. d Upshot using Roberts method. e Upshot using LoG method. f Upshot using Morphological method. g Upshot using Canny method. h Upshot using proposed method

Table 1. Performance evaluation of proposed and other methods as per PSNR (dB) and SNR (dB) values Method

BVNP-1

BVNP-2

BVNP-3

BVNP-4

BVNP-5

PSNR SNR

PSNR SNR

PSNR SNR

PSNR SNR

PSNR SNR

Sobel

19.32

14.62 21.38

19.23 21.06

18.66 25.64

25.12 23.21

22.04

Prewitt

19.34

14.63 21.38

19.23 21.08

18.68 25.63

25.12 23.19

22.02

Roberts

19.32

14.63 21.36

19.19 21.05

18.64 25.65

25.14 23.17

21.99

LoG

19.38

14.78 21.47

19.38 21.15

18.83 25.78

25.29 23.38

22.28

Canny

19.31

14.56 21.52

19.45 21.19

18.92 25.75

25.27 23.65

22.63

Morphological 19.46

15.03 21.64

19.68 21.28

19.08 25.94

25.47 23.58

22.55

Proposed

15.29 21.92

20.14 21.57

19.54 26.11

25.68 24.11

23.23

19.55

K. K. Jena et al.

PSNR(dB) and SNR(dB)

32 30 25

BVNP-1-PSNR

20

BVNP-1-SNR

15

BVNP-2-PSNR

10 5

BVNP-2-SNR

0

BVNP-3-PSNR BVNP-3-SNR BVNP-4-PSNR Method

BVNP-4-SNR

Fig. 7. Correlation representation of PSNR (dB) and SNR (dB) values of proposed and other methods

References 1. Jagadamba G, Shrinivasacharya P, Chayashree G (2019) Campus vehicle monitoring through image processing. In: Emerging research in electronics, computer science and technology. Springer, Singapore, pp 305–315 2. Tejas K, Reddy KA, Reddy DP, Bharath KP, Karthik R, Kumar MR (2019) Efficient license plate recognition system with smarter interpretation through IoT. In: Soft computing for problem solving, Springer, Singapore, pp 207–220 3. Yonetsu S, Yutaro I, Yen WC (2019) Two-stage YOLOv2 for accurate license-plate detection in complex scenes. In: 2019 International Conference on Consumer Electronics, IEEE, pp 1–4 4. Shemarry A, Meeras S, Yan L, Shahab A (2019) An efficient texture descriptor for the detection of license plates from vehicle images in difficult conditions. In: IEEE transactions on intelligent transportation systems, pp 1–12 5. Xiang H, Yong Z, Yule Y, Guiying Z, Xuefeng H (2019) Lightweight fully convolutional network for license plate detection. Optik 178:1185–1194 6. Angara S, Melvin R (2019) License plate character recognition using binarization and convolutional neural networks. In: Science and Information Conference. Springer, Cham pp 272–283 7. Yousefi E, Amir H, Nazem D, Jafar JA, Mohsen KK (2019) Real-time scale-invariant license plate detection using cascade classifiers. In: Multimedia information processing and retrieval conference. IEEE, pp 399–402 8. Kessentini Y, Mohamed DB, Sourour A, Achraf C (2019) A two-stage deep neural network for multi-norm license plate detection and recognition. In: Expert systems with applications, pp 159–170 9. Babbar S, Saommya K, Navroz D, Kartik S, Sanjeev P (2018) A new approach for vehicle number plate detection. In: Eleventh international conference on contemporary computing, IEEE, pp 1–6 10. Imaduddin H, Muhamad KA, Muhammad IP, Indra AS, Anhar R (2018) Indonesian vehicle license plate number detection using deep convolutional neural network. In: International electronics symposium on knowledge creation and intelligent computing, IEEE, pp 158–163

Vehicle Number Plate Detection: An Edge Image Based Approach

33

11. Lee Y, Jae WY, Yoojin H, Juhyun L, Moongu J (2018) Accurate license plate recognition and super-resolution using a generative adversarial networks on traffic surveillance video. In: International conference on consumer electronics-Asia, IEEE, pp 1–4 12. Saini MK, Saini S (2017) Multiwavelet transform based license plate detection. J Vis Commun Image Represent 44:128–138 13. Prakash AR, Jagannath M, Adalarasu K, Babu GM (2017) Vehicle license plate detection and recognition using non-blind image de-blurring algorithm. In: International conference on Nextgen electronic technologies: silicon to software. IEEE, pp 46–49 14. Mayan JA, Kumar AD, Kumar M, Livingston A, Reddy SP (2016) Number plate recognition using template comparison for various fonts in MATLAB. In: International conference on computational intelligence and computing research, IEEE, pp 1–6 15. Azam S, Islam MM (2016) Automatic license plate detection in hazardous condition. J Vis Commun Image Represent 36:172–186 16. Wang M, Jesse SJ, Yifei J, Xianfeng H, Lei G, Liping X (2016) The improved canny edge detection algorithm based on an anisotropic and genetic algorithm. In: Chinese conference on image and graphics technologies, Springer, Singapore, pp 115–124 17. Xin G, Chen K, Hu X (2012) An improved Canny edge detection algorithm for color image. In: 10th international conference on industrial informatics, IEEE, pp 113–117 18. Lahani J, Sulaiman HA, Muniandy RK, Bade A (2018) An enhanced edge detection method based on integration of entropy—Canny technique. Adv Sci Lett 24:1575–1578 19. Othman Z, Azizi A (2017) An adaptive threshold based on multiple resolution levels for canny edge detection. In: International conference of reliable information and communication technology. Springer, Cham, pp 316–323 20. Shanmugavadivu P, Kumar A (2014) Modified eight-directional canny for robust edge detection. In: International conference on contemporary computing and informatics, IEEE, pp 751–756 21. Yu-Qian Z, Gui WH, Chen ZC, Tang JT, Li LY (2006) Medical images edge detection based on mathematical morphology. In: Engineering in medicine and biology 27th annual conference, IEEE, pp 6492–6495 22. Gupta S, Mazumdar SG (2013) Sobel edge detection algorithm. Int J Comput Sci Manage Res 2:1578–1583 23. Chaple G, Daruwala RD (2014) Design of Sobel operator based image edge detection algorithm on FPGA. In : International conference on communication and signal processing. IEEE, pp 788–792 24. Alshorman MA, Junoh AK, Wan, Zuki A, Wan MM, Hafiz Z, Afifi M (2018) Leukaemia’s cells pattern tracking via multi-phases edge detection techniques. J Telecommun Electr Comput Eng 10:33–37 25. Wan J, Xiaofu H, Pengfei S (2007) An Iris image quality assessment method based on Laplacian of gaussian operation. In: MVA, pp 248–251 26. Patel J, Patwardhan J, Sankhe K, Kumbhare R (2011) Fuzzy inference based edge detection system using Sobel and Laplacian of Gaussian operators. In: Proceedings of the international conference and workshop on emerging trends in technology. ACM, pp 694–697 27. Gao W, Xiaoguang Z, Lei Y, Huizhong L (2010) An improved Sobel edge detection. In : 3rd international conference on computer science and information technology, vol 5. IEEE, pp 67–71 28. Romani L, Milvia R, Daniela S (2019) Edge detection methods based on RBF interpolation. J Comput Appl Math 349:532–547 29. Nayak SR, Mishra J, Palai G (2019) Analysing roughness of surface through fractal dimension: a review. Image Vis Comput 89:21–34 30. Jena KK, Mishra S, Mishra SN, Bhoi SK, Nayak SR (2019) MRI brain tumor image analysis using fuzzy rule based approach. J Res Lepid 50:98–112

34

K. K. Jena et al.

31. Nayak SR, Mishra J (2018) A modified triangle box-counting with precision in error fit. J Inf Optim Sci 39:113–128 32. Eser SERT, Dery AVCI (2019) A new edge detection approach via neutrosophy based on maximum norm entropy. Expert Syst Appl 115:499–511 33. Jena KK, Mishra S, Mishra SN, Bhoi SK (2019) 2L-ESB: a two level security scheme for edge based image steganography. Int J Emerg Technol 10:29–38 34. Nayak SR, Khandual A, Mishra J (2018) Ground truth study on fractal dimension of color images of similar texture. J Text Inst 109:1159–1167 35. Quan S, Deliang X, Boli X, Gangyao K (2019) Edge detection for PolSAR images integrating scattering characteristics and optimal contrast. IEEE Geosci Remote Sens Lett, pp 1–5 36. Nayak SR, Mishra J, Jena PM (2018) Fractal analysis of image sets using differential box counting techniques. Int J Inf Technol 10:39–47 37. Lv C, Kaifu W, Guoqing G, Yun P (2019) Accurate full-edge detection and depth measurement of internal defects using digital speckle pattern interferometry. NDT E Int 102:1–8 38. Nayak SR, Mishra J, Palai G (2018) A modified approach to estimate fractal dimension of gray scale images. Optik 161:136–145 39. Qin Y, Yanli W, Qifu W, Suping Y (2019) Method for pests detecting in stored grain based on spectral residual saliency edge detection. Grain Oil Sci Technol 69:1282–1290 40. Nayak SR, Mishra J, Padhy R (2016) An improved algorithm to estimate the fractal dimension of gray scale images. In: International conference on signal processing, communication, power and embedded system, IEEE, pp 1109–1114

A High-Precision Pixel Mapping Method for Image-Sensitive Areas Based on SVR Huang Jing1 , Amit Yadav1 , Asif Khan3,4(B) , and Dakshina Yadav2 1 Department of Computer Science and Technology, Chengdu Neusoft University, Chengdu

611844, China 2 College of Engineering, IT and Environment, Charles Darwin University, Darwin, Australia 3 Crescent Institute of Science and Technology, Vandalur, Chennai 600048, India

[email protected] 4 UESTC, Chengdu, China

Abstract. It is necessary to monitor the grain size characteristics of particles at production site to control the production equipment, for the assurance of product quality. In this respect, prior research finds that it is critical to evaluate the accuracy of tiny particles since the current practice indicates that the existing methods illustrate multiple shortcomings including large measure error, low accuracy and poor repeatability. Therefore, to improve the accuracy of particles, monitoring this study proposed a calibration method based on SVR algorithm to predict the accurate pixel size of the particles. Results revealed that high-precision pixel mapping of the sensitive area transforms the pixel mapping of the particle image closer to the actual size and improves the measurement precision of the whole system. Keywords: Image processing · SVR · Forecast · High-precision pixel mapping

1 Introduction Recently, the rapid development of science and technology and precision of tiny objects has become instrumental in production of fine products [1]. In this regard, image recognition processing technique has gained much academic interest since existing techniques demonstrate many shortcomings including both lack of good repeatability and detection features [2–4]. Thus, it is clearly evident that existing image recognition method of tiny particles is unable to meet the high-precision requirements [5–7]. Currently, the image processing algorithms based on SVR technology in the study of camera calibration, rectifying distorted images and the transformation of real objects to 2D image, provide robust results [8]. The pixel mapping proposed in this paper used the pixel size of the image to calculate the size of the object. However, the traditional mapping method calculates the pixel size of a calibrated object by taking the average of whole image pixels; nonetheless, it provides biased results. Since prior research used the average of one part of the image to predict the object size of other area, thus produces errors. In this study, particles of 0.23–0.32 mm are taken and regression support vector machine (SVR) method is introduced to predict the pixel size of the object [9–11]. © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_4

36

H. Jing et al.

2 Regression Support Vector Machines Regression support vector machine (SVR) is a method developed on the basis of classification problems [12, 13]. However, the objective of SVR problem is to find the appropriate real value function f (x) = w · φ(x) + b to fit the training point, so as to minimize  R[f ] = c(x, y, f )dP(x, y) (1) where c is the loss function [14]. The common SVR methods are ε-SVR and v-SVR [15]. ε-SVR uses insensitive coefficients ε to measure the error between observed and predicted values. When the error between the observed value and the predicted value of a certain point does not exceed ε, the fitting error of the function to these sample points is considered to be zero. As shown in Fig. 1, when the sample point falls between the two dotted lines ±5, it is assumed that there is no loss at that point, and the area formed by these two dotted lines is called the ε band [16, 17].

Fig. 1. ε band

The optimization problem can be expressed as: ⎧ l ⎪ ⎪ min 21 (w · w) + C 1  (ζi + ζi ∗) ⎪ ⎪ l ⎪ ⎨ w,ζi ,ζi ∗,b i=1 (w · φ(xi ) + b) − yi ≤ ε + ζi ⎪ ⎪ ⎪ s.t. yi − (w · φ(xi ) + b) ≤ ε + ζi ∗ ⎪ ⎪ ⎩ ζi , ζi ∗ ≥ 0

(2)

ε-SVR needs to set ε value in advance, but it is difficult to determine the excellent ε value in practical application. Therefore, v-SVR is proposed that v-SVR can

A High-Precision Pixel Mapping Method

automatically calculate ε value [18], and its expression is: ⎧ l ⎪ 1 ⎪ 1  min ⎪ (ζi + ζi ∗)) 2 (w · w) + C(νε + l ⎪ w,ζ ,ζ ∗,b,ε ⎪ ⎨ i i i=1 (w · φ(xi ) + b) − yi ≤ ε + ζi ⎪ ⎪ ⎪ s.t. yi − (w · φ(xi ) + b) ≤ ε + ζi ∗ ⎪ ⎪ ⎩ ζi , ζi ∗ ≥ 0 , ε ≥ 0

37

(3)

3 High-Precision Pixel Mapping for Sensitive Areas 3.1 The Implementation Steps of the Algorithm In this paper, a regression support vector machine (SVR) is proposed to predict the pixel size of the particle area, including the following steps: Step 1: The calibration plate is placed within the field of view, as shown in Fig. 2, where the serial number 1 is the field of view of the camera, the serial number 2 is the information of the visual calibration plate, and the serial number 3 is the particle information.

1 2 3

1 Fig. 2. Schematic of imaging

Step 2: To calculate the pixel size of the feature points in the calibration block area, such as angular points, by using the known calibration board information. Calculation example: Suppose each checkerboard in the calibration board size is 0.5 mm × 0.5 mm, if a certain corner of the calibration plate in the image area are the coordinates of the point (x 1 , y1 ), direction of u and v direction on adjacent corner point coordinates respectively (x 2 , y2 ) and (x 3 , y3 ), then the angular point (x 1 , y1 ) for the pixel size is:  0.05 0.05 1 + (4) S1 = 2 [x2 − x1 ] [y3 − y1 ] In the above equation, [x2 − x1 ] represents the integer, and the physical meaning of S1 is: the average pixel size of the point in the directions of u and v.

38

H. Jing et al.

Step 3: Take the grayscale value, pixel coordinates and pixel sizes of angular points in the visible area of the calibration block as the training sample set: {(xi , yi )|xi ∈ Rn , yi ∈ Rn , i = 1, . . . , n}. The grayscale value and coordinate are independent variables, and the pixel size is dependent variable. Then, it is normalized. The normalized mapping method is as follows: f: x →y =

(ymax − ymin ) × (x − xmin ) + ymin (xmax − xmin )

(5)

 ymin , ymax represents the scope of the mapping, for example, [0, 1] represents the normalization of the original data between 0 and 1, and xmax and xmin represent the maximum and minimum value of the original data. Step 4: To construct the following regression function according to SVR principle [19]: f (x) = ω · φ(x) + b

(6)

In the above equation : Rn → , including  represents high-dimensional feature space, ω and b are the weight coefficient vector and the bias vector, respectively, and they can be obtained by solving the following optimization problem [20]: 1 ||w||2 + C (ξi + ξi∗ ) 2 l

minP = ω,b

(7)

i=1

The constraint condition of the above formula is: yi − (ω · φ(x) + b) ≤ ε + ξi (ω · φ(x) + b) − yi ≤ ε + ξi∗ ξi , ξi∗ ≥ 0, i = 1 · · · l

(8)

where C is the equilibrium coefficient. The yi is the dependent variable value of training data. The ξi and ξi∗ are the penalty function, and its value formula is

0 |f (xi ) − yi | < ε (9) ξi∗ = |f (xi − yi )| − ε |f (xi ) − yi | ≥ ε To construct the Lagrangian function, ω is obtained by solving the saddle point: ω=

l

(αi − αi∗ )φ(xi )

(10)

i=1

The α i is the Lagrangian multiplier, and according to the above formula, the following support vector machine model can be obtained [21]: f (x) =

l i=1

(αi − αi∗ )K(xi , x) + b

(11)

A High-Precision Pixel Mapping Method

39

In this formula, the kernel function K(xi , x) selected in this paper is the radial basis kernel function, and its expression is as follows:  |x − xi |2 (12) K(xi , x) = exp − σ2 Step 5: After more than three steps to complete to build up the forecast model on the sensitive area pixel mapping, the model has been input in Fig. 3. 2 area of gray value, image point coordinate, to get to the pixel size of particle area, so as to realize the high precision of calibration area to particle area pixel mapping.

Fig. 3. Schematic of calibration plate and particles

4 Results Analysis In order to demonstrate the accuracy of the above method, this paper uses the method of prediction calibration board information to verify. The verification steps are as follows: (1) Take calibration board pictures at any angle; (2) All angular points except the last row and the last column are extracted, as shown in region 1 in Fig. 3. The angular point coordinates and their grayscale values are calculated. (3) Calculate corner pixel size according to checkerboard specification; (4) Classify sample values: The sample values of the upper part of the picture are training data, and the sample values of the lower part are test data; (5) The SVR technique is used to train the sample set data, and after obtaining the prediction model, the data information of the test set is input for prediction; (6) Analyze and compare the error between the predicted size and the actual size.

40

H. Jing et al.

In order to make the result more accurate, the 15 images used in this paper have been corrected by distortion. For correction methods, see correction of distorted images based on binary global interpolation. The calibration board used for the image is 13 rows and 13 columns, and a total of 121 corner points were removed from the last row and last column. Take the first 50% of these sample points as the training set (1–60) and the last 50% as the test set (61–121). The ε-SVR regression model was selected, and the model parameters were set as follows: kernel functions RBF, insensitive coefficient ε 0.01, penalty factor c and RBF weight coefficient g is obtained by the cross method. After SVR prediction, the error results of 15 images are shown in Table 1. Table 1. Error analysis Image number

1

2

3

4

5

Mean square error

0.002026

0.000091

0.001569

0.001131

0.000376

Correlation coefficient (%)

99.473

99.872

99.772

99.896

99.734

Image number

6

7

8

9

10

Mean square error

0.000168

0.000133

0.000652

0.000636

0.000965

Correlation coefficient (%)

99.888

99.861

99.812

99.186

99.472

Image number

11

12

13

14

15

Mean square error

0.007845

0.000933

0.009034

0.000102

0.000884

Correlation coefficient (%)

99.275

99.724

99.197

99.864

99.851

Table 1 shows that the method proposed in this paper has high accuracy, and the mean square error of the prediction results is lower than 0.01. In this method, multiple shooting angles are used for image sampling, so there is no strict requirement for the positioning of the calibration plate. Thus concluded that the method can effectively overcome the distortion, angle, azimuth and other factors on the influence of the pixel size, make no longer biased when measuring small particle size, solve the problem of sensitive area pixel size which is difficult to obtain precision and thus improve the measuring precision of small objects. In order to make the above results more intuitive, four groups of test data and regression data of images 2, 4, 6 and 8 are selected for drawing in this paper. The drawing results are shown in Fig. 4. In addition, the image was preprocessed in this paper, and the image distortion was corrected. To a certain extent, the accuracy of the camera imaging model was improved, and the prediction results were more accurate. In order to further verify the necessity of image preprocessing, this paper compares the prediction results of the original image and the preprocessing image of the first image, as shown in Table 2. Therefore, it can be concluded that the image without distortion correction has low prediction accuracy and large error. For the convenience of observation, the comparison results are drawn as follows. In Fig. 5, the image (a) used has been corrected by distortion, with a minimum absolute error of 0.0335, a maximum absolute error of 0.2002, an average absolute error

A High-Precision Pixel Mapping Method

41

Fig. 4. Results comparison of original and regression data Table 2. Error comparison of original and preprocessing image The original image MSE Correlation coefficient (%)

0.3265 34.942

The preprocessed image 0.002026 99.473

of 0.0909, and the error is smaller. The image (b) used has been the original unprocessed graph, with a minimum absolute error of 0.0016, a maximum absolute error of 1.4320, an average absolute error of 0.4443, and the error is larger.

5 Conclusion In this paper, the regression support vector machine (SVR) technology is used to predict the pixel size of particle area. To this end, this paper firstly introduces the basic idea and principle of the support vector machine regression (SVR), and describes in detail the

42

H. Jing et al.

Fig. 5. Schematic of calibration plate and particles

pixel mapping algorithm by using the SVR, implementation steps, finally analyzed the accuracy and feasibility of the method from various angles. Through the result analysis, it is proved that the method has high accuracy and practicability and can effectively improve the measurement precision of the system.

References 1. Xu GTDFF (2012) Camera calibration under small field of view. Chin J Lasers 8:34 2. Tsatsoulis C, Fu K-s (1985) A computer vision system for assembly inspection. In: Intelligent robots and computer vision, vol 521, pp 352–357. International Society for Optics and Photonics 3. Sternberg SR, Sternberg ES (1983) Industrial inspection by morphological virtual gauging. In: Proceedings of IEEE computer society workshop on computer architecture for pattern analysis and image database management, pp 237–47 4. Li X, Zhang T, Zhang S, Nan B, Guo X (2008) Small objects dimension measure and threedimension reconstruction system based on laser triangulation. Opt Instrum 30(6):21–26 5. Mills R (1991) Development of a line-scan camera for 2D high accuracy measurement. Machine Vis 267–277 6. Angrisani L, Daponte P, Pietrosanto A, Liguori C (1999) An image-based measurement system for the characterisation of automotive gaskets. Measurement 25(3):169–181 7. Chen M-C (2002) Roundness measurements for discontinuous perimeters via machine visions. Comput Indus 47(2):185–197 8. Ni KS, Nguyen TQ (2007) Image superresolution using support vector regression. IEEE Trans Image Process 16(6):1596–1610 9. Smola AJ, Schölkopf B (1998) On a kernel-based method for pattern recognition, regression, approximation, and operator inversion. Algorithmica 22(1–2):211–231 10. Cortes C, Vapnik V (1995) Support-vector networks. Mach Learn 20(3):273–297 11. Han H, Dang J, Ren E (2012) Comparative study of two uncertain support vector machines. In: 2012 IEEE fifth international conference on advanced computational intelligence (ICACI), pp. 388–390. IEEE

A High-Precision Pixel Mapping Method

43

12. Okujeni A, Van der Linden S, Jakimow B, Rabe A, Verrelst J, Hostert P (2014) A comparison of advanced regression algorithms for quantifying urban land cover. Remote Sens 6(7):6324– 6346 13. Zhang W, Du Y, Yoshida T, Wang Q, Li X (2018) SamEn-SVR: using sample entropy and support vector regression for bug number prediction. IET Softw 12(3):183–189 14. 任俊, 胡晓峰, and 李宁 (2018) 基于 SDA 与 SVR 混合模型的迁移学习预测算法. 计算 机科学 45(1): 280–284 15. Utkin LV, Coolen FPA (2018) A robust weighted SVR-based software reliability growth model. Reliab Eng Syst Saf 176:93–101 16. Amraei S, Mehdizadeh SA, Sallary S (2017) Application of computer vision and support vector regression for weight prediction of live broiler chicken. Eng Agric Environ Food 10(4):266–271 17. Mukherjee A, Ramachandran P (2018) Prediction of GWL with the help of GRACE TWS for unevenly spaced time series data in India: analysis of comparative performances of SVR, ANN and LRM. J Hydrol 558:647–658 18. Liu Y, Liu Y (2010) Incremental learning method of least squares support vector machine. In: 2010 international conference on intelligent computation technology and automation, vol 2, pp 529–532. IEEE 19. Huang C-L, Tsai C-Y (2009) A hybrid SOFM-SVR with a filter-based feature selection for stock market forecasting. Exp Syst Appl 36(2):1529–1539 20. Castro-Neto M, Jeong Y-S, Jeong M-K, Han LD (2009) Online-SVR for short-term traffic flow prediction under typical and atypical traffic conditions. Exp Syst Appl 36(3):6164–6173 21. 王杰, 刚轶金, 石成辉 (2008) SVM—RBF 网络在混沌时间序列预测中的应用. 微计算 机信息 24(33):136–137

Nucleus Segmentation from Microscopic Bone Marrow Image Shilpa, Rajesh Gopakumar(B) , and Vasundhara Acharya Department of CSE, Manipal Institute of Technology (MIT), Manipal Academy of Higher Education (MAHE), Manipal 576104, India [email protected], [email protected], [email protected]

Abstract. Acute myeloid leukemia (AML) is a type of cancer that originates in the bone marrow and moves into the blood quickly. The disease progresses rapidly if not treated. Therefore identification of the disease at an early stage is very important. In the proposed work, different image segmentation techniques like K-medoids and watershed, region growing and active contour have been applied for the segmentation of the nucleus from the blast cells and the results are analyzed.

Keywords: Blood smear images

1

· Leukemia · Segmentation

Introduction

Image Segmentation is the first and most important step in image processing. It is the process in which a digital image is divided into different partitions. The partitions are commonly termed as pixels. Using these pixels, information about the objects is extracted from the image based on certain properties. Choice of an efficient segmentation algorithm is very important since feature extraction depends on it and efficient feature extraction is very important for the proper identification and classification. Microscopic morphological examination of the bone marrow smear is often the first step in the detection of leukemia. However, the procedure is difficult and the accuracy is limited. The procedure is also timeconsuming and suffers from inter-observer variability. Hence, image processing and machine learning techniques are used by researchers to overcome this problem. In the case of AML, proper segmentation of the blood smear image into nucleus and cytoplasm is very important for the identification and classification of the disease. In this work, the emphasis is on the segmentation of the nucleus. The AML blood smear Images for the work are obtained from Kasturba Hospital, Manipal. In the proposed work nucleus is segmented from the blood smear images using different segmentation techniques like K-medoids and watershed, active contour and region growing and the results are analyzed. c Springer Nature Singapore Pte Ltd. 2021  C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_5

Nucleus Segmentation from Microscopic Bone Marrow Image

2

45

Literature Review

The input RGB image is converted to CIELAB color space. Segmentation is performed on the image using the K-means clustering algorithm. Four clusters of nucleus, cytoplasm, background and RBCs are obtained. From the segmented nucleus and cytoplasm, the shape and texture-based features are extracted. The extracted features are classified using SVM [1]. The input RGB image is enhanced using global contrast stretching and median filtering. The enhanced image is divided into marker and mask images for image reconstruction in segmentation of the whole cell. K-means is applied to the mask image. For clear cell separation, watershed distance transform is applied. For nucleus segmentation, the RGB image is converted to HSV color space. Image pixel multiplication is performed and the nucleus is obtained [2]. Active contour is applied to the input RGB image. From the segmented nucleus, features are extracted. Using these features nucleus is grouped into normal or abnormal type [3]. The input RGB image is pre-processed. K-means clustering is performed for segmentation of the nucleus. The bounding box is drawn around each nucleus and the rectangular image of the nucleus is cropped out of the original image. The shadowed c-means clustering algorithm is performed for lymphocyte segmentation. Features are then extracted from the segmented nucleus and cytoplasm. Using the extracted features classification is performed by an ensemble of classifiers like naive Bayes (NB), K-nearest neighbor (KNN),support vector machines (SVM), radial basis functional network (RBFN), and multilayer perceptron (MLP) [4]. To the input RGB image, the median filter is applied. The resulting image is then converted to LAB color space. The a* and b* components are selected as features. K-means clustering algorithm is applied. Morphological operations are performed on the nucleus cluster. The connected nuclei are then separated using the region growing algorithm [5]. The input RGB image is converted to grayscale.WBCs are segmented using sub imaging. The edges are detected. The nucleus is extracted using GVF snake. Holes in the nuclei are filled. The segmented nucleus is subtracted from the gray image. The cytoplasm is extracted using zack thresholding. The proposed approach efficiently segments both nucleus and cytoplasm [6]. The input RGB image is first converted to grayscale. Opening and closing operations are performed on the image with ball as the structuring element and a radius and height of 5 to initiate the estimate of the cell foreground. The zero level set is initialized. The level set function is initialized as the signed distance from pixel position to zero level set. The level set function is evolved using GAC. Euler time step is updated. The resulting image is converted to binary [7]. Image segmentation is performed using automatic seeded region growing and instance-based learning. The seeds are generated using histogram analysis. For each band, the histogram is divided into subintervals. The magnitude of each subinterval is obtained. The subintervals are grouped based on their amplitude and non-representative subintervals are deleted. The amplitude is reduced and

46

Shilpa et al.

consecutive intervals are fused. Next, image segmentation is performed using seeded region growing algorithm. Instance-based learning is used as distance criteria [8]. The input RGB image is converted to Lab color space. On the resulting image, segmentation is performed using the K-means clustering algorithm. From the segmented images, GLCM features and fractal dimensions are obtained. The extracted features are classified using SVM. Cross-validation is performed after classification. The system is then tested with 50 blood smear images and an accuracy of 93.5% is obtained [9]. The input RGB image is pre-processed. Segmentation is performed using two different approaches-K-means clustering and color-based segmentation algorithm. Features extracted from images segmented by K-means is classified using SVM. Features extracted from images segmented using color based clustering is classified using KNN classifier. The results of these approaches are analyzed [10]. The input RGB image is pre-processed. The pre-processed image is segmented using the K-means clustering algorithm. The features extracted are classified using SVM [11]. The input RGB image is pre-processed. The image is then converted to Lab color space. Segmentation is performed using the K-means clustering algorithm. Then global thresholding is applied. The watershed algorithm is applied to separate overlapping cells [12].

3 3.1

Methodology Image Acquisition

AML M1-M5 (Fig. 1) blood smear images were obtained from the Kasturba Hospital, Manipal. The equipment that is used for acquiring the images is an Olympus U-CMAD3 camera that is mounted on an optical microscope having a magnification of 4 * 100. The U-CMAD3 camera is a 5.24 megapixels single CCD digital color camera. The live frame rate is 8 frames per second at a live resolution of 2560 * 1920. The bone marrow blood slide is stained using Leishman stain. The images that were captured were approved by the hematologist. 3.2

Image Segmentation

K-medoids with modified watershed In K-medoids clustering each pixel is assigned to a cluster. Each pixel has a cluster index. The cluster index is found using the kmedoids function. Based on these cluster indices pixels are clustered together. The mean value for each cluster is found. Nucleus has the minimum value for the mean. Any over segmentation occurring in the segmented nucleus is removed using the watershed algorithm.

Nucleus Segmentation from Microscopic Bone Marrow Image

47

Fig. 1. AML M1-M5

Active contour Active contour also known as snakes is a segmentation technique in which the curve evolves in order to detect the object, when subjected to constraints. The curve initially starts around the object, moves towards the interior normal and finally stops at the boundary [13]. In the proposed work, the nucleus is extracted from the image, holes in the image are filled and erosion is performed. Modified watershed algorithm is performed in order to separate touching cells. To the resulting image, active contour is applied. The nucleus is obtained. Region growing Region-based segmentation is based on the principle that pixels belonging to the same neighborhood are more similar to each other and dissimilar to pixels belonging to other regions. Region growing is a type of regionbased segmentation in which pixels are grouped together into regions based on their similarities [14]. In the proposed work, the regiongrowing function is applied for the segmentation of the nucleus.

4

Results

Figure 2 depicts the output of the different steps required for the clear separation of the nucleus using K-medoids and watershed algorithm. Figure 3 depicts the output of the different steps required for the clear separation of the nucleus using active contour. Figure 4 depicts the output of the different steps required for the clear separation of the nucleus using region growing algorithm.

48

Shilpa et al.

Fig. 2. a Input image, b Image converted to lab color space, c Background image, d Image containing RBCs, e Image containing nucleus, f Image containing clearly separated nucleus

Fig. 3. a Input image, b Image containing nucleus

5

Conclusion

K-medoids with modified watershed clearly separates the nucleus than active contour and region growing as seen in Fig. 5. The modified watershed algorithm removes any tiny local minima occurring in the image. Therefore clear separation of the nucleus could be achieved.

Nucleus Segmentation from Microscopic Bone Marrow Image

49

Fig. 4. a Input image, b Image containing nucleus

Fig. 5. Nucleus obtained after segmentation using a K-medoids, b Active contour, c Region growing

Acknowledgements. The authors would like to thank Dr. Sushma Belurkar, Associate Professor, Department of Pathology, Kasturba hopsital, Manipal for helping in capturing the medical images required for the research.

References 1. Kumar P, Udwadia SM (2017) Automatic detection of Acute Myeloid Leukemia from microscopic blood smear image. In: International conference on advances in computing, communications and informatics (ICACCI) 2. Setiawan A, Harjoko A, Ratnaningsih T, Suryani E, Wiharto Palgunadi S (2018) Classification of cell types in Acute Myeloid Leukemia (AML) of M4, M5 and M7 subtypes with support vector machine classifier. In: 2018 international conference on information and communications technology (ICOIACT) 3. Marzuki NIC, Mahmood NH, Azhar M, Razak A (2015) Segmentation of white blood cell nucleus using active contour. J Teknologi 4. Mohapatra S, Patra D, Satpathy S (2014) An ensemble classifier system for early diagnosis of acute lymphoblastic Leukemia in blood microscopic images. Neural Comput Appl 5. Sarrafzadeh O, Dehnavi AM (2015) Nucleus and cytoplasm segmentation in microscopic images using K-means clustering and region growing. Adv Biomed Res

50

Shilpa et al.

6. Sadeghian F, Seman Z, Ramli AR, Kahar BHA, Saripan M (2009) A framework for white blood cell segmentation in microscopic blood images using digital image processing. Biol Procedures Online 7. AL-Dulaimi K, Tomeo-Reyes I, Banks J, ChandranV (2016) White blood cell nuclei segmentation using level set methods and geometric active contours. Ministry of Higher Education and Scientific Research. Iraq 8. G´ omez O, Gonz´ alez JA, Morales EF (2007) Image segmentation using automatic seeded region growing and instance-based learning. LNCS 9. Madhukar M, Agaian S, Chronopoulos AT (2012) Deterministic model for acute myelogenous Leukemia classification. In: IEEE international conference on systems, man, and cybernetics (SMC) 10. Sen NB, MathewM (2016) Automated AML detection from complete blood smear image using KNN classifier. Int J Adv Res Electr Electron Instrum Eng 11. Mulik V, Alhat S, Bhilare P, Bhoge V (2016) Analysis of acute lymphoblastic Leukemia cells using digital image processing. Int J Sci Res Dev 12. Savkare SS, Narote SP (2015) Blood cell segmentation from microscopic blood images. In: International conference on information processing 13. Chan TF, Vese LA (2001) Active contours without edges. IEEE Trans Image Process 10(2) 14. Fatma M, Sharma J (2014) A survey on image segmentation techniques used in Leukemia detection 4(5):66–71

Developing a Framework for Acquisition and Analysis of Speeches Md. Billal Hossain(B) , Mohammad Shamsul Arefin, and Mohammad Ashfak Habib Department of Computer Science & Engineering, Chittagong University of Engineering & Technology, Chattogram 4349, Bangladesh [email protected] {sarefin,ashfak}@cuet.ac.bd

Abstract. Speech plays a vital role for human communication. Proper delivery of speech can enable a person to connect with a large number of people. Nowadays, a lot of valuable speeches are being provided by many popular people throughout the world and it will be very helpful if important information can be extracted from those speeches by analyzing them. An automatic speech-to-text converter can facilitate the task of speech analysis. There have been carried out a lot of works for the conversion of speech to text in the last few decades. This paper presents a framework for the acquisition of speech along with the location of the speaker and then conversion of that speech into text. We have worked with speeches containing three different languages. To evaluate our framework, we collected speeches from several locations and the result shows that the framework can be used for efficient collection and analysis of the speeches.

Keywords: Recording recognition

1

· Location tracking · Database · Speech

Introduction

Speech is the most natural form of communication and interaction between humans. It enables us to exchange knowledge without experiencing it directly. It is a very important part of human development. Speech can convince its audience to some particular agenda [1]. To attract and motivate people speeches are delivered by many people. Good, accurate and authenticate speeches can guide people to the right direction and enrich their knowledge. It can play an effective role in changing the mentality of large number of people or strengthening their belief in the speaker. For example, we can think about the speech given by Abraham Linclon [2] or Martin Luther King [3] in order to inspire their people. They used their motivational speaking skill for spreading their vision. c Springer Nature Singapore Pte Ltd. 2021  C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_6

52

M. B. Hossain et al.

However, misguided speeches can lead to a crime and is a threat to the country. It often directs people in the wrong direction and increases risks in the society. Due to the rapid growth of internet technology, any misleading speech can be easily distributed among different groups of people and by the influence of that speech they can create huge problems for the society. It does not only threaten the life and livelihood of the people of a country but also interrupt in the peacekeeping throughout the world. This types of threats have got the concern in both national and international security and steps are being taken to stop such types of crimes. So, a good speech repository and speech monitoring system can assist in spreading valuable speeches and prevent spreading any misleading information. By observing the speeches provided by the speakers, it is possible to stop any potential crime. Besides that, if the audio of speech can be converted into text then the speech can be processed automatically to extract information from that text which will also help many search engines that uses computer readable text as data and offers large amount of data ranked by similarity. However, due to difficulties in recording and varieties of speeches, it is difficult to accurately extract information from these speeches. In this paper, a framework is proposed for the acquisition of speech efficiently. The location of the speaker is also tracked. Then this speech is stored in a database as an audio file along with the location information. After that, the language of the speech is detected and then the speech is converted into text using a speech recognition API. We considered three different languages in our framework: English, Bengali and Arabic. The remaining section of the paper is organized as follows, Sect. 2 shows the previous work related to this paper, Sect. 3 describes about the methodology of the proposed framework, Sect. 4 represents the implementation and performance evaluation of the system and in Sect. 5, a conclusion is drawn with future scope of improvements.

2

Related Work

The usage of mobile phone has increased drastically over the last few years. It is approximately 3.5 times larger than PCs [4]. Nowadays, mobile phone is not only being used as a tool for making call and writing SMS, but also act as a mean for personal entertainment and communication with the world [5]. Almost every features that is available in PC, can also be found in a smartphone. Smartphones are available to almost everyone and one of the most popular operating systems being used in those devices is Android, developed by Google [6]. Even a normal android device can perform varieties of tasks [7], like recording audio or video, capturing image, detecting location etc. Android applications can be developed by using these features which can be applied in many fields [8]. For example, location service and audio recording feature of the android smartphone can be used for efficient recording of speech along with the location information for effective analysis of the speech that is provided by many valuable speakers around the world.

Developing a Framework for Acquisition . . .

53

Analysis of speech manually is a lengthy process and requires a lot of time. Speech recognition or simply known as speech-to-text conversion may facilitate this task, since analysis of text is lot easier than the direct analysis of speech. Recently, many works have been carried out in this field and got a lot of improvements [9,10]. Nowadays, Automatic speech recognition systems are quite well established and can operate at accuracy of more than 90% [11]. The advancement of algorithms and modern processors enabled the optimization of almost all the phases involved in the speech recognition task [12]. There are many commercial and open source systems like CMU Sphnix, Microsoft Speech API, AT&T Watson, Google Speech API etc. [13]. CMU Sphnix [14] is an open-source speech recognition system that was developed at Carnegie Mellon University (CMU). It consists of large vocabulary and speaker independent speech recognition codebase which is available to download and use. It has different versions and packages for different applications. The latest version uses Hidden Markov Model (HMM) [15] and a strong acoustic model that is trained with a large set of vocabulary. In [16], Pytorch-kaldi toolkit is used for automatic speech recognition. Microsoft has developed an speech recognition API known as Speech Application Programming Interface or SAPI for Windows applications [17]. It used Context Dependent Deep Neural Network Hidden Markov Model (CD-DNN-HMM) that is trained with a huge volume of dataset for achieving a good recognition accuracy. Currently it is being used in Microsoft Office, Microsoft Speech Server and Microsoft Agent. Google has developed their own speech-to-text (gSTT) conversion API [18], using deep learning techniques that achieved less than 8% word error rate. This API is being used in many applications like voice search, voice typing, translation, navigation, YouTube transcription etc. Among all the above described systems, gSTT API gives better result than others in terms of both word error rate [19] and conversion time [20]. The conversion of speech-to-text in Google speech API takes place as soon as it receives the first voice packets which saves the time for further processing. For these reasons, Google speech API is chosen in our framework for the automatic conversion of speech into text.

3

Methodology

The framework is divided into three modules: (i) Speech Acquisition Module, (ii) Speech Storage Module, and (iii) Speech Recognition Module. In Fig. 1 the overall graphical structure of the framework is shown. First the speech will be recorded along with the location information. Then this speech will be stored in a database where it will be processed for automatic speech recognition. There is also a user friendly website where anyone can see all of the recorded speeches along with their location information. Since the number of speeches will be increased with the progression of time, there has been provided a way to filter the total number of speeches by searching among the speeches in the database based on various criteria.

54

M. B. Hossain et al.

Sound Wave

Speech

Storage of Speech Details

Recording

User

Database Storage of Audio Files

Location Information

Text File

Speech-to-text Conversion

Audio File

Detect Language

Map Storage

Acquisition

(a)

Recognition

(c)

(b)

Fig. 1. Overall graphical structure of the system: a Speech acquisition module; b Speech storage module; c Speech recognition module

3.1

Speech Acquisition Module

Speech acquisition contains two parts. One for recording of speech and another for tracking the location. Figure 2 shows the architecture of the speech acquisition module. A Microphone is used to capture the sound of the speaker and it is stored as an audio file in the local storage. To track the speaker’s location first latitude and longitude need to be calculated. Using the latitude and longitude we can calculate the actual address of the speaker. When the speaker finishes his speech, it is uploaded into the cloud database where further processing takes place.

Recording Audio Using Microphone

Save Audio Wi-fi Detailed Speech

Get Latitude and Longitude Using Location Service API

Upload to Database

Calculate Address from Latitude and Longitude

Fig. 2. Framework architecture of the speech acquisition module

3.2

Speech Storage Module

Table 1 shows the structure of the table in the database. For each incoming audio file a row is created in the database with a unique identifier. The file name of the incoming audio file may be same as some other audio files in the database.

Developing a Framework for Acquisition . . .

55

So, the file name is renamed as: Unique ID + ‘.’ + File Extension. Initially, we do not know what is the language and the text contained in that audio. So, they are set to null. After analyzing the audio file we update the database table with its corresponding language and text file. The name of the text file is same as the audio file name except that its extension is set as ‘txt’. Table 1. Structure of the table in the database unique id audio file name Location

Language text file name

001

001.wav

Pahartoli, Chattogram, 4349, Bangladesh

Bengali

001.txt

002

002.wav

Habiganj, Sylhet, 3310, Bangladesh

English

002.txt

003

003.wav

Durgapur, Chandpur, 3640, Bangladesh

Arabic

003.txt

004

004.wav

Dhanmondi, Dhaka, 1208, Bangladesh

Mixed

004.txt

From Table 1 it can be seen that, the audio file name and text file name can be obtained from the unique id. So we can omit this extra two columns. Besides that, instead of using full name of the languages, we can encode them with numeric digits. i.e: 1 for Bengali, 2 for English, 3 for Arabic and 0 for mixed language (that contains one or more language in the speech). The updated database table is shown in Table 2. Table 2. Structure of the updated table in the database

3.3

unique id

Location

Language

001

Pahartoli, Chattogram, 4349, Bangladesh

1

002

Habiganj, Sylhet, 3310, Bangladesh

2

003

Durgapur, Chandpur, 3640, Bangladesh

3

004

Dhanmondi, Dhaka, 1208, Bangladesh

0

Speech Recognition Module

In the recognition module the language of the incoming speech is detected and the speech is converted into text. Figure 3 shows the overall architecture of the conversion process. We use a speech recognition API for converting audio into

56

M. B. Hossain et al.

text. But the problem is that for long duration of speeches the accuracy of the conversion is very low. To solve this problem we can either split the speech into smaller chunks of constant size or we can split them based on the silence presented in the audio. If we split by keeping a constant duration then a word within the speech might get splitted and that word will not be detected. So, we choose to split the speech based on the silence presented. It is possible because when we speak, we pause for a small duration after finishing a sentence. That means, we can consider each chunk as a sentence. The speech is splitted if the amplitude level in the audio is less than −16 dBFS for more than 0.5 s.

Sentence-1

Chunk-1

Sentence-2

Chunk-2 Input Audio File

Conversion Module

Chunk-n

Merging (+)

Output Text File

Sentence-n

Fig. 3. Framework architecture of the speech recognition module

Another problem is that, we need to specify the language to which we want to convert it. For example, if a speech contains English language then we need to specify language parameter as English in the API. So, if we want to recognize speech containing multiple language we need to specify the language parameter manually. Besides that, most of the time speech contains more than one language and the system will not give good result for such type of mixed language speeches. To solve this, we send three requests in the API for three different languages for each sentence or chunk. Then the converted text is splitted into words and then checked in the dictionary of that language if that word exists. If the word is found in the dictionary then we increase the counter for that language. The language for which the counter value is maximum, is chosen as the language of that sentence. Algorithm 1 shows the process of detecting language and conversion of speech chunk into text. After the conversion the database is updated with the text file.

4

Implementation and Evaluation of the Framework

For implementing the framework we have developed an android application that can record speech along with the location information. We used Google speech-to-text (gSTT) converter as our speech recognition API. Three different languages: Bengali, English and Arabic are considered in our framework. For recording using android app an instance of the media recorder class is created and then the mic is initialized for the recording purpose. There are three buttons, first one for starting the recording, second one for stopping the recording and third one for uploading the recorded speech. The recording is started when

Developing a Framework for Acquisition . . .

57

Input: chunk Output: language, text max ← 0; for i ← 1 to 3 do counter ← 0; converted text ← recognitionAP I(audio ← chunk, language ← i); for each word in converted text do if word is in dictionary of language i then counter ← counter + 1; end end if counter > max then max ← counter; language ← i; text ← converted text end end

Algorithm 1: Detecting language and converting speech chunk into text

the user presses the start recording button and then the recording of the speech is started and continues until the stop button is pressed. After the end of the recording the audio file is stored in the local storage of the phone. Now this speech can be uploaded by pressing the upload button. When the upload button is pressed the location of the device is tracked by using the location service of the android device and then it is uploaded to the server. To get the location first latitude and longitude need to be calculated which can be obtained by using Google Play Service. From the latitude and longitude, the actual address of the speaker is calculated which contains city, postal code, state and country name. When the uploading is complete, it will be available in the website and anyone can have access to that speech. After it is uploaded to the website it is passed to gSTT API for speech recognition. To verify the proposed framework, we collected 100 speeches from 10 different locations in Bangladesh. The speeches were delivered by 10 people (10 speeches each) among which 6 of them were male and 4 of them were female. Each speaker installed the android app and gave permission to record their speech. After the speech is uploaded to the server it is verified with the actual data. Table 3 shows the performance of the speech acquisition process for 10 different locations. From the table it can be seen that some location information was not available in some region (represented by null). It is because of the variation in geocoding detail where the address line can vary. However, we can say that the acquisition of speech is done accurately with a percentage of 100%. The performance of the speech recognition is dependent on the quality of the speech. If the environment where the speech is recorded is very noisy or if the sound of the speech is low then it gives comparatively low performance. Google API has automatic noise reduction in the recognition process which helped in

58

M. B. Hossain et al. Table 3. Performance evaluation of speech acquisition process Speaker Actual location

Detected location

Speech quality

1

Pahartoli, Pahartoli, Highly satisfactory Chattogram, Chattogram, 4349, Bangladesh 4349, Bangladesh

2

Habiganj, Sylhet, Habiganj, Sylhet, Highly satisfactory 3310, Bangladesh 3310, Bangladesh

3

Durgapur, Chandpur, 3640, Bangladesh

Durgapur, Chandpur, null, Bangladesh

4

Dhanmondi, Dhaka, 1208, Bangladesh

Dhanmondi, Dhaka, 1208, Bangladesh

5

Gulshan, Dhaka, Gulshan, Dhaka, Highly satisfactory 1213, Bangladesh 1213, Bangladesh

6

Dinajpur, Rangpur, 5262, Bangladesh

7

Jamalpur, Jamalpur, Highly satisfactory Mymensingh, Mymensingh, 2030, Bangladesh 2030, Bangladesh

8

Bogura, Rajshahi, 5892, Bangladesh

9

Barguna, Barisal, null, Barisal, 8730, Bangladesh 8730, Bangladesh

10

Bagerhat, Khulna, 9301, Bangladesh

Dinajpur, Rangpur, 5262, Bangladesh

Bogura, Rajshahi, 5892, Bangladesh

Bagerhat, Khulna, null, Bangladesh

Satisfactory

Highly satisfactory

Highly satisfactory

Highly satisfactory

Satisfactory Highly Satisfactory

achieving a good recognition accuracy of the speech. Google speech recognition API achieved as low as 8% error rate in speech recognition [11]. Table 4 shows the performance of the recognition module. We can see that for speeches containing one language the recognition accuracy is comparatively higher and for mixed language the accuracy is also satisfactory if each sentence consists of a single language. However, if a sentence contains multiple language the system gives low recognition accuracy. For a speech if the total number of word is T and total number of missing word is M and total number of incorrect word is W , then the accuracy of the recognition is computed as,   M +W × 100% (1) Accuracy = 1 − T

Developing a Framework for Acquisition . . .

59

Table 4. Recognition accuracy of the framework for different languages Audio file

Actual speech

001.wav

|

Converted text

Detected language

Number of missing words

Number of wrong words

Accuracy %

Bengali

0

1

90

English

0

0

100

Arabic

0

0

100

Mixed

1

1

83.33

Mixed

2

0

84.61

? 002.wav

Birds are flying in the sky. birds are flying in the It seems so beauƟful sky it seems so when they fly. beauƟful when they fly

003.wav

004.wav

Fearlessness is like a muscle. I know from my own life. |

hear is a music I know from my own life |

How are you?

How are you

005.wav

The overall feasibility of the proposed framework is shown in the Table 5. It can be seen that the framework is feasible for many applications and it may help in performing many tasks very efficiently. Table 5. Overall feasibility of the proposed framework

5

Evaluation metric

Comments

Availability

Android phone is available to almost everyone

Implementation cost

Low

Recording quality

Very satisfactory

Location tracking

Nearly 100% accurate

Speech recognition

Less than 8% error rate

Applicability

Speech collection, speech monitoring, automation etc.

Conclusion

In this paper a framework is shown which can record speech and facilitate the task of speech analysis. Speech is recorded using an android app and the conversion of text is done by using Google speech recognition API. The system can be further applied in many fields like automatic subtitle creation, car driving using voice command, hearing impaired people etc. which can reduce a lot of manual works. Besides that, the collection of speech will enable people to access to the

60

M. B. Hossain et al.

speech anytime from anywhere. It can also be used for suspicious content detection in the speech which may prevent from occurring many unexpected events in the society. For such importance of a speech recognition framework, it can be applied in many ways which might ease the way we lead our daily life. Acknowledgments. This research is supported by the project ‘Establishment of CUET IT Business Incubator’.

References 1. Suedfeld P, Bluck S, Ballard EJ, Baker-Brown G (1990) Canadian federal elections: motive profiles and integrative complexity in political speeches and popular media. CJBS 22(1):26–36 2. Holzer H (2004) Lincoln at Cooper Union: the speech that made Abraham Lincoln president. Simon and Schuster 3. Vail M (2006) The “Integrative” Rhetoric of Martin Luther King Jr.’s’ I Have a Dream” speech. Rhetoric & Public Affairs 9(1):51–78 4. Gandhewar N, Sheikh R (2010) Google android: an emerging software platform for mobile devices. IJCSE 1(1):12–17 5. Rice RE, Katz JE (2003) Comparing internet and mobile phone usage: digital divides of usage, adoption, and dropouts. Telecommun Policy 27(8–9):597–623 6. Kaur P, Sharma S (2014, March) Google android a mobile platform: a review. In: 2014 recent advances in engineering and computational sciences (RAECS), pp 1–5. IEEE, Chandigarh, India 7. Ableson F, Sen R, King C, Ortiz CE (2011) Android in action. Manning Publications Co 8. Rogers R, Lombardo J, Mednieks Z, Meike B (2009) Android application development: programming with the Google SDK. O’Reilly Media, Inc 9. Petridis S, Stafylakis T, Ma P, Cai F, Tzimiropoulos G, Pantic M (2018) Endto-end audiovisual speech recognition. In: 2018 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, Calgary, AB, Canada, pp 6548–6552 10. Krishna G, Tran C, Yu J, Tewfik AH (2019) Speech recognition with no speech or with noisy speech. In: ICASSP 2019–2019 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, Brighton, United Kingdom, pp 1090–1094 11. Isaacs D (2010) A comparison of the network speech recognition and distributed speech recognition systems and their effect on speech enabling mobile devices. Doctoral dissertation, University of Cape Town 12. Khilari P, Bhope VP (2015) A review on speech to text conversion methods. IJARCET 4(7) 13. Gaida C, Lange P, Petrick R, Proba P, Malatawy A, Suendermann-Oeft D (2014) Comparing open-source speech recognition toolkits. Technical Report of the Project OASIS 14. Sphnix API. https://cmusphinx.github.io/ 15. Mukherjee S, Mandal SKD (2014) A Bengali HMM based speech synthesis system. arXiv preprint arXiv:1406.3915 16. Ravanelli M, Parcollet T, Bengio Y (2019) The Pytorch-Kaldi speech recognition toolkit. In: ICASSP 2019–2019 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, Brighton, United Kingdom, pp 6465–6469

Developing a Framework for Acquisition . . .

61

17. Brown R (2008) Exploring new speech recognition and synthesis APIs in Windows Vista. Talking Windows, MSDN Magazine 18. Cloud Speech-to-text. https://cloud.google.com/speech-to-text/ 19. K¨epuska V, Bohouta G (2017) Comparing speech recognition systems (Microsoft API, Google API and CMU Sphinx). Int J Eng Res Appl 7(3):20–24 20. Assefi M, Liu G, Wittie MP, Izurieta C (2015) An experimental evaluation of apple Siri and google speech recognition. In: Proceedings of the 2015 ISCA SEDE, pp 1–6

Gait-Based Person Identification, Gender Classification, and Age Estimation: A Review Rupali Patua(B) , Tripti Muchhal, and Saikat Basu Maulana Abul Kalam Azad University of Technology West Bengal, Kolkata, India {rupali.patua,21tripti,saikatbasu}@gmail.com

Abstract. In this era where both techniques and technology are going digital, there is a need of designing and developing such a system that automatically identifies the person and also the person’s attributes with the help of the biometric system. The biometric of every person is unique. Therefore, in this context, the use of the biometric system to identify the identity of an individual is a popular approach. Gait is a biometric approach that helps to verify the identity of a person by their walking patterns. Owing to the advantages of gait as a biometric, its popularity among the researchers is amplified in recent years. In this paper, a review is represented, where several papers with different methods, have been mentioned which recognize the person and its attributes such as gender and age based on the gait of the person. Keywords: Gait

1

· Kinect · SVM · KNN · PCA

Introduction

In recent years, the researchers’ major area of inclination is towards biometric identification owing to its potential application in surveillance systems, social security, etc. [1]. Biometric helps in identifying an individual using their body parts. There are some characteristics in a human being, those are unique and can be used as biometric like the fingerprint, the retina of the eyes and many more [2]. Biometric can be classified as physiological and behavioral biometrics. Under the physiological biometrics the face, palm, ears, and eyes are considered, whereas the behavioral biometric includes voice, gait, speech, handwriting, and signature. Gait is a biometric approach that is becoming popular in recent years. The popularity of gait is due to its advantages, such as it can work in low resolution as well as does not need any person’s cooperation as compared to other biometrics, such as face where high resolution is required to collect the person’s face image as well as it requires the person’s cooperation [3]. Gait is defined by the walking pattern of an individual. It works as a biometric because c Springer Nature Singapore Pte Ltd. 2021  C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_7

Gait-Based Person Identification, Gender Classification

63

every person has a unique walking pattern that cannot be imitated. The gait as a biometric differs from the others as it can be captured from distance as well as it does not need any cooperation from the individual. The gait features depend on various attributes such as the anatomy, gender, age, muscle mass, etc. [4]. One of the main advantages of using gait is that no one can imitate the walk of an individual [5]. Also in recent years, gait information has been used to help in various diseases like Parkinson’s disease, Neurodegenerative Diseases and many more [6]. The approaches for gait recognition are: one is the modelfree approach, and the other is the model-based approach. In the former, the silhouette of a person is extracted to find the attributes of that person while walking. The model-free approach is computationally efficient, but at the same time, it has few drawbacks due to the following reasons, such as changes in viewpoint, lighting, and clothing, whereas in the latter, there is a pre-defined model from which the dynamic characteristics of an individual are extracted. It is computationally expensive but at the same time, gains more information and also forceful to changes in lighting, clothing, and viewpoint in contrast to the model-free approach [7–9]. The inclination towards gait as a biometric and its huge implementation to identify the individual, the gender, and to estimate the age has been increased widely in recent years, and this has been the motivation to make this paper. In this paper, different methods of gait recognition from various recent research works have been studied and compared. The remainder of the paper is organized as follows: Sect. 2 briefly elaborates on sensors that have been used to recognize the gait of an individual. Sections 3, 4, 5, 6, and 7 represent the comparison of the methods from different papers. Apart from that, in each subsection, the feature extraction, classifiers, and the pros and cons have been discussed. There is a table which includes all the papers that have been compared. Section 8 represents the conclusion.

2

Literature Review

Gait as a biometric has an important role in many application areas such as surveillance systems, social security, senior monitoring, disease detection, identification and verification of an individual etc. This survey is organized by taking only a part of the gait recognition that includes person identification, gender classification and age estimation. In this regard, a classification tree is constructed in the Fig. 1, and it also includes whether the Kinect sensor is used or not. Kinect sensor is one among the other sensors, which is very effective to extract the skeleton points of an individual as well as to extract features. There is hardware, embedded in the Kinect sensor [4]. They are: RGB camera, Depth sensor, and Multiarray microphone. One of the major application for which the Kinect sensor is developed by Microsoft is the gaming application for the Xbox 360. Kinect has been used in biometrics, such as facial recognition, gait analysis, etc. [10]. There are two versions of Kinect sensor, one is Kinect v1 sensor, and the other is Kinect v2 sensor. In comparison to the Kinect v1 sensor, the skeleton

64

R. Patua et al.

Fig. 1. Classification tree

tracking from Kinect v2 sensor is more precise and also at a faster rate. The resolution of the Kinect v2 sensor is huge enough but there is a disadvantage that the Kinect v2 sensor is not lightweight compare to the Kinect v1 sensor. Both the sensors have different methods to calculate the depth.

3

Person Identification

This section elaborates on how the different types of features have been extracted through different Kinect sensors and the different classifiers have been used to identify the person. Finally, the pros and cons of different papers are discussed and compared. 3.1

Feature Extraction

All the papers in the table where a person is identified have used different methods and different types of Kinect sensors. In [11], the Kinect v1 sensor has been used to extract the 20 skeleton joint points, and from each point the distance feature vector is calculated with respect to the other points, and then for each joint the mean and the variance are calculated as features whereas in [1], the Kinect sensor has been used for the extraction of 20 skeleton joint points and then both the static and dynamic features are extracted. However, in [1] a single value of the feature is taken from a sequence of frames. In [12], also Kinect sensor is used for the extraction of 20 different skeletal points and here 11 static features and two dynamic features are extracted. In [13], the 3d skeletal data is

Gait-Based Person Identification, Gender Classification

65

extracted using the Kinect v2 sensor. Here two new features have been introduced, one is the JRD (Joint Relative Distance) and the other is the JRA (Joint Relative Angle). From these two features, the relevant features are selected by the genetic algorithm. In [14], the poses are extracted from the gait sequences with the help of the Kinect sensor. After these poses are acquired, they are used for the feature extraction and it is done by smoothing the voxel volume. In [15], the Kinect RGBD sensor is used to extract the skeleton data of a person. In this method, all the joints are taken in order to examine the entire video gait sequence and to extract the motion and anthropometric features. 3.2

Classifier

In [11], the KNN (K-nearest neighbor) is applied for the classification. It is a simple classifier and it classifies according to the similarity measured by the distance function. In [1], there are two classification methods used, one is the Levenberg-Marquardt back propagation and the other is the correlation algorithm. From the experiment, it has been seen that the recognition rate in the Levenberg-Marquardt back propagation is better than the correlation algorithm. In [12], three classifiers are used, one is the C 4.5, second is the Decision Tree and the third is the Naive Bayes classifier. The Naive Bayes classifier is a probabilistic classifier, performs better than the other two classifiers. In [13], a new classification method has been introduced, which is known as DTW (Dynamic Time Warping) based kernel. It is a non-linear time alignment technique. In [14], K-means clustering is used for the recognition method of an individual. In [15], two classification methods have been used. Firstly, SVM is used to identify the actions walk and run. After the respective action is identified, the person is recognized by calculating the identity cost of that person. 3.3

Discussion

In [11], the distance feature vector is calculated, which helps in the significant enhancement of the recognition rate as well as the reduction of computational time. In [1], only the single value of the feature is taken from the series of frames of a video sequence and it, in turn, becomes advantageous, since it reduces the training time as well as the classification time. Also, the compulsion of the gait cycle detection is eliminated. In [13], JRD and JRA methods have been used, which can handle speed variation and also it is robust to view and pose variation. In [14], the silhouette recorded by the Kinect gathers the depth information which in turn, preserves the vital gait information. In [15], the method identifies both the walk and run activities and the person. When the motion pattern and anthropometric features are used, the recognition rate is more accurate. In [12], the feature can be easily extracted by the use of the Kinect sensor. In [11], it has some limitations in terms of the accuracy rate. The method that uses mean or variance has a lower accuracy rate compared to the method that uses both mean and variance. In [1], some of the features extracted like the length of both the hands, the length of both the legs do not offer distinguishing

66

R. Patua et al.

features amid different individuals, which in turn becomes a disadvantage. In [13], it is susceptible to the changes that happen in the clothing and carrying object of an individual such as a handbag, etc. In [14], the method has higher response time. In [15], when biometric is used individually, it can be sensitive in some respect. In [12], the disadvantage is that the approach cannot fit for the identity of an individual among the crowd. Comparing [1] and [12] with [11], it has high precision and recognition rate. The approach used in [13] is more accurate than the approach used in [15], as only those joint pair is taken which are relevant.

4

Gender Classification

In this section, the methods used to extract the feature as well as the methods used to classify the gender have been elucidated and at the last, the pros and cons of different papers are discussed and compared. 4.1

Feature Extraction

The gait can be used for gender classification. Various papers for gender classification have been detailed in the table. In [16], the data of the human skeleton is extracted with the help of the Kinect sensor and along with that 20 joint points are captured. Here, the dynamic features are extracted differently by calculating the dynamic distance feature. In [17], the silhouette is segmented from the frames of a video, and then GEI and AEI are calculated. The feature space becomes better by applying the feature selection and resampling method. In [18], poses are taken from the video sequence, and then the feature extraction is done with the help of histogram. In [14], the new database is generated known as d gait database that contains the depth information of the subject. Here both the 2D and 3D gait features are extracted. 4.2

Classifier

In [16], three different types of classification methods have been used separately; those are the NN (Nearest Neighbor) classifier, LDC (Linear Discriminant Classifier), and SVM (Support Vector Machine). Out of these three, the accuracy rate of NN is better than the other two. In [17], the KNN is used for classification purpose. As the value of K = 1 in KNN, the performance degrades because KNN does not work well in the intrinsic data characteristics. In [18], for the classification purpose, the RBF kernel SVM is applied. In [19], a kernel SVM is used for the classification. 4.3

Discussion

In [16], the distance between the ankles is measured by which the accuracy to classify the gender becomes better. In [17], the feature selection reduces the

Gait-Based Person Identification, Gender Classification

67

dimensionality so that the classifier can work more rapidly and also more efficiently. In [18], the method is neither dependent on viewing angle, nor on clothing or carrying objects nor on the number of frames, which is an advantage of this method. In [19], the depth information gained by the Kinect is useful and also robust when the view of a person changes with respect to the camera. In [16], it has a disadvantage to the database as it doesn’t have a variant in clothing and bags. In [17], the gait considers not only the movement but also the appearance. Due to this fact, the approach is comprehensively affected and the dataset SOTON attains a poor result. In [18],in real world the challenge of tracking people in the mass gathering area still remnants. The variance calculated in the paper [19], shows that the 3D GF has a lower variance than 2D GF. Due to this reason, the 2D GF is less robust compared to the 3D GF in case of view variation. The method in [16] gets the accuracy above 90% whereas the method in [17] has an accuracy of 87% which shows the result of [16] is better than [17]. Since there is no detection of the gait cycle in [18], the method is better than [19], where gait cycle is needed.

5

Age Estimation

In this section, different method names that include a calculation to extract features and also different classifiers to estimate the age of a person are discussed. In the end, the pros and cons of various papers are illustrated. 5.1

Feature Extraction

In [20], the feature of an individual gait is captured by calculating the gait energy image (GEI). It represents both the static and dynamic features. In [21], three types of gait features are taken, which are based on silhouette. These gait features are the gait energy image (GEI), the frequency domain features and the gait periods. In [22], the feature extraction is done using gait energy image (GEI). The extracted feature dimension is reduced by applying a multi-label guided subspace. 5.2

Classifier

In [20], the problem of multiple age group classification is solved by the DAGSVM, which is a directed acyclic graph SVM. The DAGSVM is the solution to the multi-class classification problem by incorporating multiple binary SVM classifiers. In [21], the mean absolute error (MAE) and the cumulative score are taken for age estimation. In [22], the ML-KNN classifier is used for recognition and age decoding.

68

5.3

R. Patua et al.

Discussion

In [20], a new method that is the age group dependent is used. The computational time is low in this approach. In [21], the database contains approximately all the ages in the range from 2 to 94 years. This helps in performance evaluation and also it is reliable for the training model. In [22], the approach of multiple feature fusion is used, which constantly beats the single feature approach for age estimation this implies that the multiple feature fusion can give harmonizing information to represent features, which results in better estimation.. In [20], it refers that, when one of the hyperparameter contribution rates is lower, MAE (Mean Absolute Error) increases. In [21], there are two failures due to two reasons: (1) the testing of some subjects fails because the gait features of those subjects deviate from the gait of their actual age, (2) the test gait features and the training gait features are different from each other. In [22], the only single feature that is the GEI feature is extracted for age estimation. Though the single feature is less robust, the performance is reduced as compared to the multiple feature fusion. The approach in [20] is better compare to [21,22] as it uses the age group dependent approach, whereas others use the single age independent approach. In the approach of [20], the estimation error is increased as there is an increase of variation in age. The method in [4] is better for estimation of age than in [21,22] as it uses spatial proximity and multi-task learning on CNN.

6

Age Estimation and Gender Classification

Here, the methods of feature extraction and the classifier that helped to estimate age and classify gender are discussed. Also, the pros and cons of various papers have been illustrated. 6.1

Feature Extraction

In [23], the frequency domain feature is selected for feature extraction as it contains both 2D spatial information and dynamic information. 6.2

Classifier

In [23], the KNN classifier is applied for the classification approach. 6.3

Discussion

Here, some papers are referred, that worked on both gender and age. In [23], the database contains multiple views therefore; the multi-view application can be supported. In [18], when the diversification in gender and age increases, the classification becomes tough.

Gait-Based Person Identification, Gender Classification

7

69

Person Identification and Gender Classification

This section refers the feature extraction method and the classifiers to identify both the person and its gender. Apart from that, the pros and cons of several papers are discussed and compared. 7.1

Feature Extraction

In [24], the poses are taken from the video. The poses which are captured are represented by the Eulers angle which is the vectors of eight selected limbs, and then the dynamic features are extracted. The dissimilarity space is measured between the walking and the training sequence and it is represented as dissimilarity vector. In [25], the silhouette is obtained by the background subtraction, and then the cluster is formed. For every cluster, the feature is represented by calculating the average gait image of that cluster. 7.2

Classifier

In [24], the sparse representation is measured, and with the help of it the recognition and classification task are performed. In [25], the SRML is used as a discriminative distance metric in order to perform the recognition method. 7.3

Discussion

In [24], every gait can be represented by the vector of dissimilarity space, and this approach is efficient. In [25], the method recognizes both the person and gender in an arbitrary walking direction. This method can be applied to the outdoor environment. In [24], when partial body parts such as arms and legs are incorporated, there exists unpredictability due to the narrow training sample in the training set. Due to this reason, a faster deterioration in the performance happens when all the features are incorporated. In [25], the disadvantage is that the method will lead to failure when the training and testing sequences are different from each other. Comparing with [17,23,25] executes better recognition, classification and verification. Now in the below table, various papers are incorporated that have used a person’s gait and several methods to identify the person, classify the gender, and to estimate the person’s age. Table 1 contains the papers in the year range between 2010 and 2018.

70

R. Patua et al. Table 1. Comparison table Ref

Identification

Sensor

Method

Feature

Activity

Year

Database

[1]

Person

Kinect

LevenbergMarquardt back propagation and correlation algorithm

20 joint points detected using skeleton data in 3D coordinates. From these 20 joints points, the static and dynamic features are extracted

Action [walk]

2015

N.A.

[11]

Person

Kinect V1

Euclidean distance, meanv Variance, KNN

20 joint points is detected . The distance feature vector is calculated for each one of the joint with respect to other 20 joints

Action [walk]

2017

N.A.

[12]

Person

Kinect

1R, C4.5 decision tree, Naive Bayes

20 different points are detected from the skeleton and from there 13 biometric features are extracted out of which 11 are static and 2 are dynamic feature

Action [walk]

2012

N.A.

[13]

Person

Kinect V2

DTW, rank level fusion, genetic algorithm

Detection of the gait cycle is done using Kinect v2 sensor and then JRD (Joint Relative Distance), JRA (Joint Relative Angle) features are calculated

Action [walk]

2015

N.A.

[14]

Person

Kinect

Coordinate system transformation, K-Means clustering

FrontoParallel silhouette is generated and from these set the key poses is derived

Action [pose]

2014

N.A.

Gait-Based Person Identification, Gender Classification

Table 1. (continued) Ref

Identification

Sensor

Method

Feature

Activity

Year

Database

[15]

Person

Kinect RGBD

SVM,PCA, MDA

Motion pattern and anthropometric features

Action [walk, run]

2012

N.A.

[16]

Gender

Kinect

DDF, LDC, SVM, mean, standard deviation and Skew

From the 3D skeleton data, the 20 joint point is extracted and the gait signature is extracted using dynamic distance feature

Action [walk]

2017

N.A.

[17]

Gender

N.A.

GEI, AEI, KNN

The silhouette is being segmented from the frames of a gait video sequence and then the GEI and AEI are calculated

Action [walk]

2013

CASIA dataset A and SOTON dataset B

[18]

Gender

Kinect RGD Sensor

EulerAngle, PCA, SVM

The feature extraction is done with the help of histogram

Action [Pose]

2013

UPCV Gait

[19]

Gender

Kinect

PCA, LDA, SVM

2D and 3D gait features

Action [walk]

2012

D-Gait

[20]

Age

N.A.

DAGSVM, age-group dependent manifold learning, nonlinear SVR, L2 distance

GEI feature

Action [walk]

2018

OULP -Agr Dataset

[21]

Age

N.A.

Gaussian process regression, baseline algorithm

Three silhouette based features are taken such as averaged silhouette, frequency domain feature and gait period

Action [walk]

2011

Whole Generation gait

71

72

R. Patua et al. Table 1. (continued)

8

Ref

Identification

Sensor

Method

Feature

Activity

Year

Database

[22]

Age

N.A.

Label encoding algorithm, MLG (Multi-Label Guided), MLKNN, Gabor Magnitude, Gabor phase

Gabor feature

Action [walk]

2010

USF Database

[23]

Age+ Gender

N.A.

KNN

Frequency domain features

Action [walk]

2010

Multi View Gait Database

[24]

Person + Gender

Kinect

Euler’s angle, dissimilarity measure, Sparse representation

Dynamic features are extracted

Action [pose]

2015

N.A.

[25]

Person + Gender

Kinect

Background subtraction, SRML

The feature taken is the clusterbased averaged gait image

Action [walk]

2014

USF and CASIA-B Database

Conclusion

Gait has many advantages that make it more robust compared to the other biometric. But in some perspective, it is not robust when the changes in clothing and shoe types are considered. This is the limitation of gait that still needed to be approached. In this context, a database is needed, which should include all the perspectives like different view angles, changes in clothing etc., which is required for the more accurate recognition method. The gait as a biometric can be used in various places like in airports, shopping malls, etc. Gait recognition can be combined with many other research areas such as face recognition, security, and identification of an individual in crowd, which can be more tighten. In the recent years, gait recognition is used in smart phones with the help of the sensor like accelerometer to enhance the privacy of the smart phone. Nowadays, the most popular application of gait recognition is the health monitoring and surveillance system for elders and disabled persons. Also, many other applications of gait have introduced, which make gait recognition popular and hot topic in the area of Computer Vision. Acknowledgment. We are very grateful to the honorable vice-chancellor of our university for his support and encouragement to make this survey.

Gait-Based Person Identification, Gender Classification

73

References 1. Prathap C, Sumanth S (2015) Gait recognition using skeleton data. In: International conference on advances in computing, communications and informatics (ICACCI). IEEE, New York 2. Sharif M et al (2019) An overview of biometrics methods. Handbook of multimedia information security: techniques and applications. Springer, Cham, pp 15–35 3. Choudhury SD, Guan Y, Li C-T (2014) Gait recognition using low spatial and temporal resolution videos. In: International Workshop on Biometrics and Forensics (IWBF). IEEE, New York, pp 1–6 4. Sakata Atsuya, Takemura Noriko, Yagi Yasushi (2019) Gait-based age estimation using multi-stage convolutional neural network. IPSJ Trans Comput Vis Appl 11(1):4 5. Xu C et al (2019) Gait-based age progression/regression: a baseline and performance evaluation by age group classification and cross-age gait identification. Mach Vis Appl 30(4):629–644 6. Liu W et al (2018) Learning efficient spatial-temporal gait features with deep learning for human identification. Neuroinformatics 16(3–4): 457–471 7. Wang J, She M, Nahavandi S, Kouzani A (2010) A review of vision based gait recognition methods for human identification. In: 2010 International conference on digital image computing: techniques and applications (DICTA). IEEE, New York, pp 320–327 8. Sivapalan S, Chen D, Denman S, Sridharan S, Fookes C (2011) Gait energy volumes and frontal gait recognition using depth images. In: 2011 International joint conference on Biometrics (IJCB). IEEE Press, New York, pp 1–6, 181–184 9. Dikovski B, Madjarov G, Gjorgjevikj D (2014) Evaluation of different feature sets for gait recognition using skeletal data from kinect. In: 37th international convention on information and communication technology, electronics and microelectronics (MIPRO). IEEE, New York, pp 1304–1308 10. Jianwattanapaisarn N, Cheewakidakarn A, Khamsemanan N, Nattee C (2014) Human identification using skeletal gait and silhouette data extracted by microsoft kinect. In: 15th International symposium on soft computing and intelligent systems (SCIS), 2014 Joint 7th international conference on and advanced intelligent systems (ISIS). IEEE, New York, pp 410–414 11. Rahman MW, Gavrilova ML (2017) Kinect gait skeletal joint feature based person identification. In: 2017 IEEE 16th International conference on cognitive informatics & cognitive computing (ICCI* CC). IEEE, New York, pp 423–430 12. Preis J, Kessel M, Werner M, Linnhoff-Popien C (2012) Gait recognition with kinect. In: 1st international workshop on kinect in pervasive computing, New Castle, UK, pp P1–P4 13. Ahmed F, Paul PP, Gavrilova ML (2015) Dtw-based kernel and ranklevel fusion for 3d gait recognition using kinect. Visual Comput 31(6–8):915–924 14. Chattopadhyay P, Sural S, Mukherjee J (2014) Exploiting pose information for gait recognition from depth streams. In: Workshop at the European conference on computer vision. Springer, Berlin, pp 341– 355 15. Munsell BC, Temlyakov A, Qu C, Wang S (2012) Person identification using fullbody motion and anthropometric biometrics from kinect videos. In: European conference on computer vision. Springer, Berlin, pp 91–100 16. Ahmed MH, Sabir AT (2017) Human gender classification based on gait features using kinect sensor. In: 3rd IEEE international conference on cybernetics (CYBCONF). IEEE, New York, pp 1–5

74

R. Patua et al.

17. Mart´ın F´elez R, Garc´ıa Jim´enez V, S´ anchez Garreta JS (2013) Gaitbased gender classification considering resampling and feature selection 18. Kastaniotis D, Theodorakopoulos I, Economou G, Fotopoulos S (2013) Gait-based gender recognition using pose information for real time applications. In: 18th International Conference on Digital Signal Processing (DSP). IEEE, New York, pp 1–6 ´ Igual L (2012) Depth information in human gait analysis: 19. Borr´ as, Lapedriza A, an experimental study on gender recognition. In: International conference image analysis and recognition. Springer, Berlin, pp 98– 105 20. Li X, Makihara Y, Xu C, Yagi Y, Ren M (2018) Gait-based human age estimation using age group-dependent manifold learning and regression. Multimedia Tools Appl, pp 1–22 21. Makihara Y, Okumura M, Iwama H, Yagi Y (2011) Gait-based age estimation using a whole-generation gait database. In: International joint conference on biometrics (IJCB). IEEE, New York, pp 1–6 22. Lu J, Tan Y-P (2010) Gait-based human age estimation. IEEE Trans Inf Forens Secur 5(4):761–770 23. Makihara Y, Mannami H, Yagi Y (2010) Gait analysis of gender and age using a large-scale multi-view gait database. In: Asian conference on computer vision. Springer, Berlin, pp 440–451 24. Kastaniotis D, Theodorakopoulos I, Theoharatos C, Economou G, Fotopoulos S (2015) A framework for gait-based recognition using kinect. Pattern Recogn Lett 68:327–335 25. Lu J, Wang G, Moulin P (2014) Human identity and gender recognition from gait sequences with arbitrary walking directions. IEEE Trans Inf Forens Secur 9(1):51– 61

Efficient Watermarking in Color Video Using DWT-DFT and SVD Technique Mangal Patil(B) , Preeti Bamane, and Supriya Hasarmani Department of Electronics, Bharati Vidyapeeth Deemed to be University College of Engineering Pune, Dhankawadi, Pune, Maharashtra 411043, India [email protected], [email protected], [email protected]

Abstract. With the advancement in Internet technology, a digital video can be easily modified, copied, and distributed among a large audience. Copyright protection and security become very essential aspects because of the extensive use of digital multimedia applications. Digital watermarking is being used for copyright protection, data authenticity for multimedia contents such as image, audio, and video. In this paper, a DWT-DFT-SVD-based method is opted to improve robustness and overall computational requirements. The computed PSNR between original video signal and watermarked signal is improved up to 60 db. The normalized correlation value of the original and the extracted watermark image have a high level of imperceptibility. The proposed scheme shows strong robustness against several geometric and non-geometric attacks. Keywords: Discrete Fourier transform (DFT) · Discrete wavelet transform (DWT) · Singular value decomposition (SVD) · Digital video watermarking

1 Introduction The security and unauthorized redistribution of digital contents are getting essential in the digital world. Recent advances in internet technology resulted in an increase in the utilization of the digital video. Hence, the digital data (video) can be shared, copied, distributed, and modified very easily [1, 2]. For the security of digital media in opposition to illegal distributions and manipulations, digital watermarking is being used [3, 4]. The uniqueness of digital media is obtained by extracting the embedded information. Watermarking can be used for many reasons, such as proof of ownership, copy control, broadcast monitoring, and authentication. [5, 6]. Visible watermark and an invisible watermark are two types of watermark. The invisible watermark gives more security to the multimedia like video, image, and document because human eye scan analyzes visible watermark so that attackers can attack without any difficulty on this by different attacks that may be geometric or non-geometric [7]. The watermark embedding, attack, and watermark detection are the essential components of robust watermarking [8]. Goal of watermark is to be robust enough to resist © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_8

76

M. Patil et al.

attacks, but not at the expense of altering the value of the multimedia data being protected. Digital video watermarking is achieved by inserting secret data in a video sequence to protect the video from unauthorized copying. Video string is still undisturbed with evenly time spaced images. Image data hiding methods can be applied for video watermarking as well, but video watermarking schemes require to face many challenges. Watermarked video is much sensible to plagiarize attacks, namely digital–analog conversion, frame interchanging, averaging of the frame, lossy compression, etc. The rest of the paper is organized as follows. In Sect. 2, we explained related work done in digital video watermarking. Our proposed algorithm, with its block diagram, is explained in Sect. 3. The experimental results demonstrated in Sect. 4 followed by conclusion of the proposed work in the last section.

2 Related Work Digital watermarking can be done using discrete cosine transform based on binary watermark technique [9] and QIM technique. Sridha and Arun [10] proposed a discrete wavelet transform (DWT) for security enhancement in video watermarking. Dual SVD and DWT based on selective pixel [11] give high robustness against multiple watermarking attacks. To achieve strong robustness against various signal processing operations, DWT- and PCA-based scheme is proposed [12]. DWT and PCA are used to hide the data in the digital video [13]. A digital watermarking system for video authentication using DMT is discovered by Monika et al. [14]. It presents multi-wavelet-based invisible watermarking. A hybrid DMT-SVD method is determined to be more reliable than the DWT method [15]. In [16], DWT-DCT-SVD is implemented on intravascular ultrasound (IVUS) video. Replacing biomedical signals among hospitals necessitates reliable and efficient communication. In this, binary watermark images embedded into intravascular ultrasound video. The whole video is divided into frames and application of DWT, DCT followed by SVD composes the watermark hiding technique. Different existing video watermarking technologies based on DWT [17–21] and SVD [22–24] and their properties were studied to make it easier to select an appropriate technique which provides quantitative results. To increase the security and robustness of the watermark, some cases used combined or hybrid two frequency domains [25]. Based on DWT and DCT, for example, a watermark was embedded in a selected subband of the Y-component. The required perceptual quality of the video can be achieved by combining DWT with SVD [26]. The study proposed in [27] presented DWT- and DCT-based digital video watermarking using an invisible watermarking algorithm based on the spatial frequency domain. This paper presents a video watermarking technique using DWT-DFT-SVD, and watermark is embedded in YCbCr color space.

3 Proposed Algorithm The robustness of the proposed model is improved by utilizing the features of DWT, DFT, and SVD technique. DFT is robust to Gaussian noise, shift invariance, JPEG compression, image sharpening and helps in noise removal. Inserting watermark using

Efficient Watermarking in Color Video …

77

DWT enhances the robustness against attacks, and it retains the image quality. Any change in singular values does not affect the video quality, and singular values remain unchanged to attacks. The presented method consists of two sections, namely the watermark embedding process and extraction process. 3.1 Watermark Embedding Process The process of embedding watermark in the video is depicted in Fig. 1, and steps are described as follows:

Fig. 1. Watermark embedding process

S1 S2

S3 S4

The input video is divided into frames, and it is converted from RGB to YCbCr color space. The Y-component is used for watermark embedding process. 2D-DWT (Haar wavelet) of single level is applied to the Y luma component. Four sub-bands, namely LL, LH, HL, and HH, are obtained. Out of which LL component is selected from total sub-bands. DFT is applied to LL component. To get an array of U, S, and V matrices of the video frame, SVD is performed on the output of DFT transformed LL component. A = USVT

S5 S6 S7 S8

(1)

The watermark video is converted from RGB to YCbCr color space. Here, again the Y luma component is used. The LL* component is selected after applying DWT on the Y-component. Complex valued Fourier transformed LL* component is achieved by applying DFT to LL* sub-band. To get U*, S*, and V* matrices, SVD is applied on the output component. A = U ∗ S ∗ V∗T

(2)

78

S9

M. Patil et al.

Singular values of watermark embedded with singular values of every frame. The number of singular values to be embedded in each frame is equal to α scaling factor which concludes the power of watermark. SW = S + αS∗

(3)

S10 Inverse SVD is applied on U, Sw, and V matrices to get the LLw component. S11 Untransformed LLw is obtained by applying IDFT using magnitude and phase components. S12 Watermarked Yw luma component is achieved by performing IDWT to LLw, LH, HL, and HH. S13 RGB format conversion is done after combining Cb and Cr components to Yw component. S14 Non-selected frames are combined with the watermarked frames to obtain the complete watermarked video sequence. 3.2 Watermark Extraction Process The process of extraction watermark is depicted in Fig. 2, and steps are described as follows:

Fig. 2. Watermark extraction process

S15 Initially, watermarked video is split into frames and it is converted from RGB YCbCr color space. For the watermark extraction process, Yw component is selected. S16 2D-DWT (Haar wavelet) of single level is used to the Yw component which is nothing but the luma component. Four sub-bands, namely LLw, LH, HL, and HH, are obtained. Out of which LLw component is selected from total sub-bands. S17 DFT is applied to LLw component. S18 To get an array of U, Sw and V matrices of the selected video frame, SVD is performed on the output of DFT transformed LLw part. AW = U SW VT .

(4)

Efficient Watermarking in Color Video …

79

S19 Singular matrix (S*) of watermark can be achieved by the given equation, S∗ = SW −S/α.

(5)

S20 To get the LL* component, ISVD is performed after combining S* matrix with U* and V* matrices. S21 Untransformed LL* component is obtained by applying IDFT using magnitude and phase components. S22 Unwatermarked Y*luma component is achieved by performing IDWT to LL*, LH*, HL* and HH*. S23 RGB format conversion is done after combining Cb* and Cr* components to Y* component to obtain the original watermark.

4 Experimental Results The experimentation of the presented approach is carried out using MATLAB 10. Six video samples were used having various resolutions with distinct format (.avi, .mov, .mp4, .mpg, .wmv). Color watermark with size 384 × 512 in.png format has been selected. Frame selection and embedding have done according to the proposed scheme. Table 1 shows the original and watermarked video frames of different video samples. It describes that the watermarked and original frames are indistinguishable subjectively. Peak-signal-to-noise ratio (PSNR) is calculated to assure the security of the video. PSNR values are dependent on MSE. Table 2 illustrates the extracted watermark images for all video samples without attack and after applying different attacks. From this, it is observed that the maximum correlation is obtained between the original and extracted watermark. Normalized correlation for all video samples is compared and shown in Table 3. NC is used to measure the similarity between the original watermark image and extracted watermark image. The average normalized correlation (NC) of all videos is equal to 0.99, and it gets somewhat degrade for salt and pepper attack as well as for rotation attack. Vid.avi (720 × 1280) and Vid.mpg (480 × 640) give a very low value of NC, which is equal to 0.37 for cropping attack. The robustness of the proposed algorithm is analyzed by applying different attacks on the video. Figure 3 demonstrates the PSNR with and without attacks for six different video samples, where I1—PSNR without attack, I2—Gaussian noise, I3—salt and pepper noise, I4—rotation attack, I5—cropping attack, I6—median filtering attack, I7—histogram equalization, I8—image sharpening. The investigational outcomes have validated the proposed model in terms of improved NC and PSNR results by applying different attacks on videos. Table 4 depicts the comparison analysis of the proposed work with the existing video watermarking techniques. It gives better results than the other methods. In present work, various videos of high-definition resolution are used with color watermark, and several attacks are applied to check the robustness. The given method achieved great PSNR in the range of 63–73 dB with NC value 0.99.

80

M. Patil et al. Table 1. Original and watermarked frames of six different videos

Original video frame

Watermarked video frame

Bus.avi (288 × 352)

Vid.avi (720 × 1280)

Vid.mov (720 × 1280)

Vid.mp4 (720 × 1280)

Vid.mpg (480 × 640)

Vid.wmv (720 × 1280)

5 Conclusion The primary purpose of the work recognized so far is to give accurate and precise video watermarking. The algorithm is implemented DWT in conjunction with DFT transform and SVD, which is vital for achieving better security. A color watermark has

Efficient Watermarking in Color Video …

81

Table 2. Extracted watermark for six video samples Bus.avi

Vid.avi

Vid.mov

Vid.mp4

Vid.mpg

Vid.wmv

(288 × 352) (720 × 1280) (720 × 1280) (720 × 1280) (480 × 640) (720 × 1280)

Original watermark Without attack Gaussian noise

Salt and pepper Cropping attack

Rotation attack Median filter Histogram equalization Image sharpening

been embedded into the original video. Without much loss of data and features of the host video, inserting color watermark in the low-frequency sub-band helps to improve the robustness of embedding procedure. The proposed algorithm is imperceptible and robust against several attacks, and the value of PSNR (more than 60 dB) and NC (0.99) is measured high. There are some ways to be discovered for future work. An alternate watermark can be used, such as audio or video for embedding process to check the robustness of video watermarking. The implementation of this algorithm can be done using VHDL to address the hardware efficiency of this work.

82

M. Patil et al. Table 3. Normalized correlation against attacks for different video frames

Normalized correlation (NC) Bus.avi (288 × 352)

Vid.avi (720 × 1280)

Vid.mov (720 × 1280)

Vid.mp4 (720 × 1280)

Vid.mpg (480 × 640)

Vid.wmv (720 × 1280)

W/O attack

0.9798

0.9985

0.9986

0.9925

0.9991

0.9976

Gaussian noise

0.9665

0.9665

0.9628

0.9417

0.9726

0.9698

Salt and pepper

0.9413

0.8892

0.8420

0.8491

0.8434

0.8338

Cropping (50 × 50)

0.9160

0.3741

0.9979

0.3679

0.3703

0.9967

Rotation (2˚)

0.8850

0.8708

0.9541

0.7706

0.5258

0.8024

Median

0.6496

0.9937

0.9951

0.9877

0.9965

0.9830

Histogram equalization

0.9593

0.9835

0.8117

0.9633

0.7898

0.6726

Image sharpening

0.6481

0.9106

0.9268

0.8999

0.9147

0.8863

Efficient Watermarking in Color Video …

bus.avi

PSNR in dB

64 63 62 61

I1 I2 I3 I4 I5 I6 I7 I8

PSNR in dB

AƩacks

74 73 72 71 70

vid.avi

I1 I2 I3 I4 I5 I6 I7 I8 AƩacks

vid.mov PSNR in dB

70 69 68 67 I1 I2 I3 I4 I5 I6 I7 I8 AƩacks

PSNR in dB

vid.mpg 65.5 65 64.5 64 63.5 I1 I2 I3 I4 I5 I6 I7 I8 AƩacks

Fig. 3. PSNR with and without attacks for different videos

83

M. Patil et al.

PSNR in dB

vid.mp4 74 72 70 I1 I2 I3 I4 I5 I6 I7 I8 AƩcks

vid.wmv PSNR in dB

84

71 70 69 68 I1 I2 I3 I4 I5 I6 I7 I8 AƩacks

Fig. 3. (continued)

Efficient Watermarking in Color Video …

85

Table 4. Comparative analysis Method

Host video size

Watermark

Attacks

PSNR(dB)

DCT [9]

256 × 256

Binary wm 32 × 32

Gaussian, salt and pepper noise, median filter, histogram equalization

45.98

Image

Gaussian noise adding, salt and pepper noise adding, frame dropping

45

5-level DWT [11] 256 × 256

DWT, PCA–QIM [14]

256 × 256

Image 32 × 32

Gaussian noise 45.41 addition, hist. equalization, gamma correction, contrast adjustment

DWT and PCA [15]

Not given

Binary image

Gaussian and salt pepper noise, cropping, rotation, median filtering, contrast adj., hist. equalization.

46.23

MPEG [18]

320 × 240

Not given

Frame dropping and averaging, cropping

31.5–38.5

DWT, DFT, and SVD [Proposed]

(.avi, .mpg, .mp4, .wmv) 720 × 1280

Color watermark 384 × 512

Salt and pepper, rotation, cropping, image sharpening, hist. equalization, filtering

63–73

86

M. Patil et al.

References 1. Khorasani MK, Sheikholeslami MM (2012) An DWT-SVD based digital image watermarking using a novel wavelet analysis function. In: IEEE 4th international conference on computational intelligence and communication systems and networks, pp 254–256, Phuket, Thailand 2. Mane GV, Chiddarwar GG (2013) Review paper on video watermarking techniques. Int J Sci Res Publ 3:1–5 3. Prabakaran G, Bhavani R, Ramesh M (2013) A robust QR-Code video watermarking scheme based on SVD and DWT composite domain. In: IEEE international conference on pattern recognition, informatics and mobile engineering, Tamil Nadu, Salem, pp 251–257 4. Shelke NA, Chatur PN (2013) A survey on various digital video watermarking schemes. Int J Comput Sci Eng Technol 4(12), 1447–1454 5. Husain F (2012) A survey of digital watermarking techniques for multimedia data. MIT Int J Electr Commun Eng 2(1):37–43 6. Singh P, Chadha RS (2013) A survey of digital watermarking techniques, applications and attacks. Int J Eng Innov Technol 2(9):165–175 7. Jayamalar T, Radha V (2010) Survey on digital video watermarking techniques and attacks on watermarks. Int J Eng Sci Technol 2(12):6963–6967 8. Habiba Sk, Niranjanbabu D (2014) Advance digital video watermarking based on DWT-PCA for copyright protection. Int J Eng Res Appl 4(10):73–78 9. Jeswani J, Sarode T (2014) A new DCT based color video watermarking using luminance component. IOSR J Comput Eng 16(2):83–90 10. Sridha B, Arun C (2014) Security enhancement in video watermarking using wavelet transform. J Theor Appl Inf Technol 62(3):733–739 11. Ponni Sathya S, Ramakrishna S, Arjun S, Magendran V (2013) Selective pixel based efficient video watermarking using dual singular value decomposition in the discrete wavelet transform domain. Res J Comput Syst Eng 04:720–725 12. Yassin NI, Salem NM, EI Adawy MI (2014) QIM blind video watermarking scheme based on wavelet transform and PCA. Alex Eng J 13. Jadhav Shubhashri N, Ghodke VN (2014) Data hiding in digital video by watermarking. Int J Eng Trends Technol 13(5):204–208 14. Monika S, Lavanya A, Suganya S (2014) A digital watermarking system for video authentication using DMT. IJESC 4:466–470 15. Sharma AK (2011) Simulation and analysis of digital video watermarking using MPEG-2. Int J Comput Sci Eng 3(7):2700–2706 16. Dey N, Das P, Roy AB, Das A (2012) DWT-DCT-SVD based intravascular ultrasound video watermarking. In: IEEE, information and communication technology, India, pp 224–229 17. Aparna JR, Ayyappan S (2014) Comparison of digital watermarking techniques. In: IEEE, international conference on computation of power, energy, information and communication (ICCPEIC), pp 87–92, India 18. Divecha N, Jani N (2013) Implementation and performance analysis of DCT-DWT-SVD based watermarking algorithms for color images. In: IEEE International conference on intelligent systems & signal processing (ISSP), India 19. Jha C, Mishra A (2014) Digital video watermarking using cascaded stages of discrete wavelet transforms. Int J Eng Innov Technol 4(3):87–93 20. Bedi S, Ahuja R, Agarwal H (2013) Copyright protection using video watermarking based on wavelet transformation in multiband. Int J Comp App 66(8):1–5 21. Thanki R, Borisagar K (2015) Compressive sensing based multiple watermarking techniques for biometric template protection. Int J Image Gr Signal Proces 7(1):53–60

Efficient Watermarking in Color Video …

87

22. Naved A, Rajesh Y (2013) Dual band watermarking using 2-d DWT and 2-level SVD for robust watermarking in video. Int J Sci Res (IJSR) 2(9):249–252 23. Ahuja R, Bedi SS (2013) All aspects of digital video watermarking under an umbrella. Int J Image Gr Signal Proces 7(3):54–73 24. Rao YR, Nagabhooshanam E, Prathapani N (2014) Robust video watermarking algorithms based on SVD transform. In: IEEE international conference on information communication and embedded systems (ICICES), pp 1–5, India 25. Panyavaraporn J (2017) DWT/DCT watermarking techniques with chaotic map for video authentication. In: 9th international conference on digital image processing 26. Ponnisathya S et al (2017) Chaotic map based video watermarking using DWT & SVD. In: International conference on inventive communication and computational technologies, pp 45–49 27. Panyavaraporn J, Horkaew P (2018) DCT-based invisible digital watermarking scheme for video stream. In: IEEE 10th international conference on knowledge and smart Technology, Thailand, pp 154–157

Motif Discovery and Anomaly Detection in an ECG Using Matrix Profile Rutuja Wankhedkar(B) and Sanjay Kumar Jain Computer Engineering Department, National Institute of Technology, Kurukshetra, Kurukshetra 136119, India [email protected], [email protected]

Abstract. Time Series Data mining is a popular field in data science to discover and extract useful information from the time series data. Time Series Motif discovery is one of the tasks in data mining to discover frequent patterns which are unknown previously. Motif discovery has gained a lot of attention since its advent in 2002. Many motif discovery techniques were introduced and applied in various domains like E-commerce, Weather Prediction, Seismology, etc. In this paper, we introduce a technique for anomaly detection and motif discovery in the ECG data using Matrix Profile which has been introduced recently in the literature. Anomaly detection in ECG helps to detect the abnormal heartbeats before the process of diagnosis and motif discovery helps to locate the highly similar beats in the ECG. Using Matrix Profile for the task of anomaly detection and motif discovery in our proposed technique provides our technique with properties that are inherited from Matrix Profile. Thus, the proposed technique in this paper has properties like exactness, simple and parameter-free, space-efficient, anytime, handle missing data, free from the curse of dimensionality. Keywords: Data mining · Pattern discovery · Time series data · Motif · ECG

1 Introduction Data mining is a field in computer science where we discover and extract information that can provide us with useful information. The information extracted can be used in different domains such as Ecommerce, Weather Prediction, Stock value prediction, Prognostics, etc. Data mining includes various tasks such as anomaly detection, classification, clustering, rule discovery, summarization, etc. Here, in this paper, we will focus on a task in time series data mining called time series motif discovery. The term time series motif discovery was coined in 2002 and before that motifs were used in the genetics domain to extract the DNA motifs present in the human DNA. Conservation of patterns can be seen in a time series and extracting these patterns can provide information. These patterns are nothing but subsequences in a time series that occur frequently and are called time series motifs. ECG is also a time series which consists of the electrical impulses from the heart. Thus, in this paper, we focus on time series motif discovery technique to © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_9

Motif Discovery and Anomaly Detection

89

discover motifs and anomalies in ECG dataset to detect abnormality before the process of diagnosis using Matrix Profile. Organization of rest of the paper is as follows. Section 2 discusses the various time series motif discovery algorithms in literature. Section 3 provides the main contribution of this paper where the technique for discovering motifs and anomalies in ECG dataset is discussed. Section 4 provides the implementation details. The conclusion for this paper is provided in Sect. 5.

2 Related Work Time series data are generated by several applications in different domains at an unprecedented speed. These data contain useful information that can be extracted using data mining techniques. Several representations and similarity measures for time series data have been proposed in literature [1–4]. Time series motifs are sub-sequences occurring repeatedly in time series data and are not known in advance. Motifs are useful patterns and can be used in retrieving information using data mining tasks like clustering, classification, association rules, and anomaly detection. There are several applications where motifs are discovered and used for various analysis, estimations and predictions. These applications include areas medicine, motion-capture, seismology, sensor networks, telecommunication, etc. For example, brain activities are recorded using a device called Electroencephalogram (EEG). The signals generated after monitoring the brain activities indicate whether the brain activities monitored are normal or the patient is required to be tested for any medical precondition. The signals from the brain contain motifs that help in diagnosis. Time series motifs can be useful in determining the state of the system. Many techniques are proposed for extraction of time series motifs. Thus, in this section, a brief description of the existing techniques to discover time series motifs is provided. 2.1 Time Series Motifs Discovery Algorithm Time series motifs extraction methods can be categorized as approximate motif discovery and exact motif discovery. This motif discovery methods use time series data and the data is one-dimensional or multi-dimensional. The motif discovery algorithm since their introduction in 2002 has been improved to reduce the time complexity and increase efficiency in large data sets. Patel et al. [5] propose  the first efficient time series motifs discovery algorithm having time complexity O n4 , here n denotes the length of time series, in massive databases. Also, a significant definition of time series motifs was introduced in this work. In [5] first to locate 1-motif, a brute force algorithm with time  complexity O n2 m was introduced and was further optimized to an efficient algorithm Enumeration of Motifs through Matrix Approximation (EMMA). EMMA algorithm to efficiently distinguish the true matches from the false hits used Approximation Distance Map (ADM) algorithm introduced by Shasha and Wang. The main contribution of [5] is in the method used for the construction of the small matrix. The EMMA algorithm [5] requires a Euclidean distance as a similarity measure and a symbolic representation of time series data. Another method Efficient Motif Enumeration (MOEN) [6] to speed

90

R. Wankhedkar and S. K. Jain

up the performance was proposed. MOEN [6] is the first exact algorithm which is 23 times faster when compared to the naïve method. The primary goal in [6] is minimizing similarity while finding as many as possible motifs for different lengths. Probabilistic algorithm to discover time series motifs producing approximate results was proposed by Chiu et al. [7] with quadratic time complexity. The probabilistic algorithm is based on the random projection algorithm and is capable of finding motifs from noisy time series. The algorithm by Chiu et al. [7] is an anytime algorithm and can almost immediately produce the likely candidate motifs while improving the results gradually over time. Time series data in [7] is converted to symbolic representation by SAX method and the SAX word length affects the time complexity of algorithm in [7]. Yankov et al. [8] proposes time series motifs discovery by applying uniform scaling. The algorithm in [8] detects motifs having different size but has a major disadvantage of providing the length of motif in advance. Also, for finding the proper scaling factor, several examinations need to be conducted. Mueen et al. [9] introduces a tractable algorithm Mueen-Keogh (MK) to discover exact time series motifs. The MK algorithm prunes off unnecessary search spaces by use of early abandoning. The algorithm is efficient than the previous brute force algorithm and have a time complexity O(NR), here N represents the number of time series and R represents number of reference points selected by user. The MK algorithm provides scalable exact solution by finding the non-overlapping sub sequences pair with minimal ED  to each other. The time complexity of DAME [10] algorithm by Mueen et al. is O n2 . Clustering approach is employed by Li et al. [11] to propose an anytime algorithm with O(knrD) time complexity where k and r denote the number of cluster and number of iterations, respectively and D represents number of time series. Serra et al. proposed a similar anytime algorithm which is based on PSO (Particle Swarm Optimization). Mueen et al. [12] propose the first online algorithm for time series motif discovery. The algorithm in [12] maintains and monitors time series motifs in real-time using time series data’s most recent history. The worst-case time complexity for update operation is O(w) where w represents recent window size and the extension is capable of handling complicated motif structures. Z-normalized ED is applied to compute the distance between  3  sub-sequences in [12]. Amortized space complexity of the algorithm in [12] is O w2 . Yingchareonthawornchai et al. [13] proposes a novel compression based algorithm to find proper length for motifs. The compression-based algorithm is parameter-free and is an anytime algorithm and the output contains ranked classes of motif. The Minimum Description Length (MDL) principle is adopted in [13]. The technique of early termination is used to asymptotically decreasing the  complexity in [13]. The time  time complexity of compression-based algorithm is O m2 n2 , here m represents the size of motifs. The algorithm in [13] even if interrupted or has not completed, will return a valid solution. Tanaka [14] proposes a method with quadratic time complexity, discovers motifs in multidimensional data. PCA (Principal component analysis) is applied in [11] to convert the signal to one dimension from multidimensional data. A method to accelerate the execution in [14] is provided by Anh et al. [15]. Balasubramanian et al. [16] employ

Motif Discovery and Anomaly Detection

91

the clustering-based anytime algorithm [11] to discover motifs in every dimension. To discover motifs, an amplitude multi-resolution method with linear time complexity is proposed by Castro et al. [22]. Yuhong et al. [17] propose Quick-Motif, a framework that involves a two-level approach to facilitate exact motif discovery. It is not only efficient but scalable as well also the outer level of framework enables batch pruning whereas efficient correlation calculation is enabled at the inner level. Quick-Motif jointly applies both smart brute force method (SBF) and MK techniques to discover motifs. Zhu et al. [18] shows the exact motif discovery’s scalability can be increased by combining STOMP algorithm with high-performance GPU’s. Scalable Time series Orderedsearch Matrix Profile (STOMP) algorithm is efficient version of STAMP [19] algorithm. STAMP [19] algorithm with an efficient anytime algorithm compute the sub-sequences joins of time series. In [18], one hundred million objects dataset is used while discovering motifs. STOMP is parameter-free and false negatives are not allowed in this algorithm. STOMP computes more useful information than Quick-Motif and MK [20] algorithms. Speedup in STOMP does not depend on the structure of data and length of the subsequence. Space complexity of STOMP and STAMP is O(n). STOMP algorithm is empirically and theoretically faster than STAMP [19], Quick-Motif [17] and MK [20]. Chin-Chia et al. [18] proposes mSTAMP-based framework to discover motifs in multidimensional data. The framework [18] is built on top of Matrix Profile. The framework is scalable to large datasets and has the anytime property. The approach in [18] have many application domains and works with the real world non-modified data. mSTAMP algorithm is built on top of STOMP or STAMP algorithm and easy to parallelize. The runtime complexity of mSTAMP algorithm depends only on n. mSTAMP algorithm is faster, requires less parameters and supports streaming data.

3 Motif Discovery and Anomaly Detection in an ECG Using Matrix Profile Electrocardiogram (ECG) is nothing but a time series which records the heartbeats and helps in making diagnosis for any cardiac disease. ECG data contains PQRST wave which forms a single heartbeat and also contains artifacts that might or might not be similar to PQRST and this may lead to misdiagnoses and hence the patient is at risk. Many researches done [21] are unable to detect artifacts as there is no predefined pattern for an artifact. Thus anomaly detection in an ECG might provide false results and also detect an artifact as an anomaly which is a false result. Thus, we propose the following method. 3.1 Proposed Method The aim of our method is to detect anomaly present in an ECG data even in the presence of artifacts. Here, in our proposed method we consider a single lead signal which has an inevitable possibility to contain artifacts. To detect the anomaly in ECG without any false positives we use motif discovery method which here is used to detect the highest number of ECG beats in the signal. Thus, for efficient motif discovery, we use Matrix

92

R. Wankhedkar and S. K. Jain

Profile [18, 19]. Construction of the Matrix Profile, for the motif discovery in the time series data generated by the ECG, also helps to find the anomaly present in an ECG, simultaneously. The beats similar to each other and having low values in Matrix profile are the motifs whereas the beats having high value and dissimilar are the anomalies. Thus, anomaly detection using our proposed method will be accurate and efficient. Our proposed method is simple and can be easily implemented. Procedure for anomaly detection in the proposed method is explained in the following section. 3.2 Anomaly Detection Motif Discovery algorithm is used to detect the anomaly in an ECG, and motifs to avoid any false negatives. Though many techniques are present for time series motif discovery, here we use Matrix Profile [18, 19]. Matrix Profile is a data structure which consists of the minimum distances between a subsequence and nearest neighbour to the same subsequence in the time series data (here, ECG). To compute the similarity between the subsequence and its neighbours, the most popular distance metric, Euclidean Distance [9] is used. Though Euclidean distance is a lock-step similarity measure and not an elastic measure like DTW, its performance is noted to be good when the size of the data increases. To compute the Matrix Profile efficiently, Scalable Time-series Ordered Matrix Profile (STOMP) algorithm is used. STOMP is an exact motif discovery algorithm, i.e. no false positives are allowed and it also has an anytime property. The STOMP algorithm is provided in Table 1. Using the STOMP algorithm a Matrix Profile is generated which helps to discover the anomalies and motifs, here in ECG dataset. The beats similar to each other and having low values in Matrix profile are the motifs whereas the beats having high value and dissimilar are the anomalies. Table 1. STOMP algorithm Procedure STOMP(T,m) Input : A user provided time series T and a subsequence length m Output : A Matrix Profile P and matrix profile index I 1. n ← Length(T) 2. μ, σ ← ComputeMeanStd(T, m) 3. QT ← SlidingDotProduct(T[1:m], T), QT_first ← QT 4. D ← CalculateDistanceProfile(QT, μ, σ) 5. P ← D, I ← ones 6. for i = 2 to n-m+1 7. for j = n-m+1 downto 2 8. QT[j] ← QT[j-1] – T[j-1] × T[i-1] + T[j+m-1] × T[i+m-1] 9. end for 10. QT[1] ← QT_first[i] 11. D ← CalculateDistanceProfile(QT, μ, σ, i) 12. P, I ← ElementWiseMin(P, I, D, i) 13. end for 14. return P, I

Motif Discovery and Anomaly Detection

93

4 Implementation For implementation, Python 3 and its libraries numpy, pandas, matplotlib, seaborn are used. Another library called matrixprofile is used for motif discovery and record the anomaly. Real ECG datasets are obtained from PhysioNet, a website that provides various real medical datasets. One dataset has been used for experimentation. The following are the details about the datasets used (Table 2). Table 2. Dataset used for experimentation S. No.

Data set

No. of data

1

MITDB (MIT-BIH Arrhythmia Database record 108)

21,601

Experimentation is carried out on the above-mentioned dataset by using different motif length for evaluation. The ECG anomalies are found in the dataset used and top five anomalies have been recorded. Lower values in the Matrix Profile for ECG data represent the motifs whereas the higher values represent anomaly present in ECG data. Following obtained is the best result obtained after evaluation done using different lengths. In Fig. 1, blue represents the original time series and red represents its corresponding matrix profile.

Fig. 1. Snapshot of anomaly detection and motif discovery in ECG (motif length = 554)

5 Conclusion In this paper, we introduced a technique for anomaly detection in the ECG data and motif discovery to locate the frequent heartbeats in them using Matrix Profile. Due to Matrix

94

R. Wankhedkar and S. K. Jain

Profile the anomaly beats can be located easily as they are the beats that are unique and have higher values in the profile. Similarly, motifs can be easily located as they have lower values and have similar patterns. The anomaly detection in our technique is accurate and efficient. The technique introduced in this paper has many properties that were adopted from the Matrix Profile. Thus, it is found that our technique is independent of the length of the subsequence. It is exact which means that no false positives are detected in the ECG i.e. during anomaly detection artifacts are not concluded as anomaly beats. Our technique supports the anytime property which means that even if the process is interrupted due to some reason, solution is provided and the process can continue later from that stage. It is simple and can be implemented in Python with just few lines of code. It is space-efficient and free from the curse of dimensionality, this is because the time complexity in our proposed technique is constant to the length of the subsequence. It is also capable to handle the missing data. Hence, anomaly detection using the proposed technique in the ECG data will help to locate the abnormal ECGs before the process of diagnosis in an efficient way.

References 1. Keogh EJ, Chakrabarti K, Pazzani MJ, Mehrotra S (2001) Dimensionality reduction for fast similarity search in large time series databases. Knowl Inf Syst 3(3) 2. Keogh EJ, Ratanamahatana CA (2005) Exact indexing of dynamic time warping. Knowledge Inf Syst 7(3) 3. Lin J, Keogh EJ, Wei L, Lonardi S (2007) Experiencing SAX: a novel symbolic representation of time series. Data Min Knowl Discov 15(2) 4. Morse MD, Patel JM (2007) An efficient and accurate method for evaluating time series. In: SIGMOD conference 5. Patel P, Keogh E, Lin J, Lonardi S (2002) Mining motifs in massive time series databases. In: Proceedings ICDM. IEEE, Los Alamitos, pp 370–377 6. Mueen A (2013) Enumeration of time series motifs of all lengths. ICDM. IEEE, pp 547–556 7. Chiu B, Keogh E, Lonardi S (2003) Probabilistic discovery of time series motifs. In: Proceedings of the Ninth ACM SIGKDD international conference on knowledge discovery and data mining. ACM, pp 493–498 8. Yankov D, Keogh E, Medina J, Chiu BL, Zordan V (2007) Detecting time series motifs under uniform scaling. In: International conference on knowledge discovery and data mining. ACM, pp 844–853 9. Lin J, Vlachos M, Keogh E, Gunopulos D (2004) Iterative incremental clustering of time series. In: Bertino E (ed) Advances in database technology—EDBT. Springer, Berlin, pp 106–122 10. Mueen A, Keogh E, Bigdely-Shamlo N (2009) Finding time series motifs in disk-resident data. In: 9th ICDM. IEEE, pp 367–376 11. Li Y, Lin J, Oates T (2012) Visualizing variable-length time series motifs. In: SDM. SIAM, pp 895–906 12. Mueen A, Keogh E (2010) Online discovery and maintenance of time series motifs. In: Proceedings of the 16th ACM SIGKDD international conference on knowledge discovery and data mining, KDD’10. ACM, pp 1089–1098 13. Yingchareonthawornchai S, Sivaraks H, Rakthanmanon T, Ratanamahatana CA (2013) Efficient proper length time series motif discovery. In: 13th ICDM. IEEE, pp 1265–1270

Motif Discovery and Anomaly Detection

95

14. Tanaka Y, Iwamoto K, Uehara K (2005) Discovery of time series motif from multi-dimensional data based on MDL principle. Mach Learn 58:269–300 15. Anh DT, van Nhat N (2016) An efficient implementation of emd algorithm for motif discovery in time series data. Int J Data Min Model Manage 8:180 16. Balasubramanian A, Wang J, Prabhakaran B (2016) Discovering multidimensional motifs in physiological signals for personalized healthcare. IEEE J Sel Topics Signal Process, p 1 17. Yuhong L, Leong H, Yiu, ML, Gong Z (2015) Quick-motif: an efficient and scalable framework for exact motif discovery. In: ICDE conference. IEEE, pp 579–590 18. Zhu Y, Zachary Z, Nadar SS, Chin-Chia M et al (2016) Matrix profile II: exploiting a novel algorithm and GPUs to break the one hundred million barrier for time series motifs and joins. 16th ICDM. IEEE, pp 739–748 19. Michael Yeh C-C, Yan Z, Liudmila U, Nurjahan B et al (2016) Matrix profile I: all pairs similarity joins for time series: a unifying view that includes motifs, discords and Shapelets. 16th ICDM. IEEE, pp 1317–1322 20. Mueen A, Keogh E, Zhu Q, Cash S, Westover MB (2009) Exact discovery of time series motifs. ICDM. SIAM, pp 473–484 21. Sivaraks H, Ratanamahatana CA (2015) Robust and accurate anomaly detection in ECG artifacts using time series motif discovery. Comput Math Methods Med J 22. Castro N, Azevedo PJ (2010) Multiresolution motif discovery in time series. ICDM. SIAM, pp 665–676

ISS: Intelligent Security System Using Facial Recognition Rajesh Kumar Verma1(B) , Praveen Singh2 , Chhabi Rani Panigrahi3 , and Bibudhendu Pati3 1

3

Infosys Limited, Hyderabad, India Rajeshverma [email protected] 2 HSBC Limited, Bengaluru, India [email protected] Department of Computer Science, Rama Devi Women’s University, Bhubaneswar, India {panigrahichhabi,patibibudhendu}@gmail.com

Abstract. Security is mandatory and of utmost importance for all organizations. While the legitimate people should be allowed inside enterprises, the illegal ones should be barred from entering and this can be achieved using face recognition techniques. In this work, we have come up with a robust architecture using artificial intelligence and Internet of things that can be used across different enterprises. We have also derived the methodology and solution for implementing a more advanced security system. For proper demonstration, we have also considered one of the business use cases along with proposed processing work flow. Keywords: Security · Face recognition · Artificial intelligence Internet of things (IoT) · Big data · Mobile devices

1

·

Introduction

Artificial Intelligence (AI), Machine Learning (ML) and Deep Learning (DL) technologies are broadly applied while developing any intelligent security system. In the past around 20 years back, people used to develop rule-based security system which was very much complex and there was hardly any intelligence. For every scenario, you need to define scenario and corresponding action. In this work, we have proposed the architecture for modern intelligent security system, which comprises of Internet of things (IoT), cloud technologies and AI/ML/DL along with big data. The rest of the paper is organized as follows. In Sect. 2, we will discuss about the related work, Sect. 3 discussed proposed architecture framework and steps taken to reach the solution discussed, business use case and work flow in Sect. 4 followed by conclusion in Sect. 5. c Springer Nature Singapore Pte Ltd. 2021  C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_10

ISS: Intelligent Security System Using Facial Recognition

2

97

Related Work

In this area of security using AI, a plethora of work has been done in designing the security system. In [1], the authors have proposed a car theft prevention system using face recognition in IoT-based home automation. They have used controller and RFID transmitter and receiver to secure the car in home along with controlling their home appliances through communication protocol. Face recognition system is being used to check the entered person in home and the person trying to drive a car is authorized or unauthorized. Jose et al. [2] have proposed the device fingerprinting algorithm in order to improve the home automation security, and the paper highlights various security issues that are associated with modern smart homes. In the subsequent paper Jose et al. [3], have further improvised the smart home security system by introducing the behaviour prediction algorithm, which uses the required parameters to predict the behaviour of users. The proposed algorithm learns through several weeks of training data and is based on the knowledge it gains from the na¨ıve Bayesian network. In [4], author proposed a IoT-based SmartBots using MCC and big data technology, and the paper describes the architecture which is used as a base framework for building smart and advanced city using IoT devices. It provides the intelligent solution to the users in a real-time manner by use advanced tools available in the analytical market. In Singh et al. [5], have proposed an architecture for running of resource intensive jobs on the cloud, in which off-loading to the cloud is the major feature used for faster computation. The integration of IoT devices and big data is highlighted in this paper wherein big data is used as the storage medium for the humongous data generated by the IoT devices over a period of time.

3

Proposed Architecture Framework

In this section, we present the proposed architecture framework for ISS using facial recognition, in order to discover the genuine people who work for the particular enterprise. The system then allows the person to enter into the premises of the enterprise. In Fig. 1, the high-level architecture diagram of the proposed ISS has been depicted. The input to the system comes from either the IoT devices, like camera, or any user device, like mobile phones, and subsequently sent across to the cloud. The input images are further sent to the AI/ML/DL layer where the algorithms are hosted and help in identifying the similarity between the captured image and the original image stored in the system. The big data layer is used to store huge volumes of data and require parallel and distributed computing for further processing. The security team can monitor the entire process through the data which is being monitored.

98

R. K. Verma et al.

Fig. 1. Proposed intelligent security system architecture

3.1

SIAMESE Algorithm for Face Similarity Detection

SIAMESE neural network (SNN) [6] is an artificial neural network(ANN) which uses the measure of similarity for recognizing similar faces, handwritten checks, etc. The most popular application of SNN is face recognition [8], wherein the images of people are pre-computed and subsequently compared to the image which comes from the IoT device at the security entrance of a particular company. SNN helps to recognize the legitimate person amongst a large number of persons. Popular example of SNN application is DeepFace [7] which is a deep learning facial recognition system created by Facebook and identifies the human faces in digital images. This system uses a neural network(NN) having 9 layers and 120 million weights and was trained on 4 million Facebook images. We have used SNN in our work for facial recognition and similarity comparison. 3.2

Solution Flow

Figure 2 shows the various modules in the architecture. Here, the related historical data is stored in the system and the new image is captured by the devices. Both the images are sent across the different layers and subsequently the similarity between them is detected.

ISS: Intelligent Security System Using Facial Recognition

99

Error correction system facilitates enterprises to ensure the level of security and advancement of the same by applying IoT, AI/ML/DL, facial recognition and big data.

Fig. 2. Steps of solution

The different steps followed in execution of the ISS used in making ISS are explained below: 1. The current image of the person at security gate is captured using the cameras. Historical data is also used which was captured at an earlier date. Both the current and earlier data are passed to the next stage for further processing. 2. The data captured from the camera device as well as the historical data is now made available to the next module for the subsequent processing. 3. The images are now further processed (normalized) and the CNN-based algorithm is now applied to these images, which helps to detect the similarity between the edges of the faces. 4. The important features are extracted and help to recognize the person and the percentage of similarity is given by the system. 5. Identify the positive and negative signals. (a) find edges, coordinates of existing knowledge base to real-time object on the surveillance. (b) otherwise, find the similar pattern and their matching from the central repository present in the cloud.

100

R. K. Verma et al.

6. fetch the coordinates and tag to each new object and to avoid the future load. In this paper, we focus on the feature engineering based on historical data and IoT devices data. It can be categorized as a classification problem which helps to detect the similarity between faces.

4

Business Use Case and Work Flow

Figure 3 depicts the work flow for ISS starts with the particular person who arrives at the front security gate for entry into the premises of the company. Subsequently the IoT devices will take the photograph of the person for subsequent processing to detect the similarity with the reference image stored in the enterprise system. The intelligent module then analyses the captured image in reference to the image which is originally present in the repository. Upon successful identification of the captured image, the person is permitted entry into the premises of the enterprise. In case the person is detected as a fraud case, the entry is prohibited. At regular interval(can be daily, weekly or monthly), the monitoring report is generated and tracked by the security department.

Fig. 3. Steps of solution for the business use case

5

Conclusion

Security systems using face recognition are being used at all places and are the emerging trend of the day. The uses of intelligent algorithms and big data enhance the approach of the applications for all types of organizations. The

ISS: Intelligent Security System Using Facial Recognition

101

architecture discussed in this paper is classified under the face recognition category, where the functioning of the system is mainly dependent on the image repository, applied intelligent algorithm, loading and off-loading to the cloud for processing. This architecture is robust as it can be modified and optimized based on the needs. We also provide the end-to-end process along with work flow of any security-based application.

References 1. Rajalakshmi S, Tharani R, Newlin Rajkumar M, Harshani PR (2017) IOT based smart home automation for car theft prevention using image processing. Int J Trend Sci Res Dev, pp 1307–1311 2. Jose AC, Malekian R, Ye N (2016) Improving home automation security; integrating device fingerprinting into smart home. IEEE Access 4:5776–5787 3. Jose AC, Malekian R, Letswamotse BB (2018) Improving smart home security; integrating behaviour prediction into smart home. IJSNet 28:253–269 4. Singh PK, Verma RK, Krishna Prasad PESN (2019) IoT-based smartbots for smart city using MCC and big data, smart intelligent computing and applications. In: Smart innovation, systems and technologies, vol 104. Springer, Singapore, pp 525– 534 5. Singh PK, Verma RK, Sarkar JL (2019) MCC and big data integration for various technological frameworks., progress in advanced computing and intelligent engineering. In: Advances in intelligent systems and computing, vol 714. Springer, Singapore, pp 405–414 6. Koch G, Zemel R, Salakhutdinov R (2015) Siamese neural networks for one-shot image recognition. In: Proceedings of the 32nd international conference on machine learning, Lille, France 7. https://en.wikipedia.org/wiki/DeepFace 8. https://medium.com/swlh/advance-ai-face-recognition-using-siamese-networks219ee1a85cd5

G.V Black Classification of Dental Caries Using CNN Prerna Singh1(B) and Priti Sehgal2 1 Department of Computer Science, University of Delhi, New Delhi, Delhi, India

[email protected] 2 Department of Computer Science, Keshav Mahavidyalya University of Delhi, New Delhi,

Delhi, India

Abstract. Dental caries is one of the most predominant pathologies in the world. Early detection of dental caries leads to the prevention of tooth decay. According to G.V Black classification model dental caries can be classified into six classes (Class I–Class VI) based on the location of caries. The proposed work classifies caries infected tooth based on G.V black classification using deep convolution network architecture (DCNN). In the proposed approach, feature extraction of the preprocessed images is done using Local Ternary Pattern (LTP) and feature reduction is done using Principal Component Analysis (PCA). The pretrained models used in the study are AlexNet architecture and GoogleNet architecture. AlexNet is 8 layer DCNN and classifies the tooth with an accuracy of 93%, 90% sensitivity and 92% specificity. GoogleNet is a 22 layer DCNN and classifies with an accuracy of 94%, 91% sensitivity and 93% specificity. Keywords: Deep convolutional network · AlexNet · GoogleNet

1 Introduction Dental caries is one of the most widespread tooth problems of today. Dental caries is formed by the acid that is produced by the bacteria, that breaks down the food. Most of the dental caries is formed on the occlusal surface and interproximal surface. The demineralization of the enamel and formation of the tiny holes is the first stage of dental caries. Dental caries can be classified into six classes based on G.V Black Classification. Figure 1 shows the various classes of caries based on G.V Black classification.

Fig. 1. Various classes of G.V Black classification © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_11

G.V Black Classification of Dental Caries Using CNN

103

The Class I caries affects the occlusal surface, buccal surface and lingual surface of molar and premolar teeth. The class II caries affects the near or the far surface of molar and premolar. The class III caries affects the proximal surface of incisor and canine teeth. The class IV caries affects the angle of canine and incisor teeth. The class V caries affects the one-third of anterior and posterior teeth. The class VI caries affects the tip of molar and premolar teeth. In this paper, the G.V Black classification of caries infected dental images is done using deep convolution network (DCNN). One of the most important uses of CNN is to find patterns in the images to recognize objects. CNN is one of the best algorithms for deep learning. The pretrained models of AlexNet architecture and GoogleNet Architecture have been used in the proposed work. The pretrained AlexNet architecture consists of five convolutional layers, two fully connected layers and one softmax layer. The images were classified into 6 G.V black classes. The pretrained GoogleNet architecture consists of 22 layers with various inception modules. The performance of the proposed approaches has been shown in the result section. The paper is structured as follows:- Sect. 2 contains the related work. Section 3 defines the proposed work using pretrained models AlexNet and GoogleNet. Section 4 defines the result of the proposed work. Section 5 covers the conclusion.

2 Related Work Dental caries is portrayed as multifactorial sickness that results in demineralization of the tooth. Datta et al. [1] developed an optical image technique to detect dental caries. Here the optical images are filtered, the tooth region is segmented. The model was also successful in monitoring the growth of the lesion with respect to its size. These techniques used visible light thus eliminating the risk of the patient being exposed to harmful radiation. It was observed from the results that the accuracy of the system was 93% but it failed in detecting the conditions where a tooth is broken. Also, it was unsuccessful in detecting the depth of caries. Naebi et al. [2] developed an image processing approach along with particle swarm optimization (PSO) algorithm for detecting dental caries. Prerna et al. [3] developed an automatic caries detection model based on Discrete Cosine Transformation (DCT) and Radon Transformation (RT). The extracted features in the proposed approach were applied to different classifiers such as k-Nearest Neighbor (k-NN), Decision Tree (DT), Random Forest, Radial Basis Function (RBF), AdaBoost classifiers, Naive Bayes and neural network classifiers. From the results, it was observed that the accuracy of all the classifiers was in the range of 80 to 86% and random forest classifiers gave the highest accuracy of 86%. The main shortcoming of this technique was that it was designed to classify only non-cavitated and cavitated teeth. Prajapati et al. [4] used the small dataset of 251 Radiovisiography (RVG) x-ray images for dental image classification into 3 different classes i.e.dental caries disease, periapical infection and periodontitis classes using pretrained convolutional neural network VGG 16. The overall accuracy of 88% was achieved. Miki et al. [5] investigated the use of Deep CNN for the classification of the different types of tooth on dental cone-beam CT (computed tomography) images. The CT slices were used for extracting the ROI inclusive of the single tooth. The accuracy achieved was 88%. The accuracy of classifying the augmented

104

P. Singh and P. Sehgal

data for training showed an improvement of about 5%. This technique classified the tooth into seven different types and can be used for automatically filling the dental charts for the case of forensic identification. One of the drawbacks of this technique was the amount of data considered for evaluation was less. The effectiveness of a deep CNN algorithm for detecting and diagnosing the dental cavities was assessed by Lee et al. [6]. 3000 periapical radiographic images were considered of which 80% were used as training data and the remaining 20% as test data. The accuracy of diagnosing the molar teeth, premolar teeth and both molar and premolar teeth was found to be 88%, 89% and 82%, respectively. Even though the obtained efficiency and accuracy was considerably good, however, there are some drawbacks. This technique was not designed to differentiate between the proximal, early and root caries. A technique that combined deep CNN and optical coherence tomography (OCT) imaging modality for detecting the occlusal cavity lesions was developed by Salehi et al. [7]. Fifty-one permanent tooth were collected and were categorized into the three classes, i.e. the non-carious teeth, caries extending into enamel and caries extending into the dentin. The specificity and sensitivity of differentiating among the non-carious and carious lesions were 100% and 98% respectively. This model classified the carries into just three classes, hence practical application of this technique may not help the dentists to diagnose accurately since the treatment varies with the intensity of damage to the tooth. It is observed from the literature survey that all the techniques discussed above have been used only in the process of detecting and classifying dental caries, most of them having a further scope for improving the accuracy of classification. Classification of dental caries using G.V Black classification has not been explored much by the researchers. Prerna et al. [8] performed the classification of dental images into six G.V Classes using machine learning algorithms. 400 dental x-ray images were used as dataset. The features were extracted using GIST (Graphics and Intelligence based Script Technology) descriptors. Marginal features analysis was used for feature reduction. Later, the features were subjected to the various classifiers. The accuracy of the Adaboost classifier with Marginal fisher analysis feature reduction was 88%. Later the Wilcoxon Signed Ranked test was applied for feature selection. Then, the selected features were subjected to the various classifiers. The Adaboost classifier gave the highest accuracy of 92%. In this paper, we further explore the possibility of G.V black dental caries classification using DCNN with improved performance.

3 Proposed Work The dataset used for the proposed model for classifying the dental images into six different classes i.e. Class I–Class VI consists of the collection of 1500 images. The dental images are Radio Visiography(RVG) digital x-ray images obtained from various dental clinics. 250 images for each class were considered. The proposed model for G.V Black Classification using the two pretrained model AlexNet and GoogleNet is shown in Fig. 2 and is detailed below.

G.V Black Classification of Dental Caries Using CNN

105

Fig. 2. Proposed architecture for G.V Black classification

3.1 Preprocessing The acquired images are preprocessed using median filtering. This step removes unwanted noise from the image. Figure 5 shows the result of the dental images after applying median filtering. Binary segmentation is used for segmenting the images into different segments. Figure 6 shows the result of the dental images after applying binary segmentation. 3.2 Feature Extraction The features are extracted using local ternary pattern(LTP) [9]. The neighbor pixel value were encoded into three values (−1,0,1) instead of (0,1). The upper LTP and lower LTP was obtained. Then, the local ternary pattern was determined which is the concatenation of upper LTP and lower LTP. The advantages of LTP is its robustness to noise and more accurate. The principal component analysis (PCA) [10] is a tool that reduces the data into fewer dimension while retaining most of the information. 3.3 Convolutional Neural Network for Image Classification A convolutional neural network (CNN) is a multilayer neural network [11]. CNN is one of the most popular algorithms for deep learning, a type of machine learning in which a model learns to perform classification tasks directly from image, video, text or sound. The most common use of convolutional neural networks is finding patterns in images to recognize objects, faces and scenes. We propose the use of two convolutional neural networks for Dental image classification. AlexNet and GoogleNet are used for the classification of images into six classes. AlexNet architecture was proposed by Hinton et al. [12]. The AlexNet architecture consists of five convolutional layers, two fully connected layers and one softmax layer. Rectified Linear unit follows each convolution layer. The last layer is the softmax layer which is fully connected to six output classes with help of softmax function. The dropout rate employed is 50%. The preprocessed input image is of size 227 × 227 pixels. 1500 images were used out of which 1200 images were used for training and 300 images were used for testing. Figure 3 shows the architecture of AlexNet architecture. Table 1 describes the various layers (in sequence) with the kernel, kernel size, stride and padding in detail for AlexNet architecture

106

P. Singh and P. Sehgal

Fig. 3. Architecture of AlexNet architecture

GoogleNet is a convolutional neural network proposed by Google [13]. Figure 4 describes the GoogleNet architecture. It has 22 layers deep architecture. It uses the combination of various inception modules that uses some pooling, convolutional and some concatenation operations. The inception module is the heart of the GoogleNet architecture. The architecture consists of simple convolutional layers followed by many blocks of inception modules and a layer of maxpooling which affects the spatial dimension. Each inception module contains two convolutional layers [14]. The average pooling layer at the end is connected to the fully connected layer with six neurons that classify into six G.V Black classes. Table 2 describes the Google Net architecture’s various layers, input, kernel size, stride and padding in detail.

4 Results A pertained deep convolution network was used to classify the dental images into G.V Black classes. A total of 1500 periapical images were used for the study of classifying dental images into various classes (class I–Class VI) based on G.V Black classification. Out of 1500 images, 1200 images (80%) were used to train the model and 300 images (20%) was used as testing dataset. First, the dental images were preprocessed using median filtering. Figure 5 shows the result of the periapical dental image after applying median filtering to remove the unwanted noise in the image. After removing the noise, the images were segmented. Figure 6 shows the segmented image. After segmentation, the features are extracted from the images based on local Ternary pattern (LTP) and were further reduced using Principal Component Analysis (PCA). The pretrained models used are AlexNet architecture and GoogleNet architecture. Table 3 shows the confusion matrix of classifying the dental images into six classes using AlexNet architecture. Table 4 shows the confusion matrix for classifying the dental images into six classes using GoogleNet

G.V Black Classification of Dental Caries Using CNN

107

Table 1. Detailed description of AlexNet architecture Layer’s name

Input image size

Output image size

No. of filter

Filter size

Stride

Padding

Convolution + ReLU + Normalization

227 × 227

55 × 55 × 96

96

11 × 11

4

0

Pooling

55 × 55 × 96

27 × 27 × 96

3×3

2

0

Convolution + ReLU + Normalization

27 × 27 × 96

27 × 27 × 256

5×5

1

2

Pooling

27 × 27 × 256

13 × 13 × 256

3×3

2

0

Convolution + ReLU

13 × 13 × 256

13 × 13 × 384

384

3×3

1

1

Convolution + ReLU

13 × 13 × 384

13 × 13 × 384

384

3×3

1

1

Convolution + ReLU

13 × 13 × 384

13 × 13 × 256

256

3×3

1

1

Pooling

13 × 13 × 256

6 × 6 × 256

3×3

2

0

256

Fully Connected + Relu + Dropout

4096

Fully Connected + Relu + Dropout

4096

Softmax

6

Fig. 4. GoogleNet architecture for G.V.Black classification

architecture. The performance of both the models are compared in Table 5 with AlexNet architecture showing the accuracy of 93%, sensitivity of 90% and specificity of 92%

108

P. Singh and P. Sehgal Table 2. Detailed description of GoogleNet architecture.

Layers

Input image size

Output image size

No. of filter

Filter size

Stride

Padding

Convolution

227 × 22 × 7

112 × 112 × 64

64

7×7

2

1

Pooling

112 × 112 × 64

56 × 56 ×× 64

3×3

2

1

Convolution

56 × 56 × 64

56 × 56 × 192

3×3

1

1

Pooling

56 × 56 × 192

28 × 28 × 192

3×3

2

1

Inception 3a

28 × 28 × 192

28 × 28 ×× 256

256

Inception 3b

28 × 28 × 256

28 × 28 × 480

480

Pooling

28 × 28 × 480

14 × 14 × 480

3×3

2

1

Inception4a

14 × 14 × 480

14 × 14 × 512

Inception4b

14 × 14 × 512

14 × 14 × 512

Inception4c

14 × 14 × 512

14 × 14 × 512

Inception4d

14 × 14 × 512

14 × 14 × 528

Inception4e

14 × 14 × 528

14 × 14 × 832

Pooling

14 × 14 × 832

7 × 7 × 832

3×3

2

1

Inception 5a

14 × 14 × 832

7 × 7 × 832

Inception 5b

14 × 14 × 832

7 × 7 × 1024

Avepool

7 × 7 × 1024

1 ×× 1 × 1024

Dropout

1 × 1 × 1024

1 × 1 × 1024

Softmax

1 × 1 × 1024

192

512

6

and GoogleNet architecture with accuracy of 94%, sensitivity of 91% and specificity of 93%. The performance of the proposed approach is also compared with the G.V Black classification proposed by Prerna et al. [8] in Table 5.

G.V Black Classification of Dental Caries Using CNN

109

Fig. 5. Dental images after filtering

Fig. 6. Dental image after segmentation Table 3. Confusion matrix for AlexNet architecture. Class1 Class II Class III Class IV Class V Class VI Class I

47

4

0

0

0

0

Class II

1

48

1

0

0

0

Class III 0

0

46

2

2

0

Class IV 0

0

5

43

2

0

Class V

0

0

3

0

47

0

Class VI 0

0

0

0

2

48

5 Conclusion The proposed technique was used to classify 1500 dental periapical images into various classes based on G.V Black classification model (Class I–Class VI). Features are extracted using Local Ternary Pattern and after feature reduction, the features are subjected using various classifiers. The pertained model used are AlexNet architecture and

110

P. Singh and P. Sehgal Table 4. Confusion matrix for GoogleNet architecture. Class1 Class II Class III Class IV Class V Class VI Class I

49

Class II

2

1

0

0

0

0

47

1

0

0

0

Class III 0

0

45

0

5

0

Class IV 0

1

2

42

3

0

Class V

0

0

1

0

49

0

Class VI 0

0

0

0

0

50

Table 5. Performance of Pretrained Models (AlexNet and GoogleNet) and Adaboost Classifier Accuracy (%)

Sensitivity (%)

Specificity (%)

AlexNet

93

90

92

GoogleNet

94

91

93

Adaboost

92

90

90

GoogleNet architecture with a classification accuracy of 93% and 94%. The proposed algorithm can be used by the dentist to classify the dental images based on the type of caries.

References 1. Datta S, Chaki, N (2015) Detection of dental caries lesion at early stage based on image analysis technique. In: 2015 IEEE international conference on computer graphics, vision and information security (CGVIS). IEEE, pp 89–93 2. Naebi M, Saberi E, Risbaf Fakour S, Naebi A, Hosseini Tabatabaei S, Ansari Moghadam S, Azimi H (2016) Detection of carious lesions and restorations using particle swarm optimization algorithm. Int. J. Dentis. 3. Singh P, Sehgal P (2017) Automated caries detection based on Radon transformation and DCT. In: 2017 8th international conference on computing communication and networking technologies (ICCCNT). IEEE, pp 1–6 4. Prajapati SA, Nagaraj R, Mitra S (2017) Classification of dental diseases using CNN and transfer learning. In: 2017 5th international symposium on computational and business intelligence (ISCBI). IEEE, pp 70–74 5. Miki Y, Muramatsu C, Hayashi T, Zhou X, Hara T, Katsumata A, Fujita H (2017) Classification of teeth in cone-beam CT using deep convolutional neural network. Comput Biol Med 80:24– 29 6. Lee JH, Kim DH, Jeong SN, Choi SH (2018) Detection and diagnosis of dental caries using a deep learning-based convolutional neural network algorithm. J Dent 77:106–111 7. Salehi HS, Mahdian M, Murshid MM, Judex S, Tadinada A (2019) Deep learning-based quantitative analysis of dental caries using optical coherence tomography: an ex vivo study.

G.V Black Classification of Dental Caries Using CNN

8.

9. 10.

11. 12.

13. 14.

111

In: Lasers in Dentistry XXV, vol 10857. International Society for Optics and Photonics, p 108570H Singh P, Sehgal P (2020) Decision support system for black classification of dental images using GIST descriptor. Presented at 3rd International Conference on Advanced Computing and Intelligent Engineering (ICACIE). Proceeding to be published in Advances in Intelligent System and Computing(AISC) series of Springer Tan X, Triggs B (2010) Enhanced local texture feature sets for face recognition under difficult lighting conditions. IEEE Trans Image Process 19(6):1635–1650 Phillips PJ, Flynn PJ, Scruggs T, Bowyer KW, Chang J, Hoffman K, Marques J, Min J, Worek W (2005, June). Overview of the face recognition grand challenge. In: 2005 IEEE computer society conference on computer vision and pattern recognition (CVPR’05), vol 1. IEEE, pp 947–954 Hu W, Huang Y, Wei L, Zhang F, Li H (2015) Deep convolutional neural networks for hyperspectral image classification. J Sens Krizhevsky A, Sutskever I, Hinton GE (2012) ImageNet classification with deep convolutional neural network. In: Advances in neural information processing systems, vol 25. NIPS, pp. 1106–1114 Zhou B, Khosla A, Lapedriza A, Torralba A, Oliva A (2016) Places: an image database for deep scene understanding Szegedy C, Liu W, Jia Y, Sermanet P, Reed S, Anguelov D, Erhan D, Vanhoucke V, Rabinovich A (2014) Going deeper with convolutions. Computing research repository

Electronics and Electrical Applications

Pre-emptive Spectrum Access in Cognitive Radio for Better QoS Avirup Das(B) , Sandip Karar, Nabanita Das, and Sasthi C. Ghosh Advanced Computing & Microelectronics Unit, Indian Statistical Institute, Kolkata, India [email protected], [email protected], [email protected], [email protected]

Abstract. Cognitive radio techniques are becoming very popular to overcome the spectrum scarcity. For dynamic channel access, spectrum latency is a major bottleneck to achieve good QoS. In this paper, we have proposed distributed proactive channel scanning for different channel assignment policies of primary users (PUs). From periodic channel scanning, secondary users (SUs) store recent history of channel occupancy, and based on that decide the order of channel selection when demand comes, and switch channels proactively, if any better channel appears. For various channel allocation strategies of primary users, different channel selection algorithms are presented for secondary users to improve the spectrum latency as well as the channel switching overhead per call. Through simulation, the performance of the proposed strategies is compared with that of conventional cognitive radio. Results show that by the proposed scheme, the call drop/block rate decreases by more than 30% with significant reduction in interference, and hence improving the QoS of the system. Keywords: Cognitive radio · Spectrum latency Channel selection · QoS (Quality of Service)

1

· Channel history ·

Introduction

In recent days, wireless spectrums are getting congested due to the rapid increase in the number of licensed users. The main reason for spectrum scarcity is the static channel assignment for licensed users that results both under utilization and unavailability of channels. Cognitive radio (CR) is the technology by which wireless networks can overcome the scarcity of the spectrum. Unlicensed users can utilize the unused spectrum of licensed users opportunistically. In cognitive radio, an unlicensed SU can communicate through a licensed channel after sensing it idle. However, as soon as a PU appears on the channel, the SU must release it and attempts to find another idle channel to switch. This switching procedure is called reactive switching. For this reactive switching, an SU must be c Springer Nature Singapore Pte Ltd. 2021  C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_12

116

A. Das et al.

equipped with two transceivers to enable it for simultaneous scan and transmission. This introduces noise in PU’s communication and degrades QoS for SU’s. In this paper, we have proposed a proactive channel switching technique for SU that eliminates the need of multiple transceivers in SU but still maintains better QoS for SUs and helps to minimize the interference with PU. It is obvious that though cognitive radio attempts to increase the spectrum utilization by opportunistic use of licensed spectrum, interference with PU is not desired. Our work may be a good solution to this problem. Simulation studies show a significant improvement in terms of interference, throughput, call block rate, and drop rate. In cognitive radio networks, to avoid simultaneous communication and scan, one solution may be periodic scan with one transceiver and to stop communication while scanning. But for periodic scanning, if a PU starts to transmit data on a channel which is in use by a SU, then interference will occur, and it will degrade QoS for both PU and SU. So, it would be better if an SU can predict the probability of occurring interference on a channel, and based on that can select the channel with least probability of interference during its communication time. In this work, we have addressed this problem using light weight channel usage prediction algorithms depending on different channel assignment techniques of PU network that enables an SU to preempt channels based on its dynamic estimate of idle time. The rest of the paper is organized as follows: In Sect. 2, related work on this area is presented. System model of the work is described in Sect. 3. In Sect. 4, details of algorithms and different spectrum access techniques have been described. Section 5 presents the details of the simulation studies with results. Finally, the paper is concluded in Sect. 6 with future work.

2

Related Work

The relevant work on dynamic spectrum access in cognitive radio is mostly dominated by reactive spectrum access which means that an active SU will switch transmission to a different channel only when a PU arrives in the current channel. Some works on proactive spectrum access in cognitive radio have been appeared in literature. In [1], two channel selection and switching techniques are presented to minimize interference after predicting the future spectrum availability from the past channel history. In [2], the authors propose two channel selection schemes: minimum collision rate channel selection algorithm and minimum hand-off rate channel selection algorithm based on spectrum hole prediction under the constraint that the collision probability is bounded. In [3], authors follow the partially observable Markov decision process (POMDP) framework to derive optimal sensing and access policies. Another work in [4] also proposes a POMDP framework for dynamic sensing and channel access considering that the cognitive users have perfect knowledge about the distribution of the primary signals. The work in [5,6] proposes a new access policy that utilizes the channel access opportunities by combining not only the channel state information but also channel quality information for exploring the optimal trade-off between the throughput of secondary users and the interference to the primary users. The

Pre-emptive Spectrum Access in Cognitive Radio for Better QoS

117

authors in [7,8] carry out very insightful studies on how to find the available spectrum opportunities efficiently and fast, through sensing sequence selection and sensing period optimization. The paper [9] proposes a selective opportunistic spectrum access (SOSA) scheme which estimates the probability of a channel appearing idle based on past statistics and chooses the best spectrum-sensing order to maximize spectrum efficiency. A slotted transmission protocol for secondary users using a periodic sensing strategy with optimal dynamic access is proposed in [10]. In [11]; a multi-channel selection algorithm is proposed that predicts the spectrum opportunities to limit the interference to the primary as well as to enhance the spectrum utilization. The problem of target channel sequence selection is addressed for proactive spectrum-hand off in [12]. Many other works [13] use sophisticated machine learning algorithms to predict the spectrum availability in the future time slots after training the models with past dataset long enough for accurate prediction. These models provide very high degree of accuracy, the problem with these kinds of advanced machine learning based-prediction is that it is too expensive to train the model in real time. Only off-line processing is possible for these models, thereby making it difficult to predict dynamic scenario where the spectrum usage of the PUs changes rapidly with time (Fig. 1).

Fig. 1. Spectrum access by SU

In this paper, a simple and fast on-line learning scheme is proposed based on the statistical data about the ON and OFF duration and the current status of the channel to predict which of the channels will remain idle for the longest

118

A. Das et al.

duration and SUs will switch the channels accordingly. Most of the papers in literature considers that the channels used by the PUs are independent and ON and OFF durations are exponentially distributed. But this is not always true. Our work considers three different primary assignment policies and investigates how the secondary access schemes are influenced by different primary channel assignment strategies.

3

System Model

Let us consider a system with a part of wireless spectrum containing L nonoverlapping channels which are used by the PUs. The channels are also opportunistically used by M SUs. Any of the SUs can access any of those channels if it senses the channel as idle that means if there is no activity on that channel. It is assumed that every SU is equipped with a single transceiver, and therefore, SU cannot sense and transmit simultaneously. To accumulate current knowledge of channel usage, each SU periodically senses the channels in order. Time is slotted, and each slot of duration, say T is further subdivided in two parts: sensing time of duration Ts and transmission time of duration Tt . Thus T = Ts + Tt . In general, Tt >> Ts . During the sensing time Ts in each slot, each SU suspends any transmission activity and senses the channel status on the L channels sequentially in round robin fashion. Moreover, it is assumed that the SU’s co-ordinate and co-operate among themselves such that the sensing slots of SU’s do not overlap. To minimize switching delay, it is assumed that SUs initiate switching at the end of a sensing slot only. The arrival of PU traffic is assumed to follow Poisson process with an average rate λP U and the call holding time is exponentially distributed with average holding time μP1U . Three types of channel allocation process is considered for the PUs: 1. Least ID-based channel assignment 2. Round robin channel assignment 3. Random channel assignment. In least ID-based channel assignment, the channels have a predefined ordering with identification (ID) numbers 1, 2, . . . , L decided by the base station (BS) and when a PU call arrives, the BS assigns an idle channel which has the least ID among the available channels. In round robin channel assignment also, the channels are ordered according to their ID numbers. The PU calls are assigned sequentially to channel with ID number 1, then ID number 2, then ID number 3, and so on whenever available. After the channel with ID number L, the next PU call will be assigned to channel with ID number 1 if it is free, otherwise to the next idle channel available. Finally, in random channel assignment, when a PU call arrives, the BS selects a channel randomly among the available channels and assigns the channel to the PU call. Depending on the PU usage pattern, the behaviour of individual channel activity is modeled as a continuous-time, alternating renewal process in which the channel alternates its state between ON

Pre-emptive Spectrum Access in Cognitive Radio for Better QoS

119

(busy) and OFF (idle) periods. The channel state is ON during a PU transmission on the channel and the channel state is OFF if there is no PU transmission on that channel. Let us denote by Toni and Tof fi the random variables which represent the ON and OFF duration for channel i, respectively. The SU traffic is also modeled similar to primary traffic with traffic arrival rate following Poisson process at an average rate λSU and call holding time 1 . exponentially distributed with average duration μSU

4

Spectrum Access Scheme of SU

The SU should follow different spectrum access schemes in accordance with different channel allocation strategies for the PU. It is assumed that the channel allocation policy for the PU is known in advance to the SU. 4.1

Least-ID PU Channel Assignment

To derive a proactive spectrum access strategy for SU, the channel statistics information regarding the channel status must be perfectly known to the SU in advance. But, in practice, this is not always possible especially when the PU activity varies with time. The SU must therefore estimate channel behaviour statistics such as the average idle duration for each channel i from the past data. To estimate the average idle duration, the SU must collect and update information about the number of consecutive idle encountered so far through sensing. The decision about channel selection for SU should be handled on a slot-by-slot basis to minimize interference with PUs as well as to minimize the number of switching between channels. The objective for the SU is to select the best possible channel denoted by C k for transmission at the beginning of kth slot. The sensing result in ith channel during kth slot is denoted by Zik ∈ {0, 1}, (k−1) where busy means Zik = 1, and idle means Zik = 0. Let γ0i be the length (number of zeros) in the last (or current) run of zeros obtained in the ith channel (k−1) during (k − 1)th slot, n0i represents the number of runs of zeros encountered by the SU so far in ith channel during (k − 1)th slot and the average estimated (k) idle duration of the ith channel is denoted by Tof f i . There are four possible (k−1)

(k)

→ Zi in sensing results in the ith channel that can occur in transitions Zi going from (k − 1)th slot to kth slot: (i) 0 → 0, (ii) 1 → 0, (iii) 0 → 1 and (iv) 1 → 1. In 0 → 0 transition, the SU updates the variable γ0i corresponding to the number of zeros in the current run of zeros by adding one to it. The number of runs n0i encountered so far will not change. The average OFF duration Tof f i of channel i is not changed as the current run of zeros is not considered in (k) calculating the average. The remaining idle duration ui is estimated as follows ⎛ ⎞ 1.96 ⎠ (k) (k) ⎝ (k) ui = Tof f i 1 +  − γ0i T (1) (k) n0i

120

A. Das et al.

 Here,

(k) Tof f i



1+

1.96  (k) n0

is the duration corresponding to 95% confidence

i

(k)

interval for the average idle duration Tof f i considering that it follows exponential distribution [1]. This means that 95% of the idle in ith channel will  duration  (k)

last less than this duration and therefore Tof f i

1+

1.96  (k) n0

is approximately

i

considered as the upper limit for an idle time duration in ith channel with 95% (k) confidence. The remaining idle duration ui is therefore this value subtracted by the duration already elapsed. (k) In 1 → 0 transition, a new run of zeros is started and γ0i is therefore set equal to one. The number of runs n0i and the average OFF duration Tof f i do (k) not change. The remaining idle duration ui is similarly estimated as before. In 0 → 1 transition, a run of zeros is just completed and therefore number of runs n0i encountered so far is incremented by one. The average OFF duration Tof f i is now updated to a new value according to the following update rule: (k)

Tof f i =

1 (k)

n0i

(k)

(k)

(k−1)

γ0i T + (n0i − 1)Tof f i

(2)

Finally, for 1 → 1 transition, the SU need not update any of the variables γ0i , n0i or Tof f i at the kth slot. The task of the SU is to select the channel j such that the estimated idle duration ukj is maximum subject to the condition that the channel j is currently idle in kth slot i.e. Zjk = 0. If no such channel j is found, the SU will drop any ongoing transmission till the next slot. Detailed procedure is described in Algorithm 1. 4.2

Round Robin PU Channel Assignment

For round robin PU channel assignment, the use of average run length of zeros would not provide significant information to predict the future behaviour of the channel. The reason is for round robin PU channel assignment, the average run length of zeros would be more or less equal in the long run for any channel. Instead, we use the 0 → 1 transition time from the past history of the channel usage to predict which of the channels currently idle has the least probability of being busy again. To achieve this, the SU needs to store the past sensing results upto M samples for all the L channels in form of a L × M sensing matrix SR where SR(i, m) ∈ {0, 1}, i = 1, 2, . . . , L and m = 1, 2, . . . , M indicates the sensing result for the ith channel and mth sample. The sensing matrix SR gets updated at every sensing instant. For the kth instant the sensing matrix is denoted by SR(k) . At the start of kth slot, the SU senses all the L channels and form the current sensing result matrix SR(k) by appending a new column to SR(k−1) corresponding to the current sensing result and eliminating the first column. From the sensing matrix SR(k) , the SU finds the time instant of the

Pre-emptive Spectrum Access in Cognitive Radio for Better QoS

121

Algorithm 1: SU channel selection scheme for least-ID PU channel assignment (k−1) , i

Input: Sensing period T , length of the last run of zeros, i.e., γ0 encountered so far denoted by (k−1)

i.e., Tof f

i

(k−1) n0 i

number of runs of zeros

and average OFF-duration estimated so far,

for each channel i = 1, 2, . . . , L.

Output: Channel to be selected by SU for transmission in kth slot denoted by C (k) . (k) Sense all the L channels to get the current sensing result denoted by Zi for channel i = 1, 2, . . . , L. for each channel i do (k−1) (k) if Zi = 0 and Zi = 0 then (k) (k−1) γ0 = γ 0 + 1; i i (k) (k−1) = n0 ; i i (k) (k−1) Tof f = Tof f ; i i (k) (k) (k) ui = Tof f − γ0 T ; i i

n0

⎛

(k)

ui

(k) = Tof f ⎝1 + i

(k−1)

(k)

if Zi = 1 and Zi (k) γ0 = 1;

1.96 (k) n0 i

⎛

=

(k) Tof f i

(k)

if Zi = 0 and Zi (k) (k−1) γ0 = γ 0 ;

1.96 (k) n0 i

⎠ − γ (k) T ; 0i

= 1 then

i i (k) (k−1) n0 = n0 + 1; i i  (k) (k) 1 Tof f = (k) γ0 T i i n0 i

else

0i

⎞

⎝1 +

(k−1)

⎠ − γ (k) T ;

= 0 then

i (k) (k−1) n0 = n0 ; i i (k) (k−1) Tof f = Tof f ; i i

(k) ui

⎞

(k) i

+ (n0

(k−1)

− 1)Tof f

i

;

(k) (k−1) = γ0 ; i i (k) (k−1) n0 = n0 ; i i (k) (k−1) Tof f = Tof f ; i i

γ0

(k)

Find channel j such that uk j is maximum subject to Zj if no such j is found then C (k) = null; else C (k) = j;

= 0;

most recent 0 → 1 transition for a channel i, and stores it in a variable τi . If no such 0 → 1 transition is found for any particular channel l from the sensing matrix, the variable τl is set to zero, the lowest possible value so that this channel l will not be placed in the top-most priority list. The task of the SU now is to select the channel j such that τj is maximum subject to the condition that the channel j is currently idle in kth slot. If no such

122

A. Das et al.

Algorithm 2: SU spectrum access scheme for round robin PU channel assignment Input: L × M binary sensing result matrix SR(k−1) corresponding to sensing result for L channels in the last M samples at (k − 1)th slot. Output: Channel to be selected by SU for transmission in kth slot denoted by C (k) . Sense all the L channels to get the current sensing result vector Z(k) . The current sensing result matrix SR(k) is formed as follows: SR

(k)

(i, m) ← SR SR

(k)

(k−1)

(i, m + 1); ∀i, m = M (k)

(i, m) ← Zi

; ∀i, m = M

for each channel i do Calculate the most recent sample index τi from SR(k) (i, m), m = 1, 2, . . . , M at which the channel undergoes a 0 → 1 transition; if no 0 → 1 transition is found then Set τi = 0 Find channel j such that τj is maximum and SR(k) (j, M ) = 0; if no such j is found then C (k) = null; else C (k) = j; SU channel selection scheme during slot k for round robin PU channel assignment

channel j is found, the SU will drop the call till the next slot. Detailed steps of the procedure are described in Algorithm 2. 4.3

Random PU Channel Assignment

For random PU channel assignment, the activity of PU channels does not follow any pattern that can be learned and hence it is not possible to design any access strategy for SUs compatible with random PU channel assignment. However, for comparison purpose, we follow the same access strategy for SUs as used for the round robin-based PU channel assignment.

5

Simulation Results

For the performance evaluation of our proposed scheme of proactive and preemptive spectrum access by cognitive users, we consider a network scenario consisting of 20 primary channels for our simulation environment. The traffic of the network is generated according to Poisson process and call holding time is assumed to follow exponential distribution with a mean value of 8 s. The proposed scheme is executed for 20,000 s to evaluate our proposed methods. Figure 2 shows the variation of interference on PU transmission with SU’s sensing interval. It is clear that if the sensing interval is small, the interference time is negligible. PU usage is kept at 50%. Finally, we observe that our channel assignment policy, for round robin-based PU channel assignment introduces the least amount of interference.

Pre-emptive Spectrum Access in Cognitive Radio for Better QoS

123

Fig. 2. Interference versus sensing interval

Fig. 3. Latency for call completion

Figure 3 shows the duration of call completion time in comparison with the requested call duration. Here, we compared our procedures with the conventional cognitive radio technique which uses re-active spectrum selection assuming dual transceivers, i.e, simultaneous sensing and communication. Our procedure is showing better result though with single transceiver. Figures 4 and 5 show the call block rate and call drop rate percentage with different PU usage. SU usage is 20% and scanning interval is kept 0.6 s. It is clear from the results that our procedure minimizes the call block rate and drop rate in comparison with the conventional cognitive radio spectrum allocation. Figure 6 shows the utilization of the channels by PUs and SUs with respect to the total usage of the network. SU usage is assumed to be 20% and PU usage is varying from 30 to 60%, where SU scanning interval is kept 0.6 s. It is clear that our proposed procedure gives better channel utilization than conventional cognitive radio.

124

A. Das et al.

Fig. 4. Call block rate with diff. PU usage

Fig. 5. Call drop rate with PU usage

6

Conclusion

In cognitive radio networks, spectrum mobility is a major concern that determines the spectrum selection latency. In this paper, we have proposed a proactive and pre-emptive spectrum mobility technique to reduce spectrum latency for SUs with single transceiver, eliminating the need for simultaneous scanning and communication. Here, an SU can switch the channel just before the estimated arrival of the PU in that channel, so that interference is minimized for both PU and SU, to improve QoS for the whole network. Assuming different types of channel assignment policies for PUs, based on the recent history of the channels following our proposed algorithms, an SU will decide the next channel to scan. Extensive simulation results show that even with a single transceiver, the proposed schemes enable the SUs to achieve better QoS in terms of interfer-

Pre-emptive Spectrum Access in Cognitive Radio for Better QoS

125

Fig. 6. Total utilization of the network

ence, call drop/block rate, and channel utilization compared to the conventional reactive cognitive radio technique with provision for simultaneous scanning and communication that demands a two-fold increase in transceiver hardware.

References 1. Yang L, Cao L, Zheng H (2008) Proactive channel access in dynamic spectrum networks. Phys Commun 1(2):103–111 2. Xiao Q, Li Y, Zhao M, Zhou S, Wang J (2009) Opportunistic channel selection approach under collision probability constraint in cognitive radio systems. Comput Commun 32(18):1914–1922 3. Chen Y, Zhao Q, Swami A (2007) Bursty traffic in energy-constrained opportunistic spectrum access. IEEE GLOBECOM 4641–4646 4. Unnikrishnan J, Veeravalli VV (2010) Algorithms for dynamic spectrum access with learning for cognitive radio. IEEE Trans Signal Process 58(2):750–760 5. Peng X, Zhong X, Xiao L, Zhou S, Opportunistic spectrum access based on channel quality under general collision constraints. In: IEEE 24th annual international symposium on personal, indoor, and mobile radio communications (PIMRC). IEEE, New York, pp 2633–2637 6. Peng X, Zhong X, Xiao L, Zhou S (2013) Robust opportunistic spectrum access based on channel quality information in varying multi-channel networks. In: IEEE global communications conference (GLOBECOM). IEEE, New York, pp 1240–1245 7. Kim H, Shin KG (2008) Fast discovery of spectrum opportunities in cognitive radio networks. In: 3rd IEEE symposium on new frontiers in dynamic spectrum access networks. IEEE, New York, pp 1–12 8. Kim H, Shin KG (2008) Efficient discovery of spectrum opportunities with MAClayer sensing in cognitive radio networks. IEEE Trans Mob Comput 7(5):533–545 9. Yuan G, Grammenos RC, Yang Y, Wang W (2010) Performance analysis of selective opportunistic spectrum access with traffic prediction. IEEE Trans Veh Technol 59(4):1949–1959

126

A. Das et al.

10. Zhao Q, Geirhofer S, Tong L, Sadler BM (2008) Opportunistic spectrum access via periodic channel sensing. IEEE Trans Signal Process 56(2):785–796 11. Lee J, Park H-K (2014) Channel prediction-based channel allocation scheme for multichannel cognitive radio networks. J Commun Networks 16(2):209–216 12. Zheng S, Yang X, Chen S, Lou C (2011) Target channel sequence selection scheme for proactive-decision spectrum handoff. IEEE Commun Lett 15(12):1332–1334 13. Agarwal A, Gangopadhyay R, Dubey S, Debnath S, Khan MA (2018) Learningbased predictive dynamic spectrum access framework: a practical perspective for enhanced QoE of secondary users. IET Commun 12(18): 2243–2252. https://doi. org/10.1049/iet-com.2018.5407

Design of Smart Antenna Arrays for WiMAX Application Using LMS Algorithm Under Fading Channels Anupama Senapati(B) , Pratyushna Singh, and Jibendu Sekhar Roy School of Electronics Engineering, KIIT Deemed to be University, Bhubaneswar, Odisha, India [email protected], [email protected], [email protected]

Abstract. Smart antenna technology is used for capacity enhancement and generating main beam and null along with user and interferer as desired. This paper presents the design of smart antenna for WiMAX application and the use of array synthesis methods like Tchebycheff distribution to reduce side lobe level in presence of different fading channels. Also, performance evaluation of the least mean square algorithm is done in application to linear and planar smart antenna array. Side lobe-level reduction up to 10 dB is achieved for planar array. Keywords: Smart antenna · Fading channels · WiMAX · Side lobe level · Tchebycheff distribution

1 Introduction Smart antenna is considered an emerging technology that improves the performance of wireless communication systems with larger covering areas, increased SNR and capacity, saving energy for the same performances [1–3]. Smart antenna system with its multiple antenna elements uses a digital signal processing module to perform direction of arrival estimation (DOA) and beamforming as desired. Various algorithms are used for beamforming which has different advantages and disadvantages [3, 4]. Smart antenna produces the main beam along the desired direction and null towards the undesired direction. In smart antenna system, better frequency reuse is achieved as it dynamically adjusts to its environment [5, 6]. Adaptive beamforming using LMS algorithm is done for smart antenna array where the main beam and nulls are achieved properly [7]. Side lobes consume power and also cause interference for other users. To achieve good results in terms of directivity, attention is given for the reduction of sidelobe levels (SLLs) in smart antennas. A good level reduction is not observed for ordinary antenna array, so Tchebycheff distribution is used and a good level reduction is observed. Comparative analysis of multiple modulation schemes in Rayleigh faded channels using beamforming technique [8]. WiMAX is one of the most popular broadband wireless technologies that is used enormously [9]. Knowledge about WiMAX technology its security and applications © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_13

128

A. Senapati et al.

Information about threats in different layers of WiMAX is reported in [10]. Various beamforming algorithms are available for smart antenna design [11]. Sidelobe reduction below −20 dB is achieved using Tchebycheff LMS algorithm for smart antenna [12]. Effects of antenna beamwidth, Rician-K and other channel parameters on LTEOFDM channels are investigated and presented in [13] through theoretical analysis and simulations of channel models. The characteristics of two different uniformly spaced circular arrays are analyzed, including the beam width, the sidelobe peak level, the resolution and the Cramer-Rao bound (CRB) of the array is done in [14]. The integration of analytical strategies and global optimization techniques is proposed to address the limitations of current almost difference sets (ADSs) methods for thinning planar apertures. A numerical validation, including full-wave simulations and comparisons with state-of-the-art methods, is illustrated in [15]. A numerical validation, including fullwave simulations and comparisons with state-of-the-art methods, is illustrated in [16] at different frequency bands from 2 to 6 GHz band. In [17], performance analysis in terms of beamforming, null steering and rate of convergence is implemented on LMS, NLMS and LS-CMA algorithm. The objective of this paper is to use Tchebycheff distribution for side lobe level reduction in adaptive smart antenna array for WiMAX application considering Rayleigh and Rician fading channel with different Rician factors. The paper is divided into three sections. section 1 is introduction, Sect. 2 consists of a description on Tchebycheff distribution and WiMAX application with Rayleigh fading and Rician fading channels and simulation results. section 3 consists of a conclusion and future scope.

2 Side Lobe Level Reduction for Linear and Planar Smart Antenna Array with Fading Channels In this paper, LMS algorithm is used for adaptive beamforming along with Tchebycheff distribution for side lobe level reduction. Different fading channels considered are Rayleigh and Rician channel. Smart antenna is designed for WiMAX application. 2.1 Rayleigh and Rician Channel To transmit the information from source to destination, a communication channel or a radio link is used between the transmitter and receiver. This communication channel can be either a simple line-of-sight or the one in which the transmission or reception of data is severely hurdled by the obstacles like buildings, mountains, etc. and this results in multipath fading. In the non-line of sight between transmitter and receiver, Rayleigh fading is observed. When there are a number of non-dominant signal paths between transmitter and receiver, this leads to Rayleigh fading channel. The probability density function (pdf) of Rayleigh channel is [13] f (r) = re

−r 2 2 2 2σ /σ

(1)

where f (r) is received signal’s time average power before envelope detection and ‘r’ is envelope of Rayleigh fading

Design of Smart Antenna Arrays for WiMAX Application Using LMS …

129

When a line-of-sight signal is much stronger than the others, Rician fading is observed. The probability density function of Rician channel is [13]     −r02 + A2 r0 A r0 (2) p(r0 ) = 2 exp I0 σ 2σ 2 σ2 for r0 ≥ 0, A ≥ 0 where I0 (·) represents zeroth-order Bessel function and is the modified Bessel function of zero order. A is the peak magnitude of line of sight signal component. In Rician fading channel, K is defined as K = A2 /2σ 2 , i.e. the ratio of power of line of sight distribution and multipath components. From Eq (2), for K = 0, Rician fading channel becomes Rayleigh fading. 2.2 Tchebycheff Distribution Side lobes can create interference for other users. By reducing SLL, interference can be minimized, also frequency reuse can be increased. Different techniques to reduce SLL are array synthesis method, array thinning. In this paper, one of the array synthesis methods, i.e. Tchebycheff distribution is used for SLL minimization [12]. 2.3 Worldwide Interoperability for Microwave Access (WiMAX) For wireless broadband services, WiMAX technology is widely accepted as a costeffective and reliable solution [9]. It provides high-speed connectivity that can be used for long-distance communication with higher data rates. The frequencies commonly used are 3.5 and 5.8 GHz. 2.4 Uniform Linear and Uniform Planar Smart Antenna Array Smart antenna system comprises of direction of arrival estimation (DOA) and beamforming. In this work, adaptive beamforming is done using least mean square (LMS) algorithm with N uniform linear antenna elements and MxN uniform planar antenna elements. The weight updating equation for LMS algorithm at ‘n’th iteration is [6] [12] w(n + 1) = w(n) + μ ∗ e(n) ∗ x(n)

(3)

where μ is the step size parameter, w is the weight coefficient associated with each antenna element, x(n) is the received signal and e(n) is error between desired signal and array output. Figure 1 shows an ULA with spacing between antenna elements ‘d’. Array factor of a uniform linear antenna array (ULA), to generate main beam at wavelength ‘λ’ and desired beam at ‘θ0 ’ from broad side direction with propagation phase shift ‘α’, can be expressed as [5, 6, 12] AFL =

N −1 n=0

An e





jn 2πλd sin θ+α

(4)

130

A. Senapati et al.

Fig. 1. Uniform linear array configuration

where α=

−2π d sin θ0 λ

(5)

AFL AFmax

(6)

Normalized array factor is AFnorm =

Figure 2 shows a uniform planar antenna array (UPA) of M × N elements with inter-element spacing of d x and d y along x- and y-axis, respectively.

Fig. 2. Uniform planar array geometry

The array factor of M × N uniform planar array of element spacing d x and d y along x and y direction, respectively, can be written as [5], AFP = AFxM AFyN

(7)

where AFxM =

M m=1

Im1 ej(m−1)(kdx cos ψx +βx )

(8)

Design of Smart Antenna Arrays for WiMAX Application Using LMS …

AFyN =

N

In1 ej(n−1)(kdy cos ψy +βy )

131

(9)

n=1

Here, the directional cosines, i.e. ψ x and ψ y are considered equal.Normalized Array Factor is AFnorm =

AFP AFmax

(10)

2.5 Simulation Results Simulations are done using MATLAB for ULA with inter-element spacing of 0.5 λ for 10 antenna elements considering multiple interferers. Programs are run for 100 iterations and varying fading channels. Angle of desired user (AOA) is 20° , angle of interferers (AOIs) are −10° and 0° . Smart antenna is designed at 3.5 GHz for WiMAX application. Tchebycheff distribution is used for side lobe level reduction. Figures 3, 4 and 5 show normalized array factor (AF) using LMS adaptive beamforming algorithm with and without Tchebycheff distribution for Rayleigh and Rician fading channel in application to 10 element linear array. Simulations are done for step size (μ) of 0.02.

Fig. 3. Normalized AF for Rayleigh fading channel

Table 1 shows the summarized results. Beamforming for linear antenna array is done under Rayleigh and Rician fading channel with and without Tchebycheff distribution. Side lobe level reduction up to 6 dB is achieved using Tchebycheff distribution. For all the cases, the main beam is achieved as desired direction and nulls are obtained at a deviation of less than 15% as desired with Tchebycheff distribution. Better results are obtained for Rician channel with K = 10. Figures 6, 7 and 8 show adaptive beamforming with and without Tchebycheff distribution for fading environment in application to planar array.

132

A. Senapati et al.

Fig. 4. Normalized AF for Rician fading channel (K = 1)

Fig. 5. Normalized AF for Rician fading channel (K = 10)

Table 2 shows the summarized results. Beamforming for planar antenna array is done for Rayleigh and Rician fading channel with and without Tchebycheff distribution. Side lobe level reduction upto 12 dB is achieved using Tchebycheff distribution. For all the cases, the main beam is achieved as desired direction and nulls are obtained at a deviation of less than 15% as desired with Tchebycheff distribution. Better results are obtained for Rician channel with K = 10.

3 Conclusion From the paper, Beamforming for both linear and planar antenna array is done for Rayleigh and Rician fading channel with and without Tchebycheff distribution. It is

Design of Smart Antenna Arrays for WiMAX Application Using LMS …

133

Table 1. Performance evaluation of LMS algorithm with fading environment for linear antenna array Type of Fading channel

Main Beam (°)

Null1 (°)

SLLmax (dB)

FNBW (°)

Without Tchebycheff distribution

20

−10.6

-0.6

−12.46

26.4

With Tchebycheff distribution

20

−9.2

1.2

−17.16

37.2

Rician Without channel K Tchebycheff =1 distribution

20

−10.8

−1.2

−12.36

26.7

With Tchebycheff distribution

20

−9

0.2

−18.0

37

Rician Without channel K Tchebycheff = 10 distribution

20

−10

0

−12.32

26.6

With Tchebycheff distribution

20

−10

−0.2

−18.33

36.6

Rayleigh channel

Null2 (°)

Fig. 6. Normalized AF for Rayleigh fading channel

observed that a better reduction in side lobe level is achieved in planar array as compared to linear array using Tchebycheff distribution. For all the cases, main beam is achieved as

134

A. Senapati et al.

Fig. 7. Normalized AF for Rician fading channel (K = 1)

Fig. 8. Normalized AF for Rician fading channel (K = 10)

desired and nulls are achieved at a deviation of less than 20% from desired direction. With Tchebycheff distribution, nulls are achieved at a deviation of less than 15% from desired direction. Future work includes the use of other array synthesis methods with proper adaptive beamforming algorithm for improved performance in terms of convergence rate, accuracy of main beam and null generation.

Design of Smart Antenna Arrays for WiMAX Application Using LMS …

135

Table 2. Performance evaluation of LMS algorithm with fading environment for planar antenna array Type of fading channel

Main Beam Null1 (°) Null2 (°) SLLmax (dB) FNBW (°) (°) Without 20 Tchebycheff distribution

−11.2

−1.2

With 20 Tchebycheff distribution

−9.8

0.2

Rician Without 20 channel K = Tchebycheff 1 distribution

−10.8

Rayleigh channel

With 20 Tchebycheff distribution Rician Without 20 channel K = Tchebycheff 10 distribution With 20 Tchebycheff distribution

−24.72

26.2

−39

37

−1.4

−25.37

26.7

0.8

−38.24

37

−11.4

−1.2

−25.14

26.2

−9

−0.4

−9

-37

37.4

References 1. Torlak M, Xu G (1995) Performance of CDMA smart antenna systems. IEEE, Signals, Systems and Computers, New York 2. Godara LC (1997) Application of antenna arrays to mobile communications, Part II: beamforming and direction-of-arrival considerations. Proc IEEE 85(8):1195–1245 3. Herscovicin N, Christodoulou C (2000) Smart antenna. In: IEEE, antenna and propagation magazine, vol 42, no. 3 4. Balanis BA, Foutz CAJ, Spanias AS (2002) Smart antenna systems for mobile communication network, part 1, overview and antenna design. IEEE Antennas Propag Mag 44(3):145–154 5. Balanis CA (2005) Antenna theory–analysis and design, 3rd edn. Wiley 6. Gross F (2005) Smart antenna for wireless communication. McGraw-Hill 7. Kawitkar RS, Wakde DG (2005) Smart antenna array analysis using LMS algorithm. IEEE international symposium on microwave, antenna, propagation and EMC technology for wireless communications proceeding 8. Das KJ, Sarma KK (2012) Performance comparisons of multiple modulation schemes in Rayleigh faded channels using adaptive beamforming, vol. 1. ITEEE. ISSN 2319-2577,ISS-2 9. Joshi J, Yadav KK (2014) A study on WiMAX network technology. Int J Inn Res Comput Commun Eng 10. Bhambri A, Kansal N (2014) Survey on WiMAX technology and its protocol. IJARCET 11. Tao J-W, Chang W-X (2014) Adaptive beamforming based on complex quaternion processes. Mathe Probl Eng Art ID 291249(2014):1–10

136

A. Senapati et al.

12. Senapati A, Roy JS (2017) Adaptive beamforming in smart antenna using Tchebyscheff distribution and variants of least mean square algorithm. J Eng Sci Technol 13. Baki AKM, WasifAbsar, Md. Rahman T, Ahamed KMA (2017) Investigation of Rayleigh and Rician fading channels for state of the art (SOA) LTE-OFDM communication system. In: 4th International conference on advances in electrical engineering (ICAEE),28–30 September 2017, Dhaka, Bangladesh 14. Zhang J, Chen H, Ji Z (2017) Characteristic analysis of dual circular array. IEEE international conference on signal processing, communications and computing (ICSPCC), 22–25 October 2017 15. Salucci M, Gottardi G, Anselmi N, Oliveri G (2017) Planar thinned array design by hybrid analytical-stochastic optimisation. IET microwaves, antennas & propagation, 23 October 2017 16. Tengli AC, Hadalgi PM (2018) Design of inset-fed quad-band circular microstrip antenna for 2.4/5 GHz WLAN, WiMAX and X-band satellite applications. In: AIP conference proceedings, 23 July 2018 17. Ganguly S, Ghosh J, Kumar PK (2019) Performance analysis of array signal processing algorithms for adaptive beamforming. In: URSI Asia-Pacific Radio Science Conference (APRASC), 9–15 March 2019, New Delhi, India

Performance Comparison of Variants of Hybrid FLANN-DE for Intelligent Nonlinear Dynamic System Identification Swati Swayamsiddha(B) School of Electronics Engineering, Kalinga Institute of Industrial Technology, KIIT University, Bhubaneswar, India [email protected]

Abstract. This work presents the performance comparison of different variants of hybrid Functional Linked Artificial Neural Network (FLANN) structures and Differential Evolution (DE) algorithm (FLANN-DE) for intelligent nonlinear dynamic system identification. FLANN is single-layer artificial neural network structure having less computational complexity and preferred for online applications and DE being a derivative-free metaheuristic algorithm is used as a global optimization tool. System identification finds its application in direct modelling, channel identification and estimation, geological exploration, instrumentation and control. Direct modelling is based on adaptive filtering concept and can be developed as an optimization problem. The goal of direct modelling is to estimate a model and a set of system parameters by minimizing the prediction error between the actual system output and the model output. The identification problem involves the construction of an estimated model which generates the output which matches that of desired system output when subjected to the same input signal. In this present work, hybrid FLANN-DE is proposed for direct modelling of nonlinear dynamic systems and comparative analysis is carried out for different variants of FLANN structures such as Chebyshev FLANN (CFLANN), Legendre FLANN (LFLANN) and Trigonometric FLANN (TFLANN) in terms of performance, the speed of computation and accuracy of results. Keywords: Functional Linked Artificial Neural Network (FLANN) · Differential Evolution (DE) · System Identification · Mean Square Error · Convergence Rate

1 Introduction Intelligent identification of nonlinear dynamic systems plays an important role to collaborate the mathematical modelling with realistic applications and is established as an optimization problem. It finds its application in direct modelling, channel identification and estimation, geological exploration, instrumentation and control engineering [1–3].

© Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_14

138

S. Swayamsiddha

1.1 Motivation Direct modelling is based on adaptive filtering concept and can be developed as an optimization problem. The goal of direct modelling is to estimate a model and a set of system parameters by minimizing the prediction error between the actual system output and the model output. The input–output data plays a major role in establishing a mathematical model for the unknown system. The identification problem involves the construction of an estimated model which generates the output which matches that of desired system output when subjected to the same input signal. This involves iterative minimization of the mean square error which is the difference between the desired output and model’s estimated output [4–7]. However, if the unknown system is nonlinear, derivative-based techniques do not perform satisfactorily to identify the system. So to identify the nonlinear systems, the application of evolutionary computation can be thought of. The reason for choosing Differential Evolution (DE) as an effective learning algorithm for direct modelling of nonlinear systems is that it is a global optimization tool, simple to implement having fewer tunable parameters and has faster convergence [8–10]. 1.2 Related Works Direct modelling of nonlinear dynamic systems using feed-forward Multilayer Perceptron (MLP) trained by backpropagation algorithm has been proposed by Narendra and Parthasarathi [11]. Chen et al. have studied the nonlinear system identification using the MLP neural networks [12]. Pao et al. have proposed single-layered functional link artificial neural network (FLANN) structure having low complexity compared to MLP where there are no hidden layers, as a result having less computation and faster convergence [13]. Direct modelling of nonlinear dynamic systems using feedback neural network is studied in [14]. In [15], Patra et al. have proposed trigonometric FLANN (TFLANN) for identification of nonlinear dynamic systems trained by back propagation (BP) algorithm. TFLANN is a single-layer structure where nonlinearity is introduced by expanding the input vector by trigonometric polynomials such as sin and cos and the output node contains a tan-hyperbolic nonlinearity [16–18]. The direct modelling of nonlinear dynamic systems using Legendre FLANN (LFLANN) is reported in [19– 21]. Patra and Kot [22] have proposed computationally efficient Chebyshev FLANN for nonlinear dynamic system identification trained by conventional BP algorithm. 1.3 Contributions In this present work, hybrid FLANN-DE is proposed for direct modelling of nonlinear dynamic systems and comparative analysis is carried out for different variants of FLANN structures such as Chebyshev FLANN (CFLANN), Legendre FLANN (LFLANN) and Trigonometric FLANN (TFLANN) in terms of performance, the speed of computation and accuracy of results. The rest of the paper is organized as follows: Sect. 2 presents the nonlinear dynamic system identification model. The proposed work is given in Sect. 3. Section 4 gives the simulation results and discussion. The conclusion is given in Sect. 5.

Performance Comparison of Variants of Hybrid FLANN-DE

139

2 Nonlinear Dynamic System Identification Model The direct modelling basically involves two major considerations: realization of structure and a technique to update the parameters of the unknown system [23]. Figure 1 presents the model for identification where the random input x(k) is applied to both the nonlinear system and the model. The structure of the model is considered as an adaptive filter. The output of the system y(k) is passed through some nonlinear function and is contaminated with the white Gaussian noise of known strength n(k). The actual system output subject to nonlinearity and noise is denoted as yn (k). The same input given to the model produces the estimated output denoted as yˆ (k). The error signal e(k) is computed as the difference between the actual system output and the estimated model output and is mathematically represented as e(k) = yn (k) − yˆ (k). Thus, the mean square error (MSE) which serves as the fitness function is computed which is to be minimized such that the estimated output of the model equals the desired system output. This is possible by continuously updating the parameters of the model using the bio-inspired training schemes such as DE. n(k) Nonlinear System

x(k)

Model

y(k) ∑

ŷ(k)

yn(k)

∑ e(k) Bio-inspired based update algorithm

Fig. 1. Identification of nonlinear dynamic system

3 Proposed Work The DE algorithm [24] is almost identical to GA but the differential mutation is performed prior to crossover operation. This algorithm is much faster compared to GA and its working cycle involves four basic operations namely, initialization, differential mutation, crossover and selection. The main parameters are size of population, scaling factor and probability of crossover. The identification of nonlinear dynamic channels is carried out using hybrid models comprising of CFLANN, LFLANN and TFLANN structures trained with DE algorithm. The FLANN model which is composed of single-layer of neurons has less computational complexity as compared to multilayer perceptron (MLP) which consists of additional

140

S. Swayamsiddha

hidden layers. Thus, FLANN model is preferred over MLP owing to its faster learning time [25]. In the CFLANN, LFLANN and TFLANN model, the Chebyshev, Legendre and Trigonometric polynomials are used for functional expansion of the input pattern. The generalized FLANN structure is shown in Fig. 2 where the input x(k) is expanded using Chebyshev, Legendre or Trigonometric polynomials expansion. The first few terms of Chebyshev expansion are:C0 (x) = 1.0,C1 (x) = x, C2 (x) = 2x2 − 1, similarly, the higher terms are computed using the recursive formula given by the Eq. (1).

1

h0(k)

yp(k)

x(k)

Functional Expansion

v1(k)

h1(k) ŷn(k)

h2(k)

v2(k)

∑

. . . . .

vn(k)

∑

e(k)

hn(k)

Adaptive Algorithm

Fig. 2. Generic FLANN structure

Cn+1 (x) = 2xCn (x) − Cn−1 (x)

(1)

The first few terms of Legendre expansion are: P0 (x) = 1.0, P1 (x) = x, P2 (x) = ∗ 3x2 − 1, similarly, the higher terms are computed using the recursive formula given by the Eq. (2). 1 2

(n + 1)Pn+1 (x) = (2n + 1)xPn (x) − nPn−1 (x) The trigonometric expansion terms are {1, cos(π n), sin(π n), cos(2π n), sin(2π n), . . . , cos(N π n), sin(N π n)}

(2) given

by

Performance Comparison of Variants of Hybrid FLANN-DE

141

4 Simulation Results This work simulates direct modelling using the proposed hybrid CFLANN-DE and also its performance is compared with that of TFLANN-DE and LFLANN-DE-based learning algorithms. The testing of the model is carried out by giving sinusoidal input to the identified model given by Eq. (3) x(n) = sin(2π n/250)∀n ≤ 250 = 0.8 sin(2π n/250) + 0.2 sin(2π n/25)∀n > 250

(3)

The parameters are chosen for simulation of FLANN-DE such as crossover ratio and scaling factor are CR = 0.5, F = 0.5. In this investigation, four dynamic system examples are considered for the direct modelling task and are represented by the following difference equations [15]: Example 1: yp (n + 1) = 0.3y(n) + 0.6y(n − 1) + g[x(n)]

(4)

The g[·] represents the nonlinearity which can take any form out of the following nonlinear functions: g1 (x) = 0.6 sin(π x) + 0.3 sin(3π x) + 0.1 sin(5π x) g2 (x) =

4x3 − 1.2x2 − 3x + 1.2 0.4x5 + 0.8x4 − 1.2x3 + 0.2x2 − 3

g3 (x) = 0.5 sin3 (π x) −

2 − 0.1 cos(4π x) + 1.125 x3 + 2

(5) (6) (7)

g4 (x) = x3 + 0.3x2 − 0.4x

(8)

  yp (n + 1) = f yp (n), yp (n − 1) + x(n)

(9)

Example 2:

where, f (y1 , y2 ) =

y1 y2 (y1 + 2.5)(y1 − 1.0) 1.0 + y12 + y22

(10)

Example 3:   yp (n + 1) = f yp (n) + g[x(n)]

(11)

where, f (y) =

y(y + 0.3) 1.0 + y2

(12)

142

S. Swayamsiddha

and g(x) = x(x + 0.8)(x − 0.5)

(13)

  yp (n + 1) = f y(n), y(n − 1), y(n − 2), x(n), x(n − 1)

(14)

 p1 p2 p3 p5 (p3 − 1.0) + p4  f p1 , p2 , p3 , p4 , p5 = 1.0 + p22 + p32

(15)

Example 4:

where

Figures 3, 4, 5, 6, 7, 8 and 9 show the MSE (Mean square error) learning curves for nonlinear dynamic system identification using three variants of FLANN such as CFLANN, LFLANN and TFLANN trained using DE algorithm. It is observed that TFLANN performs better than its counterparts in terms of optimality of solution and convergence speed. Table 1 presents the comparative analysis of LFLANN-DE, CFLANNDE and TFLANN-DE in terms of time elapsed and NMSE for identification of nonlinear dynamic systems. It is seen that the time elapsed is minimum for TFLANN-DE based identification for all the four types of systems as compared to LFLANN-DE and CFLANN-DE based identification. For instance, the time elapsed is 71.985 s using TFLANN-DE for Ex. 1 with nonlinearity NL-1, whereas it is 229.906 s using LFLANNDE and 75.781 s using CFLANN-DE for the same system. Moreover, the NMSE attained using TFLANN-DE is least, i.e. −40.5012 as compared to 3.7823 using LFLANN-DE and −12.9147 using CFLANN-DE for identification of system based on Ex. 1 with nonlinearity NL-1. Similarly, for other systems, the TFLANN-DE-based identification is better as compared to other variants of FLANN.

5 Conclusions Thus, in this work, the effectiveness of using hybrid FLANN-DE computing-based technique for nonlinear dynamic system identification is studied. The nonlinear dynamic system identification using hybrid models comprising of CFLANN-DE, LFLANN-DE and TFLANN-DE are proposed whose performances are compared with respect to optimality of solution and convergence speed. Owing to the less computational complexity and faster training FLANN-DE can be used for modelling real nonlinear dynamic processes for online applications. The faster and more accurate techniques for large-scale complex black-box optimization can be further incorporated into the direct modelling problem.

Performance Comparison of Variants of Hybrid FLANN-DE MSE Floor for Ex-1(NL-1) 0.2 LFLANN CFLANN TFLANN

0.18 0.16 0.14

MSE

0.12 0.1 0.08 0.06 0.04 0.02 0

0

50

100

150

250 300 200 No. of iterations

350

400

450

500

Fig. 3. MSE floor comparison for Ex-1(NL-1) MSE Floor for Ex-1(NL-2) 0.5 LFLANN CFLANN TFLANN

0.45 0.4 0.35

MSE

0.3 0.25 0.2 0.15 0.1 0.05 0

0

50

100

150

200 250 300 No. of iterations

350

400

Fig. 4. MSE floor comparison for Ex-1(NL-2)

450

500

143

S. Swayamsiddha MSE Floor for Ex-1(NL-3) 0.25 LFLANN CFLANN TFLANN 0.2

MSE

0.15

0.1

0.05

0

0

50

100

150

300 250 200 No. of iterations

350

400

450

500

Fig. 5. MSE floor comparison for Ex-1(NL-3) MSE Floor for Ex-1(NL-4) 0.35 LFLANN CFLANN TFLANN

0.3

0.25

0.2

MSE

144

0.15

0.1

0.05

0

0

50

100

150

200 250 300 No. of iterations

350

400

Fig. 6. MSE floor comparison for Ex-1(NL-4)

450

500

Performance Comparison of Variants of Hybrid FLANN-DE MSE Floor for Ex-2 0.04 LFLANN CFLANN TFLANN

0.035 0.03

MSE

0.025 0.02 0.015 0.01 0.005 0

0

50

100

150

300 250 200 No. of iterations

350

400

450

500

Fig. 7. MSE floor comparison for Ex-2 MSE Floor for Ex-3 0.35 LFLANN CFLANN TFLANN

0.3

0.25

MSE

0.2

0.15

0.1

0.05

0

0

50

100

150

200 250 300 No. of iterations

350

Fig. 8. MSE floor comparison for Ex-3

400

450

500

145

146

S. Swayamsiddha MSE Floor for Ex-4 0.4 LFLANN CFLANN TFLANN

0.35 0.3

MSE

0.25 0.2 0.15 0.1 0.05 0

0

50

100

150

200 300 250 No. of iterations

350

400

450

500

Fig. 9. MSE floor comparison for Ex-4 Table 1. Comparative analysis of LFLANN-DE, CFLANN-DE and TFLANN-DE in terms of time elapsed and NMSE TYPE OF SYSTEM

LFLANN Time elapsed(s)

EX-1

NL-1

229.906

NL-2

CFLANN NMSE 3.7823

Time elapsed(s) 75.781

TFLANN NMSE −12.9147

Time elapsed(s) 71.985

NMSE −40.5012

240.563

2.5369

88.594

−28.4574

86.171

−21.4481

EX-2

28.360

−27.3202

21.922

−29.9272

13.813

−13.9888

EX-3

92.203

−15.0277

92.953

−22.9634

58.609

−21.8552

EX-4

336.937

−15.0526

321.562

−17.2549

321.562

−13.1410

References 1. Widrow B, Stearns SD (1985) Adaptive signal processing. Prentice-Hall, Englewood Cliffs, NJ, pp 193–404 2. Haykin S (1996) Adaptive filter theory. Prentice-Hall Inc., Upper Saddle River, NJ 3. Sjoberg J, Zhang Q et al (1995) Nonlinear black-box modeling in system identification: a unified overview. Automatica 31(12):1691–1724 4. Guidorzi RP (1975) Canonical structure in the identification of multivariable systems. Automatica 11:361–374 5. Overbeek JMV, Ljung L (1989) On-Line structure selection for multivariable state space models. Automatica 18(5):529–543 6. Johansen, TA (2000) Multi-objective Identification of FIR models. Proc IFAC Symp Syst Identif. SYSID2000. 3:917–922

Performance Comparison of Variants of Hybrid FLANN-DE

147

7. Chen S, Billings SA (1989) Representation of non-linear systems: the NARMAX model. Int J Control 49:1013–1032 8. Price KV (1999) An introduction to differential evolution. In: Corne D, Dorigo M, Glover F (eds) New ideas in optimization. McGraw-Hill, London, pp 79–108 9. Swayamsiddha S, Behera S, Thethi H (2015) Blind identification of nonlinear MIMO system using differential evolution techniques and performance analysis of its variants. In: Proceedings of international conference on computational intelligence and networks. IEEE, pp 63–67 10. Swayamsiddha S, Mondal S, Thethi H (2013) Identification of nonlinear dynamic systems using differential evolution based update algorithms and chebyshev functional link artificial neural network. In: IET proceedings of the third international conference on computational intelligence and information technology, pp 508–513 11. Narendra KS, Parthasarathy K (1990) Identification and control of dynamical systems using neural networks. IEEE Trans Neural Netw 1(1):4–27 12. Chen S, Billings SA, Grant PM (1990) Nonlinear system identification using neural networks. Int J Contr 51(6):1191–1214 13. Pao YH, Phillips SM, Sobajic DJ (1992) Neural-net computing and intelligent control systems. Int J Contr 56(2):263–289 14. Hayakawa T, Haddad WM, Bailey JM, Hovakimyan N (2005) Passivity-based neural network adaptive output feedback control for nonlinear nonnegative dynamical systems. IEEE Trans Neural Netw 16(2):387–398 15. Patra JC, Pal RN, Chatterji BN, Panda G (1999) Identification of nonlinear dynamic systems using functional link artificial neural networks. IEEE Trans Syst Man Cybern Part B Cybern 29(2):254–262 16. Yamamoto Y (2008) Identification of nonlinear discrete-time systems using trigonometric polynomial neural networks. In: international conference on control, automation and systems 2008 in COEX. Seoul, Korea, pp 366–370 17. Yamamoto, Y (2010) Identification of nonlinear systems using a trigonometric polynomial neural network. In: Proceedings of the international conference on modelling, identification and control. Okayama, Japan, pp. 35–40 18. Norouzi M, Mansouri M, Teshnehlab M, Shoorehdeli MA (2011) A novel type of trigonometric neural network trained by extended Kalman Filter. Fourth International Workshop on Advanced Computational Intelligence, Wuhan, Hubei, China, pp 590–595 19. Paraskevopoulos PN (1985) Legendre series approach to identification and analysis of linear systems. IEEE Trans Autom Control 30(6):585–589 20. Patra JC, Meher PK, Chakraborty G (2008) Development of intelligent sensors using legendre functional-link artificial neural networks. In: IEEE international conference on systems, man and cybernetics. pp 1140–1145 21. Zongzhun Z, Ye Y, Yongji W, Fuqiang X (2010) Entry trajectory genetic algorithm optimization using legendre pseudospectral method. In: Proceedings of the international conference on modelling, identification and control. Okayama, Japan, pp 505–510 22. Patra JC, Kot AC (2002) Nonlinear dynamic system identification using Chebyshev functional link artificial neural networks. IEEE Trans Syst Man Cybern Part B Cybern 4:505–511 23. Yang L, Liu J, Yan R, Chen X (2019) Spline adaptive filter with arctangent-momentum strategy for nonlinear system identification. Sig Process 164:99–109 24. Storn R, Price K (1997) Differential evolution—a simple and efficient heuristic for global optimization over continuous spaces. J Global Optim 11:341–359 25. Lu L, Yu Y, Yang X, Wu W (2019) Time delay Chebyshev functional link artificial neural network. Neurocomputing 329:153–164

Load Cell and FSR-Based Hand-Assistive Device Acharya K. Aneesha1(B) , Somashekara Bhat2 , and M. Kanthi2 1 Department of Instrumentation and Control Engineering, Manipal Institute of Technology,

Manipal Academy of Higher Education, Manipal, Karnataka, India [email protected] 2 Department of Electronics & Communication Engineering, Manipal Institute of Technology, Manipal Academy of Higher Education, Manipal, Karnataka, India

Abstract. Human upper extremity has numerous functions in day-to-day life. Strokes lead to weakening in hand muscles because of which the patient is unable to hold an object properly. Stroke related problems are treated by physiotherapy and grip strength recovery is assessed by the devices such as Jamar dynamometer and pinch meter. In this paper, load cell and Force Sensing Resistor sensors are used which mimics the use of Jamar dynamometer and assesses the patient recovery quantitatively. The sensory unit consists of the load cell which is used to sense the data and give it to the Wheatstone bridge. Participant is asked to apply force on the developed hand dynamometer and the values of grip strength forces are recorded in the system. This device transmits the participant data wirelessly using the Bluetooth module. Load cell and Force Sensing Resistors (FSR) sensor data are stored in Rapid Miner and the graphs of grip strength are plotted. Thus, the developed prototype helps to determine the grip strength of a disabled person hand using low-cost devices and possible to compare the present and previous results of the assessment. The field of application of this tool is in physiotherapy and occupational therapy. Keywords: Strokes · Load cell · Jamar dynamometer · Pinch meter · Force sensing resistor sensor, Bluetooth

1 Introduction Human upper extremity has numerous functions in day-to-day life. Hands are used for expressing emotions and expressions, to gesture towards something, to take care of oneself and others, to defend oneself and for thermoregulation purposes, etc. Hand function can change as a result of time, treatment, or disease [1]. Hand function test is one of the components in hand assessment which simulates tasks of daily living. This assessment helps in defining the patient’s problem and is the foundation for selecting and directing treatment [2]. For example, Jebsen–Taylor Hand Function Test (JHFT) is a standardized outcome measure commonly used in clinical practice [3]. JHFT measures hand function based on timing criteria. The test includes unilateral activities that are designed to highlight various hand activities in day-to-day life. There are a total of 7 timed functional © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_15

Load Cell and FSR-Based Hand-Assistive Device

149

tasks which include writing a short sentence, turning over 3 × 5 inch cards, picking up small objects and placing them in a container, stacking checkers, simulated eating, moving empty large cans on a table and moving weighted large cans. Bohannon RW [4] had done research to obtain generated reference values for muscular force obtained with hand-held dynamometer and specific testing procedures. Here, gender, age and weight information were used to analyze the force measures. Reference information involved the regression equation. The multiple correlations associated with the equations favored the data obtained by other researchers who used different instrumentation. Peters et al. [5] work on the measures for isometric handgrip strength of fingers showed excellent test-retest reliability and construct validity for the load cell handle specially developed for this study, both for individuals with upper limb and hand dysfunctions and for healthy volunteers. Decostre et al. [6] performed an assessment on nine wheelchairbased hand patient. Their wrist extension and flexion are measured using developed MyoWrist dynamometer. The choice of test depends on the type of information required, the experience of the assessor, ease of conducting the test, portability and cost. There are many industrial countries which face a lot of problem because of disability in its society. This disability can be caused by a stroke which impacts human creativity and wellbeing. Ischemia or hemorrhage leads to cerebral vascular accidents which eventually causes a stroke. Many people suffer from stroke and around 10% of these people are under the retirement age. Around 65% of this total population survive but the majority have a residual disability with around 33% having a major disability in hand and limb [7]. Hemiplegia is a condition in which the patient has one stronger unimpaired hand and one weaker impaired hand. There are many other causes of disability like muscular dystrophy, arthritis and regional pain syndromes. The motor control is hampered and coordination of arm is difficult as a result of such disabilities. Attempts were made [8–10] to develop a portable dynamometer which is possible to use for the grip strength test. In this paper, to meet the handgrip strength analysis requirements, the sensory unit modelling has done in two approaches. The first is the load cell approach wherein a participant is made to hold a load cell and apply full strength on the grip structure. The second approach is to use a FSR on a cylindrical object. The main advantage of using a load cell is that the highest value of grip strength is calculated and the entire grip strength including the palm and the fingers is calculated, unlike FSR where only finger strength is calculated. The developed tool particularly helps in monitoring the process of recovery in grip strength thoroughly. This study also investigates relations between numerical parameters of load cell and FSR sensor data. The results that are obtained in this work will help the doctor save handgrip strength values accurately. This paper is divided into five sections. Section 1 gives a brief introduction to the work. The Sect. 2 part focuses on the method adopted and experimental protocol being developed. Section 3 discusses result analysis. The Sects. 4 and 5 part highlights on discuss and conclusions of the work, respectively.

2 Methods In this work, load cell and Force Sensing Resistor sensors (FSR) are used for monitoring the patient recovery. The load cell and the FSR sensors are connected to the Arduino

150

A. K. Aneesha et al.

Uno and Nano, respectively. Here, the participant is asked to apply the force on both these devices one by one and the values of these forces are recorded in the pc with the help of PLX-DAQ software. 2.1 Ethics Ethics Committee of Kasturba Hospital, Manipal (Registration No. ECR/146/Inst/KA/2013) has approved this study and has also approved using the developed device on healthy subjects. 2.2 Materials The hardware that the project utilizes is Wheatstone bridge, amplifier, microcontroller, load cell and two HS 05 Bluetooth modules. The strength test also requires a computing system for the therapist, a chair for the participant, and a table to place all the items listed above. 2.3 Experimental Protocol Participant is asked to hold the load cell setup in a comfortable position and apply the maximum possible strength. The participant’s dominant and non-dominant hand noted before the test session. Then load cell readings of both the arms are obtained as per the procedure and the average value is calculated. In this paper, FSR sensor also required which is used to determine the net force that the person will be applying on the FSR sensor. The sensing element is kept at the circumference of the can bottle. The FSR sensor is attached to Arduino Nano which is kept inside the can and the data will be transferred to the computer with the help of a Bluetooth module. Test involves unilateral tasks and performed with non-dominant hand first. Figure 1 shows a load cell based hand dynamometer which consists of hardware and software units. The main unit consists of three components: Wheatstone bridge, the amplifier and a microcontroller. The sensory unit consists of the load cell which is used to sense the data and give it to the Wheatstone bridge. This is the transducer element where the person with a despaired hand applies pressure on the dynamometer. The voltage displacement obtained from strain gauges bonded on the bending beam is measured using a full-bridge Wheatstone configuration. The Wheatstone bridge obtains small voltage. This voltage is amplified by the HX711 instrumentation amplifier and transferred to the computer using microcontroller. With repeated practice after obtaining a lot of data over a period of time, data analysis is done to obtain the amount of improvement that the patient has undergone. The transmission unit consists of a Bluetooth module that is used to transfer the data obtained on one system remotely to some other Personal Computer. This is done by using two Bluetooth modules by making one module from which the data is transferred as the slave and the other Bluetooth module which will obtain the data as the Master. Tera Term software is used to configure the Master Bluetooth module.

Load Cell and FSR-Based Hand-Assistive Device

151

Fig. 1. Block diagram of load cell set up as hand dynamometer

2.3.1 Calibration of Load cell In Fig. 2, equal weights are added one after the other on the load cell calibration setup. When the weights are being put on the load cell it will give a certain reading for a specific weight in grams. The up and down readings should match with one another. So that one can infer that the device will work correctly for any amount of force applied.

Fig. 2. Load cell calibration setup

Figure 3 shows the load cell setup which consists of a 40 kg load cell and a nonfunctional load cell which acts as a handle for support. Load cell should be held vertically with a hand resting on the table. Participant will apply the maximum effort to squeeze the load cell so that peak grip strength will be recorded as shown in Table 1. 2.3.2 Force-Sensitive Resistor sensor (FSR) The material used to manufacture flex sensors is resistive carbon elements. As a variable printed resistor, it attains a good form factor on a substrate that is thin and flexible. The bending of the substrate causes the sensor to generate resistance output which is related to the radius of the bend. Greater the radius, smaller is the resistance value. The change in deflection or bending of flex sensor causes a change in resistance. As shown in Fig. 4, change in resistance is translated into a voltage using potential divider network followed by a buffer stage. The buffer stage acts as an impedance matching network. The output voltage is usually passed to an ADC, which eventually converts the voltage into equivalent digital values.

152

A. K. Aneesha et al.

Fig. 3. Load cell setup

Table 1. Load cell values during the grip strength test Number of attempts

Force (Newton)

Trial 1

2.54

Trial 2

4.69

Trial 3

5.64

Average of three trial 4.29

Fig. 4. Basic flex sensor circuits

In this work, the components used are FSR sensors, Arduino Nano, battery and breadboard as shown in Fig. 5. Initially, pressure sensors (FSR) are connected to the Arduino Nano with the help of jumper wires then the supply to the Arduino is given by the computer. As the FSR sensor is pressed, the net amount of force that is exerted on the pressure sensor is displayed on the monitor. The software which is being used in this whole experiment is open-source Arduino software (IDE). Two FSR sensors of square 38.11 mm are used in this process. First FSR to sense the thumb finger force and another FSR to get pressure information from the index, middle and ring finger. When two different forces are exerted on this FSR sensor with the help of the fingers, the two

Load Cell and FSR-Based Hand-Assistive Device

153

different forces are added in the software. The net force is displayed in newton in the serial monitor.

Fig. 5. FSR sensor interface with Arduino Nano (Microcontroller)

Bluetooth transmission of FSR sensor data with baud rate 38,400 is shown in the serial monitor and stored in the Excel sheet with the help of PLX-DAQ as shown in Table 2. Figure 6 shown is the model of FSR sensors mounted on the CAN which is used for bottle holding tasks and monitoring the participant recovery. When the participant applies the force on the cylindrical object (CAN) which consists of FSR sensor at the circumference it gives the value of the force in newton’s which is recorded in the pc and is compared with the previous values of the participant to see whether the patient has recovered.

3 Results Data analytics of the force is done using Rapid Miner Studio. Rapid Minor is a data mining tool that is freely available on the internet. The force obtained from load cell and FSR sensor via Arduino are stored in MS Excel. The data from MS Excel is stored in Rapid Miner and the graphs are plotted by the data. The block diagram that was built on Rapid minor is used to obtain the maximum value of a data set of force obtained by two subjects shown in Fig. 7. The block diagram used as shown in Fig. 7 saves the highest value of the force. In the Rapid Miner GUI, the retrieved block obtains the data from Excel. Then the ‘SELECT ATTRIBUTE’ block is used to obtain only those attributes which are needed. The unnecessary attribute is selected and the ‘REVERSE SELECTION’ check box is checked. So only the useful attributes are selected. Then the ‘TRANSPOSE’ block is used which displays the table in transpose. Then in the ‘GENERATE ATTRIBUTES’ block, the maximum value of all the data is stored under the max sub-column. After

154

A. K. Aneesha et al. Table 2. FSR sensor data when interface with Arduino Nano

Force_in_Newton (FSR1)

Force_in_Newton (FSR2)

Net_Force_in_Newton

3

3

6

5

4

9

2

3

5

4

4

8

4

4

8

4

4

8

6

5

11

7

5

12

4

4

8

5

4

9

5

6

11

5

4

9

7

4

11

4

3

7

5

4

9

4

4

8

4

3

7

5

8

13

2

0

2

1

1

2

2

4

6

5

5

10

that in the ‘SELECT ATTRIBUTES’ block, we select only the max sub-column and it is displayed in the ‘UNION’ function. The online version of the volume will be available in LNCS Online. Members of institutes subscribing to the Lecture Notes in Computer Science (LNCS) series have access to all the pdfs of all the online publications. Non-subscribers can only read as far as the abstracts. If they try to go beyond this point, they are automatically asked, whether they would like to order the pdf, and are given instructions as to how to do so.

4 Discussion For a clinician, along with the therapeutic sessions, conducting a grip strength test is a time-consuming process. In a busy hospital environment, there may be chances of skipping outcome measurements such as the Jamar dynamometer test [11] which

Load Cell and FSR-Based Hand-Assistive Device

155

Fig. 6. FSR sensor placed on CAN which is used in bottle holding task

Fig. 7. Block diagram of Rapid Miner

deals with hand function. Successful rehabilitation of upper extremity depends on the reliability of the outcome of the hand function test. So, the selected test items in any hand function battery should be universally acceptable with good content validity. An attempt is made in selecting the test items which can be automated so that reliable and unbiased estimates of the grip strength are possible. The devices which are being used in the work have a very simple form of structure. Hand-assistive devices such as load cell and FSR are used for measuring the grip force of the participant. These devices are activated by applying the finger force on it and then the net force is displayed on the system. In the FSR sensor and load cell, data can be recorded automatically and

156

A. K. Aneesha et al.

analyzed whenever it is required. Wireless protocols such as Bluetooth is incorporated to minimize the connecting wires in set up and also records the data for future references. This could be used in the future for home-based telemetry and patient need not to go hospital every time for test sessions.

5 Conclusions Dynamometer is used for measurement of handgrip strength. Hand dynamometer widely used and considered as the gold standard is Jamar Hydraulic dynamometer. Total cost of dynamometers available in the market are high and difficult to use by the common man. The load cell and FSR devices that are used in this work have a very simple form of structure. Hence these devices have a very low cost as compared to other hand-assistive devices which is currently present in the market. These devices are capable of providing a huge contribution to the patient who wants to retain their actual hand function in terms of recovery assessment. Nevertheless, these kinds of devices still need some kind of improvement for being a better alternative as compared to other hand-assistive device. Acknowledgements. Authors are thankful to Kasturba Hospital, Manipal Institutional Ethics Committee for providing permission to study on healthy subjects.

References 1. James MH, Lawrence HS, Evelyn JM, Anne DC Rehabilitation of the hand. 2nd edn 2. Bear-Lehman J, Abreu BC (1989) Evaluating the hand: issues in reliability and validity. Phys Ther 69(12):1025–1033 3. Jebsen RH, Taylor N, Trieschmann RB, Trotter MJ, Howard LA (1969) An objective and standardized test of hand function. Arch Phys Med Rehabil 50:311–319 4. Bohannon RW (1993) Comparability of force measurements obtained with different strain gauge hand-held dynamometers. J Orthop Sports Phys Ther 18(4):564–567 5. Peters MJH et al (2011) Revised normative values for grip strength with the Jamar dynamometer. J Periph Nervous Syst 16(1):47–50 6. Decostre V et al (2015) Wrist flexion and extension torques measured by highly sensitive dynamometer in healthy subjects from 5 to 80 years. BMC Musculoskeletal Disorders 16(4) 7. Noh NM, Kadri NA, Usman J (2016) Development of arduino-based hand dynamometer assistive device. J. Mech. Med. Biol 16(3) 8. Fani S, Bianchi M, Jain S, Pimenta Neto SJ, Boege S, Bicchi A, Santello M (2016) Assessment of myoelectric controller performance and kinematic behavior of a novel soft synergy-inspired assistive device. Front Neuroro 10(11):1–15 9. Ambar R, Ahmad MS, Ali AMM, Jamil MMA (2011) Arduino based arm rehabilitation assistive device. Eng Technol 7:5–13 10. Wimer B, Dong RG, Welcome DE, Warren C, McDowell TW (2009) Development of a new dynamometer for measuring grip strength applied on a cylindrical handle. Med Eng Phys 31(6):695–704 11. Peters MJH et al (2011) Revised normative values for grip strength with the Jamar dynamometer. J Peripher Nerv Syst 16(1):47–50

Power Quality Analysis of a Distributed Generation System Using Unified Power Quality Conditioner Sarita Samal1 , Akansha Hota2 , Prakash Kumar Hota2 , and Prasanta Kumar Barik3(B) 1 School of EE, KIIT Deemed to be University, Bhubaneswar, Odisha, India

[email protected]

2 Department of EE, VSSUT, Burla, India

[email protected], [email protected] 3 Department of MEE, CAET, OUAT, Bhubaneswar, India [email protected]

Abstract. This paper deals with power quality profile analysis of distributed generation (DG) system using unified power quality conditioner (UPQC). Despite the several benefits of DG like excellent energy supply, reducing the expansion of power distribution system, environmentally friendly, and so on, there are several challenges existing due to the integration of DG with the grid or operating it in stand-alone mode. Power quality (PQ) issue is one of the main technical challenges in DG power system. In order to provide improved PQ of energy supply, it is necessary to analyze the harmonics distortion of the system as well as the voltage sag and swell. The UPQC has been extensively useful and it is verified to be the best solution to diminish this PQ issue. This paper explores the detail of PQ impacts in a DG (comprising of Solar PV and Fuel cell) system operates in stand-alone mode. The voltage sag compensation with current and voltage harmonics are estimated at varying load conditions with different control scheme like the synchronous reference frame (SRF) and modified SRF technique. The proposed model is developed in MATLAB/SIMULINKR and the result obtained validates the superiority of the proposed technique over others in terms of harmonics elimination and sag compensation. Keywords: Distributed generation · Power quality · Harmonics · Sag · MSRF

1 Introduction Distributed generation (DG) can be represented as a small-scale power system that contains loads, energy sources, energy storage units and control, and protection systems [1]. Using DG is more attractive as it improves the system quality, decreases the carbon emission and reduces the losses in transmission and distribution systems [2]. When DG is connected to the utility grid, the control systems required to maintain the active and reactive power output from the energy sources connected to DG is simple. However, under autonomous operation, the DG is disconnected from the utility grid and operates © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_16

158

S. Samal et al.

in islanded condition. Usually, a stand-alone DG system used to supply power to isolated areas or places interconnected to a weak grid. The application of the above DG other hand reduces the probability of energy supply scarcity. The proposed DG consists of renewable energy sources (RES) based power sources (i.e., solar PV, and fuel cell) and storage device as battery along with controllable loads [3, 4]. Solar PV is depended upon climatic conditions, hence to get an uninterrupted power supply at any time and maintaining the continuity of load current, one of the most developed energy sources like fuel cell is combined with these RES [5]. However, electric power system is mostly affected by nonlinear loads, mostly arc furnaces, power electronics converters, and household electronic equipment plays a key role in polluting the supply voltages and currents. The increase of power electronics-based equipment in household appliances and industries are the main cause of pollution of power system [6]. The research in the area of power electronics makes sure that a unified power quality conditioner (UPQC) plays a vital role in achieving superior power quality levels. In the present scenario, the series active power filters (APFs) and shunt APF normally termed as SAPF, alone do not meet the requirement for compensating the PQ distortions. A UPQC consists of two inverters integrated with the DC-link capacitor where the series APF is integrated though a series transformer and the shunt is through interfacing inductor. The series inverter acts as a voltage source whereas the shunt one is acts as a current source. Simultaneous compensation of voltage and current related PQ distortions using UPQC is achieved by proper controlling of series APF and shunt APF [7]. The shunt APF is employed for providing compensating currents to PCC for generation/absorption of reactive power and harmonics suppression. Moreover, the operation of SAPF is depended upon three main parts which are momentous in its design; these consist of the control method used for generation of reference current, a technique used for switching pulses generation for the inverter and the controller used for DC-link capacitor voltage regulation. Different control strategy is explained in the literature as follows. The use of SAPFs for current harmonic compensation typically in domestic, commercial, and industrial applications has been explained in Montero et al. [7]. The experimental study and simulation design of a SAPF for harmonics and reactive power compensation is explained by Jain et al. [8]. The power balance theory for active and reactive power compensation has developed by Singh et al. [9]. The instantaneous reactive power techniques of three-phase shunt active filter for compensation of source current harmonics have been explained by Akagi et al. [10]. Sag is the most significant PQ problem facing lots of industrial consumers. The control for such a case can be analyzed by protecting sensitive loads in order to preserve a load voltage without a sudden phase shift [11]. Different control strategies for series APF are analyzed by Benachaiba et al. [12] with importance on the reimbursement of voltage sags with phase jump. Different control techniques to reimburse voltage sags with phase jump are also projected and compared by Jowder et al. [13]. To ensure stable operation and improve the system performance of DG in island mode, a comparative study of two different control techniques used in UPQC like reference current generation, i.e., synchronous reference frame (SRF) method and modified synchronous reference frame (MSRF) method in conjunction with pulse width modulation based hysteresis band controller is proposed in this paper by using Matlab simulation software [7, 14]. The PQ

Power Quality Analysis of a Distributed Generation System

159

issues like voltage sag compensation, current and voltage harmonics were analyzed both at linear and nonlinear load are the main contribution of this paper.

2 Proposed System The projected DG system (comprising of solar and fuel cell based energy sources) is shown in Fig. 1 where DG system generates DC power to the DC bus and by using a power inverter this DC power is converted to AC. The AC bus delivers the power to the load which may be linear or nonlinear. The UPQC is located in between the DG and nonlinear load which manage the power quality of the system by using different control techniques.

DG

VI Measurement

VI Measurement

DC to AC inverter

Non linear Load

Interfacing Inductor 3-Phase Transformer

UPQC Series Active Filter

Control Scheme for Series Active Filter

Shunt Active Filter

Control Scheme for Shunt Active Filter

Fig. 1. Basic block diagram of DG with UPQC

2.1 Modeling of Solar PV A single diode model based PV cell is used for design of DG]. Figure 2 represents the single diode equivalent model of solar PV system. The basic equation for design of PV system is given below [15].     q × VPV + IPV × Rse −1 (1) IPV = NP × IPh − NP × IO exp Ns × AkT Figure 3 shows the MATLAB simulation of PV with MPPT and boost converter and Fig. 4 shows its corresponding output voltage where the required voltage of 230 V is achieved. The parameters required for the design of solar PV system is illustrated in Table 1 [16].

160

S. Samal et al. IL Ish

Series Resistance

Id

Voc

Shunt Resistance

IS

Fig. 2. Solar cell single diode model

Boost converter

25 Solar Panel

DC Output

Temp

400 Insolation MPPT

Fig. 3. Simulation of solar PV system

Fig. 4. Output voltage of boost converter

2.2 Modeling of Fuel Cell System Proton exchange membrane (PEM) fuel cell is considered as another energy source of the DG. The fuel cell consists of two electrodes, i.e., positive cathode, negative anode, and an electrolyte. The pressurized hydrogen gas enters as the anode of the fuel cell and oxygen enters the cathode. In a basic PEM fuel cell diagram is shown in Fig. 5, and its chemical reactions are shown in Eqs. (2–4). (2H+ + 2e− )→H2

(2)

(2H+ + 2e− + 1/2O2)→H2O

(3)

Power Quality Analysis of a Distributed Generation System

161

Table 1. Different parameters and their ratings to carry out the simulation work of solar PV Different parameters

Ratings

No. of Cells in series (NP )

72

Cells in parallel (Ns )

01

Short circuit current (Isc )

10.2 A

Open circuit voltage (Voc )

90.5 V

Voltage at maximum Power (Vmp ) 81.5 V Current at maximum Power (Imp )

8.6 A

Output voltage

230 V

Electric circuit

Hydrogen inlet

O2

H2

Oxygen inlet

Electrolyte

H2O Fig. 5. Fuel cell model

H2 + 1/2O2 → H2O

(4)

The simulation fuel cell with boost converter is shown in Fig. 6 and the output voltage which matches with the output voltage of other DGs is shown in Fig. 7. Table 2 represents different parameters of fuel cell. V+

PEM Fuel Cell

V-

Boost converter

+ Voltage Measurement

Scope

Fig. 6. Fuel cell with boost converter

162

S. Samal et al.

Fig. 7. Output voltage of boost converter

Table 2. Parameters of the Fuel cell Different parameters

Values

Load resistance (Rl )

5

Oxygen percentage in air (O2 )

59.3%

Each cell voltage (vs )

1.128 V

Cell resistance (Rt )

0.70833 

Number of cell (k)

65

Hydrogen percentage in fuel (H2 ) 99.56% Fuel cell voltage (Vfc )

230 V

2.2.1 Modeling of UPQC This chapter begins with system configuration and a detailed description of UPQC. The basic structure of UPQC is shown in Fig. 8 which consists of two inverters connected to a common DC-link capacitor. The series inverter is connected through a series transformer and the shunt inverter is connected in parallel with the point of common coupling. The series inverter acts as a voltage source whereas the shunt one is acts as a current source. The main function of UPQC is to control the power flow and reduce the harmonics distortion both in voltage and current waveform. The series APF topology is shown in Fig. 9. The series APF protects the load from the utility side disturbances. In case of series APF Park’s transformation method is used for generation of unit vector signal. A PWM generator, generating synchronized switching pulses, is given to the six switches of the series converter. Figure 10 shows the basic structure of shunt active filter. The shunt active power filter injects compensating current to the PCC such that the load current becomes harmonics free. The SAPF generates compensating current which is in opposition to the harmonic current generated by nonlinear load. This compensating current cancel out the current harmonics caused and makes the load current sinusoidal. So the SAPF is used to eradicate current harmonics and reimburse reactive power at the source side so as to make load current harmonics free.

Power Quality Analysis of a Distributed Generation System AC Source

VAPE

163

Nonlinear Load IC UPQC

Shunt APF

Series APF

Fig. 8. Basic UPQC system block diagram Transformer

Source Impedance

Nonlinear Load

Source

Active Filter

Fig. 9. Block diagram of series active filter

Non Linear Load IL2

Is

IL1 L

S

PCC

L

Linear Load

C

VSI

Fig. 10. Block diagram of shunt active filter

The Eqs. (5) and (6) show instantaneous current and the source voltage. Is(t) = IL(t) − IC(t)

(5)

Vs(t) = Vm sin ωt

(6)

164

S. Samal et al.

Fourier series method is used for expressing the nonlinear load current as shown in Eq. (7). Is(t) = I 1Sin(ωt + Φ1) +

ε 

InSin(nωt + Φn)

(7)

n=2

The compensation current of the active filter should be expressed by Ic(t) = IL(t) − Is(t)

(8)

Hence, for the exact compensation of reactive power and harmonics, it is essential to determine Is(t). The instantaneous value of source, load, and compensation current can be expressed by, Is(t), IL(t) & IC(t) where Vs(t) and Vm corresponds to instantaneous value and peak value of source voltage. 2.3 Control scheme of UPQC The MSRF controller scheme works in steady-state as well as in dynamic conditions exquisitely to manage the active, reactive power and reduce the harmonics in load current. The literature in review reveals that MSRF technique has much more advantages as compare to SRF scheme, so the authors have selected this control scheme for UPQC operation. The control scheme not uses the PLL circuit as used by SRF scheme, which makes the system more compatible and may be operated in load changing condition. The MSRF scheme with its control algorithm is given below. 2.3.1 Modified Synchronous Reference Frame (MSRF) Method Figure 11 shows the block diagram of modified SRF method for unit vector generation. The unit vector is generating by vector orientation method, not by PLL. Figure 12 shows the block diagram to generate a unit vector by sensing the supply voltage.” The unit vector generation is defining by the following equation. Vα   cosθ = √ (Vsα 2) + Vsβ 2 Vβ   sinθ = √ (Vsα 2) + Vsβ 2

(9) (10)

2.3.2 Hysteresis Band Current Controller Figure 13 shows the block diagram of hysteresis current regulator which generates the required pulses for inverter. In the current regulator, the error signal is generated by comparing the reference current I *sa and actual current I sa . The switching pulses required for the inverter is design in such a way that when the error signal goes beyond the upper band of hysteresis loop the lower switches of inverter are ON and upper switches are OFF and similarly the upper switches are ON and lower switches OFMatlab simulation software he lower band [17]. So the actual current is always tracked with respect to reference current inside the hysteresis band.

Power Quality Analysis of a Distributed Generation System

Vref

PI Controller

Vdc

Ila Ilb Ilc

Vsa Vsb Vs\c

165

abc

α-β d-q

α-β

d-q

LPF

α-β

Isa*

abc

Isb * Isc*

α-β

LPF

Unit Vector Fig. 11. Block diagram of MSRF method

w 1/S Vsa

abc

Vsb Vsc

α-β

÷

Cos

Vα Est. Mag.

w 1/S

Vβ

÷ Sin

Fig. 12. Unit vector generation block diagram Upper Band Reference Current

Hystersis Band isa* isa isb* isb isc* isc

Switching Analogy

S1 S2 S3 S4 S5 S6 VDC -VDC

Fig. 13. Hysteresis current controller scheme

Actual Current Lower Band

166

S. Samal et al.

3 Simulation Results and Discussion 3.1 Performance Analysis of DG Connected to Nonlinear Load With MSRF Based UPQC In this case, the system performance is analyzed by connecting nonlinear load with the DG system first without UPQC and then with MSRF based UPQC. The performance of series APF can be evaluated by introducing voltage sag into the system. The profile of load voltage shown in Fig. 14a conforms that voltage sag is introduced from 0.1 s to 0.3 s of the load voltage waveform. For sag condition, the series APF detects the voltage drop and inject the required voltage through the series coupling transformer. It maintains the rated voltage across the load terminal. In order to compensate for the load voltage sag, UPQC (employing MSRF scheme) is turned on, which injects compensating voltage at the PCC as displayed in Fig. 14b as a result the load voltage is the same as that of source voltage. The load voltage after compensation is shown in Fig. 14c. In general, the operation of the series part of the UPQC can be described as rapid detection of voltage variations at source and it injects the compensation voltage which maintains rated voltage across the load terminal. The shunt VSI in the UPQC is realized as shunt APF and is applied to solve the current related PQ distortions current harmonic distortion, reactive power demand, etc. In order to investigate the performance of shunt APF a rectifier based nonlinear load is introduced into the system and the level of harmonics is checked. It is observed from Fig. 15a that the source current waveform has a total harmonic distortion (THD) of 16.60% as per the FFT analysis of the source current shown in Fig. 15b. In order to make source current to be sinusoidal the shunt APF of the UPQC with conventional MSRF technique is turned on, at t = 0.1 s which injects compensating current as displayed in Fig. 15c. Hence, the THD level comes down to 2.54% as shown in Fig. 15d.

4 Conclusion The research reveals that MSRF technique of UPQC makes it possible for improving the power quality of a DG system connected with nonlinear load. The advantage of MSRF technique is that the production of sine and cosine angles for synchronization purpose instead of using PLL circuit it uses a basic unit vector generation scheme. The suggested method delivers superior output than the existing method in terms of harmonic mitigation and compensation of active and reactive power. In future, the work may be extended by integrating different control approaches for reference current generation of UPQC and DC-link voltage control algorithm may be implemented for better PQ mitigation purpose.

Power Quality Analysis of a Distributed Generation System

167

Fig. 14. Profile obtained under (sag compensation) a Load voltage before compensation, b Compensating voltage injected by UPQC, c Load voltage after compensation

168

S. Samal et al.

Fig. 15. Profile obtained under (Harmonics Mitigation) a Source current before compensation, b Harmonics content before compensation, c Compensating current injected by UPQC, d Harmonics content after compensation

Power Quality Analysis of a Distributed Generation System

169

References 1. Badoni M, Singh B, Singh A (2017) Implementation of echo-state network-based control for power quality improvement. IEEE Trans Industr Electron 64(7):5576–5584 2. Mahmoud MS, Rahman MSU, Fouad MS (2015) Review of microgrid architectures–a system of systems perspective. IET Renew Power Gener 9(8):1064–1078 3. Samal S, Hota PK (2017) Design and analysis of solar PV-fuel cell and wind energy based microgrid system for power quality improvement. Cogent Engineering. 4(1):1402453 4. Suresh M, Patnaik SS, Suresh Y, Panda AK (2011) Comparison of two compensation control strategies for shunt active power filter in three-phase four-wire system. In: Innovative smart grid technologies (ISGT), IEEE PES, pp. 1–6 5. Tang Y, Loh PC, Wang P, Choo FH, Gao F, Blaabjerg F (2012) Generalized design of high performance shunt active power filter with output LCL filter. IEEE Trans Industr Electron 59(3):1443–1452 6. Hosseinpour M, Yazdian A, Mohamadian M, Kazempour J (2008) Desing and simulation of UPQC to improve power quality and transfer wind energy to grid. J Appl Sci 8(21):3770–3782 7. Xu Q, Hu X, Wang P, Xiao J, Tu P, Wen C, Lee MY (2017) A decentralized dynamic power sharing strategy for hybrid energy storage system in autonomous DC microgrid. IEEE Trans Industr Electron 64(7):5930–5941 8. Jain SK, Agarwal P, Gupta HO (2003) Simulation and experimental investigations on a shunt active power filter for harmonics and reactive power compensation. IETE Tech Rev 20(6):481– 492 9. Samal S, Hota PK (2017) Power quality improvement by solar photo-voltaic/fuel cell integrated system using unified power quality conditioner. Int J Renew Energy Res (IJRER) 7(4):2075–2084 10. Dixon JW, Venegas G, Moran LA (1997) A series active power filter based on a sinusoidal current-controlled voltage-source inverter. IEEE Trans Industr Electron 44(5):612–620 11. Ferdi B, Benachaiba C, Dib S, Dehini R (2010) Adaptive PI control of dynamic voltage restorer using fuzzy logic. J Electr Eng Theory Appl 1(3) 12. Samal S, Hota PK (2017) Power quality improvement by solar photo-voltaic/wind energy integrated system using unified power quality conditioner. Int J Power Electron Drive Syst 8(3):14–24 13. Akagi H, Kanazawa Y, Nabae A (1984) Instantaneous reactive power compensators comprising switching devices without energy storage components. IEEE Trans Industr Electron Appl 20(3):625–630 14. Hatziargyriou CS, Liang T (2014) The microgrids concept, microgrid: architectures and control. Wiley-IEEE Press, Chichester, pp 1–24 15. Olivares DE et al (2014) Trends in microgrid control. IEEE Trans Smart Grid 5(4):1905–1919 16. Altas IH, Sharaf AM (2007) A photovoltaic array simulation model for Matlab-Simulink GUI environment. In; Clean Electrical Power, ICCEP’07, 341–345 17. Karuppanan P, Mahapatr, KK (2010) A novel control strategy based shunt APLC for power quality improvements. In: IEEE International Conference on Power, Control and Embedded, systems (ICPCES), pp 1–6

A Hybrid: Biogeography-Based Optimization-Differential Evolution Algorithm Based Transient Stability Analysis P. K. Dhal(B) Department of Electrical and Electronics Engineering, Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology, Chennai, India [email protected]

Abstract. A hybrid optimization technique is used to improve the stability and voltage profile in multi-machine systems. The hybrid biogeography-based optimization (BBO)-differential evolutionary (DE) algorithm application is to reduce the system loss and the voltage profile and stability increases when the devices are tuned by hybrid BBO-DE technique. It works using the eigen value based objective function to tune the parameters of the static var compensator (SVC) and power system stabilizer (PSS). In this research paper, eigen value grounded objective function is practiced to gain stability. Many optimization techniques are used to attain a solution to tune the parameters or to place the device in a better location. Here, a hybrid optimization technique is used to tune the parameters of the SVC and PSS after clearing three-phase fault. Keywords: Hybrid optimization · SVC · PSS · Transient stability

1 Introduction The evolutional hybrid computational algorithm is used in the power system to get a better solution. One of the device, i.e. FACTS are an idea primarily constructed on powerelectronic controllers which improve the worth of conduction networks by growing the usage of their capability. As these controllers function very quickly, the power system networks must be enlarged the secure working bounds of a conduction system without risking stability [1, 2]. For assured position, the period of the FACTS was triggered by the event of recent solid-state electrical switching units. Progressively, the usage of the FACTS has given a specified rise to new controllable techniques. The very quick energy controllability in FACTS programs prepared them, applicants, for particular purposes in back-to-back formations to regulate the facility alternate between the networks they related. The speedy management of energy directed to the added use of FACTS for augmenting the transient stability of linked programs by way of active power damping. The enhancement in stability was achieved by including auxiliary indicators within the present controllers of the converters [3, 4]. There are various kinds of FACTS devices that are used worldwide. But, in this research work, static VAR compensator is used © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_17

A Hybrid: Biogeography-Based Optimization-Differential ...

171

to analyze transient stability. Increase the consistency of AC grids, reducing power supply costs, supplying inductive or capacitive reactive power to the grid are some advantages of FACTS devices. Limited capacity of power can be led over a transmission line. Conductors and equipment may be damaged by overheating if too much current is drawn. The hybrid algorithm has been used in nine bus system to identify the weak bus. In this weak bus, the SVC controller is introduced and tuned with a hybrid algorithm. It has produced better results which has discussed in the result and discussion section. Here the author has been organized in this paper as like (1) introduction, (2) proposed objective and constraints, (3) hybrid based flow chart, (4) convergence characteristics of hybrid algorithm and (5) results and discussions.

2 Proposed Objective Functions and Constraints One is the minimization of the real part of the eigen value and the other one is the maximization of the damping ratio. J1 = Min · (σ0 − σi )2

(1)

J2 = Max · (ζ0 − ζi )2

(2)

The united objective function J = J 1 + αJ 2 is used to have a closed-loop eigen values. The value of α is deliberated as 7 when subjected to (K min < K < K max , T min < T < T max ) of the SVC and PSS(Kw, pss (min) < Kw, pss < Kw, pss(max), T 1, pss(min) < T 1, pss < T 1, pss(max). Here J 1 = first part of objective function, J 2 = second part of objective function, σ0 = 0th eigen value, σi = ith eigen value, ζ0 = 0th damping ratio, ζi = ith damping ratio, α = constant, k min = minimum gain value of the SVC, k max = maximum gain value of the SVC, T min = minimum time constant of the SVC, T max = maximum time constant of the SVC, K w,pss (min) = minimum gain value of the PSS, K w, pss (max) = maximum gain value of the PSS, T 1, pss(min) = minimum time constant of the PSS and T 1, pss(max) = maximum time constant of the PSS. 2.1 Analysis of Constraints Voltage magnitude limits-Too high or too low voltages could origin problems in the power network with respect to end-user power device damage or unpredictability in the ≤ V i ≤ V max , where i = 1 … N, power system [5, 6]. It is shown in Eq. (3). V min i i V − the magnitude of the voltage

(3)

Generator active power limits-The active power of a generator i is defined to be the ≤ Pi ≤ Pmax, where real part of the complex variable. It is shown in Eq. (4). Pmin i i i = 1 . . . N , P = real power

(4)

Generator reactive power limits-The reactive power of a generator i is defined to be ≤ Qi ≤ Qmax , the real part of the complex variable. It is shown in Eq. (5). Qmin i i Where i = 1 . . . N , Q = reactive power

(5)

172

P. K. Dhal

3 Hybrid Based Flow Chart See Fig. 1.

Fig. 1. Flow chart of hybrid algorithm

A Hybrid: Biogeography-Based Optimization-Differential ...

173

4 Convergence Characteristics of BBO-DE Algorithm The characteristics graph between generation and fitness value for WSCC nine bus system. The best fitness value of the population meets after 30 objective function evaluations [7, 8]. It will produce the optimal value after this evaluation. It is shown in Fig. 2.

Fig. 2. Convergence characteristics of BBO for WSCC nine bus systems

The best fitness value of the population meets after 15 objective functions evaluations. It will produce the optimal value after this evaluation. It is shown in Fig. 3.

Fig. 3. Convergence characteristics of DE for WSCC nine bus systems

4.1 Boundaries and Optimal Solution After running the BBO-DE algorithm in the background of MATLAB, the BBO-DE solutions are obtained. Based on the required VAR ratings, the boundaries are selected from 0.01 to 100 for gain value and from 0.01 to 10 for the time constant [9, 10]. It is shown in Table 1. The optimal solution of BBO-DE for 9 bus system is shown in Table 2. Here, the optimal solution of the gain is 2.18 and the optimal solution of the time constant is 0.03.

174

P. K. Dhal Table 1. Boundaries of SVC for WSCC nine bus system Boundaries

Gain value of SVC Time constant of SVC

Lower bound 0.01

0.01

Upper bound

10

100

Table 2. optimal solution of SVC for WSCC nine bus system Gainl Time constant 2.18

0.03

5 Results and Discussions The time-domain simulations are completed in the PSAT software program which is practiced to calculate and plot the graphs of the system. The efficiency of the SVC is evaluated via the test system. A 9 bus system (WSCC—Western Science Coordinated Council) with six transmission lines, three generators, three loads and a local load D is taken into account to review. The WSCC nine bus system is presented in Fig. 4.

Fig. 4. WSCC nine bus systems

A Hybrid: Biogeography-Based Optimization-Differential ...

175

Fig. 5. Voltage profiles of WSCC nine bus systems without SVC

Fig. 6. Real power profile of WSCC nine bus system without SVC

Fig. 7. Reactive power profile of WSCC nine bus system without SVC

When a fault is applied at bus 6 for 0.10 s, the system’s voltage profile is decreased. Here, system stability is not affected. The voltage magnitude profile of WSCC nine bus systems without SVC is shown in Fig. 5. The real power and reactive power profiles of WSCC nine bus systems without SVC are shown in Figs. 6 and 7. The voltage waveforms of bus 1 to bus 9 of the WSCC nine bus system without SVC are presented in Fig. 6.5. The voltage magnitudes from bus1 to bus 9 of the WSCC nine bus system without SVC are 1.040,1.025,1.025,0.979,0.917,0.934,0.992,0.965 and 1.002 p.u., respectively, when the system undergoes fault (Fig. 8). The system parameters of the WSCC nine bus systems are presented in Table 3.

176

P. K. Dhal

Fig. 8. Voltage waveforms of WSCC nine bus systems without SVC

Table 3. System parameters of WSCC nine bus system without SVC Bus

V (p.u.)

Phase (rad) P Gen (p.u.) Q Gen (p.u.)

Bus 1 1.040000 12.028000

4.624880

2.015690

Bus 2 1.025000 11.874000

1.630000

0.624320

Bus 3 1.025000 11.796000

0.850000

0.425580

Bus 4 0.979000 11.857000

0.000000

0.000000

Bus 5 0.917000 11.711000

0.000000

0.000000

Bus 6 0.934000 11.720000

0.000000

0.000000

Bus 7 0.992000 11.774000

0.000000

0.000000

Bus 8 0.965000 11.686000

0.000000

0.000000

Bus 9 1.002000 11.747000

0.000000

0.000000

Fig. 9. Voltage profile of WSCC nine bus system with SVC using BBO-DE

If SVC is applied at bus 6, the system’s voltage profile is improved and it reduces the losses. The voltage profiles, real, reactive power profile of the system with tuned SVC are presented in Figs. 9, 10 and 11. The voltage profile of the system has been improved.

A Hybrid: Biogeography-Based Optimization-Differential ...

177

Fig. 10. Real power profile of WSCC nine bus system with SVC using BBO-DE

Fig. 11. Reactive power profile of WSCC nine bus system with SVC using BBO-DE

Fig. 12. Voltage waveforms of WSCC nine bus systems with SVC

The voltage waveforms of bus 1 to bus 9 of the WSCC nine bus system with SVC are presented in Fig. 6.10. The voltage magnitudes of the bus 1, bus 2, bus 3, bus 4, bus 5, bus 6, bus 7, bus 8 and bus 9 of the WSCC nine bus system with SVC are 1.040,1.025,1.025,1.003,0.939,1.000,1.000,0.977 and 1.017 p.u., respectively when the system undergoes fault. Here, the voltage profile is improved when compared to the system without SVC (Fig. 12). The system parameters of the WSCC nine bus systems with SVC are presented in Table 4. The voltage profile of the system with SVC (BBO-DE) is better when compared to without SVC. The comparison of the voltage profile is given in Table 5. Here, the tuned SVC by BBO-DE increased the voltage profile when compared to without SVC.

178

P. K. Dhal Table 4. System parameters of WSCC nine bus system with SVC using BBO-DE Bus

V (p.u.)

Phase (rad) P Gen (p.u.) Q Gen (p.u.)

Bus 1 1.040000 8.849000

4.614000

1.563000

Bus 2 1.025000 8.709000

1.630000

0.487000

Bus 3 1.025000 8.631000

0.850000

0.156000

Bus 4 1.003000 8.682000

0.000000

0.000000

Bus 5 0.939000 8.544000

0.000000

0.000000

Bus 6 1.000000 8.549000

0.000000

0.657000

Bus 7 1.000000 8.610000

0.000000

0.000000

Bus 8 0.977000 8.524000

0.000000

0.200000

Bus 9 1.017000 8.584000

0.000000

0.000000

Table 5. shows the comparison of the voltage profile of WSCC nine bus systems Bus

Without SVC (p.u)

With SVC (BBO-DE) (p.u)

Bus 1

1.0400628

1.0400037

Bus 2

1.0250433

1.0250389

Bus 3

1.0250508

1.0250278

Bus 4

0.9788825

1.0034483

Bus 5

0.9172847

0.9386596

Bus 6

0.9342051

1.0000109

Bus 7

0.9919748

1.0003078

Bus 8

0.9650012

0.9769005

Bus 9

1.0019088

1.0172907

Table 6. Total Generation and losses of WSCC nine bus system Total generation

Without SVC (p.u) With SVC-(BBO-DE) (p.u)

Real power (p.u.)

7.1048778

7.0937549

Reactive power (p.u.) 3.0655817

2.8622587

Total load Real power (p.u.)

7

Reactive power (p.u.) 2.8

7 2.8

Total loss Real power (p.u.)

0.1048778

0.0937549

Reactive power (p.u.) 0.2655817

0.0622587

A Hybrid: Biogeography-Based Optimization-Differential ...

179

The generation and losses of WSCC nine bus systems are given in Table 6. Here, the tuned SVC reduces the losses.

6 Conclusion A hybrid approach to boost the steadiness of the system is studied. In this study, BBO-DE algorithm is used to catch suitable parameters to stabilize the bus voltages and system. Through time-domain simulations and comparative results, it has been found that power loss minimization and the augmentation of voltage profile is achieved by suitable tuning of parameters in standard power system networks. In WSCC nine bus systems, the system with SVC using BBO-DE has improved the magnitude of the voltage from 0.9172847 to 0.9386596 p.u in bus 5 and it has reduced the reactive power loss from 0.2655817 to 0.0622587 p.u when compared to the system without SVC.

References 1. Zhang X, Kang Q, Cheng J, Wang X (2018) A novel hybrid algorithm based on biogeographybased optimization and grey wolf optimizer. Appl Soft Comput 67:197–214 2. Guesmi T (2018) Improvement of power system stability using BBO algorithm. Int J Eng Res Technol 7(2):134–139 3. Jadon SS, Tiwari R, Sharma H, Bansal JC (2017) Hybrid artificial bee colony algorithm with differential evolution. Appl Soft Comput 58:11–24 4. Adachi T, Yokoyama A (2016) Improvement of small signal stability of power system by controlling doubly fed induction generators of a large-capacity wind farm. J Int Council Electr Eng 6(1)–117–125 5. Sun H, Zhou F, Wang Y (2016) Simulation and Control Strategy of a 5.6 kV 17level STATCOM Under SVG Condition. J Control Measure Electron Comput Commun 57(4):893–901 6. Rout UK, Sahu RK, Panda S (2013) Design and analysis of differential evolution algorithm based automatic generation control for interconnected power system. Ain Shams Eng J 4(3):409–421 7. Mirjalili S, Mirjalili SM, Hatamlou A (2016) Multi-verse optimizer: a nature-inspired algorithm for global optimization. Neural Comput Appl 27(2):495–513 8. Samrajyam K, Prakash RBR (2012) Optimal location of STATCOM for reducing voltage fluctuations. Int J Modern Eng Res (IJMER), 2(3):834–839 9. Abd-Elazim SM, Ali ES (2016) Optimal location of STATCOM in multimachine power system for increasing loadability by Cuckoo Search algorithm. Int J Electri Power Energy Syst 80:240–251 10. Hemmati R, Boroujeni SMS, Behzadipour E, Delafkar H (2011) Power system dynamic stability enhancement using a new PID type PSS. Australian J Basic Appl Sci 5(5):721–727

Restraining Voltage Fluctuations in Distribution System with Jaya Algorithm-Optimized Electric Spring K. Keerthi Deepika1(B) , J. Vijayakumar2 , and Gattu Kesava Rao3 1 Department of Electrical Engineering, VIIT, Duvvada, Visakhapatnam, Andhra Pradesh, India

[email protected]

2 Departtment of EEE, ANITS, Visakhapatnam, Andhra Pradesh, India

[email protected]

3 Department of EEE, KLEF, Vaddeswaram, Andhra Pradesh, India

[email protected]

Abstract. With the widespread development of green technologies like wind, photovoltaic and other renewable energy sources into the distribution network, voltage stability problem has gained prominence. Electric spring, a new power electronic-based voltage regulating device can effectively maintain voltage constant at the critical loads. This is done by coordinating the load demand to track the power generation source. In account of the voltage fluctuations caused by the renewable power source, this paper deals with optimization of gains of the PI controller using Jaya algorithm proposed. Simulations carried out demonstrate that the adaptive PI-based ES restrains the voltage fluctuations in reduced settling time and less peak overshoot. Keywords: Jaya algorithm · Electric spring · PI control

1 Introduction Dispersed distributed generation in the distribution network causes frequency and voltage instability and voltage fluctuations due to the intermittent power of the renewable sources [1]. Possible solutions offered to these problems are increase in rating of the DG unit, deployment of storage units and demand side management. Former solutions involve in increased investment and operational cost. In demand side management, the load is made to follow the generation [2, 3]. Electric springs (ESs) proposed in [4] are designed to provide voltage support at the critical load points. Dynamic modelling of ES is detailed in [5], and ES is successfully implemented for voltage stability [6], reduction of battery storage [7]. It is also implemented in microgrids [8] and nanogrids [9]. In all these, PI controller parameters are fixed. But for varying conditions, performance of ES can be enhanced by making PI controller to be adaptive [10]. © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_18

Restraining Voltage Fluctuations in Distribution System

181

The gain constants of PI controller, K p, K I , for ES are optimized using a linearized mathematical model. Such operations are iterative, time-consuming and may not yield satisfactory response. Thus, heuristic algorithms are applied for tuning of controller parameters. Jaya algorithm proposed in [11] is advantageous as it does not require any algorithm specific parameters. In [12], tuning of a PID controller for automatic generation control is performed by Jaya algorithm. Objective is to minimize the peak overshoots and settling time in the output, when input signal contains noise elements. In [13], Jaya algorithm is modified with weight parameters assigned by fuzzy logic. It is then applied to online tuning of parameters for load frequency control in a smart grid with wind source. Superior operating performance was claimed in terms of faster convergence and reduced computational burden. Performance analysis of a PV-fed DSTATCOM with Jaya algorithm is compared with grenade explosion method and TLBO method in [14]. Works elaborated in the remaining sections in this paper are as follows. Section 2 describes the operation and control of electric spring. Section 3 gives the outline of Jaya algorithm followed by its implementation to the test system in Sect. 4. Finally, concluding remarks are mentioned in Sect. 5.

2 Electric Springs Electric spring is a capacitor, and when embedded within a non-sensitive load, it forms a smart load. As per the variations in voltage at the point of connection of the nonsensitive load and critical load, the overall operation of a smart load can be analyzed in three modes as illustrated through Fig. 1. Voltage across capacitor C f is meticulously controlled according to the control circuit detailed in Fig. 3 (Fig. 2).

Fig. 1. Operation of ES a neutral b voltage boosting c voltage suppression

3 Adaptive Electric Spring with Jaya Algorithm The Jaya algorithm may be implemented to reach the maximum or minimum of an objective function, f with d control variables. Amongst the considered initial population,

182

K. Keerthi Deepika et al. Line impedence

R1

L1

Vs Lf

+ Ves VG

INC +

VNC -

Cf

PWM Power Inverter

Controller

-

+

Electric Spring NonCriƟcal Load

Vs_ref CriƟcal Load

Fig. 2. Electric spring in a distribution network

Fig. 3. Block diagram of controller of electric spring

best and worst solutions are represented by indices b and w, respectively. m, n, k are the iteration count, variable position and candidate solution. In mth iteration, A(m, n, k) represents the nth variable of kth candidate. STEP 1: Initialize population size, variable size & termination condition. STEP 2: Amongst the population available, determine the best solution & worst solution. STEP 3: Modify the solution as per the identified in step 2: A(m+1,n,k)=A(m,n,k)+r(m,n,1){A(m,n,b)-|A(m,n,k)|}-r(m,n,2){A(m,n,w)|A(m,n,k)|} r(m,n,1), r(m,n,2) are randomly generated within [0,1]. STEP 4: Obtain solutions with A(m+1,n+k) and A(m,n,k). Identify the better one. STEP 5: If solution with A(m+1,n+k) is better, then accept & replace Best solution with this previous solution. STEP 6: If NO, retain previous solution. STEP 7: Verify if the termination condition satisfied? STEP 8: If YES, finalize it as optimum solution. STEP 9: If NO, repeat solution process from STEP 2-8.

Restraining Voltage Fluctuations in Distribution System

183

In this paper, Jaya algorithm is instigated to reach the finest K p and K i values of the controller to electric spring. Minimization of the mean square error in iter number of iterations is used as the objective function: MSE =

iter 

e2

i=1

where e =

    Kp∗ − Kp + Ki∗ − Ki iter

Jaya algorithm-based adaptive PI controlled electric spring uses lower rating of switches, operates at reduced DC link voltage value with better compensation and exhibits faster convergence to optimum solution. Thus, adaptive PI controller for ES is implemented on the system with the parameters as shown in Table 1. Number of populations = 50 and number of control variables = 2 in Jaya algorithm. Table 1. System parameters Items

Values

Regulated mains voltage (Vc ) 220 V DC bus voltage (Vdc )

400 V

Carrier amplitude (Vtri )

1V

Line resistance (R1 )

0.1 

Line inductance (L1 )

1.22 mH

Critical load(R)

53 

Non-critical load (Ra )

50.5 

Inductance of filter(Lf )

0.25 mH

Capacitance of filter (Cf )

1.32 µH

Switching frequency (fs )

5 kHz

4 Results and Discussion Tuning of PI controller of electric spring with the considered algorithm is illustrated schematically in Fig. 4. To validate the adaptive nature of the controller, a system is established, with the values given in Table 1. DG output power is imitated by a prerecorded programmable voltage source, and both the critical and non-critical loads are assumed of linear type. DG has strong fluctuation which has great influence on the voltage of active distribution network. As shown in Fig. 4, the voltage of DG is 220 V from 0 to 0.5 s. At 0.5 s,

184

K. Keerthi Deepika et al.

V JAYA ALGORITHM

ref

IniƟal parameters

Design of Kp, kI, Kd

Vs PWM GATE DRIVE

V

PWM INVERTER

DC

PID

LC FILTER

NON CRTICAL LOAD

Voltage (in Volts)

Fig. 4. ES control circuit with Jaya algorithm Voltage of Distributed Generation Source

270 260 250 240 230 220 210 200 190 180 170 0

0.5

1

1.5

2

2.5

Time (in sec)

Fig. 5. Voltage of distributed generation source

Voltage (in volts)

the voltage falls from 220 V to 180 V, and it continues to till 1 s; i.e., there is a voltage sag of about of 60 V from 0.5 to 1 s (Fig. 5). As we know that, real power (P) α square of Voltage (V 2 ), variations in real power are similar to that in voltage as shown in Fig. 6 and correspondingly real power in Fig. 7. Voltage across Non-critical Load

270 260 250 240 230 220 210 200 190 180 170 0

0.5

1.5

1 Time (in Sec)

Fig. 6. Voltage across non-critical load

2

2.5

Restraining Voltage Fluctuations in Distribution System

185

Real Power of Non-Critical Load

12000

Power (in Watts)

11000 10000 9000 8000 7000 6000 5000 0

0.5

1.5

1

2

2.5

Time (in Sec)

Fig. 7. Real power of non-critical load

Reactive Power (in VAr)

Non-critical load is the one which can withstand the voltage fluctuations caused by DG. The voltage graph, real and reactive power graphs of the non-critical load are shown in Figs. 6, 7 and 8 respectively. As the non-critical load can withstand the fluctuations triggered by DG, active and reactive powers absorbed by the non-critical load are comparable to that of DG. 850 800 750 700 650 600 550 500 450 400 350

Reactive Power of Non-Critical Load

0

0.5

1

1.5

2

2.5

Time (in Sec)

Fig. 8. Reactive power of non-critical load

The smart load ensures the line voltage and line current to be essentially constant. From Fig. 9, it is clear that from 0 to 0.5 s, ES is not activated as the voltage across DG is constant at 220 V. But at 0.5 s, ES is activated to compensate the voltage sag. ES boosts up the line voltage again to 220 V (Fig. 10). During sag condition, i.e. from 0.5 to 1 s, ES injects the real power of about 1000 W into the system and absorbs the reactive power of 1000 Var from the system; i.e., ES acts in inductive mode. During the swell condition, i.e. from 1.5 to 2 s, ES absorbs the real power of 1000 W from the system and delivers the reactive power of 1000 Var into the system; i.e., ES acts in capacitive mode (Fig. 11). Figure 12 shows the RMS value of the line voltage with PI controller and adaptive PI controller technique. At 0.5 s, a voltage sag of 60 V is simulated and the ES with

Voltage (in Volts)

186

K. Keerthi Deepika et al.

50 40 30 20 10 0 -10 -20 -30 -40 -50

Voltage across Electric Spring

1.5

1

0.5

0

2.5

2

Time (in Sec)

Fig. 9. Voltage appearing at ES terminals Instantaneous Voltage across ES

50 40 30 20 10 0 -10 -20 -30 -40 -50 0

0.5

1

1.5

2

2.5

Time (in Sec)

Power (in Watts)

Fig. 10. Instantaneous voltage appearing at ES terminals Real power and Reactive power of Eelctric spring

7000 6000 5000 4000 3000 2000 1000 0 -1000 -2000

Reactive Power Real Power

0

0.5

1

1.5

2

2.5

Time(in secs)

Fig. 11. Powers absorbed by electric spring

controller reinstates the voltage to 220 V. At t = 1.5 s, voltage swell is simulated and the ES with controller restores the system to initial voltage condition, i.e. 220 V. The aboveexplained adaptive capability of ES is supported with the instantaneous line voltage, and instantaneous line current is illustrated in Fig. 13 and Fig. 14, respectively.

Restraining Voltage Fluctuations in Distribution System RMS value of Line voltage

240

PI Controller

235

Voltage (in volts)

187

Adaptive PI Controller

230 225 220 215 210 205 200 0

0.5

1.5

1

2

2.5

2

2.5

2

2.5

Time (in sec)

Voltage (in volts)

Fig. 12. RMS value of line voltage Intantaneous line voltage

250 200 150 100 50 0 -50 -100 -150 -200 -250 0

0.5

1.5

1

Time (in sec)

Fig. 13. Instantaneous line voltage Instantaneous cline currents

Current (in Amps)

6 4 2 0 -2 -4 -6 0

0.5

1

1.5

Time (in sec)

Fig. 14. Instantaneous line current

Disparity between the conventional controller- and adaptive controller-based electric springs concludes that adaptive PI controller exhibits superior characteristics compared to that of PI controller. From that graph, we obtain Tables 2 and 3. Data presented in Table 2 clearly indicates that ES with adaptive PI controller has less settling time when compared to that of ES with PI controller. So it can be concluded

188

K. Keerthi Deepika et al. Table 2. Settling time (in s) of PI and adaptive PI Controllers Condition

PI controller (s) Adaptive PI controller (s)

Sag initiation

0.1

0.05

Sag restoration

0.08

0.05

Swell initiation

0.11

0.08

Swell restoration 0.12

0.08

Table 3. %MP of the line voltage derived from ES with PI and adaptive PI controllers Condition

PI controller (%) Adaptive PI controller (%)

Sag initiation

8.18

5.45

Sag restoration

8.18

5

Swell initiation

8.64

5.45

Swell restoration 8.18

4.09

that ES with Jaya algorithm-based adaptive PI controller can restore the line voltage from sag and swell conditions very quickly. From Table 3, it is clear that the % MP value is small for adaptive PI controller. Figure 15 shows the sag condition in a three-phase waveform below 1 s, and after 1 s, the ES with adaptive PI controller restores the sag. The waveform after 1 s is the normal three-phase waveform. Figure 15 shows the swell condition in a three-phase waveform below 2 s, and after 2 s, the ES with adaptive PI controller restores the swell. The restoration of sag and swell in line the voltage is a very quick process with adaptive PI controller-based ES. Figure 16 shows harmonic content in the current injected by ES in terms of FFT analysis. It is evident that THD is 0.04% at the initiation of sag and a THD of 0.03% at the initial conditions of swell in the line voltage; i.e., the output obtained from the ES with adaptive PID control is harmonic free.

5 Conclusion In this paper, the need for the paradigm from fixed gains of PI controller to adaptive gains is discussed. Then, importance of the new power electronic device for voltage regulation in distribution systems is presented. In view of the parameter-less advantage with the considered heuristic algorithm, it is applied in tuning of PI controller for ES. Simulation results highlight that Jaya algorithm-optimized PI controller regulates voltage fluctuation at the point of common coupling effectively than the conventional PI controller.

Restraining Voltage Fluctuations in Distribution System Instantaneous cline currents

5

Current (in Amps)

189

0

-5

1

Time (in sec) Instantaneous cline currents

Current (in Amps)

6 4 2 0 -2 -4 -6

2

Time (in sec)

Fig. 15. Restoration from sag and swell, respectively

Fundamental (50Hz) = 3.571 , THD= 0.04%

Fundamental (50Hz) = 5.225 , THD= 0.03%

Mag (% of Fundamental)

Mag (% of Fundamental)

0.25

0.15

0.1

0.05

0

0.2 0.15 0.1 0.05 0

0

200

400 600 Frequency (Hz)

(a)

800

1000

0

200

400 600 Frequency (Hz)

800

1000

(b)

Fig. 16. FFT analysis of the injected current by ES under sag and swell conditions, respectively

References 1. Yazdanian M, Mehrizi-Sani A (2014) Distributed control techniques in microgrids. IEEE Trans Smart Grid 5(6):916–924 2. Bian D, Pipattanasomporn M, Rahman S (2015) A Human expert-based approach to electrical peak demand management. IEEE Trans Power Deliv 30(3):1119–1127 3. Rahiman FA, Zeineldin HH, Khadkikar V, Kennedy SW, Pandi VR (2014) Demand response mismatch (drm): concept, impact analysis, and solution. IEEE Trans Smart Grid 5(4):1734– 1743

190

K. Keerthi Deepika et al.

4. Hui SYR, Lee CK, Wu FF (2012) Electric springs—a new smart grid technology. IEEE Trans Smart Grid 3(3):1552–1561 5. Chaudhuri NR, Lee CK, Chaudhuri B, Hui SYR (2014) Dynamic modeling of electric Springs. IEEE Trans Smart Grid 5(5):2450–2458. https://doi.org/10.1109/tsg.2014.2319858 6. Soni, J, Panda, S (2017) Electric spring for voltage and power stability and power factor correction. In: IEEE transactions on industry applications, pp. 1–1. https://doi.org/10.1109/ tia.2017.2681971 7. Wang M, Yang T, Tan S, Hui SY (2019) Hybrid electric springs for grid-tied power control and storage reduction in AC microgrids. IEEE Trans Power Electron 34(4):3214–3225. https:// doi.org/10.1109/tpel.2018.2854569 8. Cherukuri, S.C, Saravanan, B (2018) A Novel DSM strategy for micro grids consisting of higher penetration of water heater loads. 1–6. https://doi.org/10.1109/npec.2018.8476701 9. Pilehvar SM, Shadmand, Mohammad MB (2018) Analysis of smart loads in nanogrids. IEEE Access pp 1–1. https://doi.org/10.1109/access.2018.2885557 10. Ma G, Guchao X, Yixi C Rong J (2018) Voltage stability control method of electric springs based on adaptive PI controller. Electr Power Energy Syst 95:202–212 11. Rao RV (2016) Jaya: a simple and new optimization algorithm for solving constrained and unconstrained optimization problems. Int J Ind Eng Comput 7:19–34 12. Sugandh P.S, Tapan Prakash VP, Singh M, Ganesh B (2017) Analytic hierarchy process based automatic generation control of multi-area interconnected power system using Jaya algorithm. Eng Appl Artif Intell 60:35–44 13. Pradhan, C, Bhende N, Chandrashekhar (2019). Online load frequency control in wind integrated power systems using modified Jaya optimization. Eng Appl Artif Intell 77:212–228. https://doi.org/10.1016/j.engappai.2018.10.003 14. Mishra S, Ray PK (2016) Power quality improvement using photovoltaic Fed DSTATCOM based on JAYA optimization. IEEE Trans Sustain Energy 7(4):1672–1680. https://doi.org/10. 1109/tste.2016.2570256

An Experimental Setup to Study the Effects of Switcing Transients for Low Voltage Underground Cable Sanhita Mishra1(B) , A. Routray2 , and S. C. Swain1 1 School of Electrical Engineering, KIIT Deemed to be University, Bhubaneswar, India

[email protected] 2 Department of Electrical Engineering, Indian Institute of Technology, Kharagpur, India

Abstract. Successful delivery of electrical power depends upon the reliability of system such as protection of electrical equipment, delivery media and condition of environment. Unlike overhead transmission lines, an underground cable is not exposed to the variable weather condition, so the reliability of electrical power delivery is more in comparison with overhead lines. An experimental set-up has been developed to conduct switching transient of power cable with induction motor acting as a load, and the current and voltage data are saved in a digital storage oscilloscope during the transient operation and analysed properly. Switching operation mainly creates surges and moves through the cable circuit. R, L and C parameters of the cable have been calculated using traditional method of determination. The aim of this paper is mainly to energise a low-voltage distribution cable and to study the behaviour of switching transient. In addition to switching transient, different types of fault such as line to ground fault and double line to ground fault have been created in an unloaded cable. A MATLAB/Simulink platform has been used to study the cable parameters and its characteristics. Keywords: Experimental set-up · Underground cable · Distributed pi network of underground system · Switching transient · Fault analysis

1 Introduction Underground distribution system is for maintenance and for the installation of a new one. For safety point of view in highly density populated area, uses of underground cable have seen sharp hike in recent times. To enhance the relibility of power system it is really a challenging task for the power engineer to detect and locate various types of faults in underground cable. There are various traditional fault location methods [1] such as Murray loop method and acoustic detection method which helps in locating the faults, still for accuracy fault impedance calculation is an essential factor for power cables. The power is being transmitted through both overhead line and UG cable. Unlike overhead lines [2], the special characteristics of underground cable have larger capacitance value © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_19

192

S. Mishra et al.

in comparison with inductance. So while analysing various types of cable, a lot of complications [3] in calculating cable size may come. A cable system mainly consists of core, sheath and ground which may create extra complications in calculating impedance in comparison with overhead lines. This cable connects the generating stations to the load end through the transmission system. As proper model of cable describes accurate results of switching transient, so proper statistical analysis [4] of simulated model has been proposed in this paper. Due to the introduction of distributed generation [5], the traditional power system network is getting segregated into small section or cluster where more importance is given to the renewable sources to inject their power through LT line. So, the transient analysis of LT line is a major task nowadays. Impact of frequency dependence of cable parameter, length and number of pi sections [6] helps in improving the system accuracy. Overload and steady overvoltage [7] along with transient behaviour of mixed cable system have been well described by the author. Low-voltage distribution system is mostly composed of compound sections of line and cables with various types, sizes and length. Arcing and spitting phenomena mainly occur due to switching of unloaded cable by load break elbow or circuit breaker. Energisation [8] and de-energisation operations create multiple re-strikes in voltage and current waveforms. Basically in an electrical power system, current blockage or unwanted creation of conducting path is known as fault. Fault is mainly an accidental and unwelcome parameter for power system, so voltage and current analyses [9] with proper protection are important. For locating fault in an underground cable, proper line model analysis is highly required. It is foremost duty of the power engineers to evaluate the parameters such as voltage and current during fault condition. So that as per the fault level, damaging effects of the fault can be reduced by using protective devices. Various types of offline and online methods help in detecting the location of fault. It is also difficult to trace incipient fault [10], so a various incipient fault detection method is to be studied. To deliver electricity to the end user, electrical distribution is the final stage. For curtailment of transmission line losses and to model the transmission system efficiently, the generated voltage is stepped up to higher voltage but this transmission voltage is not used by the consumer. So, the transmission power line will move into a distribution substation where step down of voltage occurs and which can be utilised by various industrial, commercial and residential consumers. Nowadays as underground system plays a major role due to its advantages compared to overhead lines, residential and commercial distribution is processed through underground cable. DG penetration in low voltage distribution system mainly creates voltage instability issue in the system. So transient analysis is a challenging task for low-voltage network consists of underground cables. For protection of any system, an actual model is always helpful for analysis, design and control. In this paper, an experimental set-up has been developed and transient analysis has been done by switching the cable through a circuit breaker of an unloaded and loaded LV cable. The transient nature of voltage and current has been analysed which has captured through a DSO. 1.1 Experimental Set-up for Four-Core Cable In this paper, experimental data of underground cable have been analysed. So, switching transient of a low voltage 1.1 kV four-core cable has been created. The experimental set-up has been designed as shown in Fig. 1b. A 400 V supply is given to a four-core

An Experimental Setup to Study the Effects of Switcing

193

armoured PVC cable having length 50 mt. A number of case studies has been analysed by considering unloaded and loaded cable for transient analysis. The specification of the cable which we have used for our experiment is a 10 mm2 , four-core, PVCinsulated armoured cable, circular solid aluminium conductor having approximate AC resistance 3.95 /km, approximate capacitance 0.6 mfd/Km and approximate reactance 0.091 /Km. The system has been developed with proper protection by using relay and circuit breaker to avoid unnecessary damage to the system. We also have used a threephase, one HP, 50 Hz, 415 V squirrel cage induction motor as a load for experimenting the switching transient.

Fig. 1. a Approximate diagram of a four-core cable [13], b experimental set-up, c switching transient analysis of R phase in DSO, d switching transient analysis of R phase, e switching transient analysis for three phases in DSO, f switching transient analysis of R phase, g current during L-G fault occur, h current during L_L_G fault occur

Fig. 1. (continued)

S. Mishra et al.

Fig. 1. (continued)

2 1.5

current in amp

194

1 0.5 0 -0.5 0

500

1000

1500

2000

-1 -1.5 -2

No of Sample

Fig. 1. (continued)

Fig. 1. (continued)

2500

3000

An Experimental Setup to Study the Effects of Switcing

195

Fig. 1. (continued)

Fig. 1. (continued)

Figure 1a depicts regarding a four-core cable which is used mainly for the distribution purpose. In some cases also three and half core cables are used. The cable in this paper has four cores with PVC insulation as a layer. A conducting layer is present named as armour whose main function is to give mechanical protection to the cable. For modelling of the cable, it is highly essential to calculate [11, 12] the R L C matrix of the cable. The capacitance and impedance matrix calculations are highly important to do fault analysis. The data have been used to simulate the four-core cable. In the above figure, the total number of sample is 2500 and during the switching the transient current is very high. Figure 1c and d shows how the current during the starting period increases rapidly.

196

S. Mishra et al.

Fig. 1. (continued)

In the above figure, the switching transient of three-phase induction motor has been analysed and it has been seen that the starting current is quite high after few seconds the system becomes steady one.

2 Analysis of Three and Half Core Cable In Fig. 1e, a three and half core cable having length 7 mt has been used to do the analysis. The induction motor has operated with different load conditions, and the current, voltage and power measured in three phases are given in Table 1. Table 1. Analysis of three and half core cable Data

R

Y

B

Current (Amp)

1.80

1.67

1.63

Real power (W)

75

42

75

Power factor (lag)

0.19

0.11

0.20

Voltage (sending end) (V) 395.3 391.6 394.6

The specification details of the cables are 16 mm2 armoured cable having approximate capacitance 0.830 microfarad/Km. The current waveform is shown in the Fig. 1f. Switching action mainly starts when either load break switches, circuit breaker operates or fuse operates. As we know, switching action defines both closing operation and opening operation only. Transient current will flow through the system subsequent to closing operation. We have analysed the transient current for distribution cable.

An Experimental Setup to Study the Effects of Switcing

197

3 Fault Analysis of Underground Cable Using MATLAB/Simulink We have considered a three-phase, 11 kV source with four numbers of single-core cable. The fault has been created at the receiving end of the unloaded cable, and the voltage and current waveforms have been analysed using MATLAB/Simulink. The R, L and C parameters of the cable have been calculated and used in the cable Simulink model, R, L, C parameters are calculated in matrix form, and these matrix data act as input to cable model. Sheath voltage has also been observed during fault. A relay circuit has been designed which sends a signal to the breaker after an immediate fault occurs in the system. Both L-G and L-L-G have been created in the system, and results have been examined properly. Figure 1g gives entire information for L_G fault occurred in the R phase of an unloaded cable. Initially, the current is zero across the receiving end but when a L-G fault occurs after 0.1 s, a high current flows through the short-circuited path and the relay senses the fault, the circuit breaker trips. The current in three phases becomes zero after immediate effect of the breaker. Figure 1h represents the current when the fault occurs in both R phase and Y phase and ground.

4 Conclusion In this paper, various tests of underground cable have been verified in laboratory. Switching transient in electrical power system is a major issue, so experimental results have been analysed properly for a low-voltage distribution cable. A simulation model has been developed for pi equivalent distribution cable, and the transient current graph has been analysed for both line to ground fault and double line to ground fault. The amplitude of current is suddenly increased at the instant of fault occur. The R, L, C parameter calculation plays a vital role for simulating the fault. Hardware implementation can be done to analyse the fault analysis in future to validate the simulation model, and also DWT analysis can be done for the transient voltages and currents analysis. Acknowledgements. The authors would like to thank the School of Electrical Engineering, KIIT University, for providing the necessary laboratory facility and library for successful modelling and testing of the proposed system.

References 1. Yang X, Choi M, Lee S, Ten C, Lim S (2008) Fault location for underground power cable using distributed parameter approach. IEEE Trans Power Syst 23(4):1809–1816 2. Filomena AD, Resener M, Salim RH, Bretas AS (2008) Extended impedance-based fault location formulation for unbalanced underground distribution systems. In: 2008 IEEE power and energy society general meeting—conversion and delivery of electrical energy in the 21st century. Pittsburgh, PA, pp 1–8 3. Georgiev D, Stanchev P, Kamenov Y, Rangelov Y (2018) Modelling of underground transmission systems for steady state and transient studies. Sensitivity analysis on the precision of parameters

198

S. Mishra et al.

4. Daud MZ, Ciufo P, Perera S (2013) A study on the suitability of cable models to simulate switching transients in a 132 kV underground cable. Australian J Electri Electron Eng 10(1):45–54 5. Ackermann T, Andersson G, Söder L (2001) Distributed generation: a definition. Electric Power Syst Res 57(3):195–204. ISSN 0378-7796 6. Hoshmeh A, Schmidt U, Gürlek A (2018) Investigations on the developed full frequencydependent cable model for calculations of fast transients. Energies 11:2390. https://doi.org/ 10.3390/en11092390 7. Hoogendorp G (2016) Steady State and transient behavior of underground cables in 380 kV transmission grids. https://doi.org/10.4233/uuid:2ecf0e07-58c8-42b9-bbf1-67878a3f6018 8. Walling RA, Melchior RD, McDermott BA (1995) Measurement of cable switching transients in underground distribution systems. IEEE Trans Power Delivery 10(1):534–539 9. Yindeesap P, Ngaopitakkul A, Pothisarn C, Jettanasen C (2015) An experimental setup investigation to study characteristics of fault on transmission system. In: Proceedings of the international multiconference of engineers and computer scientists 2015 vol II, IMECS 2015, March 18–20, 2015. Hong Kong 10. Xu Z (2011) Fault location and incipient fault detection in distribution cables. Electronic Thesis and Dissertation Repository, 319 11. Nasser DT, Tleis A (2008) Power systems modelling and fault analysis: theory and practice. Published by Elsevier Ltd. ISBN-13: 978-0-7506-8074-5 12. Da Silva FF, Bak CL (2013) Electromagnetic transients in power cables. Springer, London, UK. https://doi.org/10.1007/978-1-4471-5236-1 13. Shafieipour M, Chen Z, Menshov A, De Silva J, Okhmatovski V (2018) Efficiently computing the electrical parameters of cables with arbitrary cross-sections using the method-of-moments. Electric Power Syst Res 162:37–49

Effect of Distributed Generator on Over Current Relay Behaviour Tapaswini Biswal(B) School of Electrical Engineering, KIIT Deemed to be University, Bhubaneswar, Odisha 751024, India [email protected]

Abstract. Integration of Distributed Generator (DG) to the distribution network feeder causes its power flow to bidirectional in place of unidirectional influencing the feeder protection. The analysis of chapter presents the influence of Distributed Generator on the over current relay behaviour. The behaviour of the scheme is evaluated for an 8-bus radial distribution feeder in PSCAD/EMTDC software and the characteristics of over current relay are tested on MATLAB software. The simulation result indicates the effects of DG on feeder protection as the current from Distributed Generator reduces relay reach. Keywords: Distributed generator · Distribution network feeder · Over current relay

1 Introduction Distributed generators are small size units that directly connected to the distribution feeder or on the consumer site which provides an alternate solution for delivering power to some customers [1–3]. Generally distributed generators are synchronous generators, induction generators powered by wind, fuel cells, hydro, and photo-voltaic. They offer various applications including backup generation, utmost shaving, and smart metering. Apart from the applications, the benefits include voltage support, a decrease of energyloss, and release of system capacity and enhancement of reliability [4, 5]. Previously the distribution feeder was incorporated for the transmission of power from the transmission feeder to the load centre. There was no provision for incorporating the DGs directly to the feeder. The protection of the distribution feeder is well equipped considering one-way power flow, i.e. point of transmission to the load centre. The distribution system feeder causes the power flow to be two-way in place of oneway affecting the behaviour and stability of the network in so many methods with the connection of DGs to it [6]. The tremendous effects that remain unresolved deploying traditional methods found to be the influence of Distributed Generator on regulation and feeder protection [7–9]. The chapter emphasizes the influence of Distributed Generator on protection of the feeder. Protection scheme essentially consists of protective relay and circuit breaker for © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_20

200

T. Biswal

the protection of power system elements against unusual faulty conditions. Protective relay is a sensing device that directs the circuit breaker to open when current through the relay is more than the set current of the relay. Circuit breaker after getting a signal from the protective relay isolates the faulty part from the rest of the feeder. The motivation behind the research is relay mal-operates if the fault current magnitude is less than the set current of the relay in-cooperating the over current protection. But it becomes more difficult for the relay if distributed generators are connected to it as it reduces the reach of the relay. But the integration of a DG affects the over current relay behaviour that has been discussed in this chapter. This chapter is organized as follows. The mathematical modelling of an 8 bus radial distribution system is simulated in PSCAD/EMTDC software has been discussed in Sect. 2. The detailed description of the impact of DG on over current relay behaviour and simulation results are provided in Sect. 3. Finally, the conclusion part is discussed in Sect. 4.

2 System Modelling An 11 kV, 8 bus radial distribution system with DG integrated at bus 6 is simulated in PSCAD/EMTDC software in this chapter. A linear load of each 0.851 MVA at a power factor of 0.76 per phase is equally distributed along the feeder. Distribution line Segment connected are of R = 0.38 , X = 0.4084 . The DG connected to bus 6 is a constant current source and a line to ground fault of fault resistance 10 WÄ is applied to bus 8. The schematic diagram of the test model is shown in Fig. 1.

DG

Relay

AC

Bus1

i1 Bus2

load 1

i2 load 2

Bus3

Bus4 i3 load 3

i4

load 4

Bus5

Bus6

i5

i6

load 5

Bus7 Bus8

load 6

Fault

i7

load 7

load 8

Fig. 1. Line diagram of 8-bus radial distribution network

3 Effect of Distributed Generator on Over Current Relay Behaviour Apart from others, one reason affecting the DG on feeder protection is “reach” of relay which gets affected due to current contribution from DG. OC relays are used for protecting a definite portion of the feeder, which is called “reach”. It is calculated by the

Effect of Distributed Generator on Over Current Relay Behaviour

201

pick-up current which is the minimum fault current at which the relay operates. The “reach” of OC relay will get lessened due to Distributed Generator subsequently, faults at the feeder end remain undetected. The main cause for reduction in reach is due to Distributed Generator which contributes current, which causes a decrease in fault current sense by the relay. For understanding the effect of Distributed generator on over current relay behaviour of the feeder, Fig. 1 is simulated in PSCAD/EMTDC software with and without DG with a fault resistance of 10  at the feeder end. Fault current at bus1, bus4, bus6, bus7 are calculated for a fault resistance of 10  at the feeder end before connecting DG. When DG is connected at bus 6 fault current at bus1, bus4, bus6, bus7 are calculated for a fault resistance of 10  at the feeder end with DG. Figure 2 shows fault current seen by the relay at the grid after few transients reaches to a constant value of 1186 amp after 20 ms when there is no DG connected to the feeder. Figure 3 shows that the fault current subsidies to a constant current of 768 amp few milliseconds after when DG is connected to bus 6.

Fig. 2. Current at different buses without DG

Fig. 3. Current at different buses with DG

202

T. Biswal

As seen from Fig. 4 the fault current seen by the relay is 1186 amp when the feeder is not connected by DG. But, however, current decreases at relay point when the feeder is connected to DG. The relay characteristics algorithm is tested on MATLAB software to see the relay nature.

Fig. 4. Current sense by relay before and after integration of DG with fault at the feeder end

Figure 5 shows the relay characteristic is rectangular hyperbola whose characteristic is inverse in the initial part and as the current becomes very high it operates at a definite minimum value. This is due to the fact at high values of current in the electromagnetic relay the flux saturates and maintains a constant value due to saturation. So if the fault current is higher, relay takes a minimum time to operate.

Fig. 5. Relay I–T characteristics

4 Conclusion With the integration DG in the feeder no doubt it provides various operational benefits to the system at the same time it affects the feeder protection. The analysis of this chapter

Effect of Distributed Generator on Over Current Relay Behaviour

203

presents the effect of Distributed Generator on feeder over current relay behaviour. Simulation results clearly show the integration of DG reduces the over current relay reach. One algorithm is to be developed which will assist over current relay with and without distributed generators under faulty condition. Acknowledgements. I am thankful to the school of electrical engineering, KIIT Deemed to be University for providing the requisite laboratory facility and library for successful modelling and testing of the proposed system.

References 1. Chandraratne C, Woo WL, Logenthiran T, Naayagi R (2018) Adaptive overcurrent protection for power systems with distributed generators, pp 98–103. https://doi.org/10.1109/icpesys. 2018.8626908 2. Singh B, Mishra DK (2018) A survey on enhancement of power system performances by optimally placed DG in distribution networks. Energy Rep 4:129–158 3. Chandraratne C, Logenthiran T, Naayagi RT, Woo WL (2018) Overview of adaptive protection system for modern power systems, ISGT-ASIA 4. Baran M, El-Markabi I (2004) Adaptive over current protection for distribution feeders with distributed generators. In: IEEE PES power systems conference and exposition, vol 2. New York, pp 715–719 5. Motabarian F, Golkar MA, Hajiaghasi S (2013) Surveying the effect of distributed generation (DG) on over current protection in radial distribution systems. In: 18th electric power distribution conference, Kermanshah, pp 1–6 6. Shih MY, Enriquez AC, Leonowicz ZM, Martirano L (2016) Mitigating the impact of distributed generation on directional overcurrent relay coordination by adaptive protection scheme. In: 2016 IEEE 16th international conference on environment and electrical engineering (EEEIC), pp 1–6 7. Coffele F, Booth C, Dy´sko A (2015) An adaptive overcurrent protection scheme for distribution networks. IEEE Trans Power Deliv 30(2):61–568 8. Jain DK, Gupta P, Singh M (2015) Overcurrent protection of distribution network with distributed generation. In: 2015 IEEE innovative smart grid technologies—Asia (ISGT ASIA). Bangkok, pp 1–6 9. Li Z, Tong W, Li F (2006) Study on adaptive protection system of power supply and distribution line—IEEE conference publication. www.Ieeexplore.ieee.org

Advanced Network Applications

Detection and Prevention from DDoS Attack Using Software-Defined Security Sumit Badotra(B) , Surya Narayan Panda, and Priyanka Datta Chitkara University Institute of Engineering and Technology, Chitkara University, Raipura, Punjab, India {sumit.badotra,snpanda,priyanka.datta}@chitkara.edu.in

Abstract. The network which is able to accommodate today’s real-time need is growing in a very fast manner. But simultaneously also occurs an increase in the rate of network attacks and threats. Distributed Denial of Service (DDoS) is one of the attacks in which intruder attempts to disrupt normal network traffic by flooding huge traffic into the network and ultimately halt the network services and resources. There are numerous solutions available for the detection and prevention of DDoS attacks in traditional networks but making use of Software-Defined Security (SDS) is a new way of securing the network. The basic principle of separating the intelligence of the network from the infrastructure can be considered as the new hope for securing the network. This chapter aims to provide the need for SDS in networks with related literature survey we have also found out the research gaps from research done till now or going on. A method to prevent a network from DDoS attacks is also proposed using SDS. Keywords: Traditional networks · DDoS attack · Software-defined security

1 Introduction In order to build many of the network devices and middleboxes like network switches, routers, load balancers for network, firewalls, Network Address Translation (NAT), etc. used in the network, each and every device needs to be manipulated individually. It is very difficult to make any changes in the traffic with the help of such intermediated. This change in traffic is very complex as compared to simple packet forwarding. Multiple complex network protocols are constituted by these intermediated network devices [1]. These devices are vendor-specific and hence it becomes a tedious job for a network administrator to configure these devices individually. Traditional networks are not only suffering from aforementioned challenges but also suffer from security attacks and threats as well. Although there are many solutions proposed until now to overcome these threats but with the complexity that these networks are comprised of is very difficult to overcome from network attacks. Network attacks can be categorized into two types: Active and Passive. Active Attacks are those types of attacks in which square measure are those within which the hacker makes an attempt to change knowledge or data traveling from sender to receiver within the network. A number of the active attacks square © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_21

208

S. Badotra et al.

measure spoofing attack; Spoofing or Hollow attack, Modification, Denial of services (DoS), Sinkhole, and Sybil attack [2]. Passive Attacks are those kinds of attacks within which hackers or unwelcome person don’t make changes or modify the information traveling in between the sender and receiver. The intention behind this attack is to browse and analyze the information. A number of passive attacks are traffic analysis, Eavesdropping, and observance [3]. In the chapter, we have taken DDoS attack as a point of study. It is an attempt to disrupt the normal traffic by flooding a huge traffic to the targeted server and ultimately halts the services provided by the server for the legitimate users as well. To overcome the situation in traditional networks Software-Defined Networking (SDN) has come into act [4]. The complexity of today’s real-time network is increased at a huge rate. In order to accommodate the alterations making the network programmable is the only solution. This will help in meeting the various requirements of the users. By segregating the intelligence of the network from the proprietary hardware is making it simple to incorporate the various amendments in the network. This segregation is achievable with the help of SDN. The new modified applications and techniques for network management are very easy to implement and use [5]. By making the simplified management and view of the network, new security features can be easily implemented, and hence SDN based network architecture is able to cope with the network attacks. The main contributions of this chapter are as follows: • To review current security issues and limitations in networks. • To find out the research gaps from the previous work done related to securing the network from a DDoS attack. • Based on these identified gaps, a method or framework is also proposed, which is making use of SDS for detection and prevention of DDoS attacks. The remainder of the chapter is organized as follows: Sect. 2, related survey regarding the security techniques/approaches used to prevent network from DDoS attack is given. Section 3 is comprised of research findings from the literature studied. In Sect. 4 is comprised of SDN controller whereas in Sect. 5 a discussion is given on the proposed approach to diminish the DDoS attacks by using SDN and finally conclusion is stated in Sect. 6.

2 Literature Survey Previous literature lays a foundation to formulate the objective of the research. The existing literature shows state of art technologies used to prevent network against DDoS attack connected works show that each of DDoS attack and countermeasures is kept evolving and growing. It can be observed that traditional methods of mitigating DDoS attacks are mostly on the basis of IP traceback, anomaly detection, filtering (ingress/egress), ISP defense, and network self-similarity as shown in Fig. 1. Braga et al. [6] have presented a method in order to detect the DDoS attack which is a lightweight. This method is based on network traffic flow features, in which taking out or withdrawal of such information about the attack is made with very low overhead compared to old approaches used in the respective domain. They have made use of

Detection and Prevention from DDoS Attack

209

Fig. 1. Traditional methods for militating against DDoS attack

NOX SDN controller, which was providing an interface to fetch such information from network switches. Another major contribution they have made is to include the high rate of detection and very low rate of false alarms obtained by analyzing the network flow using Self Organizing Maps (SOM). The drawback of their work is that they have made use of NOX controller, which is just a reference SDN controller. This work is not executed with a practical based SDN controller such as RYU, Floodlight, etc. Jun et al. [7] have proposed another method to mitigate the DDoS attack using the flow entropy- and packet sampling-based mechanism which is used to detect DDoS attack. To differentiate between normal traffic and network traffic generated by a DDoS attack, they had used OPNET simulation results. The limitation of their work was that they had only proposed a model using a simulator, the results of using a simulator may vary when implemented in the real world. Jyothi et al. [8] designed a framework to detect DDoS attack called Behavior-based Adaptive Intrusion detection in Networks (BRAIN). According to the proposed method, various multiple applications behavior is making use of low-level network hardware events. The approach had added advantages for already available hardware performance counters. The combination of previous network traffic statistics and modeled network application behavior for detecting DDoS attacks by making use of machine learning is achieved in this case. Li et al. [9] have proposed a new system called Drawbridge, which is used to address and manage the network traffic. They have made use of SDN in their research and ultimately, they have investigated a solution through which enabling end hosts to make use of their knowledge of network traffic which they desire in order to improve traffic flow during DDoS attacks. These approaches [9] had shown many limitations to mitigate DDoS attacks and therefore provide a viable solution for that using SDN. In order to differentiate between authorized users from intruders the Trust Management Helmet (TMH) model achieves this by the difference in registering four styles of trusts accustomed and transferred as a part of a license at the multiple users for session reference

210

S. Badotra et al.

to the targeted server firmly [10, 11]. However, this approach also suffers from many disadvantages such as license forgery, replay, deletion attacks, and either by sharing or using the duplicate or previously issued licenses attackers may cheat and ultimately the model fails. To overcome the disadvantages in the previous model on the network routers, approach having synchronization comprised of two-tier network traffic filters from distrustful traffic. The monitoring of the network traffic with a mechanism called as unique RED/Drop tail is also depicted in [11]. However, in this case, also spoofed addresses generated by attackers won’t be captured by the routers in the network and therefore it will give an open chance for attackers to launch a DDoS attack. A new approach called a blacklist approach is used later on to overcome the shortcomings. This approach uses a communications protocol. It additionally adds CAPTCHA to differentiate among legitimate users and botnets [12]. Therefore, the protocol to provide the communication between user and server is rejected and the mechanism of CAPTCHA was planned only to provide the mitigation of DDoS attacks specific to application layer botnet. On the other hand managing flash crowd events [13] still remained unaddressed. The use of CAPTCHA may create a hindrance as well to most of the legitimate users and will create a negative impact on various operations which work online. Among all solutions to mitigate DDoS attacks, entropy-based solutions have gathered a lot of attention [14]. However, in spite of various solutions provided and stated the aforementioned related to entropy, these solutions lack to detect low-rate and high-rate DDoS attack. The point of considering entropy to various options from traffic flows was accustomed to kind traditional patterns victimization clump analysis algorithmic rule and determine the deviations from the models that are created [15]. This proposed approach suffers from many cons and it needs an effective algorithmic rule to overcome the issues such as back process time and memory usage in a very high volume and at a very high-speed network. In order to overcome the various limitations in this approach a quick entropy technique that follows the victimization of traffic flow which is based on network analysis was planned [16]. However, this approach is not able to search out the offender and agents responsible for DDoS attack underneath this approach. To make use of cloud computing to defend against the DDoS attacks is not a new technique. Resource distribution of resources dynamically was planned and supported through the queuing theory [17]. Still, servers that are hosted by cloud are vulnerable to the DDoS attacks [18]. Making use of honeypots can be considered as a brand-new effort in providing the mechanism for defense in the network. In the approach of honeypots, on physical servers, a network of virtualized honeypots was deployed and then observation of incoming traffic or malicious activities all together with flooding packets was observed [17]. However, this approach is also vulnerable because the network routers have already been flooded with multiple malicious requests before the honeypots come into play. Another method to mitigate DDoS attacks was ant-based. This technique was victimized by virtual honeypots [18]. Attackers, in this case, may be able to identify the honeypots and it can become a launching pad for attackers to launch DDoS attacks either on the system itself or network and thus worm from one honeypot may spread to other networks as well. Shin et al. [19] have considered two aspects of one of the most common communication protocols which is used to interact between infrastructure plane and control plane, i.e., OpenFlow. The first aspect is the bottleneck or full memory space of controller

Detection and Prevention from DDoS Attack

211

and the second aspect is that of enabling the control plane to expedite both detection of, and responses to, the changing flow dynamics within the data plane. They have provided solutions to both aspects as well but the limitation of their work was that in SDN based network attack between data-to-control plane, sending bogus packet request continuously which increases the network latency and ultimately can lead to DDoS attack. Shoeb et al. [20] have proposed a method in order to control the network traffic communication flow between the control layer and the infrastructure layer. This enables the key principle to execute the network amendments in a very efficient way. Based on the multiple OpenFlow devices requests sent or received by the underlying networking devices, using the priority method or the traffic flow configuration the compatibility of the network switch are maintained as well. The time out value for network flow between control plane and data plane is considered by increasing the efficiency of both controller and switch by proposing a method that is feasible and efficient. Wang et al. [21] have presented an architecture to mitigate DDoS attack that facilitates to make the network programmable and flexible in their method they have used graphical model-based attack detection method which can overcome the problem of the dataset. They have used simulation tools to perform the experimentation. The limitation was high latency, low scalability. Zheng et al. [22] proposed a real-time DDoS Defense using COTS SDN switches via adaptive correlation analysis, it is used to detect DDoS attacks via adaptive correlation analysis on COT SDN switches. The disadvantage of the proposed work was new emerging sophisticated DDoS attacks (e.g., Crossfire) constructed by low rate and short-lived “benign” traffic are even more challenging to capture. Tseng et al. [23] have described a protocol PATMOS which was proposed to mitigate against DDoS attack in multi-controller environment using clustering. The main advantage of their work is that they have eliminated overloaded dependency on a single controller which ultimately reducing the CPU usage rate and hence increasing throughput. The proposed work suffers from some cons which is an analysis of network traffic and an increase in the computational cost of the network. Badotra et al. [24] have implemented an SDN based firewall using RYU controller which works on both the transport layer and application layer. In [25] SDN based Collaborative Scheme for Mitigation of DDoS attack is proposed. This work has made use of RYU and POX controller which is not being used in industries. It is just used for experimentation purposes only. Therefore, from the literature survey discussed above, we can conclude that although there are a number of solutions available to mitigate the DDoS attack in traditional networks, these methods are inadequate because nowadays attackers are making use of dynamic methods of DDoS attack. Therefore, we need a practical and intelligent solution to implement security into the networks. By making use of SDN with its open-source controllers and the basic principle of separating the intelligence and data plane of the network many customized APIs can be built which are open-source and can be used for providing security. We no longer need any middlebox and dedicated hardware which is vendor-specific and non-configurable.

212

S. Badotra et al.

3 Research Gaps SDN can be considered as an active and vast research area in the field of networking. A large number of researchers are working in various domains of SDN but very few researchers have tried to unfold the security feature of SDN and making it to the most of use. Providing security to the network by using SDN can overcome the various challenges faced by traditional networks. For implementing the security, until now researchers have used only those SDN controllers which are not being implemented in the real world such as NOX, POX, RYU, and Floodlight, etc. These controllers are only used for experimentation. Based on the literature studied following major research gaps have been identified: • Currently, in order to mitigate DDoS attack various SDN based collaborative schemes and solutions are only making use of such SDN based controllers which are not being used in industries [6, 9, 24, 25]. • Early detection of each low-rate and high-rate DDoS attack remains to be selfaddressed [14]. • No Graphical User Interface (GUI) feature and platform support for windows and MAC are supported by currently used SDN controllers. Most of these controllers (POX, NOX, RYU, and Floodlight) are based on Linux based platform only and possess a traditional DDoS mitigation method which has more network computational cost [15, 23]. • Development of such framework which can handle and overcome DDoS attack, make use of open-source API’s and can support multiple vendors is still lacking [25, 26]. Summarization of various identified gaps is shown in Table 1.

4 SDN Controllers Almost every network activity in SDN based network revolves around the centralized controller. It is located at the control layer and hence acts as the intermediate between the underlying infrastructure layer and application layer. Through the bare-metal switches, the controller sends the specific instructions on how to send the data and also on which path to select [27, 28]. Being the vital and important component of SDN based network, SDN controller needs to have reliability and security for a better SDN based environment. The use of multiple controllers must be used for critical application missions. In this case, if one controller is targeted by the attackers (leader controller), other follower controllers come into play to maintain the proper functionality of the entire network. Aside from path selection, other different policies like security, Quality of Service (QoS), network traffic engineering continued by SDN controllers [29, 30]. All the correspondence is possible with the assistance of Southbound and Northbound APIs. Controllers provide the intelligence, cost-efficient mechanism, automation to the network. The SDN architecture heart are controllers, Nicira Networks made the first SDN controller in 2009 and named it as Nox which was developed also with the first version of OpenFlow’s [29]. Further, its revised version was developed along with Python support and was called POX controller [31–34]. After that ONIX platform was developed, a distributed platform for

Detection and Prevention from DDoS Attack

213

Table 1. Gaps identification Gaps identified

Description

Various SDN based collaborative schemes and solutions are only making use of such SDN based controllers which are not being used in industries for mitigating DDoS attack [6, 9, 24, 25]

Though firewall with SDN solves many traditional firewall drawback but still it lacks behind as it don’t have open API’s which can be combined with multiple applications and hence can be used by different enterprises having heterogeneous vendors

Every high-rate as well as low-rate DDoS It had been observed that the use of attack which is detected early still needs to be appropriate data helps to magnify the spacing self-addressed [14] between attack traffic and legitimate for both high-rate and low-rate. This advantage can only take if it is detected in early-stage only Recently used SDN controllers, does not support Graphical User Interface (GUI) feature for MAC and windows. Most of these controllers (POX, RYU, Floodlight and NOX) are based on Linux based platform only and possess a traditional DDoS mitigation method which have more network computational cost [15, 23]

POX, NOX, RYU, and Floodlight still now are used for experiments only. There is no evidence of using such controllers in industries. These controllers are based on LINUX, as well as they don’t support GUI features of windows and MAC. So, experiments on those controllers are not required

Development of such framework which can handle and overcome DDoS attack, make use of open-source API’s and can support multiple vendors is still lacking [25, 26]

The absence of non-commercial API’s which can be used by any enterprise and then reconfigured accordingly by adding security rules in the network accordingly

the data center with vast scale networks and Google had developed it, a few years later, NTT and Nicira, become the foundation of VMware’s SDN controller which is the most used and famous SDN controller in the commercial industry [27]. Some of the popular and most used SDN controllers are defined below: • OpenDayLight (ODL)—It is the most and widely used SDN controller [35] which is an open-source controller project. It is controlled by the Linux Foundation. It is comprised of a huge number of vendors/ enterprises in its group. ODL has successfully made a big change in the commercialization of the SDN sector [36]. • Open Networking Operating System (ONOS)—It is the SDN controller platform by Linux Foundation which has the ability to transit from traditional “brown field” networks to new SDN based “green field” networks which help in faster deployment and lowering the cost of deployment [35]. • Floodlight—It is developed by open community of developers mostly from Big Switch Networks and used OpenFlow protocol. It was initially offered by Big Switch Networks as part of ODL project. Big Switch then stepped out of this project because of some conflicts with Cisco Systems and now Floodlight is not a part of ODL project [37, 38].

214

S. Badotra et al.

• Ryu—Ryu Controller is an open standard SDN controller that is specifically designed to strengthen network agility with easy manageability. This controller is used by NTT in their cloud-based data centers. Ryu bring well-defined APIs along with various software components. Ryu source code is on Github and is managed and maintained by Ryu developer community. It is written in Python and the source code is available under Apache 2.0 license [32]. • POX—It is an SDN controller written entirely in Python [31]. It was created after the Nox and become much more popular than Nox. It supports the same graphical user interface as Nox and performs better than Nox in the real world.

5 Proposed Approach and Discussion As mentioned before as well that SDN acts as a brain of the network, centralized SDN controller is the one who is managing the whole network and has a global view. In the proposed scenario as shown in Fig. 2 SDN architecture is implemented in a network and in this, a control layer constitutes the controller, for example,. ODL, controller is having communication with an application such as a firewall through an API (Application Programming Interface). As the controller is the single point of failure so, to overcome this, we can also add another SDN controller at the edge of the network, for example, ONOS. Whenever there is flooding of traffic from multiple botnets and DDoS attack is launched on a targeted server, at that time edge controller will be handling DDoS attack and another controller will be able to maintain the functionality and working of the network. Both controllers will work simultaneously. Open-source API can be created to work with any security-based application and this API can be reconfigured easily by any enterprise as per their need.

Fig. 2. Proposed framework to defend DDoS attack using SDN based architecture

6 Conclusion IoT (Internet of Things) is a big buzz nowadays and number of devices connected to the internet is growing at an exponential rate and ultimately increases the number of

Detection and Prevention from DDoS Attack

215

sources from which DDoS attacks can be launched. Many companies such as CISCO, Juniper, etc. are already making many applications such as Checkpoint, Palo Alto, etc. in order to detect and prevent DDoS attacks but these applications are commercial and one has to pay to get the benefits. Another disadvantage of these applications is nonconfigurability; one cannot modify it as per their need. In this chapter, need for SDN for securing the network is described. Approaches that are used by traditional networks to secure the networks are discussed with their limitations. A method has also been proposed to detect and prevent the DDoS attack by using SDS with various available SDN controllers’ illustration.

References 1. McKeown N, Anderson T, Balakrishnan H, Parulkar G, Peterson L, Rexford J, Turner J et al (2008) OpenFlow: enabling innovation in campus networks. ACM SIGCOMM Comput Commun Rev 38(2):69–74 2. Sharma K, Khandelwal N, Prabhakar M (2011) An overview of security problems in MANET. In: ISEM international conference 3. O’Brien WJ, Formoso CT, Ruben V, London K (2008) Construction supply chain management handbook. CRC Press 4. Xia W, Wen Y, Foh CH, Niyato D, Xie H (2014) A survey on software-defined networking. IEEE Commun Surv Tutor 17(1):27–51 5. Nunes BAA, Mendonca M, Nguyen XN, Obraczka K, Turletti T (2014) A survey of softwaredefined networking: past, present, and future of programmable networks. IEEE Commun Surv Tutor 16(3):1617–1634 6. Braga R, de Souza Mota E, Passito A (2010) Lightweight DDoS flooding attack detection using NOX/OpenFlow. In: LCN, vol 10, pp 408–415 7. Jun JH, Lee D, Ahn CW, Kim SH (2014) DDoS attack detection using flow entropy and packet sampling on huge networks of ICN, pp 185–190 8. Jyothi V, Wang X, Addepalli SK, Karri R (2016) Brain: behavior based adaptive intrusion detection in networks: using hardware performance counters to detect ddos attacks. In: 2016 29th international conference on VLSI design and 2016 15th international conference on embedded systems (VLSID). IEEE, pp. 587–588 9. Li J, Berg S, Zhang M, Reiher P, Wei T (2014) Drawbridge: software-defined ddos-resistant traffic engineering. ACM SIGCOMM Comput Commun Rev 44(4):591–592 (ACM) 10. Wang X, Chen M, Xing C (2015) SDSNM: a software-defined security networking mechanism to defend against DDoS attacks. In: 2015 ninth international conference on frontier of computer science and technology. IEEE, pp 115–121 11. Yu J, Fang C, Lu L, Li Z (2010) Mitigating application layer distributed denial of service attacks via effective trust management. IET Commun 4(16):1952–1962 12. Singh KJ, De T (2015) DDOS attack detection and mitigation technique based on Http count and verification using CAPTCHA. In: 2015 international conference on computational intelligence and networks. IEEE, pp 196–197 13. Al-Ali Z, Al-Duwairi B, Al-Hammouri AT (2015) Handling system overload resulting from DDoS attacks and flash crowd events. In: 2015 IEEE 2nd international conference on cyber security and cloud computing. IEEE, pp. 512–512 14. Bhuyan MH, Bhattacharyya DK, Kalita JK (2015) An empirical evaluation of information metrics for low-rate and high-rate DDoS attack detection. Pattern Recogn Lett 51:1–7

216

S. Badotra et al.

15. Qin X, Xu T, Wang C (2015) DDoS attack detection using flow entropy and clustering technique. In: 2015 11th international conference on computational intelligence and security (CIS). IEEE, pp 412–415 16. David J, Thomas C (2015) DDoS attack detection using fast entropy approach on flow-based network traffic. Proc Comput Sci 50:30–36 17. Yu S, Tian Y, Guo S, Wu DO (2013) Can we beat DDoS attacks in clouds? IEEE Trans Parallel Distrib Syst 25(9):2245–2254 18. Alqahtani S, Gamble RF (2015) DDoS attacks in service clouds. In: 2015 48th Hawaii international conference on system sciences. IEEE, pp 5331–5340 19. Shin S, Yegneswaran V, Porras P, Gu G (2013) Avant-guard: scalable and vigilant switch flow management in software-defined networks. In: Proceedings of the 2013 ACM SIGSAC conference on computer & communications security. ACM, pp 413–424 20. Shoeb A, Chithralekha T (2016) Resource management of switches and controller during saturation time to avoid DDoS in SDN. In: 2016 IEEE international conference on engineering and technology (ICETECH). IEEE, pp. 152–157 21. Wang B, Zheng Y, Lou W, Hou YT (2015) DDoS attack protection in the era of cloud computing and software-defined networking. Comput Netw 81:308–319 22. Zheng J, Li Q, Gu G, Cao J, Yau DK, Wu J (2018) Realtime DDoS defense using COTS SDN switches via adaptive correlation analysis. IEEE Trans Inf Forensics Security 13(7):1838– 1853 23. Tseng Y, Zhang Z, Naït-Abdesselam F (2016) Controllersepa: a security-enhancing SDN controller plug-in for openflow applications. In: 2016 17th international conference on parallel and distributed computing, applications and technologies (PDCAT). IEEE, pp 268–273 24. Badotra S, Singh J (2019) Creating firewall in transport layer and application layer using software defined networking. In: Innovations in computer science and engineering. Springer, pp 95–103 25. Hameed S, Ahmed Khan H (2018) SDN based collaborative scheme for mitigation of DDoS attacks. Future Internet 10(3):23 26. Pal C, Veena S, Rustagi RP, Murthy KNB (2014) Implementation of simplified custom topology framework in Mininet. In: 2014 Asia-Pacific conference on computer aided system engineering (APCASE). IEEE, pp 48–53 27. Shalimov A, Zuikov D, Zimarina D, Pashkov V, Smeliansky R (2013) Advanced study of SDN/OpenFlow controllers. In: Proceedings of the 9th central & eastern European software engineering conference in Russia. ACM, p 1 28. Chen M, Qian Y, Mao S, Tang W, Yang X (2016) Software-defined mobile networks security. Mob Netw Appl 21(5):729–743 29. Badotra S, Panda SN (2020) SNORT based early DDoS detection system using Open daylight and open networking operating system in software-defined networking. In: Cluster Computing 30. Oktian YE, Lee S, Lee H, Lam J (2017) Distributed SDN controller system: a survey on design choice. Comput Netw 121:100–111 31. Kaur S, Singh J, Ghumman NS (2014) Network programmability using POX controller. In: ICCCS international conference on communication, computing & systems, vol 138, IEEE 32. Ryu SDN Controller https://osrg.github.io/ryu/. Accessed on 29 Apr 2019 33. Shalimov A, Zuikov D, Zimarina D, Pashkov V, Smeliansky R (2013) Advanced study of SDN/OpenFlow controllers. In: Proceedings of the 9th central & eastern European software engineering conference in Russia. ACM, p 1 34. Open Networking Foundation: ONF (2019). https://www.opennetworking.org. Accessed on 01 May 19 35. Badotra S, Panda SN Evaluation and comparison of OpenDayLight and open networking operating system in software-defined networking. Cluster Comput 1–11

Detection and Prevention from DDoS Attack

217

36. Badotra S, Singh J (2017) Open daylight as a controller for software defined networking. Int J Adv Res Comput Sci 8(5) 37. Badotra S, Panda SN (2019) A review on software-defined networking enabled iot cloud computing. IIUM Eng J 20(2):105–126 38. Morales LV, Murillo AF, Rueda SJ (2015) Extending the floodlight controller. In: 2015 IEEE 14th international symposium on network computing and applications. IEEE, pp 126–133

Dynamic Resource Aware Scheduling Schemes for IEEE 802.16 Broadband Wireless Networks M. Deva Priya1(B) , A. Christy Jeba Malar2 , S. Sam Peter1 , G. Sandhya1 , L. R. Vishnu Varthan1 , and R. Vignesh1 1 Department of Computer Science & Engineering, Sri Krishna College of Technology,

Coimbatore, Tamil Nadu, India {m.devapriya,sampeter.s,sandhya.g,18tucs253, 18tucs248}@skct.edu.in 2 Department of Information Technology, Sri Krishna College of Technology, Coimbatore, Tamil Nadu, India [email protected]

Abstract. The scheduling algorithms for IEEE 802.16 standard are designed with the predominant goals of throughput optimization, ensuring fairness and Quality of Service (QoS) provisioning. In this work, enhancements are proposed to the existing Weighted Fair Queuing (WFQ) and Deficit Weighted Round Robin (DWRR) scheduling algorithms to efficiently utilize the unused units. In WFQ, additional units may be assigned to a queue, thus reducing the service rate. Instead in Enhanced WFQ (EWFQ), multiple queues are served in a round by effectively utilizing the unexploited units. In DWRR, a queue is not serviced if the size of the packet at the front of the queue exceeds the available quantum. Enhanced DWRR (EDWRR) checks for packets with sizes less than the Deficit Counter (DC), sorts the queue and services a smaller packet in the current round. Further, if the queue that is currently served becomes empty, the DC is transferred to the ensuing active queue instead of making it zero. This helps in servicing more number of packets in a round. The proposed scheduling schemes are proficient in servicing specific traffic flows. Keywords: WiMAX · Scheduling · WFQ · DWRR

1 Introduction Wireless Interoperability for Microwave Access (WiMAX) supports Broadband Wireless Access (BWA) in Metropolitan Area Networks (MANs). It is suitable for applications that demand high bandwidth. The WiMAX framework includes information about the types of service flows and the schedulers supported by the standard. As the scheduling mechanisms are not defined, the service providers can choose their own scheduling mechanisms based on their requirements to schedule packets.

© Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_22

Dynamic Resource Aware Scheduling Schemes for IEEE 802.16

219

The Mobile Stations (MSs) have the updated details of their queues and hence they can make scheduling decisions concerning service flows associated with them. The service flows include information about the requests sent by the MS for Uplink (UL) bandwidth allocation and the method of working of the Base Station-UL(BS-UL) scheduler. The BS is responsible for assigning time slots to the MS. In this chapter, enhancements are propounded to the existing Weighted Fair Queuing (WFQ) and Deficit Weighted Round Robin (DWRR) scheduling algorithms to support different classes of traffic. Enhanced WFQ (EWFQ) assigns a queue’s unexploited resource to the subsequent queues. In the same way, Enhanced DWRR (EDWRR) accommodates more number of requests in a round by allocating a queue’s outstanding Deficit Counter (DC) to the subsequent queues. The chapter is organized as follows. Section 2 discusses about the work done by various authors. Section 3 and Sect. 4 detail the steps of DWRR and EDWRR respectively. Similarly, Sect. 5 deals with WFQ, while Section 6 details about EWFQ. Section 7 discusses the performance of the scheduling schemes and Sect. 8 gives the conclusion.

2 Related Work Several scheduling algorithms are proposed by various authors. Some schedulers such as WFQ [1], Weighted Round Robin (WRR) [2], Packet-by-packet Generalized Processor Sharing (GPS) scheme [3], DWRR [4], Earliest Deadline First (EDF) [5] and their alternates are used for intra-class scheduling [6]. Lakkakorpi et al. [1] have compared the performance of various scheduling algorithms like Deficit Round Robin (DRR), Proportional Fair (PF) and DWRR. When channel conditions are taken into consideration in PF and WDRR, throughput is improved. Belghith and Nuaymi [7] have concentrated on rtPS service flows. In the proposed modified maximum Signal-to-Interference Ratio (mmSIR) technique, the mean sojourn time and the throughput are improved. Ghosh et al. [8] have proposed a dynamic scheduler for IEEE 802.16j networks. To deal with NP-Hard problem, a scheduling heuristic is proposed for Orthogonal Frequency-Division Multiple Access (OFDMA) based relay networks. The zone boundaries are computed by OFDMA Relay Scheduler (ORS) depending on the count of relays and MSs, bandwidth needs and state of links. Time slots and subchannels are allocated depending on the demands of nodes. Ting et al. [9] have propounded Random Early Detection based Deficit Fair Priority Queue (RED-based DFPQ) scheduling technique. It modifies the Deficit Counter (DC) of rtPS based on the queue length. The low priority service classes starve to a minimal level in contrast to other scheduling schemes. Chung et al. [10] have proposed Dynamic Clique-based Link Scheduling algorithm (DCLS), wherein the number of parallel transmissions is improved by effectively allocating bandwidth. Mardin and Alfool [11] have propounded Modified Weighted Round Robin (MWRR) and have compared its performance with WRR, Strict Priority (SP) and WFQ. Chandur et al. [12] have compared the performance of PF, Modified Longest Weighted Delay First (MLWDF) and Exponential Rule (ER) scheduling algorithms. A variant of ER called EXPQW is proposed, wherein the weights are assigned to subscribers based on the waiting time and queue length so as to ensure fairness and QoS for non-real-time applications. Hierarchical schedulers that

220

M. Deva Priya et al.

use a combination of ER for waiting time and queue length along with other scheduling rules are proposed. EXPQW and hierarchical schedulers yield better results in contrast to the PF and MLWDF schedulers in moderately and heavily loaded scenarios. El-Shennawy et al. [13] have propounded Variably Weighted Round Robin (VWRR) to assign bandwidth to users. The performance is compared with that of WRR and MDRR. The proposed scheme offers better reliability and scalability. Chern et al. [14] have proposed Enhanced RED-based Scheduling, wherein the initial DeficiCounter is taken as a constant when the queue length exceeds the threshold. Heavily loaded service flows involve more delay and hence assigned a larger initial DC. Chang et al. [15] have designed Scheduling Algorithm with Dynamic Programming (SADP), an optimal scheduling scheme that deals with chances of spatial reuse and improves throughput based on bandwidth demands and network topology of the subscriber. Further, a Heuristic Scheduling Algorithm (HSA) is propounded to deal with the computational complexity. SADP yields the maximum throughput and involves less transmission time, while HSA offers optimal results. Safa and Khayat [16] have proposed a scheduling scheme based on preemption that concentrates on improving the QoS requirements of real-time service flows. Separate schedulers are hosted at both the MS and the BS. Patel and Dalal [17] have designed Dynamically Weighted Low Complexity Fair Queuing (DWLC-FQ), an enhancement of WFQ and Worst-case Fair WFQ+ (WF2Q+). Weights are dynamically adjusted to handle the dynamics of data traffic. Teixeira and Guardieiro [18] have designed scheduling algorithms that are directly applied to bandwidth request queues to support real-time applications. Dighriri et al. [19] have compared the performance of PQ, FIFO and WFQ for data traffic in 5G networks. Performance degradation occurs in WFQ due to multiple flows in a single queue. Shareef et al. [20] have presented a Class-Based QoS Scheduling (CBS) to ensure QoS guarantee in downlink stream communication. It supports diverse types of traffic and offers considerable throughput. Yadav et al. [21] have propounded a hybrid scheduler based on WFQ and RR which has a splitter for File Transfer Protocol (FTP) traffic. Proper load distribution is provided by two-stage priority scheduling. It is suitable for heterogeneous traffic applications, but the dynamic nature of wireless networks demand throughput optimization. Khanna and Kumar [22] have designed a hybrid scheme comprising of the First come First Serve (FCFS) and RR algorithms to handle heterogeneous traffic services in Long Term Evolution (LTE) based networks. Signal-to-Noise-Ratio (SNR) and the Bit Error Rate (BER) are measured. Though the proposed algorithm is hybrid in nature, it is not capable of dealing with dynamic loads and congestion. Ahmed et al. [23] have designed a QoS framework by providing a two-level scheduling algorithm for video traffic. They have dealt with ensuring QoS by avoiding starvation and improving scalability.

3 Deficit Weighted Round Robin In Deficit Weighted Round Robin (DWRR), queues with packets of different sizes are considered. In contrast to DRR, the flows are assigned constant weights [24]. The weight of a queue is based on the bandwidth assigned to it. Deficit Counter (DC) is the number

Dynamic Resource Aware Scheduling Schemes for IEEE 802.16

221

of bytes transferred by a queue in a round. Quantum depends on the queue’s weight stated in bytes. DWRR permits transmission of whole packets. If DC is not sufficient, then the packet is ignored until the ensuing round. The maximum permissible packet size is deducted from the offered DC. The queue is ignored and the quantum is added to the credit. The DC for the subsequent round is based on the computed value to service the packet at the front of the queue. As packets are dequeued, the DC is reduced by the transmitted packet size until the size of the packet at the front of the queue is larger than the DC or there are no packets in the queue. Once the queue becomes empty, the DC is made zero and the queue is deactivated, which in turn is removed from the ActiveList. Based on the number of active queues, bandwidth is effectively allocated.

4 Enhanced Deficit Weighted Round Robin In the existing DWRR, if the quantum is not sufficient, the packet at the front of a queue cannot be served in the current round. The queue is ignored in the present round, and the packet is held until the ensuing round. There are chances for a queue to contain a smaller packet that can be served in the current round, which in turn will reduce the delay. To handle this challenge, Enhanced DWRR (EDWRR) algorithm is propounded to deal with packets of sizes less than the DC. The queues are sorted independently and packets smaller in sizes are serviced in the current round. Figure 1 shows the enqueue and dequeue algorithms of EDWRR. In EDWRR, if a queue becomes empty, its DC is transferred to the subsequent live queue. The increase in DC of the currently served queue enables servicing number packets in a round. Algorithm IDWRR initialize(Q) for every queue Set Deficit Counter=0 Compute Quantum = Weight * Bandwidth End End enqueue(Q, P) Select the queue of the service flow pertaining to the incoming packet if (!InActiveList(Q)) then activate(Q) initialize(Q) End Add P to Q End

dequeue(Q) while (!isEmpty(ActiveList)) then Select Q from the Active List Deficit Counter[Q]+= Quantum[Q] while(!isEmpty(Q)) if (Deficit Counter[Q] ≤ Size(P)) then Arrange packets in the increasing order of Size(P) else Transmit P Deficit Counter[Q]-= Size(P) End End if (isEmpty(Q)) then Transfer the outstanding Deficit Counter[Q] to the ensuing Queue in the Active List deactivate(Q) else activate(Q) End End End

Fig. 1. EDWRR algorithm

222

M. Deva Priya et al.

5 Weighted Fair Queuing GPS [25, 26] is an overview of unvarying processor sharing. Packet-based form is termed as WFQ [27]. WFQ supports varied priorities and assured bandwidth to traffic flows, and connects numerous sessions to a common link. Resources are impartially assigned to flows. Bursty flows do not demand too much of bandwidth. The buffer consists of multiple queues with packets of diverse flows [28]. The active list contains a list of active queues. An arriving packet is added to the respective queue and stamped with the finish time that is based on the link bandwidth, number and weight of queues and packet size. The queue length is based on the time taken for packet transmission. The queue containing packets with early finish time is served first [29, 30]. Every queue has its weight and every flow has its own bandwidth demands, circumventing take-over of resources by a flow [31, 32]. Queues with a more weighted portion of available resources are serviced regularly than insignificant ones [33]. As weights are dynamically controlled, WFQ can be used for monitoring the QoS. The computational overhead is reduced by finding the virtual time of a packet which is based on the finish time of the previous packet and the size of the packet that is transmitted. The order of transmission is based on the virtual finish time.

6 Enhanced Weighted Fair Queuing In WFQ, the amount of available bandwidth is constant. There may be situations wherein, the quantity of demanded bandwidth may be less in contrast to the amount of active bandwidth. Additional amount of resources may be allocated than demanded. The rate of service may be reduced as these excess units are not made available to other queues. In the proposed Enhanced WFQ (EWFQ) algorithm, the unexploited bandwidth is passed to the subsequent queue, promising similar QoS without involving added delay. As several queues are served in a round, the throughput is improved. On servicing a queue, its demanded bandwidth is made zero and the queue is deactivated. Figure 2 shows the enqueue and dequeue algorithms of EWFQ. The units left after a queue is served are included in the set of accessible units, consequently improving the service rate of ensuing queues. If a queue cannot be served due to insufficient number of units, then a portion equivalent to the amount of active bandwidth is served until the entire packet is done. The bandwidth assigned to flow is updated in each round. A queue is deactivated once it becomes empty; else it is added to the active list.

7 Results and Discussion The system is simulated using ns2 and the performance of the scheduling algorithms for all classes of traffic is analyzed in terms of Packet Delivery Ratio (PDR), throughput, Packet Loss Ratio (PLR), delay and jitter. Various works are available to deal with scheduling different traffic classes, authentication and efficient allocation of resources in WiMAX networks [34–38].

Dynamic Resource Aware Scheduling Schemes for IEEE 802.16 Algorithm EWFQ initialize (i) for i = 1 to q Finish [i] = 0 Rlink = k Roundcurrent [i] = 0 Ractive [i] = 0 End End enqueue(i, k) Select the appropriate queue for the packet if (!InActiveList(i)) then activate(i) initialize(i) q+=1 End if (isempty(i)) Finish[i] = Rk(t) + Pk / Wi else Finish[i] = Finishk-1[i] + Pk / Wi End Insert the packet into Queue ‘i’ Rreq [k,i] = Finish [i] End

223

dequeue() while (!isemptyActiveList) then n=q, i =1 do Ractive [i] = Rlink / q if (Rreq [k,i] ≤ Ractive[i]) Service queue ‘i’ Rallocated[i] = Ractive[i] Rleft = Ractive[i] - Rreq[k,i] Roundcurrent [i] += Rreq[k,i] Rreq[k,i] = 0 deactivate(i) q - =1 for (j = i +1; j ≤ q && Rleft !=0; j+=1) if (Rreq [k,j] ≤ Rleft ) Sevice queue ‘j’ Rleft - = Rreq [k,j] Roundcurrent [j] += Rreq[k,j] Rreq[k,j] = 0 deactivate(j) q - =1 End End Rlink = Rlink - Rallocated [i] + Rleft else Sevice queue ‘i’ Roundcurrent [i] + = Rreq[k,i] Rreq [k,i] - = Ractive [i] End i=i+1 while (q!=0) for every queue if (!isemptyQueue(i)) then activate(i) else deactivate(i) End End End End

Fig. 2. EWFQ algorithm

In this section, the existing and proposed algorithms are applied to schedule the different traffic flows-Unsolicited Grant Service (UGS), real-time Polling Service (rtPS), nonreal-time Polling Service (nrtPS) and Best Effort (BE). It is seen that EWFQ offers better results for UGS and rtPS service classes. Correspondingly, EDWRR yields improved results for nrtPS and BE service classes. Simulation parameters are listed in Table 1. Appropriate scheduling schemes can be chosen for each service class. BE services demand minimal service assurances, while nrtPS support loose delay demands. Hence, EDWRR is suitable for nrtPS and BE traffic flows. The outstanding DC is moved to the subsequent queue, thus improving the service rate. Similarly, as the available bandwidth is proficiently used in EWFQ, it is capable of dealing with delay-sensitive rtPS and UGS service classes.

224

M. Deva Priya et al. Table 1. Simulation parameters Parameter

Value

MAC protocol

IEEE 802.16e

Routing protocol

DSDV

Modulation scheme

OFDM_QPSK

Queue length

50

Queue type

Drop tail/WFQ

Bandwidth

50 Mbps

Packet size UGS and nrtPS 1024 BE and rtPS

512

Transmission range

250–400 m

Number of mobile stations

100

Speed

1–40 ms−1

Simulation time

80 s

7.1 EWFQ for UGS Service Class EWFQ yields better results for UGS type of traffic. EWFQ yields 1.13, 1.24 and 1.1 time(s) improved PDR, 10.21, 31.42 and 3.13 times better throughput, 11.71, 33.14 and 2.91 times reduced delay, 13.62, 34.14 and 3.23 times lesser jitter and 5.34, 15.63 and 2.42 times reduced PLR in contrast to WFQ, DWRR and EDWRR respectively (Fig. 3). 7.2 EWFQ for rtPS Services EWFQ is suitable for rtPS traffic. EWFQ yields 1.14, 1.23 and 1.24 times better PDR, 6.32, 8.23 and 1.24 times improved throughput, 9.42, 13.43 and 1.61 times reduced delay, 10.23, 13.41 and 2.92 times less jitter and 10.31, 13.42 and 3.12 times reduced PLR in contrast to WFQ, DWRR and EDWRR respectively (Fig. 4). 7.3 EDWRR for nrtPS Services EDWRR is suitable for nrtPS traffic. EDWRR yields 1.23, 1.14 and 1.23 times improved PDR, 14.21, 5.81 and 1.13 times better throughput, 11.32, 7.23 and 2.94 times reduced delay, 12.72, 8.23 and 2.72 times lesser jitter and 12.41, 9.44 and 2.53 times reduced PLR in contrast to WFQ, DWRR and EWFQ respectively (Fig. 5). 7.4 EDWRR for BE Services EDWRR is suitable for BE traffic. EDWRR yields 1.22, 1.25 and 1.3 time(s) better PDR, 14.32, 6.34 and 1.52 times improved throughput, 9.42, 8.47 and 1.63 times lesser delay,

Dynamic Resource Aware Scheduling Schemes for IEEE 802.16

225

Fig. 3. Performance of the Scheduling Schemes for UGS traffic

10.73, 8.62 and 2.31 times reduced jitter and 14.32, 10.71 and 3.45 times lesser PLR when compared to WFQ, DWRR and EWFQ respectively (Fig. 6). It is evident that EDWRR is suitable for scheduling BE and nrtPS traffic, while EWFQ is appropriate for UGS and rtPS classes of traffic.

8 Conclusion Weighted Fair Queuing (WFQ) and Deficit Weighted Round Robin (DWRR) scheduling algorithms are enhanced to schedule delay-sensitive and delay tolerable services respectively. From the results, it is evident that Enhanced WFQ (EWFQ) is best suited for real-time traffic, while Enhanced DWRR (EDWRR) is suitable for non-real-time flows. In EWFQ, once a flow is served, the remaining bandwidth is made available to

226

M. Deva Priya et al.

Fig. 4. Performance of the Scheduling Schemes for rtPS traffic

the subsequent active queues. Similarly, in EDWRR, if the Deficit Counter (DC) is not adequate for the packet at the front of the queue, the queue is sorted based on the packet size. Further, the remaining DC is moved to the next queue instead of making it zero. These enhancements enable the enhanced schemes to serve more requests in a round.

Dynamic Resource Aware Scheduling Schemes for IEEE 802.16

Fig. 5. Performance of the Scheduling Schemes for nrtPS traffic

227

228

M. Deva Priya et al.

Fig. 6. Performance of the Scheduling Schemes for BE traffic

References 1. Lakkakorpi J, Sayenko A, Moilanen J (2008) Comparison of different scheduling algorithms for WiMAX base station: deficit round-robin versus proportional fair versus weighted deficit round-robin. In: IEEE wireless communications and networking conference, pp 1991–1996 2. Shin J, Kim J, Kuo CCJ (2000) Content-based packet video forwarding mechanism in differentiated service networks. In: IEEE packet video workshop 3. Guesmi H, Maaloul S, Tourki R (2011) Design of scheduling algorithm for QoS management on WiMAX networks. J Comput Sci Eng 1(2):43–50 4. Kwon TG, Lee SH, Rho JK (1998) Scheduling algorithm for real-time burst traffic using dynamic weighted round robin. In: IEEE international symposium on circuits and systems, vol 6, pp 506–509

Dynamic Resource Aware Scheduling Schemes for IEEE 802.16

229

5. Singla S, Kamboj MS (2012) Analysis of packet scheduling in WiMAX network. In: International conference on recent advances and future trends in information technology 6. Puranik SR, Vijayalakshmi M, Kulkarni L (2013) A survey and analysis on scheduling algorithms in IEEE 802.16e (WiMAX) standard. Int J Comput Appl 79(12) 7. Belghith A, Nuaymi L (2008) Comparison of WiMAX scheduling algorithms and proposals for the rtPS QoS class. In: 14th European wireless conference, pp 1–6 8. Ghosh D, Gupta A, Mohapatra P (2009) Adaptive scheduling of prioritized traffic in IEEE 802.16j wireless networks. In: IEEE international conference on wireless and mobile computing, networking and communications, pp 307–313 9. Ting PC, Yu CY, Chilamkurti N, Tung-Hsien W, Shieh CK A proposed RED-based scheduling scheme for QoS in WiMAX networks. In: 4th IEEE international symposium on wireless pervasive computing, pp 1–5 (2009) 10. Chung TY, Fu HL, Lee HC (2010) Dcls: a fair and efficient scheduling with bandwidth guarantee in wimax mesh networks. J Inf Sci Eng 26(4):1429–1442 11. Mardini W, Alfool MA (2011) Modified WRR scheduling algorithm for WiMAX networks. Netw Protocols Alg 3(2):24–53 12. Chandur P, Karthik RM, Sivalingam KM (2012) Performance evaluation of scheduling algorithms for mobile WiMAX networks. In: IEEE International Conference on Pervasive Computing and Communications Workshops, pp 764–769 13. El-Shennawy NM, El-Derini MN, Youssef MA, Fahmy MM (2013) A dynamic uplink scheduling scheme for WiMAX networks. In: 2013 8th IEEE International Conference on Computer Engineering & Systems (ICCES), pp 111–115 14. Chern HT, Liu Y, Jhou JS (2013) Enhanced red-based scheduling (ERBS) scheme in WiMAX network. In: IEEE International Computer Science and Engineering Conference (ICSEC), pp 276–280 15. Chang CY, Li MH, Huang WC, Lee SC (2013) An optimal scheduling algorithm for maximizing throughput in WiMax mesh networks. IEEE Syst J 9(2):542–555 16. Safa H, Khayat S (2015) A preemption?based scheduling algorithm for WiMAX networks. Wirel Commun Mob Comput 15(3):584–600 17. Patel Z, Dalal U (2016) Implementation and evaluation of dynamically weighted low complexity fair queuing (DWLC-FQ) algorithm for packet scheduling in WiMAX networks. China Commun 13(5):128–140 18. Teixeira MA, Guardieiro PR (2013) Adaptive packet scheduling for the uplink traffic in IEEE 802.16e networks. Int J Commun Syst 26(8):1038–1053 19. Dighriri M, Alfoudi ASD, Lee GM, Baker T, Pereira R (2017) Comparison data traffic scheduling techniques for classifying QoS over 5G mobile networks. In: 31st IEEE international conference on advanced information networking and applications workshops, pp 492–497 20. Shareef ZA, Hussin M, Abdullah A, Muhammed A (2018) Class-based QoS scheduling of WiMAX networks. J High Speed Netw 24(4):345–362 21. Yadav AL, Vyavahare PD, Bansod PP (2018) Proposed WiMAX hybrid scheduler with split FTP traffic and its performance evaluation. Int J Wirel Microwave Technol 6:1–14 22. Khanna R, Kumar N (2019) Quality of service aware traffic scheduling algorithm for heterogeneous wireless networks (Doctoral dissertation) 23. Ahmed Z, Hamma S, Nasir Z (2019) An optimal bandwidth allocation algorithm for improving QoS in WiMAX. Multimed Tools Appl 1–40 24. Shreedhar M, Varghese G (1996) Efficient fair queuing using deficit round-robin. IEEE/ACM Trans Netw 4(3):375–385 25. Parekh AK, Gallager RG (1993) A generalized processor sharing approach to flow control in integrated services networks: the single-node case. IEEE/ACM Trans Netw 1(3):344–357 26. Parekh AK, Gallagher RG (1994) A generalized processor sharing approach to flow control in integrated services networks: the multiple node case. IEEE/ACM Trans Netw 2(2):137–150

230

M. Deva Priya et al.

27. Demers A, Keshav S, Shenker S (1989) Analysis and simulation of a fair queueing algorithm. ACM SIGCOMM Comput Commun Rev 19(4):1–12 28. Taniguchi S, Kawate R, Sato K, Horiuchi E, Yokotani T (2012) Performance evaluation of the simplified WFQ to multiplex a huge number of queues. In: Proceedings of the IEEE international workshop technical committee on communications quality and reliability, pp 1–6 29. Georges JP, Divoux T, Rondeau E (2005) Strict priority versus weighted fair queueing in switched ethernet networks for time critical applications. In: 19th IEEE international parallel and distributed processing symposium, p 141 30. Kautz R, Keh R, Chua KC, Leon-Garcia A (2000) A distributed fair queueing (DFQ) architecture for wireless ATM local access networks. Int J Wirel Inf Netw 7(4):221–229 31. So-In C, Jain R, Tamimi AK (2009) Scheduling in IEEE 802.16e mobile WiMAX networks: key issues and a survey. IEEE J Sel Areas Commun 27(2):156–171 32. Chuck D, Chang JM (2010) Bandwidth recycling in IEEE 802.16 networks. IEEE Trans Mob Comput 9(10):1451–1464 33. Zhang H (1995) Service disciplines for guaranteed performance service in packet-switching networks. IEEE 83(10):1374–1396 34. Priya MD, Valarmathi ML, Sundarameena V (2013) Efficient QoS-based Scheduling Mechanisms for IEEE 802. 16e Networks. Int J Comput Appl 72(20):34–41 35. Priya D, Valarmathi M, Prithviraj D (2014) Performance analysis of scheduling mechanisms in WiMAX networks. SOP Transactions Wirel Commun 1(1):40–49 36. Priya MD, Janakiraman S, Sandhya G, Aishwaryalakshmi G (2019) Efficient preauthentication scheme for inter-ASN handover in high mobility MANET. Wirel Networks 1–15 37. Priya MD, Sangeetha M, Malar ACJ, Dhivyaprabha E, Kiruthiga N, Rajarajeswari PL (2019) Fair adaptive cross-layer resource allocation scheme for IEEE 802.16 broadband wireless networks. Wirel Pers Commun 1–22 38. Priya MD, Suganya T, Malar ACJ, Dhivyaprabha E, Prasad PK, Vardhan LV (2020) An efficient scheduling algorithm for sensor-based IoT networks. In: Inventive communication and computational technologies, pp 1323–1331 Springer, Singapore

Evaluation of the Applicability and Advantages of Application of Artificial Neural Network Based Scanning System for Grid Networks Shubhranshu Kumar Tiwary(B) , Jagadish Pal, and Chandan Kumar Chanda Electrical Engineering Department, IIEST, Shibpur, Howrah, West Bengal 711103, India [email protected], [email protected], [email protected]

Abstract. This chapter presents the application of an artificial neural networkbased monitoring system power grid network. Neural net modules used for this study are of two kinds, a distributed separate artificial neural net (ANN) module to monitor all lines individually from separate points in the network and central common multiple-input, multiple-layer ANN to monitor all lines together. Only the active power flowing on all the lines of the utility network were monitored using the ANN’s. This work elaborates and evaluates the technical repercussions of both the modules. The ANN model employed was a feed-forward net with backpropagation of error. The aspiration of the task is to deliberate on the opportunities and obstacles of the various configurations of ANN models employed. Keywords: Artificial neural network · Grid system · Power system computing · Power system control · Power system security

1 Introduction The radical and astonishing novel progress in the field of computation during the previous four decades have inspired engineers and researchers to consider machine intelligence and pattern recognition with revived vitality, as deduced from Refs. [1–5] and many similar articles available in many pioneer journals [6–10]. Utilization of pattern recognition in any area alleviates such intricate complications that are difficult to identify otherwise [11, 12]. And the speed of response is also comparatively much faster for a machine learning unit when compared to other modes [13, 14]. ANN is the building block of pattern recognition and machine intelligence and its utilization to obtain any objective is very difficult unless it is properly trained [15]. Application of ANN methodologies to any field has its own intrinsic complexities for which an engineer has to be properly educated [16]. Pal et al. have shown in [16, 17], how the ANN models were used for identifying the critical contingency cases after the security assessment cycle was completed, but it serves no practical purpose and the time consumption is still too much for the ANN © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_23

232

S. K. Tiwary et al.

to be considered as a viable alternative to the currently practiced methods. In [18] the authors have gone one step ahead and have developed a monitoring method for the crucial power system parameters such as active power (MW) and reactive power (MVAR) using designated ANN for each line separately, which can be used to identify the contingencies of any line or any disturbance in the power system, immediately. Similarly, in [19], the authors have used a common ANN to monitor all the similar parameters using a common ANN, which can be useful for the Load Dispatch Centers to identify the stability of the power network as a whole. In this chapter, neural net models have been applied to track the MW power circulating on the transmission lines of Damodar Valley Corporations’ (DVC) power grid [41–44], and to identify the insecure or secure state of the active power flowing [16, 17]. As elaborated [18, 19] in two modes of ANN monitoring modules, “distributed, single-input small neural net to monitor each line separately, from different points in the power system, and multiple-input, multiple-layer neural net, to monitor all lines together from a common point”, had been examined. In this work, the opportunities and obstacles of both the modules have been discussed and then some conclusions were derived from the work and the advantage and disadvantage of the methods are discussed. This chapter is standardized as follows. Section 1 proposes the work, Sect. 2 elaborates the network under study, Sect. 3 describes the neural network model and its architecture, Sect. 4 explains the software used for simulation and the models developed, Sect. 5 deliberates the outputs of simulation and ANN monitoring and Sect. 6 concludes the chapter, followed by acknowledgements and references.

2 DVC Grid Network The diagram for the DVC’s grid network is as shown in Fig. 1. For a given loading condition, based on day-ahead load forecast, the unit commitment, and scheduling, a load flow program is run to calculate the loading conditions of the transmission lines [20, 21]. After the line-loadings are calculated, it is given a differential calculative-error margin of ±10%, to generate the apt amount of data for training the artificial neural network that will be used for monitoring the lines of the network [20–25]. The contingency analysis is also performed for the event of an outage of any of the lines in the above-mentioned load forecast scenario, to identify the critical flow gates which may get overloaded causing a cascade event, which may then lead to the blackout of a large part of the power grid [26–30]. The predominant hitch in the traditional process of load flow methods-based security monitoring procedures is that it is prevalently iterative and computation-intensive in nature, which exhausts most of the time allotted to the power system operator for protective and control actions [31–40].

3 Types and Description of ANN As explained previously, two different modes ANN modules, dispersed, individual input simple neural net for observing each line individually, from distinct places in the power network, and a multiple-input, multiple-layer complex neural net, for tracking all lines

Evaluation of the Applicability and Advantages of Application …

233

Fig. 1. Damodar valley corporation’s DVC-46 bus network

collectively from a central point, have been performed [16–19]. They are elucidated below. 3.1 Multiple-Input Multiple-Layer Common ANN As mentioned in Sect. 2, taking into consideration the differential calculative-error margin of ±10%, the secure range training datasets for the ANN’s are developed, based on load forecasting, unit commitment, scheduling, and transmission line limits. The power flows beyond the aforementioned range is assumed to be insecure for ANN training purposes. It is presumed that the MW power flow beyond the calculated range will provoke instability and insecurity in the power grid [19]. After taking into consideration the previously mentioned prerequisites, a training data set is established for ANN development of each line. All these datasets are then tabulated into a single datafile, which is used to develop the supervised ANN model. The input nodes in this ANN will be equal to line parameters to be monitored and to preclude any confusion, one multiple-input-multiple-layer ANN will be used to monitor only one type of data. For this case, we develop one ANN for the purpose of monitoring MW power flow on all the network lines. This way the overall trend of MW power flow deviation on the whole power network can be monitored. The common ANN monitor is as shown in Fig. 2.

234

S. K. Tiwary et al.

Fig. 2. Common-ANN monitor for DVC-46 bus network

3.2 Distributed Single-Input Small ANN In this case, every line is monitored by dedicated specifically trained ANN [18]. The procedure for the training and development for the separate ANN is similar to the previous method, with the only difference that there is separate datafile containing training data for each line to be monitored. After the training converges, we have separate ANN monitors for each line. It is as shown in Fig. 3.

Fig. 3. Separate ANN monitor for DVC-46 bus network

Thereafter, the ANN monitor is simulated in both the scenarios, and the results are discussed in the subsequent section [45–51].

4 Simulation The simulation was performed with the help of the Neural Network Toolbox on Simulink and MATLAB (R2015a) using the OPAL RT’s OP-5600 simulator. The device had the frequency step-size of 50 µs [18, 19], and the simulation was run in real-time with hardware in the loop for a duration of 30 min. Three major faults were brought about on crucial lines of the network, in the duration of the test. They were, a two-line fault on line-3 at 10 min, a three-line-to-ground fault on line-9 at 18 min, and a three-line fault on line-21 at 25 min, respectively. The Simulink model is similar to the one as shown in Fig. 4.

Evaluation of the Applicability and Advantages of Application …

235

Fig. 4. Simulink model of power network with one ANN monitoring each line’s MW power. In common ANN the network model is the same but there is only one ANN monitoring all lines

5 Results of Monitoring The developed network was simulated for a period of 30 min using both ANN modules separately and their responses were recorded in memory. Thereafter, to observe and analyze the performance of the ANN modules, the recorded data was mapped out in a graph. The details of the ANN’s recorded outputs have been elucidated in the following segments.

236

S. K. Tiwary et al.

Fig. 4. (continued)

5.1 Multiple-Input-Multiple-Layer Common ANN The output plot of the central common ANN tracking and monitoring model has been shown in Fig. 5 [19].

Fig. 5. Common ANN monitoring response

As shown in Fig. 4, the plot of the common ANN monitoring response, it is evident that the two-line fault on line-3 at 10 min and line-9’s, three-line-to-ground fault at 18 min affect the stability and the security of the whole power network for several seconds, but then the power flows rebound to stable state. The three-phase fault on line-21 at 25 min does not affect the power network as a whole, albeit there are some localized effects that die down promptly. The meticulous performance of these lines during the fault events can be seen more distinctly by the separate ANN monitor as discussed in the succeeding sections. In the plot diagrams, “the X-axis shows time in seconds and the Y-axis shows the magnitude of ANN output, between 0 (Secure) and 1 (Insecure)”, as shown in [19]. 5.2 Distributed Single-Input Small Individual ANN The response of the separate ANN is more detailed and the output of the ANN monitor gives an intricate insight into the behavior of the power flows on each line. The response

Evaluation of the Applicability and Advantages of Application …

237

of the distributed single-input small ANN monitoring modules, tracking the MW power flowing on each of the lines is elucidated below [18]. Since there are 56 lines in the DVC-46 bus system, to preserve page-space, here the similar types of responses have been represented under the same figure as elaborated below [18]. Figure 6, shows ANN response of lines affected by all faults.

Fig. 6. Separate ANN monitoring response for lines affected by all faults

From Fig. 6, the ANN monitoring response of the lines getting affected by all three faults is shown. Here, lines 1, 2, 3, 4, 5, 21, and 29 are affected by all three faults and the response of the ANN is more or less similar to the one shown in Fig. 6. The ANN response of the lines affected by only the first two faults is as shown in Fig. 7. These lines are lines 6, 8, 10, and 56.

Fig. 7. Separate ANN monitoring response for lines affected by the first two faults

The ANN response of the lines affected by the last two faults are as shown in Fig. 8. These lines are lines 17, 18, 19, and 20. The ANN response of the lines affected by the first and the last faults are as shown in Fig. 9. These are lines 38, 40, 43, and 44. The ANN response of the lines affected by only the first fault is as shown in Fig. 10. These lines are lines 11, 12, 13, and 14. The ANN response of the lines affected by only the second fault is as shown in Fig. 11. Only line 16 is affected by the second fault. The ANN response of the lines affected by only the last fault is as shown in Fig. 12. Only line 15 is affected by the last fault.

238

S. K. Tiwary et al.

Fig. 8. Separate ANN monitoring response for lines affected by the last two faults

Fig. 9. Separate ANN monitoring response for lines affected by first and last faults

Fig. 10. Separate ANN monitoring response for lines affected by only the first fault

The ANN response of the lines not affected by any fault is as shown in Fig. 13. Here, we can see that the ANN response stays at 0, which signifies that the active (MW) power flow on such lines remains in a secure state. They are lines 7, 9, 22, 23, 24, 25, 26, 27, 28, 30, 31, 32, 33, 34, 35, 36, 37, 39, 41, 42, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, and 55. From all of the above responses, it can be seen that the ANN’s monitoring each line gives the response of the effect of the line faults, immediately. These kinds of prompt forewarn can allow the Power System Operator (PSO) to take corrective actions punctually, thereby reducing the risk and dangers of uncontrolled equipment outage and by keeping the power system in a stable operational state [52].

Evaluation of the Applicability and Advantages of Application …

239

Fig. 11. Separate ANN monitoring response for lines affected by only the second fault

Fig. 12. Separate ANN monitoring response for lines affected by only the last fault

Fig. 13. Separate ANN monitoring response for lines not affected by any fault

6 Conclusion The above analysis has emphasized the subsequent points: (a) ANN Model Development and its Time Consumption: It takes more time to develop separate ANN monitoring modules to observe each line individually in comparison to developing a common ANN monitoring module. For example, assuming a power network has ‘n’ transmission lines and ‘t’ is the time consumption of each module generation, then the gross time expenditure for generating separate neural net models will be (n * t) but the time taken for the development of a common neural net model is only ‘t’.

240

S. K. Tiwary et al.

(b) Labor Involved: It is burdensome and time-consuming for generating disperse numerous ANN monitor modules to observe every line individually, but it’s simpler, quick, and less complex for generating the common ANN for observing all lines collectively. (c) ANN Output Observation Contrast: It may be swift and simpler to generate a common ANN model, but, while in practice it’s very arduous to single out the network lines where a fault had happened. Even the response of the multiple-input multiple-layer common ANN’s may not be factual enough. Whereas, while utilizing a separate ANN mode of observing the network, the ANN’s monitoring each line individually provide response corresponding to the state of their respective lines power flow immediately. Also, the separate ANN’s provide better summarization and extra awareness on their respective lines (faulty and otherwise) compared to the common ANN. (d) ANN Application Suitability: The denouement of this study demonstrates the economical suitability of the common ANN monitoring mode for a central, power network monitoring facility or a center for load despatch (CLD) but the separate ANN monitoring mode for each line individually could be more economical and better suited for the distinct sub-stations of network understudy or the various Grid Operation and Administration Divisions (GOAD) while applying it. (e) Contingency Analysis: It has been observed from this work that performing contingency studies on a power network while utilizing separate ANN mode of monitoring is much easier and less time-consuming compared to the application of a common ANN monitoring mode for the same purpose. This work provides an elaborate analysis of the relevance of the two configurations of supervised learning feed-forward ANN’s to monitor a power network. The results could be reproduced by utilizing other numerous types of ANN models for a better understanding and judgement of the endeavor. The training algorithms for ANN’s may also be altered to obtain a fine-tuning of the results. The end results are very inspiring and demonstrate the practicability of the ANN’s for real-time. This scheme of ANN application may also help machine learning trainees to grasp an understanding of the application of ANN for power network security. For future scope, automatic learning, decision trees, and joint ensemble of ANN’s may be used to reduce the errors incurred due to human interaction during training. Acknowledgements. This project was technically supported by the professors and staff at the Electrical Engineering Department at Jadavpur University, Kolkata, since February 2017. The financial support was arranged by the Ministry of Human Resource Development under the Indian Government and the Technical-Education-Quality-Improvement-Programme, Phase-2, under grant to registration number R/2016/0010, from February 2015.

References 1. Jain MB, Srinivas MB, Jain A (2008) A novel web based expert system architecture for on-line and off-line fault diagnosis and control (FDC) of power system equipment. In: 2008

Evaluation of the Applicability and Advantages of Application …

2. 3. 4.

5. 6.

7. 8. 9. 10.

11.

12.

13.

14.

15.

16.

17.

18.

19.

241

joint international conference on power system technology and IEEE power India conference. IEEE, pp 1–5 Sachdev MS, Agarwal R (1988) A technique for estimating transmission line fault locations from digital impedance relay measurements. IEEE Trans Power Delivery 3(1):121–129 Koushanfar F, Potkonjak M, Sangiovanni-Vincentelli A (2003) On-line fault detection of sensor measurements. In: Sensors, IEEE, vol 2. IEEE, pp 974–979 Slootweg JG, Polinder H, Kling WL (2001) Dynamic modelling of a wind turbine with doubly fed induction generator. In: 2001 Power engineering society summer meeting. Conference proceedings (Cat. No. 01CH37262), vol 1. IEEE, pp 644–649 Padulles J, Ault GW, McDonald JR (2000) An integrated SOFC plant dynamic model for power systems simulation. J Power Sources 86(1–2):495–500 Glover F (1997) Tabu search and adaptive memory programming—advances, applications and challenges. In: Interfaces in computer science and operations research. Springer, Boston, MA, pp 1–75 Bottou, L., 2010. Large-scale machine learning with stochastic gradient descent. In Proceedings of COMPSTAT’2010 (pp. 177–186). Physica-Verlag HD Rosten E, Porter R, Drummond T (2008) Faster and better: a machine learning approach to corner detection. IEEE Trans Pattern Anal Mach Intell 32(1):105–119 Jordan MI, Mitchell TM (2015) Machine learning: trends, perspectives, and prospects. Science 349(6245):255–260 Klopf AH, Morgan JS (1990) The role of time in natural intelligence: implications for neural network and artificial intelligence research. In: Advanced neural computers. North-Holland, pp 201–207 Klopf AH (1987) Drive-reinforcement learning: a real-time learning mechanism for unsupervised learning. In: Proceedings of the IEEE first annual international conference on neural networks, vol 2, pp 441–445 Zhang H, Qin C, Luo Y (2014) Neural-network-based constrained optimal control scheme for discrete-time switched nonlinear system using dual heuristic programming. IEEE Trans Autom Sci Eng 11(3):839–849 Cheng L, Hou ZG, Lin Y, Tan M, Zhang WC, Wu FX (2011) Recurrent neural network for non-smooth convex optimization problems with application to the identification of genetic regulatory networks. IEEE Trans Neural Networks 22(5):714–726 Hadjidemetriou L, Nicolaou G, Stavrou D, Kyriakides E (2016) Low-cost real-time monitoring of a laboratory scale power system. In: 2016 18th mediterranean electrotechnical conference (MELECON). IEEE, pp 1–6 Zhai J, Wang Y, Liu X (2017) Optimal reset controller designed for induction machine drive with hardware in the loop test. In: 2017 IEEE applied power electronics conference and exposition (APEC). IEEE, pp 506–511 Tiwary SK, Pal J (2017) ANN application for MW security assessment of a large test bus system. In: 2017 3rd international conference on advances in computing, communication & automation (ICACCA) (Fall). IEEE, pp 1–4 Tiwary SK, Pal J (2017) ANN application for voltage security assessment of a large test bus system: a case study on IEEE 57 bus system. In: 2017 6th International Conference on Computer Applications In Electrical Engineering-Recent Advances (CERA). IEEE, pp 332–334 Tiwary SK, Pal J, Chanda CK (2017) Mimicking on-line monitoring and security estimation of power system using ANN on RT lab. In: 2017 IEEE Calcutta conference (CALCON). IEEE, pp 100–104 Tiwary SK, Pal J, Chanda CK (2019) Application of common ANN for Similar datatypes in on-line monitoring and security estimation of power system. In: Emerging technologies in data mining and information security. Springer, pp 3–11

242

S. K. Tiwary et al.

20. Khazaei J, Piyasinghe L, Miao Z, Fan L (2014) Real-time digital simulation modeling of single-phase PV in RT-LAB. In: 2014 IEEE PES general meeting|conference & exposition. IEEE, pp 1–5 21. Wu W, Wu X, Jing L, Li J (2016) Design of modular multilevel converter hardware-in-loop platform based on RT-lab. In: 2016 IEEE 8th International Power Electronics and Motion Control Conference (IPEMC-ECCE Asia) (pp. 2350–2355). IEEE 22. Huang, H., Liu, X., Wang, H. and Wu, H., 2017, June. Fault analysis of RT-Lab semi-physical simulation for east fast control power supply. In 2017 IEEE 3rd international future energy electronics conference and ECCE Asia (IFEEC 2017-ECCE Asia). IEEE, pp 1369–1374 23. Hsu YY, Chen CR (1991) Tuning of power system stabilizers using an artificial neural network. IEEE Trans Energy Convers 6(4):612–619 24. Srinivasan D, Chang CS, Liew AC, Leong KC (1998) Power system security assessment and enhancement using artificial neural network. In: Proceedings of EMPD ’98, 1998 international conference on energy management and power delivery (Cat. No. 98EX137), vol 2. IEEE, pp 582–587 25. Aggoune ME, Atlas LE, Cohn DA, Damborg MJ, El-Sharkawi MA, Marks RJ (1989) Artificial neural networks for power system static security assessment. In: IEEE international symposium on circuits and systems. IEEE, pp 490–494 26. Sobajic DJ, Pao YH (1989) Artificial neural-net based dynamic security assessment for electric power systems. IEEE Trans Power Syst 4(1):220–228 27. Pao YH, Sobajic DJ (1992) Combined use of unsupervised and supervised learning for dynamic security assessment. IEEE Trans Power Syst 7(2):878–884 28. Aggoune M, El-Sharkawi MA, Park DC, Dambourg MJ, Marks RJ (1989) Preliminary results on using artificial neural networks for security assessment (of power systems). In: Conference papers power industry computer application conference. IEEE, pp 252–258 29. Zhou Q, Davidson J, Fouad AA (1994) Application of artificial neural networks in power system security and vulnerability assessment. IEEE Trans Power Syst 9(1):525–532 30. Jain T, Srivastava L, Singh SN, Erlich I (2005) New parallel radial basis function neural network for voltage security analysis. In: Proceedings of the 13th international conference on, intelligent systems application to power systems. IEEE, pp 7 31. Nakawiro W, Erlich I (2008) Online voltage stability monitoring using artificial neural network. In: 2008 third international conference on electric utility deregulation and restructuring and power technologies. IEEE, pp 941–947 32. Saeh IS, Khairuddin A (2008) Static security assessment using artificial neural network. In: 2008 IEEE 2nd international power and energy conference. IEEE, pp 1172–1178 33. Ding N, Benoit C, Foggia G, Bésanger Y, Wurtz F (2015) Neural network-based model design for short-term load forecast in distribution systems. IEEE Trans Power Syst 31(1):72–81 34. Bulac C, Tri¸stiu I, Mandi¸s A, Toma L (2015) On-line power systems voltage stability monitoring using artificial neural networks. In: 2015 9th international symposium on advanced topics in electrical engineering (ATEE). IEEE, pp 622–625 35. Bahmanyar AR, Karami A (2014) Power system voltage stability monitoring using artificial neural networks with a reduced set of inputs. Int J Electr Power Energy Syst 58:246–256 36. Basnet SM, Aburub H, Jewell W (2016) An artificial neural network-based peak demand and system loss forecasting system and its effect on demand response programs. In: 2016 Clemson University power systems conference (PSC). IEEE, pp 1–5 37. Shi HF (2010) Short-term load forecasting based on Bayesian neural networks learned by Hybrid Monte Carlo method. In: 2010 international conference on machine learning and cybernetics, vol 3. IEEE, pp 1494–1499 38. Chakrabarti S, Jeyasurya B (2004) On-line voltage stability monitoring using artificial neural network. In: 2004 large engineering systems conference on power engineering (IEEE Cat. No. 04EX819). IEEE, pp 71–75

Evaluation of the Applicability and Advantages of Application …

243

39. Panasetsky D, Tomin N (2013) Using of neural network technology and multi-agent systems to preventing large-scale emergencies in electric power systems. In: 2013 4th international youth conference on energy (IYCE). IEEE, pp. 1–8 40. Wang Z, Martí J, Zuo C (2017) On-line loading limit monitoring and reliability must-run development using artificial neural networks. In: 2017 IEEE power & energy society general meeting. IEEE, pp 1–5 41. Lahiri-Dutt K (2003) People, power and rivers: experiences from the Damodar river, India. Water Nepal J Water Res Devel 9(10):251–267 42. Ghosh S (2011) Hydrological changes and their impact on fluvial environment of the lower Damodar basin over a period of fifty years of damming The Mighty Damodar River in Eastern India. Proc Social Behav Sci 19:511–519 43. Lahiri-Dutt K (2012) Large dams and changes in an agrarian society: gendering the impacts of Damodar valley corporation in eastern India 44. Lahiri-Dutt K (2012) Large state and the community in water management case of the Damodar valley corporation, India 45. Yu Y, Lin W (2006) Study on the security assessment platform for electric power secondary system. In: 2006 international conference on power system technology. IEEE, pp 1–6 46. Bi T, Qin X, He R, Yang Q (2008) Risk theory based on-line power system security assessment. In: 2008 third international conference on electric utility deregulation and restructuring and power technologies. IEEE, pp 704–708 47. Cheng XR, Li M, Zhao HI (2007) Fuzzy security assessment of entropy-weight coefficient method applied in electric power information systems. In: 2007 international power engineering conference (IPEC 2007). IEEE, pp 784–787 48. Marsadek M, Mohamed A, Nizam M, Norpiah ZM (2008) Risk based static security assessment in a practical interconnected power system. In: 2008 IEEE 2nd international power and energy conference. IEEE, pp 1619–1622 49. Min Y, Lin Z, Qiao J, Jiang X (2005) On-line steady state security assessment of power systems by SMES. IEEE Trans Appl Supercond 15(2):1923–1926 50. Wang L, Howell F, Morison K (2008) A framework for special protection system modeling for dynamic security assessment of power systems. In: 2008 joint international conference on power system technology and IEEE power India conference. IEEE, pp 1–6 51. Hou Z (2010) Application of GB/T20984 in electric power information security risk assessment. In: 2010 international conference on measuring technology and mechatronics automation, vol 1. IEEE, pp 616–619 52. Tiwary SK, Pal J, Chanda CK (2020) Multi-dimensional ANN application for active power flow state classification on a utility system. In: 2020 IEEE Calcutta conference (CALCON). IEEE, pp 64–68. https://doi.org/10.1109/CALCON49167.2020.9106479

A Realistic Sensing Model for Event Area Estimation in Wireless Sensor Networks Srabani Kundu1(B) , Nabanita Das2 , and Dibakar Saha3 1

3

Guru Nanak Institute of Technology, Kolkata, India [email protected] 2 Indian Statistical Institute, Kolkata, India [email protected] Indian Institute of Technology, Guwahati, Assam, India [email protected]

Abstract. A lot of research works have been reported so far for event area localization and estimation in self-organized wireless sensor networks deployed to monitor a region round the clock. In most of the works, it has been assumed that a node is affected whenever it lies within the event region. But in reality, each node does not sense just its point of location but covers a region defined by its sensing range and extracts an aggregated view of the sensed region. Unfortunately, so far no sensing model takes into account this fact. In this paper, a new realistic model of sensing is proposed for continuous event region, and based on that a lightweight localized algorithm is developed to identify a minimal set of boundary nodes based on 0/1 decision predicates to locate and estimate the event area in real time with high precision. Extensive simulation studies and testbed results validate our proposed model and also show that using only elementary integer operations and limited communication, the proposed scheme outperforms existing techniques achieving a 5–10% precision in area estimation with 75–80% reduction in network traffic even for sparse networks. Keywords: Wireless sensor networks (WSN) · Affected area Boundary node · Uniform area sensing model · Digital circle

1

·

Introduction

Recent technological advances have made the use of small and low-cost sensor devices technically and economically feasible for the purpose of sensing ground data from a region of interest (RoI). Typically, a sensor node is capable of sensing data from an area within its sensing range rs and also can communicate via wireless links with other neighboring nodes within its transmission range rc . After deployment, each node periodically senses data and cooperatively forwards it through the network to a gateway node, often referred to as the sink node, c Springer Nature Singapore Pte Ltd. 2021  C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_24

A Realistic Sensing Model for Event Area Estimation

245

thus forming a wireless sensor network (WSN). WSN can provide a fine global view through the collaboration of many sensors, each capturing a local snapshot at regular intervals. In critical situations, like forest fire, chemical spills, natural disasters, it is crucial to report the event to the sink immediately to estimate the affected area and its location. It is obvious that in case, all affected nodes attempt to send data to the sink, preferably via multihop paths to ensure energy efficiency, the network will get congested increasing the end-to-end packet delay and energy consumption. Hence, it is a great challenge to select a minimal subset of affected nodes, using lightweight in-node computation to define the event region boundary satisfying the precision requirement of the application. So far, a lot of research works have been reported addressing the event boundary detection and area estimation problem in 2D WSN. Most of the approaches reported till now are based on either the computation-intensive classical techniques of computational geometry, or some simpler greedy heuristics. In [1], a boundary face detection technique has been proposed by adopting the planarization algorithm. They have also mentioned that the estimation of area affected by any event is more desirable rather than the detection of exact boundary of the event region. Authors, in [2–4], have used the graph-theoretic models, relative neighborhood graph, and Gabriel graph methods to detect the boundary of an event area. In [5], authors proposed an angle based boundary detection algorithm to detect the event boundary. In most of the earlier works, on event boundary detection, it has been assumed that a sensor node gets affected if and only if it lies within the event area. But in reality, a sensor node senses a region around it and is affected only if a substantial portion of its sensed region lies within the event area. Though a large number of abstract sensing models are already in place and widely being used to solve the problems related to coverage, boundary detection, event area estimation etc., unfortunately, so far no sensing model has captured this fact [6]. To alleviate the problem, in this paper, a realistic model of uniform area sensing has been proposed considering the fact that a sensor actually captures an aggregated view of its sensed region, not just a point under its coverage, as has been assumed in the earlier models. In this paper, following the proposed realistic model of sensing, a lightweight distributed algorithm has been proposed that involves elementary integer operations only to localize and estimate the event region without compromising the accuracy. Extensive simulation studies have been done to compare the performance of the proposed technique with earlier works. Simulation results and testbed experiments show that the proposed technique performs significantly better in terms of selection of boundary nodes and area estimation accuracy even for sparse networks with reduced computation and communication overhead. The rest of the paper is organized as follows. Section 2 presents the proposed sensing model. Section 3 describes the lightweight area estimation technique under uniform area sensing model. Section 4 evaluates the performance of the proposed method with earlier works by simulation and testbed results. Finally, Sect. 5 concludes the paper.

246

2

S. Kundu et al.

Proposed Model

It is assumed that n number of homogeneous sensor nodes S : {s1 , s2 , . . . , sn } with uniform sensing range rs and communication range rc are randomly distributed over a 2D region A. In most of the earlier works, following the existing models of sensing, it is assumed that a node gets affected if and only if it lies within the event area, as if a node just senses its point of location. But in reality, each node covers not just its point of location but a region around it determined by its sensing range. With homogeneous nodes, the area covered by a node in 2D is assumed to be a circular region with sensing radius rs . Given an irregular-shaped event area A as shown in Fig. 1, a node s12 lying outside A may report that it is affected since a major portion of its coverage area lies within the event region. On the contrary, a node s5 lying within A may remain unaffected since most of the area covered by it lies outside the even area. Hence, the condition that a node should lie within the event region is neither necessary nor sufficient to affect it, as has been considered so far. Appropriate sensing model is required to represent the scenario.

A  s3 A s4

s2 s1 s10

s5

s9

s8

s7

s11 s12

A



s13

s6

Affected node Unaffected node

Fig. 1. An event area (shaded) and the affected and unaffected sensors, assuming uniform circular coverage

2.1

Sensing Models

In the literature, so far, various abstract sensing models, either directional or omnidirectional, have been proposed and used widely [6]. Considering only the omnidirectional ones here follows a classification of the existing sensing models.

A Realistic Sensing Model for Event Area Estimation

247

Fig. 2. Actual event area and estimated event area with false positive and false negative

• Deterministic Sensing Model: The probability that a sensor at location S detects an event at point P is represented by the coverage function given by:  1 if d(S, P ) ≤ rs C(S, P ) = 0 otherwise where d(S, P ) is the Euclidean distance between S and P [7]. • Probabilistic Sensing Model: Considering the fact that the quality of sensing actually depends on various parameters like the distance d(S, P ), the presence of obstacles, various probabilistic sensing models have been proposed so far [8]. – Elfes Sensing Model: In this model [9], the coverage function is given by: ⎧ ⎪ if d(S, P ) ≤ rmin ⎨1 γ C(S, P ) = pr = e−λ(d(S,P )−rmin ) if rmin < d(S, P ) < rmax ⎪ ⎩ 0 if d(S, P ) ≥ rmax It is to be noted that the deterministic sensing model is a special case of this model with rmax = rmin = rs . where λ and γ are adjustable parameters according to the physical properties of the sensor. – The Shadow Fading Sensing Model: This model [10] is proposed to take into account the shadowing loss of signal due to the presence of obstacles on signal path between S and P , where rmax  ) 10βlog10 ( d(S,P ) 1 r¯ × 2πd(S, P )dr, Q C(S, P ) = A σ 0

β is the signal power decay factor, dr represents a small increment in distance d(S, P ), σ is the shadow fading parameter, and r¯ is the average sensing radius, respectively.

248

S. Kundu et al.

It is interesting to see that in each of the models, the coverage function actually assumes that an event always occurs at a point P . But in case of events like fire, smoke, oil spill, the event spreads with time over a continuous region, and each sensor actually senses an aggregated view and detects its impact over its covered area. So the above models are not adequate to represent an event spanning over an area. Hence, a new sensing model is proposed in this paper. Definition 1. Uniform Area Sensing Model: In this proposed model, it is assumed that each sensor S senses a uniform circular area A of radius rs , its sensing range. In case of an event spanning a continuous region A  , the sensor data is given by:  1 δa .da, D(S, A  ) = A A

where

 1 δa = 0

if a ∈ A  otherwise

Now, a sensor is affected, if and only if D(S, A  ) ≥ p, a predetermined threshold, 0 < p ≤ 1. It is obvious that this model is more appropriate to estimate the event area in case of events like fire, gas pollution, oil spill, where not a single point but a continuous area is affected. It is to be noted that under this model, given an event area as shown in Fig. 1, the sensor nodes s12 and s5 can be identified accurately as affected and unaffected nodes, respectively, provided the threshold is decided appropriately during design time. 2.2

Topology Graph and Data Model

It is obvious that for collaborative computing and data gathering, the nodes need to communicate with each other and they forward their data to the sink node via multihop paths for energy-efficient data forwarding. However, it has been already proved that with rc ≥ 2rs , connectivity is guaranteed in case of full coverage [11]. So, we assume rc = 2rs throughout the paper. For data gathering, the underlying topology graph must remain connected, as defined below. Definition 2. Topology Graph: A wireless sensor network is represented by an undirected topology graph G(V, E), where V is the set of nodes and E is the set of edges such that an edge e(i, j) ∈ E, if and only if sensor node si can communicate with node sj directly, i.e., d(si , sj ) ≤ rc , where d(si , sj ) is the Euclidean distance between nodes si and sj , and vice versa, with si , sj ∈ V . In our proposed model, sensor nodes do not require the actual data value within the event area. It is assumed that in case, the sensed data crosses a predetermined threshold p, the node is considered to be affected.

A Realistic Sensing Model for Event Area Estimation

3

249

Event Area Estimation Based on Uniform Area Sensing Model

Under the uniform area sensing model, proposed above, a sensor node is affected if and only if substantial part of its covered area is within the event region. Hence given the set of all affected nodes, the event boundary can be identified as a sequence of intersecting real circles covered by a minimal set of boundary nodes. However, with real circles, it is not easy to estimate the bounded area [12]. Some concepts of digital geometry were applied to ease the computation in [13]. For completeness, an outline of the scheme is included in the following subsection. 3.1

Real Circle Versus Digital Circle

In [12], from computational geometry approach, authors proposed an O(n log n) algorithm to compute the area covered by a random set of real circles. Since complex computations are involved, the technique is not feasible for sensor nodes with elementary computing and storage capacities. Later, in [13], authors have shown that complex computation can be avoided if the real circles are approximated by digital circles following the concepts of digital geometry. Authors in [14] studied the performance of the digital circle approach (DCA) to solve the boundary detection and event area estimation problem for an irregular-shaped event region. Though the proposed area estimation technique based on digital circle simplifies the computation significantly, the boundary node detection and intersection point computation require complex computation and data structures. To make the computation even simpler, in this paper we propose a lightweight distributed approach outer angular algorithm (OAA) involving elementary integer operations only. 3.2

Outer Angular Algorithm (OAA)

To simplify the computation of event area estimation based on the proposed realistic sensing model, we assume that the sensor nodes are equipped with directional antennae, and knows the angle of arrival of the signal received from its adjacent neighbors. With that information, each node (having some affected and some unaffected neighbors) creates a circular sorted neighbor list with its state, based on the angle of arrival of signal. During traversal through the circular list, in a certain direction (clockwise/anticlockwise), if it identifies a state transition between successive neighbors, it selects the unaffected node as the reporting node to define the boundary of the event region. After selection, the reporting nodes, send their ID’s to the sink node to compute the area of the polygonal region defined by the reporting nodes. An outline of the procedure is given in Algorithm 1.

250

S. Kundu et al.

Algorithm 1: OAA Input: circular sorted list of neighbors N L(si ) Output: List of reporting nodes Bo 1 2 3 4 5 6 7 8 9 10

for each node si ; broadcast a hello message with status flag at regular interval; if node si receives a hello message from its neighbor sj then update the (0/1) flag bit in the circular list N L; end After receiving hello message from all neighbors, node si traverses through the circular list and if any transition found, then include the unaffected node sj model, such as in Bo ; broadcasts selected(Bo ) message; if receives selected(Bo ) message then if the node id is present in Bo then send its location with flag bit to sink via multihop path; end

In [5], based on the point coverage sensing model, an angle-based approach was followed for event area estimation. But it always underestimates the event area. In this paper, a new angle-based algorithm based on the realistic uniform area sensing model is presented and the performance comparison by simulation shows that OAA significantly outperforms [5] in accuracy of event area estimation.

4

Performance Evaluation

For performance comparison with Algorithm 1, some distributed techniques proposed earlier have been simulated under the proposed uniform area sensing model, such as inner angular algorithm (IAA) [5], digital circle algorithm (DCA) [13], and BDCIS algorithm [2] respectively. For simulation, n number of sensor nodes are uniformly and randomly distributed over an area A. An irregular-shaped event area A also termed as True Event Area (TA) is generated by diffusion process presented in [15]. Figure 2 shows an arbitrary-shaped event region TA and the estimated event region EA that defines both False Negative (FN) and False Positive (FP) areas. 4.1

Simulation Results

For simulation studies over 200 × 200 region, different event regions are created by changing the source cells randomly and the experiments are repeated for different networks. The algorithms are implemented using Java. The simulation parameters are presented in Table 1. Here the performance of OAA,DCA, IAA and BDCIS methods has been compared under the Uniform Area Sensing Model with different values of threshold p. Variation of Estimated True Area (ETA): Figures 3, 4 and 5 show DCA and OAA methods are always able to report almost 100% of the TA compared to other methods discussed in this paper.

A Realistic Sensing Model for Event Area Estimation

251

Fig. 3. Percentage of estimated true area for p = 0.50

Fig. 4. Percentage of estimated true area for p = 0.60

Variation of Error: Figures 6 and 7 show the variation of error (False Positive + False Negative ) in percentage. From the previous results, it is clear that OAA and DCA methods can always estimate almost 100% of the event area for p = 0.5 to 0.7 but error is high if the value of p is low. Number of Nodes Reported: Figures 8 and 9 show the variation in the number of reporting nodes with n. It is obvious that with the increase in node density the number of reporting nodes also increases, resulting best performance by OAA and DCA with almost 70–85% reduction in the number of nodes reported, and hence in network traffic. Computation Overhead: By observing the percentage of estimated area, error percentage and number of reporting nodes we can say, both OAA and DCA methods perform well but in terms of execution time and number of computation as

252

S. Kundu et al.

Fig. 5. Percentage of estimated true area for p = 0.70

Fig. 6. Percentage of error for p = 0

shown in Figs. 10 and 11, OAA method performs significantly better than DCA. OAA method needs significantly less number of integer operations compared to DCA method. Moreover, DCA technique needs floating-point operations, which is not required for OAA method. 4.2

Testbed Implementation

To validate our proposed model, we set up a simple indoor experimental testbed. Fourteen JN5168-001-M00 sensenut modules with light sensor are deployed randomly over a 400 × 400 square unit region. From device specifications, the communication range rc in indoor varies from 25 to 30 m and at outdoor 75–80 m. An irregular-shaped event region is generated using light sources and shadows. Some existing algorithms for boundary detection and area estimation with the

A Realistic Sensing Model for Event Area Estimation

253

Fig. 7. Percentage of error for p = 0.70

Fig. 8. Number of nodes reporting for p = 0.60

proposed one are implemented on this testbed. This experiment is executed 20 times, and each time it runs for 300 s. Initially, during the switch-off condition, each sensor node measures a light intensity of 0 Lux. But, as the light is switched on, the illumination increases rapidly and the maximum data value in nodes is 280 Lux. For this experiment, we consider the threshold value p = 0.6 and 0.7, respectively. Table 2 shows the percentage of the estimated area and the percentage of estimated true area for different methods. Though the testbed experiment has been done with small number of sensor nodes only, still it shows the same trend as observed by simulation. Hence, it is evident that with significantly less computation and communication overhead the proposed OAA technique with the uniform area sensing model is more suitable for event area estimation in wireless sensor networks, even in a sparse network.

254

S. Kundu et al.

Fig. 9. Number of nodes reporting for p = 0.70

Fig. 10. Execution time per node in milliseconds

5

Conclusion

In this paper, it has been shown that for wireless sensor networks monitoring a region against events like forest fire, oil spill, chemical pollution, where the event spreads over a continuous area, the existing sensing models are not adequate. Hence, a new, more realistic model, namely the Uniform Area Sensing model is proposed here. Based on that model, a localized lightweight technique is developed to estimate the event area using only elementary integer operations. Performance comparison by simulation and experimental results show that the proposed algorithm outperforms the earlier works achieving a 5–10% precision in area estimation with 75–80% reduction in reporting traffic even for sparse networks, provided the threshold p is determined appropriately.

A Realistic Sensing Model for Event Area Estimation

Fig. 11. Number of computation per node

Table 1. Simulation parameters Parameter Value A

200 × 200

n

1250–2500

rs

4 unit

rc

8 unit

l

1 (unit cell)

p

0.50–0.70

Table 2. Performance summary for different methods on testbed Method p = 0.60 p = 0.70 % of EA % of ETA % of EA % of ETA OAA

99.54

90.39

94.44

87.51

DCA

104.69

87.98

100.06

92.37

IAA

82.17

82.17

79.8

75.23

BDCIS 89.58

74.19

84.49

70.39

255

256

S. Kundu et al.

References 1. Ping ZSH, Zhou Z, Rahaman T (2018) Accurate and energy-efficient boundary detection of continuous objects in duty-cycled wireless sensor networks. Pers Ubiquit Comput 22(3):597–613 2. Beghdad R, Lamraoui A (2016) Boundary and holes recognition in wireless sensor networks. J Innov Digital Ecosyst (Elsevier) 3(1):1–14 3. Zhang LMY, Wang Z, Zhou Z (2018) Boundary region detection for continuous objects in wireless sensor networks. Wireless Commun Mob Comput 13 4. Zhou Z, Zhang Y, Yi X, Chen C, Ping H (2019) Accurate boundary detection and refinement for continuous objects in IoT sensing networks. IEEE Commun Mag 57(6):93–99 5. Kundu S, Das N, Roy S, Saha D (2016) Irregular shaped event boundary estimation in wireless sensor networks. In: International conference on advanced computing, networking and informatics. Springer, Berlin, pp 423–436 6. Elhabyan R, Shi W, St-Hilaire M (2019) Coverage protocols for wireless sensor networks: review and future directions. J Commun Networks 21(1):45–60 7. Yazid Boudaren ME, Senouci MR, Senouci MA, Mellouk A (2014) New trends in sensor coverage modeling and related techniques: a brief synthesis. In: 2014 international conference on smart communications in network technologies (SaCoNeT), 2014, pp 1–6 8. Zou Y, Chakrabarty K (2004) Sensor deployment and target localization in distributed sensor networks. ACM Trans Embed Comput Syst 3(1):61–91 9. Elfes A (1989) Using occupancy grids for mobile robot perception and navigation. Computer 22(6):46–57 10. Tsai Y (2008) Sensing coverage for randomly distributed wireless sensor networks in shadowed environments. IEEE Trans Veh Technol 57(1):556–564 11. Zhang H, Hou J (2005) Maintaining sensing coverage and connectivity in large sensor networks. Wireless Ad Hoc and Sensor Network 1(1–2):89–124 12. Zhang C, Zhang Y, Fang Y (2009) Localized algorithms for coverage boundary detection in wireless sensor networks. Wireless Netw 15(1):3–20 13. Saha D, Pal S, Das N, Bhattacharya B (2017) Fast estimation of area-coverage for wireless sensor networks based on digital geometry. IEEE Trans Multi-Scale Comput Syst 3(3):166–180 14. Kundu S, Das N, Saha D (2018) Boundary detection and area estimation of an event region in wireless sensor networks using digital circles. In: Proceedings of the workshop NWDCN of the 19th international conference on distributed computing and networking. ACM, New York 15. Lian J, Chen L, Naik K, Liu Y, Agnew GB (2007) Gradient boundary detection for time series snapshot construction in sensor networks. IEEE Trans Parallel Distrib Syst 18(10):1462–1475

Generic Framework for Privacy Preservation in Cyber-Physical Systems Rashmi Agarwal1(B) and Muzzammil Hussain2 1

Department of Computer Science, Central University of Rajasthan, Ajmer, India [email protected] 2 Department of Computer Science and Engineering, Central University of Rajasthan, Ajmer, India [email protected]

Abstract. Cyber-physical system (CPS) is an evolving technology, and as usual, security is a vital issue in its adaptation. Privacy is a primary security requirement in CPS and can cause havoc if unresolved. Much work is done in the area of privacy preservation in CPS, but they are domain-specific. There is no generic mechanism for privacy preservation in CPS. Here, we design a framework for privacy preservation in CPS. The proposed study aims to integrate separate privacy protection mechanisms in different levels of the CPS architecture, addressing different kinds of privacy as information contents, locations, identities, dates and times, addresses, etc., within a common structure.

Keywords: Cyber-physical systems (CPSs) Privacy · Security · Framework

1

· Internet of things ·

Introduction

Cyber-physical systems are assimilation of cyber-systems and physical system, which are bound by a communication framework. Examples of such systems are smart cities, smart grids, high confidence medical devices and systems, assisted living, etc. In other words, cyber-physical systems are integrated, networked computer systems that work together to compute, interact and regulate natural or personalized systems [1]. Security and privacy are the essential interests for every modern system. Protecting its private data from any manipulation is a challenging problem. Security is assured at many CPS levels, like content, location and individuals. Security at every stage is challenging, as the balance between safety and necessary results is important [2]. Privacy of CPS includes privacy of both physical and cyber-system, which makes CPS system’s privacy very complicated. We need to know the entire system and its components to understand privacy complexity. One problem with CPS is that it need not be always connected to networks or the Internet. While c Springer Nature Singapore Pte Ltd. 2021  C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_25

258

R. Agarwal and M. Hussain

using any system, consumers are more worried about data that is being collected. Notably, in the era of CPS, data is probably shared secretly. Previously, there was manual data entry system introduced, but nowadays, data is collected directly from the devices which are connected to the Internet. Even if user poses this information, they are unaware of privacy complexity. CPS technology combines many different types of physical objects that should perform a specific task [3]. The absence of a generic structure to define the privacy of cyber-physical systems is one of the greatest issues with this sort of embedded system. Many CPS systems obtain private information including human activity, patterns of sleep, health information, use of home automation, etc. Breach of privacy to that data may lead to misused and can cause harm to individual. So, there is a need to have a generic framework for privacy preservation. In our contemplation of the literature, we found that there are few frameworks designed for privacy preservation in CPS, but they were also intended for particular application fields and requirements. We could not identify any generic framework which could be tailored to any CPS application. The proposed framework includes data privacy, identity privacy and location privacy, which in turn eliminates unauthorized access to resources and ensures data confidentiality and reliability. This framework also defines a layered model of CPS, which classifies privacy according to its attributes. This layered approach also facilitates scalability and compatibility. The associated research is described in Sect. 2, the framework suggested in Sect. 3 and theoretical assessment in Sect. 4. In Sect. 5, we conclude this paper.

2

State of the Art

An exhaustive study on various frameworks of CPS and privacy mechanism has been done. After which, we are able to choose an appropriate framework for privacy in CPS. 2.1

Framework I (For Reliable Sensing)

The CPS framework for accurate sensing and optimization planning was proposed by the researchers [4]. It has five layers, and it proposes a network layer safety mechanism. • Access layer: This layer consists of sensors and actuators. • Apperceive layer: Function of this layer is information sensing and processing. • Networks layer: Function of this layer is data transmission, routing, access control and security, etc. • Data processing layer: This layer extracts the hidden information. • Service layer: This layer takes care of scheduling and monitoring of tasks.

Generic Framework for Privacy Preservation

2.2

259

Framework II (Service-Oriented Architecture)

Authors [5] have suggested a general CPS architecture based on service-oriented architecture (SOA), and the primary benefit of this suggested architecture is the ease of inclusion of services and elements. In safety certainty, information and device safety are discussed here. • Perceive tier: It includes interaction with external environment and sensing the data. • Data tier: Computational and storage of data is provided at this layer • Service tier: The main function of this tier is to interact with the whole system for task scheduling and monitoring. • Execution tier: This level executes the controls that the system receives. • Security assurance: This portion is in the entire scheme that is accountable for safety of access, safety of information and safety of devices. 2.3

Framework III (Component Based)

This framework [6] includes CPS domains, facets and aspects. This framework is the base of our proposed privacy framework. • Domains: It consists of areas of deployment of CPS. For example, manufacturing, transportation, energy, health care, etc. • Aspects: It consists of conceptually equivalent groups. There are nine defined aspects: functional, business, human, trustworthiness, timing, data, boundaries, composition and life cycle. • Facets: It contains activities for addressing concerns. There are three facets: conceptualization, realization and assurance. 2.4

Literature Review of Privacy in CPS

Cyber-system privacy is one part of CPS. Encryption is used for confidentiality. Integrity can be achieved by digital signatures and secure hashes. Authentication mechanism includes certificates and passwords that ensure only authorized users may access resources. For physical privacy, every entity including person and device should be identified and monitored. Few researchers [7] have suggested some access control mechanism through data structure which contains device identity and access rights. The author [8] has suggested that the user should use the schema where access is provided on the grounds of contextual data gathered by the surroundings. If an individual can match the access rights stored in the device, then access is granted. Others have [9] suggested scheme based on token which contains identity of users, devices, etc. Jan et al. [10] used Advanced Encryption Standard (AES) mechanism without any device participation in communication. Wang et al. [11] proposed various schemes to protect data content privacy by using pairwise key establishment. Few researchers have [12] discussed about privacy of data and identity in healthcare systems. Florian

260

R. Agarwal and M. Hussain

kerschbaum et al. [13] proposed public key encryption and distribution of keys for smart meters. Heng Zhang et al. [14] used mechanism known as differential privacy in which the original data is mapped on to a domain through two differential parameters. The mapped data is transmitted and cannot be interpreted by any adversary. The data is transmitted by adding noise to it, and any adversary cannot know the accurate data. Katewa et al. [15] proposed that if agents are private, they should trust and cooperate with each other. For privacy protection, hardware-based schemes [16] are proposed, though password-based schemes are popular which are implemented using hash locks or hash chain [17]. Some researchers [18] argued about data privacy in three aspects: information sharing, collection and management. Authors [19] have described privacy in four respects: physical privacy, mental privacy, privacy of decisions and privacy of data. Few more recent framework has been proposed for PrivacyPreserving based Anomaly Detection for Cyber-Physical Systems [20] and for Privacy-Preserving Data Publishing with Enhanced Utility for Cyber-Physical Systems [21].

3

Proposed Framework

We have created a generic framework for privacy preservation in the proposed research work. We have conducted surveys of appropriate frameworks and privacy mechanisms. In the proposed work, we address privacy mechanism for data, identity and location. Further, these privacy classifications are mapped to layered architecture of CPS. 3.1

Generic Framework for Privacy Preservation

We have adopted CPS framework [6] which is component based. In the model shown in Fig. 1, domain, aspects and facets are defined. Aspect is categorized into various sub-components described above, and among those, trustworthiness includes many functionalities like reliability, security, safety, confidentiality and privacy. CPS privacy means preventing people or machines from gaining access to data [6]. There can be many attributes in CPS privacy, but for our framework, we have considered identity privacy, data/content privacy and location privacy as described in Fig. 2. 3.2

Data/Content Privacy

• Encryption: When encryption is not used, privacy of data is compromised. Generally, symmetric encryption algorithms like elliptic curve cryptography (ECC) and advanced encryption standard (AES) block cipher are used for encryption and decryption [22]. • Change Encryption Key: To protect the data, encryption can be used. It can be effective unless anyone knows key to decrypt it. Changing keys frequently is an efficient way to protect the data.

Energy

Healthcare

Transportation

Manufacturing

Domain

Fig. 1. CPS framework [6]

Trustworthiness

Composition

Boundaries

Data

Timing

Lifecycle

Human

Business

Functional

Aspects

Privacy

Assurance

Realization

Location Privacy

Data Privacy

Identity Privacy

Conceptualization

Facets

Cyber Physical Systems Framework

Generic Framework for Privacy Preservation 261

Privacy

Pseudonyms

Encryption

Key Generation List

Access Control

Fig. 2. Generic framework for privacy preservation in CPS

Model

Change ID

Data Classification

Encryption Mechanism

Perceptual Layer

Path

Mechanism

Location Privacy

Change Routing

User Authentication

Change Encryption key Security Manager

Network Layer

Hiding Identity

Layer

/Distribution Method

Support Layer

Multi hop Routing

ation

Identity Privacy

Fake Path

Applic

Management

Control

Access

Management

Protection

Data Privacy

Privacy

Generic Framework for Privacy Preservation in Cyber Physical Systems

262 R. Agarwal and M. Hussain

Generic Framework for Privacy Preservation

263

• Key Generation and Distribution: Key generation will be done at support layer. Key management should be able to generate and distribute required keys between communication entities of CPS. It should follow requirements such as scalability, freshness and accountability. 3.3

Location Privacy

• Multi-hop routing Path: Multi-hop routing can prevent from determining the routing paths of communication and source. • Fake Path: To prevent from tracking the routing path, fake paths can be used. • Change Path: Lastly, altering routing routes can thwart jamming attacks on opponents. 3.4

Identity Privacy

• Hiding Identity: Identity of data/message could be in encrypted form in order to hide the identity. • Pseudonyms: It can be used to cover the real identity. • Change ID: Changing identities can often thwart attackers from disclosure of identity. 3.5

Layers of Framework

• Application layer: Application layer consists of user applications and mechanisms to support and interact with user applications. • Support Layer: Support layer consists of key generation and distribution mechanisms, which is a part of data privacy. • Network Layer: The network layer is the second level. The network layer is accountable for trustworthy transmission of perceptual layer info. In this layer, identity privacy and data privacy are covered. The main attributes covered in this are change ID, pseudonyms and hiding identity which comes under identity privacy. Encryption and changing encryption also fall under this layer which comes under data privacy. • Perceptual Layer: This layer interacts with physical world and collects data from physical devices such as sensors, RFID and GPS. This layer takes care of location privacy. This can be achieved by changing routing path, fake path or multi-hop routing. 3.6

Security Manager

Function of security manager [23] includes user authentication, data classification, access control and encryption. • User Authentication: User authentication is achieved by digital certificate.

264

R. Agarwal and M. Hussain

• Data Classification: For data classification, tag-based approach is used. Here, data is assigned some value based on its type. Tag value 00 is assigned to general data, 01 is assigned to general and private data and 10 is assigned to private data. • Access Control: It is achieved through access control list. By listing the permissions connected to an item, it indicates the access rights. • Encryption: Data Encryption Standard (DES), a mechanism for symmetric key encryption, is used for encryption and decryption as it requires less time and resources.

4

Theoretical Evaluation

The proposed framework for privacy preservation in CPS is generic and not domain-specific. It ensures data privacy, identity privacy and location privacy. It adapts user data classification and access control mechanisms to prevent unauthorized access of crucial data and also eliminates illegitimate access to resources. The proposed framework ensures and achieves: • Data privacy: Data privacy is achieved through user data classification and access control lists. • Data Confidentiality: Data confidentiality is achieved through encryption of user data through secret key of user. • User authentication: Unauthorized access of resources is eliminated through efficient authentication mechanism using digital certificates. The proposed framework ensures privacy, confidentiality and authentication at a little expense of storage space and delay.

5

Conclusion

Privacy preservation is a major challenge in public networks like cyber-physical systems, and there is no generic framework for achieving privacy in CPS. In this paper, we designed and proposed a generic framework for preservation of privacy in CPS, and the proposed framework is layered for ease of compatibility and scalability. The proposed framework ensures data privacy, data confidentiality, location privacy, identity privacy, anonymity and eliminates unauthorized access of data and resources. The suggested research aims to incorporate distinct mechanisms for preserving privacy in distinct layers of CPS architecture, addressing various types of privacy like data content, location, identity, date and time, address, etc., under a common framework. This generic framework can be used to design privacy preservation mechanism specific to a network, application or domain. The proposed work can be adapted in design and development of secured CPS. In the future, the proposed framework can be extended to accommodate heterogeneous data sources and systems.

Generic Framework for Privacy Preservation

265

References 1. Al Faruque MA, Ahourai F (2014) A model-based design of cyber-physical energy systems. In: 19th Asia and South Pacific design automation conference (ASPDAC), Singapore, pp 97–104 2. Zhang H, Shu YC, Cheng P, Chen JM (2016) Privacy and performance trade-off in cyber-physical systems. IEEE Network 30:62–66 3. Sztipanovits J, Ying S, Cohen I, Corman D, Davis J, Khurana H, Mosterman PJ, Prasad V, Stormo L (2012) Strategic R and D opportunities for 21st century cyber-physical systems. Technical Report for Steering Committee for Foundation in Innovation for Cyber-Physical Systems 4. Lun YL, Cheng LL (2011) The research on the framework of cyber-physical systems for the reliable sensing and optimization scheduling. In: Applied mechanics and materials, vol 65. Trans Tech Publications, pp 451–454 5. Hu L, Xie N, Kuang Z, Zhao K (2012) Review of cyber-physical system architecture. In: 2012 15th IEEE international symposium on object/component/serviceoriented real-time distributed computing workshops (ISORCW). IEEE, New York, pp 25–30 6. Griffor ER, Greer C, Wollman DA, Burns MJ (2017) Framework for cyberphysical systems, vol 1, Overview. Special Publication (NIST SP) - 1500-201; NIST Rockville, MD 7. Mahalle PN, Anggorojati B, Prasad NR, Prasad R (2012) Identity establishment and capability based access control (IECAC) scheme for internet of things. In: 2012 15th international symposium on wireless personal multimedia communications (WPMC). IEEE, New York, pp 187–191 8. Liu J, Xiao Y, Chen CP (2012) Authentication and access control in the internet of things. In: 2012 32nd international conference on distributed computing systems workshops (ICDCSW). IEEE, New York, pp 588–592 9. Butkus P (2014) A user centric identity management for Internet of things. In: 2014 international conference on IT convergence and security (ICITCS). IEEE, New York, pp 1–4 10. Jan MA, Nanda P, He X, Tan Z, Liu RP (2014) A robust authentication scheme for observing resources in the internet of things environment. In: 2014 IEEE 13th international conference on trust, security and privacy in computing and communications (TrustCom), pp 205–211 11. Wang H, Li Q (2006) Distributed user access control in sensor networks. In: International conference on distributed computing in sensor systems. Springer, Berlin, pp 305–320 12. Haque SA, Rahman M, Aziz SM (2015) Sensor anomaly detection in wireless sensor networks for healthcare. Sensors 15:8764–8786 13. Kerschbaum F, Lim HW (2015) Privacy-preserving observation in public spaces. In: European symposium on research in computer security. Springer, Cham, pp 81–100 14. Zhang H, Shu Y, Cheng P, Chen J (2016) Privacy and performance trade-off in cyber-physical systems. IEEE Network 30(2):62–66 15. Katewa V, Pasqualetti F, Gupta V (2017) On privacy versus cooperation in multiagent systems. Int J Control, pp 1–15 16. Thales Hardware based scheme (2015). https://www.thales-esecurity.com/ products-and-services/products-and-services/hardware-security-modules

266

R. Agarwal and M. Hussain

17. Granjal J, Monteiro E, De Silva J (2013) Security issues and wireless M2M systems. In: Wireless networks and security. Springer, Heidelberg, pp 133–164 18. Mocana-NanoDTLS. https://mocana.com/products.html. Accessed Nov 2012 19. Roman R, Alcaraz C, Lopez J, Sklavos N (2011) Key management systems for sensor networks in the context of the Internet of things. Comput Electr Eng 37(2):147– 159 20. Keshk M, Sitnikova E, Moustafa N, Hu J, Khalil I (2019) An integrated framework for privacy-preserving based anomaly detection for cyber-physical systems.IEEE Trans Sustain Comput 21. Sangogboye FC, Jia R, Hong T, Spanos C, Kjærgaard MB (2018) A framework for privacy-preserving data publishing with enhanced utility for cyber-physical systems. ACM Trans Sens Networks (TOSN) 14(3–4):30 22. Tawalbeh LA, Mowafi M, Aljoby W (2013) Use of elliptic curve cryptography for multimedia encryption. IET Inf Secur 7(2):67–74 23. Kaliya N, Hussain M (2017) Framework for privacy preservation in IoT through classification and access control mechanisms. In: 2nd international conference for convergence in technology (I2CT), pp 430–434

A Novel Authentication Scheme for VANET with Anonymity Harshita Pal(B) and Bhawna Narwal Department of IT, IGDTUW, New Delhi, Delhi, India [email protected], [email protected]

Abstract. VANET is an essential key technology for building up an intelligent transportation system (ITS) that combines current wireless technologies to vehicles. Real-time information sent to driver or vehicle ensures smooth traffic flow in order to ensure safe driving and traffic management while avoiding accidents. To maintain smooth functioning and protection of real-time information from alteration, data must be secured and authenticated. In this paper, a novel anonymous authentication scheme for VANET has been presented which tries to provide strong non-repudiation, anti-forgery, and anonymity properties. Additionally, to avoid vehicles from abusing VANET as well as to provide strong privacy protection, a conditional tracking mechanism for vehicle tracing is developed which revokes certificates of misbehaving vehicles in an efficacious manner. Moreover, the proposed protocol protects the network from attacks such as masquerading, identity theft, and certificate replication. Keywords: Authentication · Anonymity · Conditional tracking · Non-repudiation · Security · VANET

1 Introduction Vehicular ad hoc networks (VANET), a form of mobile ad hoc networks (MANET), is a collection of more structured and cooperative mobile nodes which has a principal distinction from standard wireless ad hoc networks in terms of resources, memory, and computation power. VANET has provided an effective way for information exchange (specifically for safety reasons) between moving and smart vehicles where routes are predefined. Advancements in wireless network communication and vehicular technology lead to a great acceleration in the field of VANET research area over the past few years. This advancement leads to the development of intelligent transportation system which provides road safety, free flow of traffic, information about nearby parking area, navigation which provides pre-accidental warnings, and heavy traffic jammed areas which are very helpful [1]. Research is in active mode in this direction, and major players such as Nissan, Audi, General Motors, Toyota, BMW, etc., are primarily focusing on vehicle to vehicle communications since 2000. There are various applications of

© Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_26

268

H. Pal and B. Narwal

VANET like traffic information system, road transportation, emergency services, electronic brake lights, on-road services, etc. Three types of communication mode are supported in VANET: V2V (vehicle to vehicle), V2I (vehicle to infrastructure), and hybrid. In V2V, multi-hop/multicast connection is established between vehicles to exchange information. V2V utilizes naïve and intelligent broadcasting. In V2I, a multi-hop high bandwidth link is established between vehicles. This mode also benefits other vehicles with the exchanged information by sending information to road-side unit, which broadcasts this received information to the next nodes in the network. Hybrid is a combination of the aforementioned techniques. IEEE 1609.2 standard is used in VANET to provide security services. VANET enables vehicles to share real-time information about their surroundings (such as information about accidents, and traffic jam) and allows cooperative driving as well as access to Internet services and navigation. However, assessing the trustworthiness (trusted and verified integrity) of the received information is very important and hence the security. The attackers can insert or modify the exchanged information which can endanger the safety of people on the road. At the same time for greater efficiency, resilience toward attacks and to avoid harm to life as well as property, security aspect should be contemplated [1–5]. In comparison with MANETs, VANET has its own specific attributes which are illustrated in Fig. 1. There are many factors which impacts the performance of VANET while realizing its objective. Some of the key technical challenges are highlighted in Fig. 2. In this paper, a novel anonymous authentication scheme for VANET has been presented which tries to provide strong non-repudiation, anti-forgery, and anonymity properties. Additionally, to avoid vehicles from abusing VANET as well as to provide strong privacy protection a conditional tracking mechanism for vehicle tracing is developed which revokes certificates of misbehaving vehicles in an efficacious manner. Moreover, the proposed protocol protects the network from attacks such as masquerading, identity theft, and certificate replication. The organization of the paper is as follows: In Sect. 2, some related work is presented. Section 3 discusses problem statement and solution goals. The proposed scheme is presented in Sect. 4, and Sect. 5 concludes the paper and talks about future work.

2 Related Work VANETs face different challenges in terms of security, and in order to provide universal connectivity, there is a strong need of secure exchange of information and opinion management system between the networking entities as it affects the trust and cooperation between the entities. Many researchers addressed these problems, and some of them are discussed in this section. Mishra et al. in [6] provided a secure and effective protocol for VANET that guarantees message authentication and privacy preservation by using elliptic curve digital signature algorithm (ECDSA). The protocol achieves conditional privacy as well as provides resilience toward DoS, Sybil, and gray/black hole attack. To assure user anonymity and privacy, temporary identities are utilized (produced through secure cryptographic mechanisms). Authors were successful in providing security with the usage of minimal

A Novel Authentication Scheme for VANET with Anonymity

269

Fig. 1. Attributes of VANET

bits as ECDSA is incorporated in the proposed protocol. Cui et al. in [7] explored deep potentials of VANETs which can be very useful in building a smarter transportation network. Also, they mentioned how the government is also providing full support to this. Authors came up with edgecomputing concept and incorporated it in the centralized authentication process in order to solve the redundant authentication issue. Singh et al. in [8] explained how this new era of digitalization in the transportation infrastructure brought an enormous change in the driving as well as security experiences of the vehicles in the network. In addition to this, authors also openly discussed how a malicious vehicle owner or a driver can hide behind an anonymous identity and can abuse the network. They focused on the idea of securing the broadcasted messages in the network which can be malicious. So with the help of their proposed protocol, they are providing the idea of fraud detection, tracing back the source of anonymous vehicles that abuses the network. Jiang et al. in [9] proposed an effective batch authentication protocol (uses session and group key to achieve authentication) which utilizes HMAC to maintain the integrity of the messages communicated between the entities. Moreover, the protocol is cost-effective in terms of communication and storage.

270

H. Pal and B. Narwal

Fig. 2. Challenges in VANET

Liu et al. [10] in 2018 proposed an authentication protocol which uses identity-based cryptography and does not fully depend on TPD. In addition to this, the protocol makes efficient usage of the storage space and does not require any trusted third party. The protocol developed by Liu et al. is successful in identity preservation, mutually authenticating the entities as well as resilience toward forgery, modification, masquerading, and replay attack. Azees et al. in [11] proposed an effective authentication protocol for both vehicles and RSUs that prevent the malicious vehicle from getting an entry into the network and proffer conditional tracking for trailing the vehicles or RSUs that try to abuse the VANET. In addition to this, the protocol is successful in achieving integrity and reduces the message loss ratio, computation burden as well as certificate and signature verification cost.

3 System Model In this section, the system model, problem statement, and the goals of the proposed scheme are introduced.

A Novel Authentication Scheme for VANET with Anonymity

271

3.1 System Model There are three major components in our VANET system model (as per assumption) among which the communication takes place: Trusted authority (TA), road-side units (RSUs), and vehicles. Trusted authority (TA). TA is the most trusted authority which stores all the crucial information related to vehicles and RSUs in transportation infrastructure. Every vehicle and RSU registration is done via TA. Each geographic region has its own TA and TAs communicates among them. Road-side units (RSUs). The communication between vehicles and TA happens via RSUs which are fixed unit in transportation infrastructure. These are semi-trusted entities in a transportation infrastructure equipped with dedicated short range service (DSRC) for communication. However, if they got compromised, it can easily be detected by TA. Vehicles. Each vehicle which acts as a communicating node in a network is equipped with on-board units (OBUs) which helps in communicating with other vehicles in ITS. These OBUs are equipped with temper-proof devices (TPDs). Further, TPD stores crucial information into them such as secret keys, passwords, GPS information (to provide time and location), and event recorder information (traffic jam, vehicle crashes information, etc.). In addition to the above three components, TPD and telematics devices (TDs) are also involved. These are briefly described as follows: Temper-proof device (TPD). TPD is storage inside OBU which contains cryptographic information and cannot be hacked easily. If in case got hacked, then it is self-destructive. Telematics device (TD). These devices are installed in vehicles to provide monitored information and connection information if in case vehicles are connected via a wireless network. This device provides information about driving behavior, speed estimation, live weather information, traffic information, and parking information. 3.2 Problem Statement and Solution Goals A novel anonymous two-factor authentication scheme for VANET has been presented which tries to provide strong non-repudiation, anti-forgery, and anonymity properties. To provide anonymity to the vehicle, pseudo identities are used. Additionally, to avoid vehicles from abusing VANET as well as to provide strong privacy protection, a conditional tracking mechanism for vehicle tracing is developed which revokes certificates of misbehaving vehicles in an efficacious manner. Moreover, the proposed protocol protects the network from attacks such as masquerading, identity theft, and certificate replication. The goals of the proposed scheme are as follows: Authentication. The entities involved in the communication are valid and avoid malicious entities from getting into the system. Anti-forgery. Avoids transmission of forged messages to other vehicles. Privacy protection. The identities of the vehicles disseminating messages must be preserved, but

272

H. Pal and B. Narwal

whenever it is required by legal authorities, it must be available [10]. Unlinkability. If a malicious entity captures the broadcasted messages of the vehicle, then also the malicious entity cannot link the messages generated from the same vehicle [10]. Low computation and communication cost. All communication and computation which are needed during message transmissions should require the least computations and communication cost [12]. Revoking certificates of misbehaving vehicles. In this, the certificate of the misbehaving vehicle is revoked, and their access permissions are taken away. Strong nonrepudiation. The vehicle which generates and sends the information cannot deny later that it does not send the information [13]. Conditional traceability. Trusted authority can trace the vehicle from the vehicle identification information and bring out real identity. Minimum message loss ratio. The number of messages dropped during transmission should be minimal for better throughput. Resistance to security attacks. To protect the scheme from masquerading, certificate replication, and identity theft [14–16].

4 Proposed Scheme In this section, the proposed scheme is described in detail and illustrated through Figs. 3, 4, and 5. The notations used throughout this chapter are listed in Table 1. Bio-table for each vehicle registered at TA is given in Table 2, and their persistence table is given in Table 3.

Fig. 3. System initialization and vehicle registration phase

4.1 System Initialization and Vehicle Registration Phase During this phase, the driver submits the vehicle information and his/her biological password to the TA. Then, TA returns an access token for TDi login, a pseudo-identity

A Novel Authentication Scheme for VANET with Anonymity

273

Fig. 4. Driver identification and TPD login phase

Fig. 5. RSU and vehicle communication and verification phase

Table 1. Table of notations Notations used Description i

Vehicle

PID

Pseudo ID

VID

Vehicle ID

TDi

Telematics device number i

TPDi

Temper-proof device number i

MAC

Message authentication code

K

Shared group key

Hagg

Height of aggregated hash of all the messages

PID (randomly generated number) to the vehicle, and system public key. Moreover, TA stores a password verifier in TDi and a password keeper in TPDi.

274

H. Pal and B. Narwal Table 2. Bio-table for each vehicle registered at TA Vehicle

Original IDs Pseudo IDs assigned

Vehicle 1 ID1

PID1

Vehicle 2 ID2

PID2

Vehicle 3 ID3

PID3

Table 3. Persistence table stored at RSUs VID

PID

VID1 PID1 VID2 PID2 VID3 PID3

4.2 Driver Identity Verification and TPD Login Phase In this phase, the driver login into TDi using the access token provided by TA. As TA stored the password verifier into TDi in the initialization phase, then whenever valid driver login with the provided access token then only the TDi access can be granted, and if it is successful then only TPDi access is given to the driver. Similarly, TA stored a password keeper in TPDi (as specified in initialization phase) and after TDi login, a key is given to the driver for TPDi login. By using this key, driver can access the cryptographically strong TPDi. This assures that only a legitimate vehicle driver is given access. In order to avoid a replay attack, PID is attached with a timestamp and signatures of the sender. 4.3 RSUs and Vehicles Communication and Verification Phase In order to avoid multiple keys overhead of PKI scheme, we make use of symmetric key cryptography along with digital signatures. A secret key is shared between vehicles and RSU every time the set of vehicles enter the communication range of RSU at time t1, and then, this secret key is used for every communication that takes place between them (group key). Vehicles communicating with other vehicles during time t1 need not verify messages using PKI scheme rather they attach MAC to each message encrypted with a secret key once shared between them at time t1. The communication between RSU and vehicle and vehicle to vehicle involves VID of vehicle, message, and MAC of the message, and then, the whole message is encrypted by a shared group key K at time t2. Usage of MAC-based verification fastens the process. For a detailed view of this phase: When the vehicle enters into the range of RSU, a bidirectional authentication takes place between them. Firstly, RSU verifies that the vehicle is driven by a valid driver or not based on certificates generated by the TA. After that, a unique symmetric key K and VID is assigned to the vehicle for further communication. After that, a persistence

A Novel Authentication Scheme for VANET with Anonymity

275

table is maintained by RSU which stores VIDs of every vehicle that forms a dynamic network with it for short period of time t1 then this VID maps them to PID which ultimately maps it back to their original id in Bio-table of TA (given in Table 2). Thus, it helps in conditional privacy tracking of the misbehaving vehicle in the VANET. The timestamp is included here to avoid replay attacks, and it is worthy to note that each message consists of MAC and digital signature. The unique symmetric key and VID are only known by RSU of the particular geographic area only. It is solely responsible for the collection and distribution of messages with vehicles within its range. Whenever there is any misbehaving vehicle in the network, then using correct MAC and digital signature, it just needs to refer its persistence table (shown in Table 3) and can easily trace back the source of the misbehaving vehicle in the VANET and can report the TA along with its certificate. A vehicle can receive messages from only two entities, RSUs and vehicles. In order to verify whether the vehicle and RSU are communicating with the correct set of entities or not, hash of the overall message is generated at the end and named as an aggregated hash. And, for each dynamic group, a hash of all the message exchange is calculated with name Hagg. Calculation of Hagg is performed by both RSU and vehicle, and if it turns out the same, it means they were exchanging information in the correct group. Therefore, the vehicle takes the message along with its hash from RSUs as its arguments, whereas the message from other vehicles is stored in their local memory buffer without verifying immediately. For the verification of message by RSU, the message should be attached with a valid RSU signature otherwise message verification is not possible. If the signature is valid, then RSU matches hash generated from it with the hash of the incoming message, and if the match is valid, then only the message is accepted otherwise rejected. RSU checks the hash for two times only in order to avoid network delay in VANET. The proposed algorithm supports four types of communication: (1) message verification by RSU, (2) message distribution to vehicles by RSU, (3) message distribution from vehicle to vehicle, and (4) message verification by vehicle. The limitation of this work is that the security of the whole protocol relies on the unique pseudo-ID generated by TA. As in the proposed scheme, TA responsibilities are decentralized among telematics device (TDi) and TPDi, which is still a threat to VANET if in case they got compromised in the future.

5 Conclusion and Future Work In this paper, a novel anonymous two-factor authentication scheme for VANET has been presented which tries to provide strong non-repudiation, anti-forgery, and anonymity properties. Additionally, to avoid vehicles from abusing VANET as well as to provide strong privacy protection, a conditional tracking mechanism for vehicle tracing is developed which revokes certificates of misbehaving vehicles in an efficacious manner. Moreover, the proposed protocol protects the network from attacks such as masquerading, identity theft, and certificate replication. In the future, we will extend our approach to fully decentralize the VANET authentication scheme by keeping its accuracy, efficiency, and performance intact. Moreover, once the sender is authenticated and message integrity, as well as non-repudiation, is verified, the sender should take the responsibility

276

H. Pal and B. Narwal

of its own messages and not worrying about overhead of signatures and certificate computations during every communication session. Also in the future, we will try to extend our work for improving our protocol with very little dependency on RSU for message authentication.

References 1. Al-Sultan S, Al-Doori MM, Al-Bayatti AH, Zedan H (2014) A comprehensive survey on vehicular ad hoc network. J Netw Comput Appl 37:380–389 2. Narwal B (2018) Fake news in digital media. In: 2018 international conference on advances in computing, communication control and networking (ICACCCN). Greater Noida (UP), India, pp 977–981 3. Narwal B, Mohapatra A (2017) Performance analysis of QoS parameters during vertical handover process between Wi-Fi and WiMAX networks. In: International conference on recent developments in science, engineering and technology, Springer, p 330344 4. Rani S, Narwal B, Mohapatra AK (2017) RREQ flood attack and its mitigation in ad hoc network. In: International conference on recent developments in science, engineering and technology. Springer, Singapore, pp 599–607 5. Narwal B, Mohapatra AK (2016) Energy efficient vertical handover algorithm for heterogeneous wireless networks, vol 9, pp 8981–8984 6. Mishra B, Panigrahy SK, Tripathy TC, Jena D, Jena SK (2011) A secure and efficient message authentication protocol for VANETs with privacy preservation. In: IEEE world congress on information and communication technologies, pp 880–885 7. Cui J, Wei L, Zhang J, Xu Y, Zhong H (2018) An efficient message-authentication scheme based on edge computing for vehicular ad hoc networks. IEEE Trans Intell Transp Syst 20(5):1621–1632 8. Singh A, Fhom HCS (2017) Restricted usage of anonymous credentials in vehicular ad hoc networks for misbehavior detection. Int J Inf Secur 16(2):195–211 9. Jiang S, Zhu X, Wang L (2016) An efficient anonymous batch authentication scheme based on HMAC for VANETs. IEEE Trans Intell Transp Syst 17(8):2193–2204 10. Liu ZC, Xiong L, Peng T, Peng DY, Liang HB (2018) A realistic distributed conditional privacy-preserving authentication scheme for vehicular ad hoc networks. IEEE Access 6:26307–26317 11. Azees M, Vijayakumar P, Deboarh LJ (2017) EAAP: Efficient anonymous authentication with conditional privacy-preserving scheme for vehicular ad hoc networks. IEEE Trans Intell Transp Syst 18(9):2467–2476 12. Lim K, Manivannan D (2016) An efficient protocol for authenticated and secure message delivery in vehicular ad hoc networks. Vehic Commun 4:30–37 13. Asl FR, Samavi R (2017) SyNORM: symmetric non repudiated message authentication in vehicular ad hoc networks. In: 2017 IEEE 86th vehicular technology conference (VTCFall), pp 1–5 14. Tan H, Choi D, Kim P, Pan S, Chung I (2017) Comments on dual authentication and key management techniques for secure data transmission in vehicular ad hoc networks. IEEE Trans Intell Transp Syst 19(7):2149–2151 15. Narwal B, Mohapatra AK, Usmani KA (2019) Towards a taxonomy of cyber threats against target applications. J Stat Manage Syst 22(2):301–325 16. Dhawan S, Narwal B (2019) Unfolding the mystery of ransomware. In: International conference on innovative computing and communications, Springer, pp 25–32

A Survey on Handover Algorithms in Heterogeneous Wireless Network Mithun B Patil(B) and Rekha Patil Dept of CSE, Poojya Doddappa Appa College of Engineering, Kalburgi, Karnataka, India [email protected], [email protected] Abstract. The next generation of wireless network consists of many overlaying integrated networks which know as heterogenous network in which mobile node will be on mobility between/within these networks. During mobility of node, the ongoing call of mobile node should be transferred between/within these seamless networks. The transfer of mobile node between/within network can be done using handover algorithms. Most of the handover algorithms designed for heterogeneous wireless network are mainly based on parameters such as signal strength, SIR, distance, velocity, direction, and power consumption. For an effective handover, different approaches have been proposed. These approaches have their own advantages and disadvantages, and each of them performs better than the others under certain circumstances. The chapter classifies and discuss the different approaches for designing of vertical handoff mechanism. Keywords: Handover · Signal · Base station (BS) · Mobile station (MS) · Receiving signal strength (RSS) · QoS

1 Introduction One of the main objectives of handovers in a wireless network is to provide uninterrupted connectivity between users and the radio access network while they move across cell boundaries within/between network. Efficient handover algorithms should be designed to preserve capacity and enhance quality of service (QoS) of communication systems in a cost-effective manner [15]. There are researches working on developing efficient handover algorithms. For example, the occurrence of a handover can be attributed to several factors, which could be related to radio link quality, network management, and service options [19]. While radio link quality related handovers occur frequently and are most difficult to handle, network management and service option related handovers usually occur infrequently and are easy to handle [16]. There are different kind of algorithms to handle these handovers which are classified depending on different parameter as shown in Fig. 1 are discussed in this chapter. The chapter is organized as follows: Sect. 1 introduces handover algorithm, and Sect. 2 explains the process and phases involved in handover execution. Section 3 categorizes handover algorithm into six different categories and gives the overview of different handover algorithms. In Section 4, we have discussed benefits and drawbacks of different handover algorithms, and Sect. 5 is conclusion and future scope to work. © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_27

278

M. B. Patil and R. Patil

Fig. 1. Vertical handoff parameters

2 Execution of Vertical Handover Mechanism To achieve seamless mobility for a mobile node during mobility between/within the network in heterogenous network, the handoff algorithm should be designed such that the handover process executes without any interruption of ongoing call of user. Handover executes in the three phases as shown in Fig. 2. Phase 1 Network discovery: The mobile node needs to discover new available network for providing service. If mobile node does not satisfy with existing connected network with parameter defined for QoS, then mobile node needs to search the new network available for services. Phase 2 Handover decisions: Depending on the parameters received in network discovery and required parameter defined in the handover algorithm, the decision of handover execution needs to be taken in this phase. Handover decision can be implemented on network side or at mobile node. If handover decision is executed at the network side considering all the parameter received during network discovery, then it is referred as network controlled vertical handover (NCVHO), and if decision of handover is taken at mobile node, it is called as mobile-controlled vertical handover (MCVHO). Phase 3 Handover execution: in Phase 3, the execution of phase 2 is done. In this phase, the handover execution of mobile node, i.e., the ongoing call of mobile node is transmitted to new network. Execution of handoff can be done in two method, soft handover and hard handover. In soft handover, the new link with the new network is established, and then, the earlier networks are terminated, i.e., connect and break, whereas in hard handoff, the existing link of network is terminated, and then new link is established, i.e., break before make.

A Survey on Handover Algorithms in Heterogeneous Wireless Network

279

Fig. 2. Handover execution phases

3 Vertical Handover Algorithms Handover algorithm in wireless network are classified depending on different parameter used by mobile node for execution of handover such as mobile node receiving signal strength [RSS], available quality of service (QoS), and velocity of mobile node, and depending on such parameter handover algorthims are classified as follows, 3.1 Signal Strength-Based Handover Algorithm Signal strength-based algorithms are designed depending on receiving signal strength [RSS] of mobile node. According to the RSS criterion [1], an MS is connected with a BS from which it receives the highest signal strength. This algorithm allows the MS to be attached with the BS with the strongest signal. The disadvantage of this algorithm is that the signal strength may vary because of shadowing and propagation characteristics such as shadowing and refraction which can cause frequent handovers. Another disadvantage is that the MS may continue to stay connected to the current BS, until it receives good signal strength from other base stations even if it has crossed the designed cell boundaries. This will lead to difficulties in maintaining physically planned cell boundaries and load balance across the cells. The addition of hysteresis as an additional criterion will help to overcome these disadvantages. Although hysteresis helps in preventing frequent unnecessary handovers, it does not help in reducing call dropping as it introduces delay in necessary handover [2]. A balance between the number of handovers and delay

280

M. B. Patil and R. Patil

in handover should be achieved by averaging signal strength and taking the appropriate hysteresis. By introducing hysteresis, if the RSS of any other BS exceeds the RSS of the current BS, then a handover is performed to the new BS. This can cause ping pong effect [2]. To avoid this, a handover margin is set. The North American personal communication systems (PACS) and personal communication service (PCS) standards combine hysteresis with dwell timer [3] to decrease total number of handovers. In the absolute signal strength algorithm, a handover is requested when the RSS drops below the threshold of 100 dBm for a noise limited system and 95 dBm for an interference limited system [5]. The handover threshold can be varied to achieve better performance. The handover threshold can also be determined dynamically by the mobile velocity and path loss slope. This will help completing a handover successfully and avoid unnecessary handover. However, this algorithm has several serious disadvantages. BS increases transmission power to cope with high interference. If the RSS is very high because of high interference, then a handover will not take place, although ideally, a handover is desirable to avoid interference. If the RSS is low, a handover will take place even if the voice quality is good, although ideally, a handover is not required. In such cases, the supervisory audio tone (SAT) information is used along with the RSS by some systems to avoid unnecessary handover [4]. Some of the findings reported in reference [6] with respect to this algorithm are (1) the probability of not finding a handover candidate channel decreases as the overlap region increases, (2) the probability of not finding a handover candidate increases as the handover threshold increases, (3) the probability of a late handover (handover occurred after the optimum time of handover) decreases as the handover threshold increases, (4) the probability of unnecessary handover, i.e., the ping pong effect, increases as the handover threshold increases, and the probability of unnecessary handover decreases as the hysteresis increases. In [17], author has derived a mathematical model for vertical handoff between WLAN and cellular network. The model takes the handover decision based on RSS in this model, and the handover decision was executed based on probability of handoff [21]. It is observed that handover failure probability increases when either velocity or handover signaling delay upsurges the fixed value of the RSS. 3.2 Velocity-Based Adaptive Handover Algorithms Velocity-based adaptive handover algorithm helps in handling handover for users with different speeds. When users are moving at different speed, their handovers should also be performed in different time. A handover request from fast moving node must be processed quickly, and this can be done using a handover algorithm with short temporal averaging window. However, if the length of the average window is kept constant, optimal handover performance will be obtained only at a speed. Velocity-adaptive handover algorithm provides a good performance for an MS with different velocities by adjusting the effective length of the averaging window [5]. This algorithm can also serve as an alternative to the umbrella cell [5] approach for tackling high speed users if low network delay can be achieved. As the umbrella cell requires extra infrastructure, this alternative approach can lead to savings in terms of infrastructure cost. One of the velocity estimation techniques uses the level crossing rate (LCR) of which the threshold level should be set as the average value of the Rayleigh distribution of the RSS [6], requiring

A Survey on Handover Algorithms in Heterogeneous Wireless Network

281

special equipment to detect the propagation-dependent average receiver power. Kawabata [6] proposes a method of velocity proportionality to the Doppler frequency. The velocity estimation technique exploits the method of diversity reception. If an MS is using selection diversity, this method requires no special equipment. The characteristics of a velocity-adaptive handover algorithm for microcellular systems are presented in reference [6]. Three methods for velocity estimation are analyzed: the LCR method, zero crossing rate method, and covariance approximation method. It is found that the spatial averaging distance that is required to sufficiently reduce the effects of fading depends on the environment. This algorithm can adapt the temporal averaging window (a window with a certain time length) used to take samples of RSS value. The window can be adapted either by keeping the sampling period of LCR constant and adjusting the number of samples per window or vice versa. 3.3 Minimum Power-Based Handover Algorithm Power optimization in handover is important task, so while designing a handoff algorithm, power consumption need to be considered. In [18], the vertical handoff decision by management strongly affects the behavior of mobile terminal in terms of battery consumption, and the authors have compared the battery consumption of mobile node in WIFI and LTE which resulted LTE consumes less battery compared to WIFI; if mobile tries to be stay connected in LTE network, then mobile battery can be efficiently used for longer period [7]. Vertical handover scheme is to minimize the total power consumption required in serving a traffic flow, while guaranteeing a service rate of different access networks. Based on a Markov decision process (MDP), it uniquely captures the power consumption during the vertical handover execution as well as the transmission power and circuit power. In [8], author proposes an optimized vertical handoff algorithm based on Markov process in vehicle heterogeneous network. In this algorithm, it considers that the status transformation of available network will affect the quality of service (QoS) of vehicle terminal’s communication service. Markov process is used to predict the transformation of wireless network’s status after the decision via transition probability. In [22], the author proposes the scheme in which the performance was superior in the context of energy consumption, handover delay and time, throughput, and average stay time in a network. 3.4 Dynamic Programming-Based Handover Algorithm A dynamic programming-based handover algorithm provides systematic solution to the handover problem. However, the efficiency of the handover algorithm depends on the model used. In [9], author investigates network selection and handoff decision with the goal of maximizing user QoS. An algorithm based on Q-learning is obtained that chooses the best network based not only on the current network state but also the potential future network and device states. The method does not require the knowledge of the statistics of the wireless environment but learns an optimum policy by utilizing the mobile device’s experience. It is shown that the QoS results of the proposed dynamic handoff decision (DHD) algorithm come very close to the performance of an optimum oracle algorithm. Zhang et al. [10] proposes a vertical handoff decision algorithm based also on dynamic

282

M. B. Patil and R. Patil

programming is presented, and the model considers the user’s location and mobility information but assumes that there is no constraint on the user’s total budget for each connection. 3.5 Prediction-Based Handover Algorithm Prediction-based handover algorithm is proposed in chapter [11], for seamless mobilitybased wireless networks. That is, scheme incorporates fuzzy logic with AP prediction process in order to lend cognitive capability to handover decision making. Selection metrics, including received signal streth, mobile node relative direction toward the access points in the vicinity, and access point load are collected and considered inputs of the fuzzy decision-making system in order to select the best preferable AP around WLANs. Adjustable weight vector concept for input metrics is proposed in order to cope with the continuous, unpredictable variation in their membership degrees. In [12], a predictionbased handover model is for multiclass traffic in wireless mobile networks by using software agents. The local and global handoff decisions are made by agent in mobile node based on speed of mobile node, i.e., for high speeds, global handoff is executed. The predictions are based upon the speed and moving direction of the mobile node. The scheme predicts location and provides necessary information for advance bandwidth (resource) reservation and optimal route establishment. 3.6 Fuzzy Handover Algorithm Fuzzy logic-based algorithm is proposed in Ref. [13], and QoS-aware fuzzy rule-based vertical handoff mechanism makes a multi-criteria-based decision and is found to be effective for meeting the requirements of different applications in a heterogeneous networking environment. The QoS parameters considered are available bandwidth, endto-end delay, jitter, and bit error rate. A new evaluation model is proposed using a non-birth–death Markov chain, in which the states correspond to the available networks. This algorithm uses fuzzy logic to make handover decision. The complexities involved in this handover criterion are tackled. The algorithm performs better than the traditional signal strength-based algorithm. It weighs different parameters according to the needs of the network. The algorithm is also capable of handling heavy fading. In [14], an adaptive fuzzy logic-based handoff decision algorithm is introduced for wireless heterogeneous networks. The parameters data rate, monetary cost, RSS, and mobile speed are considered as inputs of the proposed fuzzy-based system. According to these parameters, an output value, which varies between one and ten, is produced. This output describes the candidacy level of the available access points in the vicinity of smart terminal and is utilized in the access point selection algorithm [16]. This algorithm decides whether a handoff is necessary or not, by using the handoff resolution value. In [20], it is important to extend battery and save energy without abrupting the ongoing call or services, and even with reduced QoS, the author have designed a fuzzy rule to support an energy-efficient approach with less QoS.

A Survey on Handover Algorithms in Heterogeneous Wireless Network

283

4 Comparison of Handover Algorithms As discussed in Sect. 2, handover algorithms can be designed by considering different parameters, and these handover algorithms have their own features and drawbacks as summarized in Table 1. Table 1. Features and drawbacks of different handover algorithms Algorithm

Features

Drawbacks

Signal strength based

• Less handover failure • Unnecessary handover is avoided • Best suitable for location stable mobile nodes

• Extra handover delay • The signal strength may vary because of shadowing and propagation which can results in frequent handovers

Velocity adaptive based

• Predication of handover can • More number of handover be done depending on failure due to packet loss • Extra overhead involved in velocity of mobile node • More suitable for mobile calculating the velocity of node moving with high mobile node • Velocity may not be constant velocity or speed due which predication of between/within cells handover may fail

Power consumption based

• Mobile stays for longer • High-speed networks like period in the network which WLAN generally consumes consumes less power more power, which may • Battery life of mobile node result into mobile node may will be increased never handover/get connected to high-speed network

Dynamic programming based • Many parameters are • No general formation of considered and divided to dynamic program is take decision of handover available • Speed of handoff decision • Increased complexity in and reduced handover delay designing of handover algorithm Prediction-based handover

• Predication of handover was • Predication-based handover made on available history of involves more complexity • More signaling is involved mobile node • Less handover delay and for collection of parameters better packet delivery ratio and predication

Fuzzy handover

• Algorithms are more • Low speed and increased customizable, reliable, and handover delay • For more accuracy, requires more accurate fuzzier grade which results • Consumes low power in increase in complexity • Easy transfer of ongoing call and rules

284

M. B. Patil and R. Patil

5 Conclusion and Future Work In this chapter, we have given an overview of vertical handover mechanism in heterogenous wireless network, and the focus and attention are given on existing handover algorithm proposed by different authors which we have broadly classified into six categories depending on the parameters and mechanism used in the handover execution algorithms. The features and drawbacks of each category of handover algorithms are discussed in detail. Depending on the drawback and benefit, a best algorithm with QoS can be designed for seamless transfer of ongoing call in the next generations heterogenous network.

References 1. Ahmed T, Kyamakya K, Ludwig M (2006) Architecture of a context awarevertical handover decision model and its performance analysis for GPRS WiFi handover. In: Proceedings of the 11th IEEE symposium on computers and communications 2. Lee WCY (1995) Mobile cellular telecommunications, 2 ed. McGraw Hill 3. Pollini GP (1996) Trends in handover design. In: IEEE communications magazine, pp 82–90 4. Tripathi N (1997) Generic adaptive handoff algorithms using fuzzy logic and neuralnetworks. PhD thesis, Virginia Polytechnic Institute and State University 5. Kaloxylos G, Lanlpropoulos N, Passas L, Merakos (2006) A flexible handover rmechanism for seamless service continuity in heterogeneous environments. Comput Commun 29: 717–729 6. Ransom J (1995) Handoff considerations in micro cellular systems planning. In: Proceedings PIMRC, pp 804–808 7. Song Y, Kong P-Y, Han Y (2014) Power-optimized vertical handover scheme for heterogeneous wireless networks. Commun Lett IEEE 18(2):277–280 8. Bin M, Hong, D, Xianzhong X, Xiaofeng L (2015) An optimized vertical handoff algorithm based on Markov process in vehicle heterogeneous network, Publisher: IEEE 9. Tabrizi H, Farhadi G, Cioffi J Dynamic handoff decision in heterogeneous wireless systems: Q-learning approach. In: 2012 IEEE international conference on communications (ICC), 10–15 June 2012. ISSN: 1550-3607 10. Zhang J, Chan HC, Leung V (2006) A location-based vertical handoff decision algorithm for heterogeneous mobile networks. In: Proceedings of IEEE Globecom ’06, San Francisco, CA 11. Sadiq AS, Fisal NB, Ghafoor KZ, Lloret Z (2014) An adaptive handover prediction scheme for seamless mobility based wireless networks. Sci World J Article ID 610652 12. Birje MN, Manvi SS, Kakkasageri MS, Saboji SV Prediction based handover for multiclass traffic in wireless mobile networks: an agent based approach. In: Information, communications & signal processing, 2007 6th international conference on 10–13 Dec 2007, pp 3217–3222. ISSN:1550-3607 13. Vasu K, Maheshwari S, Mahapatra S, Kumar CS (2012) QoS-aware fuzzy rule-based vertical handoff decision algorithm incorporating a new evaluation model for wireless heterogeneous networks. EURASIP J Wirel Commun Netw Springer 14. Ceken C, Arslan C An adaptive fuzzy logic based vertical handoff decision algorithm for wireless heterogeneous networks. In: Wireless and microwave technology conference, 2009. WAMICON ‘09. IEEE 10th Annual 20–21 April 2009, pp 1–9. E-ISBN:978-1-4244-4565-3 15. Ravichandra M, Kiran Gowda KN, Uday Kumar CA (2013) A survey on handover literature for next generation wireless network. Int J Adv Res Comput Commun Eng 2(12)

A Survey on Handover Algorithms in Heterogeneous Wireless Network

285

16. Chandavarkar BR, Ram Mohan Reddy G (2011) Survey paper: mobility management in heterogenous wireless networks. In: International conference on communication technology and system design 17. Kumbalavati SB, Mallapur JD (2015) A survey on vertical handover in heterogenous network. Int J Inf Futurist Res 2(9) 18. Castillo JMR (2013) Energy-efficent vertical handovers, KTH Royal Institute of technology 19. Patil MB (2011) Vertical handoff in future heterogenous 4G network. IJCSNS Int J Comput Sci Netw Security 11(10) 20. Coqueiro T, Jailton J, Carvalho T, Francês R (2019) A fuzzy logic system for vertical handover and maximizing battery lifetime in heterogeneous wireless multimedia networks. Hindawi Wirel Commun Mobile Comput 21. Malathy EM, Muthuswamy V State of art: vertical handover decision schemes in nextgeneration wireless network. Springer J Commun Inf Netw 3(1) (2018) 22. Khan M, Din S, Gohar M, Ahmad A, Cuomo S, Piccialli F, Jeon G (2017) Enabling multimedia aware vertical handover management in internet of things based heterogeneous wireless networks. Multimed Tools Appl 76(24):25919–25941, Springer

A Non-stationary Analysis of Erlang Loss Model Amit Kumar Singh1(B) , Dilip Senapati2 , Sujit Bebortta2 , and Nikhil Kumar Rajput1 1

2

Jawaharlal Nehru University, New Delhi, India [email protected] Revenshaw University, Cuttack, Odisha, India

Abstract. A complex issue in handling systems with continually changing processing demands is an intractable task. A more current example of these systems can be observed in wireless sensor networks and trafficintensive IoT networks. Thus, an adaptive framework is desired which can handle the load and can also assist in enhancing the performance of the system. In this paper, our objective is to provide the non-stationary solution of Erlang loss queueing model where s servers can serve at most s jobs at a time. We have employed time-dependent perturbation theory to obtain the probability distribution of M/M/s/s queueing model. The time-dependent arrival and service rates are assumed to be in sinusoidal form. The opted theory gives approximation for probability distribution correct up to first and second order. The result shows that firstand second-order approximations provide better approximation than the existing ones. Keywords: Erlang loss queueing model · Non-stationary queues Blocking probability · Time-dependent perturbation theory

1

·

Introduction

Queueing system is an important tool for designing and analyzing the systems ranging from IOT, blockchain technology to telecommunication systems [1–3]. Suitable queueing models have been designed for different systems. Erlang loss queueing model is applicable in many domains like telecommunication networks, transportation systems, etc. [5,6]. Such systems are designed better and analyzed on the basis of their performance measures. In order to evaluate performance measures, the systems are assumed to be in a steady state. Based on this assumption, the systems are analyzed. The reason for this assumption is that stationary solutions are easy to obtain. However, in reality, the systems are not in steady Amit Kumar Singh was a Ph.D. student at JNU, New Delhi. c Springer Nature Singapore Pte Ltd. 2021  C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_28

A Non-stationary Analysis of Erlang Loss Model

287

state and steady-state solutions are far from reality. The arrival and service rates are not constant; rather, they are a function of time. For instance, the pattern of calls or vehicles on road is different in morning and in evening [7,8]. The non-stationary solutions of the queueing models are difficult to obtain. So, the researchers are trying to obtain approximations [6,7]. Several approximations have been found in the literature to approximate and analyze the nonstationary queueing systems. Some of these are simple stationary approximation (SSA), pointwise stationary approximation (PSA), modified offer load (MOL), etc [9–13]. The arrival and service rates are assumed to be in sinusoidal form to test these approximations. In this paper, we have opted time-dependent perturbation (TDP) theory, extensively used to solve numerous problems of quantum mechanics [13–15]. TDP is applied on the non-stationary Erlang loss model with time-varying arrival and service rates. The sinusoidal rates are generally used so that we can accurately estimate the performance of the approximation scheme, although this method can also be applied for general perturbed rates. The rest of the chapter is as follows: After the introduction section, Sect. 2 presents a literature review of non-stationary queues. Section 3 provides Erlang loss model and its CR equations. Time-dependent perturbation theory is briefly discussed in Sect. 4. Section 5 presents numerical results. The last section is the conclusion.

2

Literature Review

Non-stationary queueing models developed for analyzing the systems are complex in nature [13]. Simulation techniques can predict the behavior of the queueing models but these techniques, for example, Monte Carlo simulation, take more computational power as well as time. People choose approximation methods based on steady-state results to predict the behavior of systems. These approximations are much easier and require less computation. Simple stationary approximation (SSA) is one of the techniques based on steady-state solution. The average rate is calculated over a time interval and inserted in steady-state formula. Mathematically, the average arrival rate (similarly average service rate) over a time of length T is given by: ¯= 1 λ T

T λ(t)dt 0

All Markovian queues can be evaluated by using these average arrival rates. However, SSA method ignores non-stationarity behavior and provides stationary solution for the number of jobs in the system. Green et al. [10] found that SSA gives a reasonable approximation if the arrival rate function varies by only 10% about its average arrival rate. Green et al. [10] proposed pointwise stationary approximation (PSA) method used to compute non-stationary behavior. PSA provides time-dependent behavior based on steady-state behavior of a stationary model. It uses arrival and/or service rate that prevails at the time at which we

288

A. K. Singh et al.

want to describe the performance. For example, the blocking probability PB (t) which is the probability that no server is available is calculated approximately for M (t)/M (t)/s/s queue as: s

ρ(t) /s! PB (t) ≈ s s i=0 ρ(t) /i! where ρ(t) = λ(t)/μ(t) The expected number of busy servers E[S(t)] can be given approximately as E[S(t)] ≈ (1 − PB (t))ρ(t) Massey and Whitt [12] showed that PSA is asymptotically correct as arrival rate changes less rapidly. Modified offered load (MOL) model was introduced to approximate telephone traffic [12]. MOL is applicable to only Erlang loss models and works effectively if the blocking probability is small.

3

Erlang Loss Model

Consider a queueing system with s number of servers and with zero buffer size. That is, there can be at most s jobs in the system that are under service. The arrival pattern follows Poisson process with a non-negative time-varying rate λ(t). The service of each job follows exponential distribution with rate μ(t). Let p(n, t) be the probability of n jobs in the system at time t. The Chapmann– Kolmogorov (CR) equations for M (t)/M (t)/s/s queueing system [5] are given: d p(0, t) = −λ(t)p(0, t) + μ(t)p(1, t) dt d p(n, t) = λ(t)p(n − 1, t)−(λ(t) + nμ(t))p(n, t) + (n + 1)μ(t)p(n + 1, t) dt (1) d p(s, t) = λ(t)p(s − 1, t) − sμ(t)p(s, t) dt 0

Fig. 1. Average number of customers in M (t)/M/5/5. (λ = 10 + 7sin(2t), µ = 2)

292

A. K. Singh et al. 0.4 TDP (second approx.) PSA MOL TDP(first approx.)

0.3

Relative error

0.2 0.1 0 −0.1 −0.2 −0.3 −0.4

1

1.5

2

2.5

3

3.5

4

4.5

5

5.5

6

time t−−>

Fig. 2. Relative error of TDP(second approx.) (λ = 10 + 7sin(2t), µ = 2) 0.1 TDP(second approx.) MOL

0.05

Relative error

0 −0.05 −0.1 −0.15 −0.2 −0.25

0

1

2

3

4

5

6

time t −−>

Fig. 3. Relative error of TDP(second approx.)(λ = 10 + 7sin(2t), µ = 2)

6

Conclusion

In this chapter a time-dependent perturbation theory is used for computing probability distribution and blocking probability for non-stationary Erlang loss model. The scope of this study covers a plethora of systems with dynamic organization. The numerical results show that the opted theory outperforms the

A Non-stationary Analysis of Erlang Loss Model

293

existing approximations. The theory can be applied extensively to obtain approximations for various non-stationary queueing models.

References 1. Rivera D, Cruz-Piris L, Lopez-Civera G, de la Hoz E, Marsa-Maestre I (2015) Applying an unified access control for IoT-based intelligent agent systems. In: 2015 IEEE 8th international conference on service-oriented computing and applications (SOCA), pp 247–251 2. Said D, Cherkaoui S, Khoukhi L (2015) Multi-priority queuing for electric vehicles charging at public supply stations with price variation. Wirel Commun Mobile Comput 15(6):1049–1065 3. Yan H, Zhang Y, Pang Z, Da Xu L (2014) Superframe planning and access latency of slotted MAC for industrial WSN in IoT environment. IEEE Trans Ind Inform 10(2):1242–1251; Strielkina A, Uzun D, Kharchenko V (2017) Modelling of healthcare IoT using the queueing theory. In: 2017 9th IEEE international conference on intelligent data acquisition and advanced computing systems: technology and applications (IDAACS), vol 2 4. Ozmen M, Gursoy MC (2015) Wireless throughput and energy efficiency with random arrivals and statistical queuing constraints. IEEE Trans Inf Theory 62(3):1375–1395 5. Shortle JF, Thompson JM, Gross D, Harris (2008) Fundamentals of queueing theory. Wiley, C. M 6. Abdalla N, Boucherie RJ (2002) Blocking probabilities in mobile communications networks with time-varying rates and redialing subscribers. Ann Oper Res 112:15– 34 7. Worthington DJ, Wall AD (2007) Time-dependent analysis of virtual waiting time behaviour in discrete time queues. Euro J Oper Res 178(2):482–499 8. Pender J (2015) Nonstationary loss queues via cumulant moment approximations. Prob Eng Inform Sci 29(1):27–49 9. Alnowibet KA, Perros HG (2006) The nonstationary queue: a survey. In: Barria J (ed) Communications and computer systems: a tribute to Professor Erol Gelenbe, World Scientific 10. Green L, Kolesar P (1991) The pointwise stationary approximation for queues with nonstationary arrivals. Manage Sci 37(1):84–97 11. Alnowibet, Perros (2009) Nonstationary analysis of the loss queue and of queueing networks of loss queues. Euro J Oper Res 196:1015–1030 12. Massey WA, Whitt W (1997) An analysis of the modified offered-load approximation for the nonstationary loss model. Ann Appl Prob 4:1145–1160 13. Massey WA (2002) The analysis of queues with time-varying rates for telecommunication models. Telecommun Syst 21:173–204 14. Gottfried K, Yan TM (2013) Quantum mechanics: fundamentals. Springer Science & Business Media 15. Griffiths DJ (2016) Introduction to quantum mechanics. Cambridge University Press 16. Bransden BH, Joachain (2006) Quantum mechanics. Pearson Education Ltd., C. J

294

A. K. Singh et al.

17. Singh AK(2015) Performance modeling of communication networks: power law behavior in stationary and non-stationary environments, thesis. Thesis submitted to JNU, New Delhi 18. Dilip S, Karmeshu (2016) Generation of cubic power-law for high frequency intraday returns: maximum Tsallis entropy framework. Digital Signal Process 48:276– 284 19. Tanmay M, Singh AK, Senapati D (2019) Performance evaluation of wireless communication systems over weibull/q-lognormal shadowed fading using Tsallis’ entropy framework. Wirel Pers Commun 106(2):789–803

A Novel Authentication Scheme for Wireless Body Area Networks with Anonymity Upasna Singh(B) and Bhawna Narwal Department of IT, Indira Gandhi Delhi Technical University for Women (IGDTUW), New Delhi, Delhi, India [email protected], [email protected]

Abstract. Increasing crossover of information and wireless network technologies in the medical field embarks a major revolution. Wire body area network (WBAN) is one such result of crossover. Low-cost, lightweight, multipurpose sensors can be easily integrated into a wireless communication network for health monitoring. It is a wireless networking technology based on radio frequency, consisting of small sensors, transmitting the data which could be further used for medical or safeguarding measures, and thus, there is a need to introduce better safety measures in WBAN schemes. In this paper, we proposed a novel authentication scheme for WBAN with anonymity and provided a formal security proof through BAN logic. Keywords: Authentication · Anonymity · BAN logic · Security · Wireless body area network (WBAN)

1 Introduction There has been a great growth in ubiquitous computing from the past few decades and demand for low power technologies is increasing as well. A wireless body area network (WBAN) is a network of multiple sensor nodes implanted or worn by human body. The term was first coined by Van Dam et al. in 2001 [1]. WBANs are designed in such a way that they collect patients’ physiological values such as activity, blood pressure, heart rate, blood-oxygen level, resting body heat. Typically, these nodes transmit in small range of 2 m (approximately) [2]. The physician access the recorded values through network and remotely prescribes the required treatment via Internet or on-site to patient. WBAN use case varies to various applications including workplace safety, electronics, for athletes, health care, and medical monitoring. Also, there is a significant lifestyle problem as millions of people are suffering from chronic diseases which include obesity, heart problems, etc., and thus, real-time monitoring is becoming increasingly common for general users. The future trend and current alarming situation need new technologies to facilitate affordable, easy-to-use, mobile, first-hand, and lightweight health monitoring. WBANs play an important role in pervasive communications and thus cultivating a budding market for commercial use. WBANs are open and mobile, and due to these properties, data security during the exchange is paramount [3, 4]. As WBAN is being © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_29

296

U. Singh and B. Narwal

used commercially, security attacks are common and adversaries are able to gather potential information about client and perform physical attacks and they do not need to know the context of the data or identities of the node to achieve this. The values that are being transmitted in WBAN are the basis of medical research and assessments. These values contain very sensitive data that needs to be preserved thoroughly. If any value is forged it may harm the patient. Therefore, security is a crucial problem for WBANs and the data transfer must be anonymous, confidential and nodes should be mutually authenticated. The basic WBAN illustration is provided in Fig. 1.

Fig. 1. Illustration of wireless body area network (WBAN)

2 Related Work In network and information security, authentication and key agreement mechanisms are widely used to provide secure service and user access. There are many authentication and security schemes being proposed for wireless networks. With the increased use of WBAN, IEEE proposed 802.15.6 standard which consists of three levels of security as per the application need. Later on it recommends 4 ECC-based security schemes to achieve these security levels but recent work has shown these protocols are vulnerable to several attacks [3]. Li et al. [5] proposed a secure, lightweight, anonymous, key agreement, and mutual authentication scheme for 2-hop centralized WBAN. The system consisted of 2 models, network model with two-tiered architecture and an adversary model. WBAN setup here has three phases, namely setup, registration, and authentication. In authentication phase, N (2nd level node) authenticates anonymously with the hub node (HN) through an intermediate network node (IN). The role of intermediate role is to simply forward the data to hub node after placing its identity. Hub node verifies the identity and rejects it if there is no match. It also checks the timestamp and performs

A Novel Authentication Scheme

297

the authentication check functions. The scheme was able to resist eavesdropping attack, stolen ephemeral key attack, impersonation attack, spoofing, reply, node capture attacks and provides forward and backward security with unlinkability of sessions. They used well-known Dolev–Yao threat model. Wang et al. [6] introduced an authentication scheme using bilinear pairing. Their scheme constituted of entities the client, a third party, and an application provider. However, they used bilinear pairing, which makes it unsuitable for light WBAN applications because of its high computations and in-node processing overheads. They claimed that their scheme accomplishes mutual authentication, key establishment and can defend various attacks. However, Jiang et al. [7] observed that their scheme is not resistant to client impersonation attacks and they made an improvement on [6] by introducing ECC and bilinear pairing, thus introducing the unlinkability and resistance to client impersonation attack by introducing new ‘Auth’ parameters in the authentication part of [6]. Xiong et al. [8] introduced a scalable certificateless authentication scheme which supported anonymity as well as forward security along with key escrow resilience, but is computationally heavy. Author reported reduction in the network communication overhead and cost of computation along with remote authentication [9]. Looking at some more protocols briefly, He et al. [10] proposed a new anonymous authentication (AA) scheme targeting multiple attack scenarios. Liu et al. [11] worked on a similar domain, but rather focused on selective authentication. They worked on a layered network model with 2-hop star topology. Arya et al. [12] introduced a new protocol which targeted the fake sink issue. From these studies, we noticed that: • For the prevention of false data injections, denial of service (DoS) attacks, etc., authentication is necessary. It is also required for the verification of user identity for data transfer and data access services in WBAN [4, 13, 14]. • Portable sensors have insufficient power supplies which render them rather inferior in terms of storage capacities and computations. Most of the existing authentication schemes aren’t suitable for wearable devices, and they are either computationally heavy or lack security measures. There is a trade-off between computation cost and functionality. • Sensor nodes use certain cryptography mechanisms, and they should be as lightweight as possible, offer fast computation and minimal storage overhead [15–18]. • Data injection by adversary should be prevented. • There is always a risk of client’s privacy information getting leaked because of the access policies. Therefore, client anonymity is desired.

3 Proposed Approach 3.1 Security Requirements • Protocol should fulfill basic security goals, e.g., message modification, nonrepudiation, integrity, confidentiality. • Client should be anonymous. • The communication between client and AP should be confidential. This will be achieved by symmetric key encryption techniques [19].

298

• • • • •

U. Singh and B. Narwal

Authentication scheme should preserve privacy. Scheme should be certificateless. The client and application provider should be mutually authenticated [20, 21]. Protocol should be resistant to replay attack [22]. The objective is to keep the scheme lightweight but at the same time also effectively securing the privacy of the WBAN client [22].

3.2 System Model The system model consists of three main entities, connected over a wireless network. These entities are as follows: • WBAN client. WBAN clients are issued with WBAN terminals. WBAN terminals can be wearable sensors, biosensor and lightweight, movable medical device. The sensor takes the physiological values and sends them to AP. Client is preloaded with public parameters used for communications. It is registered with the trusted third party to obtain parameters, and it accesses the services being provided by AP after the initial setup is done. • Application Provider (AP). It is a medical institute such as clinic, hospital, a thirdparty monitoring server by the hospital. APs provide several services from client’s point of view like status monitoring, archiving the long-term patient data, which is later on used by the doctors. AP is also pre-equipped with authentication parameters and registered to NM similarly to a WBAN client. Doctors and experts use the information about the client through AP to remotely treat the client. • Trusted Third Party/Network Manager (NM). The registration of WBAN clients and APs is done by fully trusted third-party network manager. The public and auth parameters are distributed for wireless communication between the nodes and AP. Since NM is fully trusted, commercial third party, the impersonation of clients from its side is very less likely (Fig. 2).

4 Proposed Scheme We propose a new protocol to achieve mutual authentication between the entities mentioned in the previous section. The protocol is divided into initialization, registration, and authentication phases. Initialization. In the initialization phase, NM generates public and private keys to communicate with client and application provider. It publishes the public key and identity IDnm and K nm , which is h (id, random number). Registration. In the registration phase, the object (client/application provider) generates a random number, calculates K o = h(randomo ⊕ IDo ), and generates a public and private key. The object then encrypts the M o and sends to NM. M0 (Ko ⊕ Knm , Keyo )EKeynm

A Novel Authentication Scheme

299

Fig. 2. System model

NM decrypts the message and replies with encrypted message to client and AP, respectively, along verification data, that is, Auth = (Kc ⊕ Knm ⊕ Kap ). The object stores the parameters. Authentication. In the authentication phase, first C and AP mutually authenticate, AP shares authentication parameters with client and then both generates a shared secret session key which is later used for data transmission. C selects a random number r c . Calculate Lc = ( Kc , Tc , Authc , rc )Keyap 3. M1 : (Lc , Tc )EKeyc . C sends M 1 to the AP. After receiving M 1 , AP decrypts and checks the freshness of Timestamp. If the value is not the latest, AP discards the auth request. AP also checks the auth values. Using its private key, AP decrypts L c and stores the remaining values. AP calculates K c = Authc ⊕ K nm ⊕ K ap . 5. AP checks K c = K c , else discard. 6. AP generates key parameters, generates a reply to client, encrypts it with Keyc, and sends the message. 7. Client checks the timestamp again, decrypts, and verifies and stores the original parameters. Using these stored parameters, AP and client obtain a session key, S k (Fig. 3 and Table 1).

1. 2. 3. 4.

5 Security Proof: BAN Logic The BAN logic [23, 24] is generally used to do an analysis of the security of key agreement and authentication protocols and models. We use BAN logic to prove that

300

U. Singh and B. Narwal

Fig. 3. Proposed authentication scheme

Table 1. Notations used in scheme Notation

Description

k nm/c/ap

Identity parameters

IDo

Identity of client/AP

Keyc/ap/nm

Public key of AP/C/NM, respectively

Keyc /ap /nm’

Private key of C/AP/NM, respectively

Rc , Z, randomo Generated random numbers Ti

Current timestamp of object i

Sk

Symmetric key for data sharing

h

Hash function

AP and client entities are mutually authenticated. Notations and rules used for security derivations are as follows. 1.

X, Y: principals; A, B: statements

A Novel Authentication Scheme

301

2. 3. 4. 5. 6. 7.

#(A): A is fresh X |≡ A: X believes A X ⇒ A: X has jurisdiction over A X |∼ A: X says A X  A: X sees A k X ←→ Y: X and Y have share key ‘k’.

8.

Message meaning Rule (1): If X |≡ X ←→ Y and X  (A)K then X believes Y once said A. Message meaning Rule (2): If K is Y s public key, and X  (A)−1 K , then X believes that A was sent by Y. Message meaning Rule (3): If B is a shared secrete [25] between X and Y, and X sees a message where B is combined with A (and X did not send the message), then X believes that it was sent by Y. The freshness conjugation rule: If X |≡ #(A), then X |≡ #(A, B). The nonce verification rule: If X |≡ #(A) and X |≡ Y |∼ A, then X |≡ Y |≡ A. The jurisdiction rule: If X |≡ Y ⇒ A and X |≡ Y |≡ A then X |≡ A.

9. 10.

11. 12. 13.

k

5.1 Goals The above-mentioned rules are applied over the annotated and idealized versions of the messages, and goals are derived as per the security requirements of the protocol. Sk

1. C |≡ (C ←→ AP) Sk

2. C |≡ AP |≡ (C ←→ AP) Sk

3. AP |≡ (C ←→ AP) Sk

4. AP |≡ C |≡ (C ←→ AP) 5.2 Protocol is Converted into Idealized Form in Order to Apply BAN Logic 1. C to AP: M 1 : (T c , Rc S k )Keyap . Sk

2. AP to C: M 2 : (T c , Rc , T ap Z, C ←→ AP)Keyc . According to the description of the protocol, the following assumptions can be made for initialization. 1. 2.

C |≡ #(TimestampAP ) AP |≡ #(TimestampC )

3.

C |≡ −→ C

4.

AP |≡ −→ AP

5.

C |≡ −→ AP .Keyap AP |≡ .. −→ C

6. 7.

Keyc

Keyap

Keyap

Keyc −1

C |≡ −→ C

302

8. 9.

U. Singh and B. Narwal Keyc −1

AP |≡ −→ AP SK C |≡ AP ⇒ (C ←→ AP) SK

10. AP |≡ C ⇒ (C ←→ AP) Based on BAN rules and above assumptions, the following analysis can be made. From message 1, AP  (T c , Rc )Keyap , where Rc is extracted after decrypting (L)Keyc . From extended public key rule, AP |≡ C |∼ (T c , Rc ). From freshness conjugate rule, AP |≡ C |≡ (T c , Rc ), which can be further deduced to AP |≡ C |≡ (Rc ). SK

Therefore, AP |≡ C |≡ (C ←→ AP) GOAL4. From the above goal, jurisdiction rule and considering the initialization assumption SK

where AP |≡ C ⇒ (C ←→ AP). SK

So, AP |≡ C ←→ AP GOAL3. Message 2 idealized form: SK

C  (T c , Rc , T ap , Z, C ←→ AP)Keyc . From public key rule, SK

C |≡ AP |∼ ((T c , Rc , T ap , Z, C ←→ AP). SK

Applying freshness conjugate rule, C |≡ AP |≡ (T c , Rc , T ap , Z(C ←→ AP)), SK

i.e. C |≡ AP |≡ (C ←→ AP) GOAL2. From GOAL 2 and initial statement 9, applying jurisdiction rule: C |≡ (C S k ←→ AP) GOAL1. From BAN logic, we proved that client and AP are mutually authenticated. 5.3 Performance Analysis The protocol was simulated on AVISPA [26], texted using OFMC back-end checker whose results are specified in Fig. 4. The proposed scheme is SAFE as per the goals specified. The simulation was run over a 62-bit operating system, ×64-based Intel(R),i37100U processor, CPU @ 2.40 GHz. 5.4 Discussion 1. Anonymous Unlinkable Session. For every session, the client node generates fresh random parameters for the communication. The client identity parameters used in each session include the random parameter (h(IDo ⊕ random)) which is included in the communication message along with Rc , T c which are session inclusive parameters, both independent, fresh, and random. Thus, it will be impossible for an adversary to link two session IDs with same N and sessions are unlinkable. 2. Replay Attack. Our scheme uses time stamps—T ap , T c —for each sessions, hence preventing the replay attack. Every time the message is received at either of the nodes, time stamps are checked for freshness, if they are not fresh, authentication request is discarded, which further helps prevent the relay attacks too.

A Novel Authentication Scheme

303

Fig. 4. OFMC simulation result

3. Forward/Backward security. It means that if any session key Keys is exposed to adversary, no previous and future sessions should be affected. Our scheme works with this because session key is generated with session-specific dynamic parameters, thus ensuring the forward and backward security. 4. Confidentiality and Authentication. Our scheme uses public key encryption for session key generation, (M1 )Kc , for first level auth with the C public key shared by NM and later on the AP matches the authc parameter after decryption of (L c )Keyap . Also as proved using BAN logic, both client and AP are mutually authenticated. 5. Our scheme also fulfills other security requirements like privacy, non-repudiation, client anonymity, certificateless, resilience to forgery.

6 Conclusion In this paper, we proposed a certificateless, lightweight, anonymous authentication scheme for WBAN. In our scheme, nodes need to store only one parameter (authc) and key details during registration phase and later on generate session-specific parameters dynamically for session key generation which implements forward and backward security. Our scheme also meets the security requirements mentioned in Sect. 3. In future, we plan to explore a broader horizon of security aspect—authentication in scalable WBANs, key escrow, etc.

References 1. Van Dam K, Pitchers S, Barnard M (2001) Body area networks: towards a wear-able future. In: Proc. WWRF kick off meeting, Munich, Germany, pp 6–7 2. Li M, Lou W, Ren K (2010) Data security and privacy in wireless body area networks. In: IEEE wireless communications, vol 17, No 1 3. Toorani M (2015) On vulnerabilities of the security association in the ieee 802.15. 6standard. In: International conference on financial cryptography and data security, Springer, pp 245–260 4. Narwal B, Mohapatra AK, Usmani KA (2019) Towards a taxonomy of cyber threats against target applications. J Stat Manage Sys 22(2):301–325

304

U. Singh and B. Narwal

5. Li X, Ibrahim MH, Kumari S, Sangaiah AK, Gupta V, Choo K-KR (2017) Anonymous mutual authentication and key agreement scheme for wearable sensors in wireless body area networks. Computer networks, vol 129, pp 429–443 6. Wang C, Zhang Y (2015) New authentication scheme for wireless body area net-works using the bilinear pairing. J Med Sys 39(11):136 7. Jiang Q, Lian X, Yang C, Ma J, Tian Y, Yang Y (2016) A bilinear pairing based anonymous authentication scheme in wireless body area networks for mhealth. J Med Sys 40(11):231 8. Xiong H (2014) Cost-effective scalable and anonymous certificateless remote authentication protocol. IEEE Trans Inform Forensics Secur 9(12):2327–2339 9. Xiong H, Qin Z (2015) Revocable and scalable certificateless remote authentication protocol with anonymity for wireless body area networks. IEEE Trans Inform Forensics Secur 10(7):1442–1455 10. He D, Zeadally S, Kumar N, Lee J-H (2017) Anonymous authentication for wire-less body area networks with provable security. IEEE Sys J 11(4):2590–2601 11. Liu J, Li Q, Yan R, Sun R (2015) Efficient authenticated key exchange protocol ls for wireless body area networks. EURASIP J Wirel Commun Networking 2015(1):188 12. Arya A, Reddy C, Limbasiya T (2017) An improved remote user verification scheme in wireless body area networks. Procedia Comp Sci 113:113–120 13. Xu C, Yang J, Gao J (2019) Coupled-learning convolutional neural networks for object recognition. Multimedia Tools Appl 78(1):573–589 14. Liu B, Wang M, Foroosh H, Tappen M, Pensky M (2015) Sparse convolutional neural networks. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 806–814 15. Sun Y, Xue B, Zhang M, Yen GG (2019) Evolving deep convolutional neural networks for image classification. IEEE Trans Evol Comput 16. Zhao ZQ, Zheng P, Xu ST, Wu X (2019) Object detection with deep learning: a review. IEEE Trans Neural Netw Learn Syst 17. Zhang X, Zhou X, Lin M, Sun J (2018) Shufflenet: an extremely efficient convolutional neural network for mobile devices. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 6848–6856 18. Iandola FN, Han S, Moskewicz MW, Ashraf K, Dally WJ, Keutzer K (2016) SqueezeNet: Alexnet-level accuracy with 50× fewer parameters and 1 this model has a long-tailed behavior which provides a better fit to the tail fluctuations in the fading signals [13]. In this paper, the importance of q-Weibull distribution [12] is portrayed to characterize the composite fading scenarios. The different performance metrics, viz. the outage probability (Pout ), average channel capacity, and amount of fading (AOF ), are expressed in closed form. It is observed that the results obtained are analytically tractable and the obtained measures are found to be in close agreement with the Monte Carlo simulation results. The remaining paper proceeds as follows: Section 2 illustrates the q-Weibull distribution. Section 3 portrays the importance of q-Weibull model in contrast to the well-known Weibull-lognormal distribution in capturing composite fading environments. The numerical and simulation results for the performance measures have been discussed in Sect. 4. Finally, Sect. 5 provides the conclusion and future works.

2

The q-Weibull Distribution

The pdf of q-Weibull distribution can be defined as [12]: 1  α  1−q f (γ) = γ α−1 αλα (2 − q) 1 − (1 − q)(λ2 γ 2 ) 2 , γ > 0, 1 < q < 2,

(1)

where α > 0 is the fading parameter, λ > 0 represents the scale parameter, and γ is the received signal-to-noise ratio (SNR). When q → 1, Eq. 1 converges to the well-known Weibull distribution. Furthermore, for q¿1 the q-Weibull distribution follows a long-tailed behavior which enables it to smoothly characterize the tail fluctuations in the fading signals. From Fig. 1, it is observed that as q approaches 2, the phenomena of longtailed distributions are obtained and similarly as q → 1, the curve becomes peaked and mimics the Weibull distribution.

3

Significance of q-Weibull Model

It is well known that shadowing effects are characterized by lognormal distribution [1,3]. However, the conventionally used lognormal-based composite models, viz. Rayleigh-lognormal, WL models are inefficacious in characterizing the tail

Performance Evaluation of Composite Fading

319

1 q=1 q = 1.1

0.8

q = 1.3

PDF

0.6

q = 1.5

0.4

q = 1.7 q = 1.9

0.2 0

0

1

2

3

4

5

6

7

8

9

10

SNR ( ) Fig. 1. Plot of (q-Weibull pdf ) against SNR (γ) for different values of q) 100

p( )

10-2

10-4

10-6

10-8

Generated synthetic signal Weibulll-Lognormal distribution 1

1.5

2

2.5

3

3.5

4

4.5

5

SNR ( ) Fig. 2. Illustration of (Weibull-lognormal ) distribution corresponding to synthetic signal

fluctuations in the fading signals [13]. In Fig.2, it is observed that the well-known WL model [4,5] fails to provide a better fit to the synthetic fading signals. The fading signals are generated using MATLAB to mimic the real-time fading scenarios. However, it is possible that the entropy-based q-Weibull distribution can characterize the entire outliers in the fading channels. In this context, the q-Weibull model provides a better fit to the outliers in the fading signals for 1 < q < 2. The generated fading signals are well characterized by the q-Weibull distribution corresponding to q = 1.3 as shown in Fig. 3.

320

T. Mukherjee et al. 100

p( )

10-1

10-2

10-3

10-4 0.5

Generated synthetic signal q-Weibull distribution 1

1.5

2

2.5

3

3.5

4

4.5

5

5.5

6

SNR ( ) Fig. 3. Illustration of (q-Weibull distribution) corresponding to synthetic signal for (q = 1.3)

4

Numerical Results and Discussions

In this section, various performance metrics in wireless communications systems, viz. average channel capacity, amount of fading, outage probability in correspondence to the q-Weibull model, have been illustrated. Furthermore, the obtained measures are found to be in closed agreement with the simulation results (Fig. 4). 4.1

Amount of Fading

In wireless fading scenarios, the amount of fading (AOF) is an important performance metric as it specifies the severity of fading and is given as [4,5]: AOF =

E[γ 2 ] (E[γ])

2

−1

(2)

The rth moment of γ is obtained as: ∞ 1  α  1−q r E[γ ] = γ r ×γ α−1 αλα (2 − q) 1 − (1 − q)(λ2 γ 2 ) 2 dγ

(3)

0

or, E[γ r ] =

λα (2−q) 1 Γ( q−1 )



   − 1+ r ((q − 1) λα ) ( α ) Γ 1 + αr Γ −1 −

r α

+

1 q−1



From Eqs. (2) and (4), analytical expression for AOF is given as:

− 1+ 2 1 1 2 Γ[ q−1 ((q−1)λα ) ( α ) Γ[1+ α ]Γ[−1− α2 + q−1 ] ] AOF = × λα (2−q) − 1 −2 1+ 2 ((q−1)λα ) ( α ) Γ2 [1+ 1 ]Γ2 [−1− 1 + 1 ] α

α

q−1

(4)

(5)

Performance Evaluation of Composite Fading

321

Amount of Fading

2

Analytical solution Simulation result

1.5

1

q = 1.01, 1.1, 1.2

0.5

0

1

1.5

2

2.5

3

3.5

4

4.5

5

Fading parameter ( ) Fig. 4. Analytical solution with simulation results for amount of fading against fading parameter (α) corresponding to different q

4.2

Average Channel Capacity

It is an important performance metric that defines the maximum transmission rate a channel achieves with a small probability of error. In ergodic sense, the average channel capacity is expressed as [5,14]: C 1 = B ln(2)

∞ ln(1 + γ)f (γ)dγ,

(6)

0

where B denotes the bandwidth of the fading channel. Equation(1) and Eq.(6) yield: 1 C = B ln(2)

∞

1  α  1−q αλα (2 − q)ln(1 + γ)γ α−1 × 1 − (1 − q)(λ2 γ 2 ) 2 dγ

(7)

0

As Eq. (7) is not analytically tractable,   using  Meijer’s G approximations of the 1,1 12 . polynomials [15]; ln(1 + γ) = G22 γ 1,0 and exp(−γ) = G10 01 [γ |0 ], Eq. (7) becomes: ∞      α αλα C 12 1,1 10 2 2 . (8) = γ α−1 G22 γ 1,0 G01 (λ γ ) 2 |0 dγ B ln(2) 0

Using Eq. (8) and the substitutions in [16], the average channel capacity is obtained as:  C αλα 31  α (−α,α),(1−α,α) (9) = H23 λ (0,1),(−α,−α),(−α,α) B ln(2)

322

T. Mukherjee et al.

Average Channel Capacity (C/B)

1.15

Analytical solution Simulation results

1.1

q = 1, 1.1, 1.2

1.05 1 0.95 0.9 0.85

1

1.5

2

2.5

3

3.5

4

4.5

5

Fading Parameter ( ) Fig. 5. Plot of average channel capacity against fading parameter (α) corresponding to different values of parameter q

In Fig. 5, it is evident that the analytical and simulation results of average channel capacity are in close agreement with each other when q → 1. 4.3

Outage Probability

It is one of the significant performance measures for communications systems over several fading channels. It is denoted as Pout and defining the probability of output SNR γ under a specified threshold value γth [4,5] (Fig. 6). It is expressed as [1] γth Pout = f (γ)dγ (10) 0

Equations (1) and (10) yield: Pout

γth α 1 = αλ (2 − q) γ α−1 [1 − (1 − q)(λγ) ] 1−q dγ α

(11)

0

Hence, outage probability can be obtained as: Pout = 1 − (1 + γth α (q − 1) λα )

5

1 1+ 1−q

(12)

Conclusion and Future Work

The theoretical results of the important performance metrics, viz. outage probability, average channel capacity, and amount of fading with respect to the q-

Performance Evaluation of Composite Fading

323

Outage Probability

100

10-5

= 1, 3, 5

10-10 -20

Analytical solution Simulation results -15

-10

-5

0

5

10

15

20

25

30

Threshold SNR [dB] Fig. 6. Analytical solution with simulation results for Pout against threshold SNR for different α

Weibull model, were derived. It is profound that Weibull distribution incorporated with the Tsallis’ non-extensive parameter ‘q’ can characterize the composite fading channels. The q-Weibull model captured both fast fading and shadowing effects corresponding to different wireless communication channels for continuous interval of the non-extensive parameter ‘q’, i.e., 1 < q < 2. It is worth noting that the q-Weibull model provided a better fit with the synthetic fading signal corresponding to q=1.3 and also characterized the tail fluctuations in the fading signal in contrast to the composite WL model. It would be interesting and challenging in characterizing fading channels and computation of the symbol error probability using the aforementioned ‘q’-Weibull model over the other existing complex models on the basis of non-extensive parameter ‘q’.

References 1. Simon MK, Alouini MS (2005) Digital communication over fading channels, vol 95. Wiley, Hoboken 2. Shankar PM (2017) Fading and shadowing in wireless systems. Springer, Berlin 3. Rappaport TS (1996) Wireless communications: principles and practice, vol 2. Prentice Hall PTR, New Jersey 4. Singh R, Soni SK, Raw RS, Kumar S (2017) A new approximate closed-form distribution and performance analysis of a composite Weibull/log-normal fading channel. Wireless Personal Commun 92(3):883–900 5. Chauhan PS, Tiwari D, Soni SK (2017) New analytical expressions for the performance metrics of wireless communication system over Weibull/Lognormal composite fading. AEU-Int J Electron Commun 82:397–405 6. Hansen F, Meno FI (1977) Mobile fading rayleigh and lognormal superimposed. IEEE Trans Veh Technol 26(4):332–335

324

T. Mukherjee et al.

7. Ismail MH, Matalgah MM (2005) Outage probability in multiple access systems with Weibull-Faded lognormal-shadowed communication links. In: 2005 IEEE 62nd vehicular technology conference, 2005. VTC-2005-Fall, vol 4. IEEE, New York, pp 2129–2133 8. Hashemi H (1993) The indoor radio propagation channel. Proc IEEE 81(7):943–968 9. Sagias NC, Karagiannidis GK (2005) Gaussian class multivariate Weibull distributions: theory and applications in fading channels. IEEE Trans Inf Theory 51(10):3608–3619 10. Senapati D (2016) Generation of cubic power-law for high frequency intra-day returns: maximum Tsallis entropy framework. Digit Signal Proc 48:276–284 11. Tsallis C, Mechanics NES (2004) Construction and physical interpretation. Nonextensive Entropy Interdiscip Appl, pp 1–52 12. Jose K, Naik SR, Risti´c MM (2010) Marshall-Olkin q-Weibull distribution and max-min processes. Stat Pap 51(4):837–851 13. Mukherjee T, Singh AK, Senapati D (2019) Performance evaluation of wireless communication systems over Weibull/q-lognormal shadowed fading using Tsallis entropy framework. Wireless Personal Commun, pp 1–15 14. Shannon CE (2001) A mathematical theory of communication. ACM SIGMOBILE Mob Comput Commun Rev 5(1):3–55 15. Adamchik VS, Marichev OI (1990, July) The algorithm for calculating integrals of hypergeometric type functions and its realization in REDUCE system. In: Proceedings of the international symposium on Symbolic and algebraic computation. ACM, New York, pp 212–224 16. Mathai AM, Haubold HJ (2008) Special functions for applied scientists, vol 4. Springer, New York

Benchmarking Performance of Erasure Codes for Linux Filesystem EXT4, XFS and BTRFS Shreya Bokare1(B) and Sanjay S. Pawar2 1

2

Principal Technical Officer, Centre for Development of Advanced Computing, Mumbai , India [email protected], http://www.cdac.in Usha Mittal Institute of Technology, SNDT Women’s University, Mumbai , India [email protected], http://www.umit.ac.in

Abstract. Over the past few years, erasure coding has been widely used as an efficient fault tolerance mechanism in distributed storage systems. There are various implementations of erasure coding available in the research community. Jerasure is one of the widely used open-source library in erasure coding. In this paper, we compared various implementations of Jerasure library in encoding and decoding scenario. Our goal is to compare codes with different filesystems data to understand its impact on code performance. The number of failure scenarios is evaluated to understand performance characteristics of Jerasure code implementation. Keywords: Erasure coding · Distributed storage BTRFS · EXT4 · Jerasure 2.0

1

· Filesystem–XFS ·

Introduction

Erasure coding for storage-intensive applications is gaining importance as distributed storage systems are growing in size and complexity. Erasure coding is an advanced version of RAID systems in the factors like fault tolerance and lower storage overhead and the ability to scale in a distributed environment. This makes erasure codes superior to RAID systems and the most suitable for storage intensive applications [1,2]. There are several implementation libraries of erasure coding, namely liberasurecode [3], Jerasure [4], Zfec [5], LongHair [6], Intel ISA-L [7], etc. Jerasure is a widely used erasure coding library in various open-source software-defined distributed storages like Ceph [8]. Jerasure is one of the stable libraries that supports a horizontal mode of erasure codes, written in C/C++ and implements several variants of Reed–Solomon and maximum distance separable (MDS) erasure codes (Vandermonde, Cauchy, Blaum–Roth, RAID-6, Liberation). Jerasure implementation uses matrix-based coding with Galois field arithmetic. GF-Complete library, which has procedures of Galois Field arithmetic, is used in Jerasure 2.0. c Springer Nature Singapore Pte Ltd. 2021  C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_32

326

S. Bokare and S. S. Pawar

In traditional enterprise architecture having structured and unstructured data types, filesystem plays an important role providing support for scalability, extendibility and optimization with respect to the storage technology. Impact of the traditional and modern filesystems needs to be evaluated with erasure coding used in distributed storage systems. Therefore, primary motivation behind the work is to evaluate the working of traditional and modern filesystems with respect to different erasure coding implementations, which can help in selecting the best suitable combination for optimal data storage and access. In this paper, encoding and decoding experiments on various code variants of Jerasure library (version 2.0 released in 2014) are performed. The rest of the paper is organized as follows. Section 2 provides various code implementations in Jerasure. Section 3 mentions about the impact of filesystem on data the reading and writing. Experimental setup is explained Sect. 4. Section 5 provides details of benchmarks and results. The paper is concluded in Sect. 6.

2

Jerasure Coding Library

The Jerasure library with the first release in 2007 and the next release Jerasure 2.0 in 2014 is one of the oldest and most popular erasure coding libraries [4]. Jerasure implements minimum distance separable (MDS) codes, where erasure coding is organized in the following manner for given dataset D. The data D is divided into k equal-sized information blocks. The k information blocks are then coded into n blocks of equal size. The n blocks consist of k information blocks and m = (n − k) parity blocks. These n blocks are written into n distributed storage nodes. The k storage nodes holding data information are called data nodes, and those having parity information are called as parity nodes. In the case of failure of any m nodes, the lost data can be reconstructed using the remaining k nodes. Another important parameter in the erasure coding is a strip unit w, which is the word size. All the code views each device as having w bits worth of data. The w ∈ {8, 16, and 32} can be considered as collection of w bits to be a byte, short word or word, respectively. The matrix-based coding used in Jerasure is based on distribution (generator) matrix whose elements are calculated using Galois field arithmetic GF (2w ) for some value of w. Figure 1 shows the distribution matrix and calculation of data and parity blocks. 2.1

Reed–Solomon Code

Reed–Solomon codes are the widely used codes and have the longest history [9]. The strip unit, i.e., w-bit word, must be large enough to satisfy n 1 which follows power law [21, 22]. 3.3 MEP in Laplace Domain There are time-dependent problems which are analytically intractable. It may happen that the solution of a system in time domain is difficult but if the problem is transformed in Laplace domain, the solution can be computed using MEP easily. A variety of systems which is described by birth and death process generally provides probabilistic description of the system in equilibrium. A challenging problem is to investigate the transient solution of such problems. For better illustration, a transient solution of M/M/1/1 queue is described below. Consider pn (t) be the number of jobs in the system at time t. The chapman-Kolmogorov equations for M/M/1/1 queue are given in [23]. If we consider Pn (s) to be the Laplace transformation of pn (t) then Pn∗ (s) = sPn (s) would be probability in Laplace domain. MEP formulation given in Sect. 2.2 can be transformed to Laplace domain in the context of M/M/1/1 queue as follows: Max S ∗ (P ∗ , s) = −

1  n=0

Pn∗ (s) log Pn∗ (s)

(16)

378

A. K. Singh et al.

subject to P0∗ (s) + P1∗ (s) = 1.

(17)

and −

λ+μ ∗ μ P0 (s) + P1∗ (s) = n0 − s s

(18)

This optimization problem yields P0∗ (s) =

1 − n0 μ + s(s + λ + μ) s + λ + μ

(19)

p1∗ (s) =

n0 λ + s(s + λ + μ) s + λ + μ

(20)

Taking inverse Laplace transform, we get μ  p0 (t) = 1 − e−(λ+μ)t + (1 − n0 )e−(λ+μ)t λ+μ λ  1 − e−(λ+μ)t + n0 e−(λ+μ)t p1 (t) = λ+μ

(21) (22)

which are the expressions of exact transient probabilities for M/M/1/1 queue [23].

4 Conclusion MEP has been used to derive both exponential family of distribution and power law distribution. Transient problems may also be solved by employing MEP on the transformed problem in Laplace domain. The probability distribution obtained by maximizing entropy subject to information set portrayed the wide range of exponential family of distributions and distributions exhibiting power law behavior. The power law distribution in terms of Hurwitz zeta function can be obtained by maximum Shannon entropy in presence of geometric mean constraint. The maximization of Shannon entropy in Laplace domain provided a transient probability distribution which characterized the behavior of M/M/1/1 queue systems [15].

References 1. Singh VP, Rajagopal AK, Singh K (1986) Derivation of some frequency distributions using the principle of maximum entropy (POME). Adv Water Resour 9(2):91–106 2. Mitzenmacher M (2003) A brief history of generative models for power law and lognormal distributions. Internet Math. 1(2):129–251 3. Leland WE, Taqqu MS, Willinger W, Wilson DV (1994) On the self-similar nature of ethernet traffic. IEEE/ACM Trans Networking 2:1–15 4. Kapur JN (1989) Maximum-entropy models in science and engineering. Wiley Eastern Ltd.

Adaptive Applications of Maximum Entropy Principle

379

5. Sharma S (2006) Queue length distribution of network packet traffic: Tsallis entropy maximization with fractional moments. IEEE Commun Lett 10(1):34 6. Senapati D (2016) Generation of cubic power-law for high frequency intra-day returns: maximum Tsallis entropy framework. Digit Signal Proc 48:276–284 7. Lampo TJ, Stylianidou S, Backlund MP, Wiggins PA, Spakowitz AJ (2017) Cytoplasmic RNA-protein particles exhibit non-Gaussian subdiffusive behavior. Biophys J 112(3):532–542 8. Karagiannis T, Molle M, Faloutsos M (2004) Long-range dependence: ten years of internet traffic modeling. IEEE Internet Comput 8:57–64 9. Clegg RG, Cairano-Gilfedder CD, Zhou S (2010) A critical look at power law modelling of the internet. Comput Commun 33:259–268 10. Goh KI, Kahng B, Kim D (2001) Universal behavior of load distribution in scale-free networks. Phys Rev Lett 87(27):278701 11. Jaynes ET (1982) On the rationale of maximum-entropy methods. Proc IEEE 70:939–952 12. Presse S, Ghosh K, Lee J, Dill KA (2013) Principles of maximum entropy and maximum caliber in statistical physics. Rev Modern Phy 85:1115–1141 13. Visser M (2013) Zipf’s law, power laws and maximum entropy. New J Phys 15:040321 14. Shannon CE (1948) A mathematical theory of communication. Bell Syst Tech J 27(3):379–423 15. Sharma S (2006) Queue length distribution of network packet traffic: Tsallis entropy maximization with fractional moments. IEEE Commun Lett 10(1):34 16. Koverda VP, Skokov VN (2012) Maximum entropy in a nonlinear system with a 1/f power spectrum. Physica A 391:21–28 17. Cover TM, Thomas JA (2012) Elements of information theory. Wiley, New York 18. Rajput NK, Ahuja B, Riyal MK (2018) A novel approach towards deriving vocabulary quotient. Digital Scholarship Human 33(4):894–901 19. Gell-Mann M, Tsallis C (eds) (2004) Nonextensive entropy: interdisciplinary applications. Oxford University Press on Demand, Oxford 20. Singh AK (2014) Power law behavior of queue size: maximum entropy principle with shifted geometric mean constraint. IEEE Commun Lett 18(8):1335–1338 21. Singh AK, Singh HP (2015) Karmeshu,: Analysis of finite buffer queue: maximum entropy probability distribution with shifted fractional geometric and arithmetic means. IEEE Commun Lett 19(2):163–166 22. Singh AK (2015) Performance modeling of communication networks: power law behavior in stationary and non-stationary environments, thesis, JNU, New Delhi 23. Shortle JF, Thompson JM, Gross D, Harris CM (2014) Fundamentals of queueing theory. Wiley, New York 24. Mukherjee T, Singh AK, Senapati D (2019) Performance evaluation of wireless communication systems over Weibull/q-lognormal shadowed fading using Tsallis’ entropy framework. Wirel Pers Commun 106(2):789–803

Identifying Challenges in the Adoption of Industry 4.0 in the Indian Construction Industry Arpit Singh and Subhas Chandra Misra(B) Department of Industrial and Management Engineering, IIT Kanpur, Kanpur, India {arpits,subhasm}@iitk.ac.in

Abstract. Industry 4.0 holds tremendous potential to transform the operational productivity of industries. Construction sector of India has fallen behind to embrace Industry 4.0 framework. Delay in project completion and lack of coordination within departments due to unavailability of real-time information hampers the effectiveness of the operations on a daily basis. The investigation of impediments in the adoption of Industry 4.0 in the construction industry of India is an urgent requirement to restore the efficiency of the sector. Based on the extant literature review and discussions with the experts, 25 key challenges were identified. Using multi-criteria decision-making (MCDM) tool, fuzzy TOPSIS, which operates with uncertain and vague inputs, the ranking of the challenges was established. Huge costs incurred in the implementation and maintenance emerged as the biggest obstacle followed closely by problems in hiring skilled people with the required expertise. Heavy lay-offs, disruptions in compensation and legal barriers are some other serious issues that hinder adoption of Industry 4.0. Through this paper, key obstacles in the adoption of digital technology are expected to surface up that can inform management and assist in the timely decision making. Keywords: Industry 4.0 · Challenges in adoption · Industry 4.0 in construction · Ranking

1 Introduction Almost every sphere of industry ranging from manufacturing, banking and automobiles has embarked on the digital era by integrating their operations and activities with digital technology. The adoption of highly sophisticated technology in everyday processes provides the competitive edge that arms the firms to stay in the competition and flourish. This aids in retaining and accelerating the growth, productivity, efficiency and improved customer satisfaction. Social media platforms such as Facebook, Netflix and Gaana.com have transformed the face of entertainment. Brick and mortar retailers are replaced substantially with giants such as Amazon and Flipkart. Various traditional automaker companies saw a decline with the introduction of automated digitally operated mobility companies. The technologies ushered as a result of industrial revolution, © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_37

Identifying Challenges in the Adoption of Industry 4.0

381

improved the technical competency and sustainability of the firms, thereby contributing immensely in increasing the productivity. Construction sector of India, however, has remained fairly untouched with the advancements in the technology. Construction sector’s reluctance to embrace technology has stagnated the operational processes for the last three to four decades [1]. It has continued to operate with the manual labor, mechanical technology and pre-existing business models. As a result of which, productivity has seen a remarkable decline [2]. It becomes a matter of concern, especially because construction is one of the prominent players in the economic development of India as it opens up investment opportunities across related sectors. It is a cornerstone of the world’s economy. Being a highly laborintensive industry, it provides employment to over 35 million people in India as per the recent statistics [2]. The primary cause of non-adoption of technology in the construction realm is the resistance due to lack of awareness and the reluctance to come out of one’s comfort zone and delve in experimentation with new processes [2]. With the advent of Industry 4.0, substantial change is anticipated in the construction sector that will be driven by development and deployment of digital technologies and processes. Industry 4.0 refers to the fourth industrial revolution that integrates smart technology, interconnectivity, automation in real-time operations and processes. It is also referred to as the industrial Internet of things (IIoT) that combines physical operations and processes with the smart technology and machine learning that increases connectivity in the ecosystem for companies involved in manufacturing and supply chain management [4]. Adoption of digital technology will enable companies to boost productivity, manage complexities and increase the operational efficiency [4]. As per Boston consulting group, digitization of processes leads to saving from $0.7 trillion to $1.2 trillion (13–21%) in the engineering and construction departments and $0.3 trillion to $0.5 trillion (10–17%) in the operations realm [3]. Thus, adopting Industry 4.0 framework seems to be the immediate need for the construction sector of India to get quality as well as financial boost that paves the way for increased customer satisfaction and overall productivity. Despite numerous potential benefits expected from the adoption of Industry 4.0, there is a dearth of literature that focuses on addressing the issues faced in the implementation of the framework in the construction sector of developing countries such as India. In the past, there has been appreciable amount of study conducted on the assessment of challenges in the manufacturing sector of countries like China, Germany and India [7–12]. But the literature is somewhat scanty for the identification of challenges in the construction sector. This study attempts to address this gap by conducting an empirical analysis to gather the major impediments in the implementation of Industry 4.0 in the Indian construction sector and assess the severity level of each challenge with the intention of establishing a working structure with Industry 4.0 framework. The paper consists of seven sections. Section 1 speaks about the introduction followed by a brief literature review in Sect. 2. Methodology is explained in Sect. 3 with the sensitivity analysis, results and conclusions in Sects. 4 and 5, respectively. Finally, paper closes by citing the limitations in Sect. 6 and references.

382

A. Singh and S. C. Misra

2 Literature Review 2.1 Industry 4.0 Originated from a high-tech strategy project in Germany, “Industrie 4.0” or simply “Industry 4.0” loosely refers the onset of fourth industrial revolution that has particularly occurred in manufacturing. It marks the transition from the first industrial revolution where water and steam power was used for mechanization, and the third which resorted to electricity for mass production and assembly lines to the era where smart technology coupled with data and machine learning leads to autonomous system where decision making is done without human intervention. Industry 4.0 or fourth industrial revolution signifies a theoretical concept of a factory where machines are augmented with wireless connectivity and automated devices with sensors that are connected to cyber physical systems (CPS). CPS replicates the real-time factory setting or the physical system and makes decentralized decisions by interacting and communicating with each other and humans over the Internet of things (IoT). Industry 4.0 has been studied and applied extensively in the manufacturing sector. The concept of intelligent manufacturing that means using equipment that automatically senses the situation and act in the smart environment was studied in the context of various nations such as European Union, China and Japan, especially from the perspective of Government’s strategic planning [8]. In order to advance the knowledge of Industry 4.0 and its applications, a detailed conceptual framework for smart manufacturing systems for Industry 4.0 was presented that demonstrated various scenarios pertaining to smart design, smart machining, smart control and smart scheduling [9]. The inclusion of a digital platform such as Industry 4.0 was critically analyzed to determine if it is a disruptive technology or a natural development of industrial production systems in Germany context. The readiness of German firms to embrace this framework was assessed by examining the drivers, goals and limitations of Industry 4.0 [10]. It is clearly observed that the manufacturing sector has tapped on the new industrial revolution by adopting Industry 4.0 in their operations and other business processes. The much needed boost to the diminishing productivity of Indian construction sector can only be made with the inclusion of Industry 4.0 framework. 2.2 Indian Construction Sector and Industry 4.0 Construction sector of India is one of the highest contributors toward the GDP. As per recent statistics, GDP of India from the construction sector stands at approximately $288 million which is higher than the GDP contribution from the automotive manufacturing sector [11]. During the past 50 years, construction has contributed about 40% of the development investment employing about 35 million people across the country in the year 2019. More than 16% of the population of India depends on construction sector for the livelihood [12]. Nearly 78% of gross capital formation is attributed solely to the sector [13]. The pivotal role played by the sector in the economic development of the nation is indicative of the fact that any inefficiency in the production and operation can sabotage the productivity which in turn harms the economy and well-being of the country. Some of

Identifying Challenges in the Adoption of Industry 4.0

383

the factors contributing in the low efficiency of the industry are hazardous sites, project complexity, economic disruption, lack of transparency and corruption [14]. Industry 4.0 framework presents a decent solution to tackle the problems affecting the efficiency of the sector by automating the production line, thereby increasing visibility across domains and real-time interaction between the digital environment and humans. The need to digitize the construction sector is demonstrated through innumerable efforts taken to study the concept of Industry 4.0 in the construction domain. The enthusiasm toward a new technology and the benefits it entails was revealed in the study originally taken up to observe the challenges companies are poised with in the wake of fourth industrial revolution [15]. Inefficiency caused due to incompetency of the workforce, glitches in planning management and poor execution of tasks can all be addressed using the digital framework that operates automatically leveraging on machine learning algorithms and sensors that act as per the environment dictates. Despite numerous advantages and potential showcased by the adoption of Industry 4.0 framework, construction sector has not been able to accommodate the technology and falls considerably behind the manufacturing sector [16]. Therefore, the primary objective of this paper is to investigate major challenges in the implementation and adoption of Industry 4.0 and subsequently rank them on the basis of magnitude of seriousness in the construction sector of India.

3 Methodology The current work investigates the major obstacles or challenges in the implementation of Industry 4.0 framework in the Indian construction sector. Subsequently, we intend to prioritize the challenges on the basis of the degree of severity associated with them in the implementation process. The subject of the study is a small construction firm operating in a remote part of Northern India. Based on extensive literature review and personal interaction with the construction management, we identified some challenges that were further validated by experts from the field of information technology (IT) and academics. The experts were people from their respective domains that had more than 8–10 years of work experience in their fields and were involved in activities related to integration of IT in construction. Following the above procedure we were able to discover 25 major challenges in the adoption of Industry 4.0 framework as shown in Table 1. 3.1 Fuzzy TOPSIS Developed by Ching-lai Hwang and Yoon in 1981 and later developed by Hwang, Lai and Liu in 1993, Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) is a modeling tool in multi-criteria decision-making (MCDM) problem. Since it fails to quantify the linguistic variable, therefore, Bellman and Zadeh in 1970 integrated it with fuzzy set theory to deal with the decision-making problems involving natural language as the inputs [17]. With the usage of fuzzy numbers to quantify the linguistic inputs rather than crisp values, the uncertainty and vagueness of the inputs were quantified in a better and practical manner.

384

A. Singh and S. C. Misra Table 1. Challenges in the adoption of Industry 4.0 Challenges in Industry 4.0 adoption

1

Heavy lay-offs due to smart processes

2

Huge initial investment and maintenance costs

3

Need to prepare for a significant organizational and process change

4

Requirement to attain advanced technical skills

5

Need to establish a strong information collection, distribution, use and management mechanism

6

Lack of management support

7

Problem in hiring qualified professionals especially at the ground level

8

Reluctance and apprehensions to resort to change in technology

9

Need to establish research and development facilities

10 Educating higher management particularly experienced professionals about the technological change 11 Providing contractors and sub-contractors the necessary skills and understanding of the process 12 Recruitment of skilled personnel to impart necessary knowledge and training to the employees 13 Non-seriousness to adopt new concept in technology 14 Unclear about the economic benefits of IoT enabled framework 15 Getting a common consensus for the adoption of new technology from the employees and management 16 Unclear comprehensibility of the advantages of IoT 17 Disruptions in the compensation policies 18 Proper Internet connectivity and other IT facilities 19 Uncertain impact on working life 20 Safety issues arising due to manhandling of devices 21 Increased protection of sensitive devices from dust and pollutants present on-site 22 Need to establish a reliable and stable machine to machine communication network 23 Need to ensure proper monitoring, inspection and validation of services in the production of key assets 24 Lack of regulation, standards and certifications 25 Legal barriers

The current work proposes to assess the viewpoint from the three perspectives namely information technology (IT), academics (ACA) and construction management (CM) on the alternatives, i.e., challenges to estimate the willingness and preparedness of the firm to accept the digital framework based on Industry 4.0. There are, therefore, three

Identifying Challenges in the Adoption of Industry 4.0

385

perspectives that form the criteria and the 25 challenges that form the alternatives that are evaluated from the three perspectives. The alternative that shows the maximum potential in the decision-making process is ranked at the top from all the possible alternatives. The following steps summarize the procedure of ranking the challenges on the three criteria. The linguistic assignment of criteria weights and evaluation of the alternatives are accomplished using triangular fuzzy number. According to fuzzy set theory, a triangular fuzzy number FUZZY~num is represented by a triplet (a1 , a2 , a3 ) where the triplet forms the points on the x-axis that generates the fuzzy set for a given linguistic evaluation. The representation of the weights used for the perspectives and the evaluation of the alternatives in the form of fuzzy numbers, as gathered from the managers, are shown in Table 2. Table 2. Fuzzy representation of criteria weights and alternative evaluation Criteria weights

Membership function

Alternative rating

Membership function

Very low (VL)

(1, 1, 3)

Least important (LI)

(1, 1, 3)

Low (L)

(1, 3, 5)

Somewhat important (SI)

(1, 3, 5)

Medium (M)

(3, 5, 7)

Neutral (N)

(3, 5, 7)

High (H)

(5, 7, 9)

Very important (VI)

(5, 7, 9)

Very high (VH)

(7, 9, 9)

Most important (MI)

(7, 9, 9)

The membership function of an element x is given as ⎧ x−a1 ⎪ ⎨ a2 −a1 , if a 3 −x ∼ = µFnum a3 −a2 , if ⎪ ⎩ 0,

µ~FUZZYnum (x),

in the fuzzy set FUZZY~num , written as a1 ≤ x ≤ a2 a2 ≤ x ≤ a3 otherwise

The following steps illustrate the rankings of the alternatives based on the three perspectives Step 1: The decision makers assign the ratings to the criteria (perspectives) and evaluate the alternatives. In total, 9 decision makers were taken which are assumed to be representative of the three perspectives. The evaluations made by the decision makers on the criteria are shown in Table 3. Step 2: Calculation of aggregate fuzzy weights for the criteria (perspectives): The criteria weights assigned by the decision makers are combined together to give an aggregate fuzzy rating which is also a fuzzy number as shown in Table 4. The fuzzy rating for the “r” decision makers is given by the following fuzzy number Vr = {xr , yr , zr } where x = min{xr }, y = 1/r r



yr and z = max{zr } r

386

A. Singh and S. C. Misra Table 3. Assignment of fuzzy weights to the criteria Criteria IT1 IT2 IT3 ACA1 ACA2 ACA3 CM1 CM2 CM3 PER-1

VH H

PER-2

H

PER-3

H

VH VH

H

VH

VH

VH

VH

VH VH H

H

VH

VH

M

H

M

M

H

L

L

H

L

L

Table 4. Aggregate fuzzy weights of perspectives Criteria IT1

IT2

IT3

ACA1

ACA2

ACA3

CM1 CM2 CM3 Aggregate fuzzy weight

PER-1

(7, (5, (7, (7, 9, 9) (5, 7, 9) (7, 9, 9) (7, 9, (7, 9, (7, 9, (5, 8.56, 9) 9, 9) 7, 9) 9, 9) 9) 9) 9)

PER-2

(5, (7, (7, (5, 7, 9) (5, 7, 9) (7, 9, 9) (7, 9, (3, 5, (5, 7, (3, 7.67,9) 7, 9) 9, 9) 9, 9) 9) 7) 9)

PER-3

(5, (3, (1, (1, 3, 5) (3, 5, 7) (5, 7, 9) (1, 3, (1, 3, (5, 7, (1, 4.78, 9) 7, 9) 5, 7) 3, 5) 5) 5) 9)

The fuzzy decision matrix for criteria weight is given by Ti = {t1 , t2 , . . . , tn }, i = 1, 2, 3, . . . , n Step 3: Evaluation of alternatives: Fuzzy decision matrix Drawing on the linguistic scale for the evaluation of the alternatives from Table 2, the fuzzy decision matrix is generated by converting the linguistic inputs from Table 3. Step 4: Computation of the aggregate fuzzy ratings for each perspective Alternative rated by the rth decision maker as V ijr = {x ijr , yijr , zijr }, then the aggregate fuzzy rating for a given criterion is given as   Vij = xij , yij , zij , where      xij = min xijr , yij = 1/r yijr and zij = max zijr r

r

as shown in Table 5. Step 5: Normalization of the fuzzy decision matrix The entries in Table 5 are normalized rendering the criteria and alternatives on the comparable scale. The normalization is made using the following   Vm = vij , where i = 1, 2, 3, . . . l and j = 1, 2, 3, . . . t

  where vij = xij /gj∗ , yij /gj∗ , zij /gj∗ and gj∗ = max gij . . . (benefit criteria) i

Identifying Challenges in the Adoption of Industry 4.0

387

Table 5. Aggregate fuzzy ratings for each perspective Challenges Aggregated scores IT

ACA

CM

C1

(5, 7.67, 9)

(5, 8.33, 9)

(3, 6.33, 9)

C2

(5, 7.67, 9)

(5, 7.67, 9)

(5, 8.33, 9)

C3

(5, 8.33, 9)

(3, 7.67, 9)

(3, 5.67, 9)

C4

(3, 7.67, 9)

(5, 7.67, 9)

(5,7.67, 9)

C5

(5, 5.67, 9)

(3, 5.67, 9)

(1, 3.67, 7)

C6

(1, 5.67, 9)

(3, 7.67, 9)

(1, 2.33, 5)

C7

(5, 7.67, 9)

(5, 7.67, 9)

(5, 8.33, 9)

C8

(1, 6.33, 9)

(5, 8.33, 9)

(3, 6.33, 9)

C9

(3, 7.67, 9)

(3, 5.67, 9)

(1, 4.33, 7)

C10

(5, 7.67, 9)

(1, 5.67, 9) (1, 3, 7)

C11

(3, 7.67, 9)

(1, 4.33, 7)

(1, 3.67, 7)

C12

(5, 7.67, 9)

(1, 4.33, 9)

(3, 5.67, 9)

C13

(1, 5.67, 9)

(1, 5.67, 9) (1, 3, 5)

C14

(5, 7, 9)

(3, 6.33, 9)

C15

(5, 7.67, 9)

C16

(3, 6.33, 9) (5, 7, 9)

C17

(3, 7, 9)

(3, 5.67, 9)

(5, 7.67, 9)

(5, 8.33, 9)

(1, 3.67, 7)

(3, 7.67, 9)

C19

(3, 6.33, 9) (1, 3, 5)

C20

(5, 7.67, 9)

C22 C23 C24 C25

(5, 7.67, 9) (3, 7, 9) (5, 8.33, 9) (5, 7.67, 9) (1, 5, 9)

(5, 7.67, 9) (1, 5, 9)

(5, 8.33, 9)

C18

C21

(1, 3.67, 7)

(1, 1.67, 5)

(5, 7.67, 9) (5, 7, 9) (5, 8.33, 9) (1, 5, 9)

(5, 8.33, 9) (5, 8.33, 9)

(1, 4.33, 7)

(5, 7.67, 9)

(5, 7.67, 9)

(5, 7.67, 9)

(7, 9, 9)

(5, 8.33, 9)

The resultant matrix is shown in Table 6. Step 6: Computation of weighted normalized matrix The normalized matrix obtained in step 5 {vij } is multiplied with the aggregated weights for each perspective from step 2 to result in the normalized weighted matrix as shown in Table 7.   Q = qij , where qij = vij (.) ti

388

A. Singh and S. C. Misra Table 6. Normalized fuzzy ratings Normalized fuzzy weights of alternatives IT

ACA

CM

C1

(0.56, 0.85, 1) (0.56, 0.93, 1)

(0.33, 0.70, 1)

C2

(0.56, 0.85, 1) (0.56, 0.85, 1)

(0.56, 0.93, 1)

C3

(0.56, 0.93, 1) (0.33, 0.85, 1)

(0.33, 0.63, 1)

C4

(0.33, 0.85, 1) (0.56, 0.85, 1)

(0.56, 0.85, 1)

C5

(0.56, 0.63, 1) (0.33, 0.63, 1)

(0.11, 0.41, 0.78)

C6

(0.11, 0.63, 1) (0.33, 0.85, 1)

(0.11, 0.26, 0.56)

C7

(0.56, 0.85, 1) (0.56, 0.85, 1)

(0.56, 0.93, 1)

C8

(0.11, 0.70, 1) (0.56, 0.93, 1)

(0.33, 0.70, 1)

C9

(0.33, 0.85, 1) (0.33, 0.63, 1)

(0.11, 0.48, 0.78)

C10 (0.56, 0.85, 1) (0.11, 0.63, 1)

(0.11, 0.33, 0.78)

C11 (0.33, 0.85, 1) (0.11, 0.48, 0.78) (0.11, 0.41, 0.78) C12 (0.56, 0.85, 1) (0.11, 0.48, 1)

(0.33, 0.63, 1)

C13 (0.11, 0.63, 1) (0.11, 0.63, 1)

(0.11, 0.33, 0.56)

C14 (0.56, 0.78, 1) (0.33, 0.70, 1)

(0.11, 0.41, 0.78)

C15 (0.56, 0.85, 1) (0.33, 0.63, 1)

(0.56, 0.85, 1)

C16 (0.33, 0.70, 1) (0.56, 0.78, 1)

(0.11, 0.56, 1)

C17 (0.33, 0.78, 1) (0.56, 0.93, 1)

(0.56, 0.85, 1)

C18 (0.33, 0.85, 1) (0.56, 0.93, 1)

(0.11, 0.41, 0.78)

C19 (0.33, 0.70, 1) (0.11, 0.33, 0.56) (0.11, 0.19, 0.56) C20 (0.56, 0.85, 1) (0.56, 0.85, 1)

(0.56, 0.78, 1)

C21 (0.56, 0.85, 1) (0.56, 0.93, 1)

(0.56, 0.93, 1)

C22 (0.33, 0.78, 1) (0.11, 0.56, 1)

(0.56, 0.93, 1)

C23 (0.56, 0.93, 1) (0.11, 0.48, 0.78) (0.56, 0.85, 1) C24 (0.56, 0.85, 1) (0.56, 0.85, 1)

(0.56, 0.85, 1)

C25 (0.11, 0.56, 1) (0.78, 1, 1)

(0.56, 0.93, 1)

Step 7: Calculation of fuzzy positive ideal solution (FPoS) and fuzzy negative ideal solution (FNoS) It is a crucial step in that it determines the reference point to decide the quality of the solution. TOPSIS suggests that a particular alternative should have minimum geometric distance from the FPIS and maximum geometric distance from FNIS. FPoS and FNoS for the given criterion are calculated as shown below     O∗ = z1∗ , z2∗ , z3∗ , . . . , zn∗ and O− = z1− , z2− , z3− , . . . , zn− ,     Where Zj∗ = max zij3 and Zj− = min zij3 i

i

Identifying Challenges in the Adoption of Industry 4.0

389

i = 1, 2, 3, . . . l and j = 1, 2, 3, . . . t Step 8: Calculation of distance of each alternative from the ideal solution The distance of each alternative from PoS and NoS is calculated using the vertex method as shown below ¯ c¯ ) = 1/3[(a1 − b1 )2 + (a2 − b2 )2 + (a3 − b3 )2 d (¯a, b, Table 7. Normalized weighted matrix of alternatives Weighted normalized weights of alternatives IT

ACA

CM

C1

(2.8, 7.28, 9)

(2.8, 7.96, 9)

(1.65, 5.99, 9)

C2

(2.8, 7.28, 9)

(2.8, 7.28, 9)

(2.8, 7.96, 9)

C3

(2.8, 7.96, 9)

(1.65, 7.28, 9) (1.65, 5.39, 9)

C4

(1.65, 7.28, 9)

(2.8, 7.28, 9)

C5

(2.8, 5.39, 9)

(1.65, 5.39, 9) (0.55, 3.51, 7.02)

C6

(0.55, 5.39, 9)

(1.65, 7.28, 9) (0.55, 2.23, 5.04)

C7

(2.8, 7.28, 9)

(2.8, 7.28, 9)

(2.8, 7.96, 9)

C8

(0.55, 5.99, 9)

(2.8, 7.96, 9)

(1.65, 5.99, 9)

C9

(1.65, 7.28, 9)

(1.65, 5.39, 9) (0.55, 4.11, 7.02)

C10

(2.8, 7.28, 9)

(0.55, 5.39, 9) (0.55, 2.82, 7.02)

C11

(1.65, 7.28, 9)

(0.55, 4.11, 7.02)

C12

(2.8, 7.28, 9)

(0.55, 4.11, 9) (1.65, 5.39, 9)

C13

(0.55, 5.39, 9)

(0.55, 5.39, 9) (0.55, 2.82, 5.04)

C14

(2.8, 6.68, 9)

(1.65, 5.99, 9) (0.55, 3.51, 7.02)

C15

(2.8, 7.28, 9)

(1.65, 5.39, 9) (2.8, 7.28, 9)

C16

(1.65, 5.99, 9)

(2.8, 6.68, 9)

(0.55, 4.79, 9)

C17

(1.65, 6.68, 9)

(2.8, 7.96, 9)

(2.8, 7.28, 9)

C18

(1.65, 7.28, 9)

(2.8, 7.96, 9)

(0.55, 3.51, 7.02)

C19

(1.65, 5.99, 9)

(0.55, 2.82, 5.04)

(0.55, 1.63, 5.04)

C20

(2.8, 7.28, 9)

(2.8, 7.28, 9)

(2.8, 6.68, 9)

(2.8, 7.28, 9)

(0.55, 3.51,7.02)

(continued)

390

A. Singh and S. C. Misra Table 7. (continued) Weighted normalized weights of alternatives IT

ACA

CM (2.8, 7.96, 9)

C21

(2.8, 7.28, 9)

(2.8, 7.96, 9)

C22

(1.65, 6.68, 9)

(0.55, 4.79, 9) (2.8, 7.96, 9)

C23

(2.8, 7.96, 9)

(0.55, 4.11, 7.02)

(2.8, 7.28, 9)

C24

(2.8, 7.28, 9)

(2.8, 7.28, 9)

(2.8, 7.28, 9)

C25

(0.55, 4.79, 9)

(3.9, 8.56, 9)

(2.8, 7.96, 9)

The calculated distance is shown in Table 8. Table 9 presents the results for the computation of the closeness coefficient of each alternative corresponding to the ideal solution. Closeness coefficient (ClCo*) is a metric that measures the relative distance of an alternative from the ideal solution. It is calculated as shown below CCo∗ = di+ /(di+ + di− )

  where di+ = j=1...n d zij , zj∗ and di− = j=1...n d zij , zj− Step 9: Ranking the alternatives Table 10 lists the final ranking of the alternatives based on the closeness coefficient scores. The alternative ranked at the top has maximum geometric distance from the negative ideal solution and minimum distance from the positive ideal solution. 3.2 Sensitivity Analysis Sensitivity analysis is the process by which the change in the ranking of the alternatives is assessed. The criteria weights are altered and the effect on the final ranking of the alternatives is observed (Awasthi et al. 2011). Since the weights assigned to the criteria or perspectives were done subjectively by the decision makers, it is considered to be in the domain of uncertainty. Therefore, the weights can be reassigned in a different way to see the effects on the alternatives evaluation. The analysis was carried out in eight different phases where the weights given to each criteria were changed every time. In the first five phases, the weights assigned to the criteria were (1, 1, 3), (1, 3, 5), (3, 5, 7), (5, 7, 9) and (7, 9, 9), respectively. In the later phases, different combinations of the “most important” (7, 9, 9) and “least important” (1, 1, 3) were used with all the criteria. The results are shown in Table 11. The results of sensitivity analysis reveal that there is no significant change in the rankings of the alternatives for all the phases except 7. There is a slight change in the ranking for the phase 7 due to the criteria weight. This validates the robustness of the ranking obtained using the fuzzy TOPSIS method.

Identifying Challenges in the Adoption of Industry 4.0

391

Table 8. Distance of alternatives from PIS and NIS Distance from PoS (d*) IT +

Distance from NoS (d − )

ACA+ CM+ IT-

ACA- CM-

C1

d1

3.71 3.63

4.59

6.37 6.62

5.84

C2

d2

3.71 3.71

3.63

6.37 6.37

6.62

C3

d3

3.63 3.63

4.73

6.62 6.27

5.66

C4

d4

3.63 3.71

3.71

6.27 6.37

6.37

C5

d5

4.14 4.73

5.93

5.77 5.66

4.11

C6

d6

5.31 3.63

6.66

5.62 6.27

2.77

C7

d7

3.71 3.71

3.63

6.37 6.37

6.62

C8

d8

5.18 3.63

4.59

5.8

6.62

5.84

C9

d9

3.63 4.73

5.75

6.27 5.66

4.26

C10 d10 3.71 5.31

6.46

6.37 5.77

2.9

C11 d11 3.63 5.75

5.93

4.26 4.26

4.11

C12 d12 3.71 5.64

4.73

5.29 5.29

5.66

C13 d13 5.31 5.31

6.46

5.62 5.62

2.9

C14 d14 3.82 4.59

5.93

6.17 5.84

4.11

C15 d15 3.71 4.73

3.71

6.37 5.66

6.37

C16 d16 4.59 3.82

5.45

5.84 6.17

5.46

C17 d17 4.45 3.63

3.71

6.06 6.62

6.37

C18 d18 3.63 3.63

5.93

4.26 6.62

4.11

C19 d19 4.59 6.46

6.87

5.84 2.9

2.67

C20 d20 3.71 3.71

3.82

6.37 6.37

6.17

C21 d21 3.71 3.63

3.63

6.37 6.37

6.37

C22 d22 4.45 5.45

3.63

6.06 3.57

6.37

C23 d23 3.63 5.75

3.71

6.62 4.26

6.37

C24 d24 3.71 3.71

3.71

6.37 6.37

6.37

C25 d25 5.45 2.96

3.63

5.46 6.99

6.37

4 Results The final ranking of the alternatives as shown in Table 10 indicates that the indicator “Huge initial investment and maintenance costs” is ranked first with the highest closeness coefficient of 0.637. It was followed closely by challenges like “problem in hiring qualified professionals especially at the ground level,” “increased protection of sensitive devices from dust and pollutants present on-site,” “requirement to attain advanced

392

A. Singh and S. C. Misra Table 9. Aggregate closeness coefficient for alternatives di+

di−

CCo*

C1

11.93 18.83 0.61

C2

11.05 19.36 0.64

C3

11.99 18.55 0.61

C4

11.05 19.01 0.63

C5

14.8

15.54 0.51

C6

15.6

14.66 0.48

C7

11.05 19.36 0.64

C8

13.4

C9

14.11 16.19 0.53

18.26 0.58

C10 15.48 15.04 0.49 C11 15.31 12.63 0.45 C12 14.08 16.24 0.54 C13 17.08 14.14 0.45 C14 14.34 16.12 0.53 C15 12.15 18.4

0.6

C16 13.86 17.47 0.56 C17 11.79 19.05 0.62 C18 13.19 14.99 0.53 C19 17.92 11.41 0.39 C20 11.24 18.91 0.63 C21 10.97 19.11 0.64 C22 13.53 16

0.54

C23 13.09 17.25 0.57 C24 11.13 19.11 0.63 C25 12.04 18.82 0.61

technical skills,” “lack of regulation, standards and certifications,” “safety issues arising due to manhandling of devices,” “disruptions in the compensation policies,” “heavy lay-offs due to smart processes,” “legal barriers,” “need to prepare for a significant organizational and process change,” “getting a common consensus for the adoption of new technology from the employees and management,” “reluctance and apprehensions to resort to change in technology,” “need to ensure proper monitoring, inspection and validation of services in the production of key assets,” “unclear comprehensibility of the advantages of IoT,” “need to establish a reliable and stable machine to machine communication network,” “recruitment of skilled personnel to impart necessary knowledge and training to the employees,” “need to establish research and development facilities,”

Identifying Challenges in the Adoption of Industry 4.0

393

Table 10. Final ranking of the alternatives based on closeness coefficient Challenges in Industry 4.0 adoption

CCo* Rank

C2

Huge initial investment and maintenance costs

0.637

1

C7

Problem in hiring qualified professionals especially at the ground level

0.637

2

0.635

3

C21 Increased protection of sensitive devices from dust and pollutants present on-site C4

Requirement to attain advanced technical skills

0.632

4

C24 Lack of regulation, standards and certifications

0.632

5

C20 Safety issues arising due to manhandling of devices

0.627

6

C17 Disruptions in the compensation policies

0.618

7

C1

0.612

8

C25 Legal barriers

0.61

9

C3

0.607 10

Heavy lay-offs due to smart processes Need to prepare for a significant organizational and process change

C15 Getting a common consensus for the adoption of new technology from the 0.602 11 employees and management C8

Reluctance and apprehensions to resort to change in technology

0.577 12

C23 Need to ensure proper monitoring, inspection and validation of services in 0.569 13 the production of key assets C16 Unclear comprehensibility of the advantages of IoT

0.558 14

C22 Need to establish a reliable and stable machine to machine communication network

0.542 15

C12 Recruitment of skilled personnel to impart necessary knowledge and training to the employees

0.536 16

C9

0.534 17

Need to establish research and development facilities

C18 Proper Internet connectivity and other IT facilities

0.532 18

C14 Unclear about the economic benefits of IoT enabled framework

0.529 19

C5

0.512 20

Need to establish a strong information collection, distribution, use and management mechanism

C10 Educating higher management particularly experienced professionals about the technological change C6

Lack of management support

0.493 21 0.484 22

C13 Non-seriousness to adopt new concept in technology

0.453 23

C11 Providing contractors and sub-contractors the necessary skills and understanding of the process

0.452 24

C19 Uncertain impact on working life

0.389 25

“proper Internet connectivity and other IT facilities,” “unclear about the economic benefits of IoT enabled framework,” “need to establish a strong information collection,

Criteria weights

all: (1, 1, 3)

all: (1, 3, 5)

all: (3, 5, 7)

all: (5, 7, 9)

all: (7, 9, 9)

P1: (7, 9, 9) and P2, P3 = (1, 1, 3)

P2: (7, 9, 9) and P1, P3 = (1, 1, 3)

P3: (7, 9, 9) and P1, P2 = (1, 1, 3)

Phase No.

1

2

3

4

5

6

7

8

C21 > C2 > C7 > C25 > C24 > C17 > C4 > C15 > C20 > C23 > C1 > C8 > C12 > C3 > C16 > C22 > C9 > C18 > C5 > C14 > C10 > C11 > C6 > C13 > C19

C25 > C21 > C17 > C1 > C2 > C7 > C24 > C20 > C4 > C8 > C12 > C18 > C16 > C3 > C15 > C14 > C6 > C5 > C9 > C22 > C10 > C23 > C11 > C19 > C13

C21 > C2 > C7 > C24 > C20 > C3 > C1 > C15 > C12 > C23 > C5 > C4 > C10 > C14 > C17 > C18 > C9 > C22 > C16 > C11 > C8 > C25 > C6 > C19 > C13

C21 > C2 > C7 > C24 > C20 > C4 > C17 > C25 > C1 > C15 > C12 > C3 > C8 > C23 > C16 > C18 > C14 > C5 > C9 > C22 > C10 > C6 > C11 > C19 > C13

C21 > C2 > C7 > C24 > C20 > C17 > C4 > C25 > C1 > C15 > C12 > C3 > C8 > C23 > C16 > C18 > C5 > C14 > C22 > C9 > C10 > C6 > C11 > C19 > C13

C21 > C2 > C7 > C24 > C20 > C17 > C4 > C25 > C1 > C15 > C12 > C3 > C8 > C23 > C16 > C18 > C22 > C5 > C14 > C9 > C10 > C6 > C11 > C19 > C13

C21 > C2 > C7 > C24 > C20 > C17 > C4 > C25 > C1 > C15 > C3 > C12 > C8 > C16 > C23 > C22 > C18 > C5 > C9 > C14 > C10 > C6 > C11 > C19 > C13

C21 > C2 > C7 > C24 > C20 > C17 > C4 > C25 > C1 > C15 > C12 > C3 > C8 > C16 > C23 > C22 > C18 > C5 > C14 > C9 > C10 > C11 > C6 > C19 > C13

Ranked alternatives

Table 11. Sensitivity analysis

394 A. Singh and S. C. Misra

Identifying Challenges in the Adoption of Industry 4.0

395

distribution, use and management mechanism,” “educating higher management particularly experienced professionals about the technological change,” “lack of management support,” “non-seriousness to adopt new concept in technology,” “providing contractors and sub-contractors the necessary skills and understanding of the process” and “uncertain impact on working life.” The ranking of the alternatives (indicators) with the variations in the criteria weights did not vary significantly except for the scenario 7 where the “most important” weight was assigned to the academics.

5 Conclusion and Discussions The paper investigates the level of acceptability and readiness of the construction firms to adopt Industry 4.0 framework in the operations and business processes. A detailed literature review and in-depth interaction with the experts facilitated in listing out the major obstacles (challenges) in the implementation of the digital platform. Opinions gathered from professionals and experts active in the disciplines of IT, academics and construction industry were sought to assess the severity assessment of each challenge and decipher insights about the future course of action to be undertaken in order to step in the realm of fourth industrial revolution. Clearly, the major hindrance in the successful implementation or adoption of the industrial Internet of things (IIoT) is the huge initial investment and maintenance cost. Particularly, in a set-up involving developing nation with limited resources, the high cost incurred during the establishment of any technical framework is perceived to be biggest impediment. Proper planning and analysis of the financial condition of the firm are required in order to chalk out an effective method for the aforementioned issue. Probably, instead of investing a huge amount right at the beginning of the installation, a part can be used which can be subsequently used in the later stages of the implementation process. The second and perhaps the insurmountable challenge is the hiring of technically skilled workers at the ground level. It is a well-known fact that most of the workers employed at the ground level in Indian construction firms are daily wage workers with minimal qualifications. It becomes a daunting task to educate the people at this level about the new technology that entails deep technical knowledge. The best way to deal with the problem is to hire skilled people with the technical expertise to impart necessary training to the ground-level workers. The training can be mostly visual-based that needs little to no reading material which can be helpful for people with less educational qualifications. Industrial revolution is still in a very nascent stage in a country such as India. There is a lot of apprehensions and confusion regarding the benefits of the proposed framework and fear of losing jobs and disruptions in the compensations. The inadequate knowledge of the potential benefits offered by adopting Industry 4.0 in construction should be increased by training sessions organized within the firms that include consultants and expert from the field. Reluctance to resort to a new method of operation can inhibit adoption on a very large scale. Since this issue is aligned more toward the psychology of humans, it can be best handled by professionals that can deal with the understanding of consumer behavior.

396

A. Singh and S. C. Misra

Replacement of workforce by better and sophisticated machines in operations and other processes will surely lead to job losses, but the benefits and profits to the firms can outweigh the loss significantly. Since the productivity rises manifold, leading to high output with less resources. Another important factor that was revealed as a prominent challenge is the improper Internet connectivity in the remote parts of India. This is, in fact, a problem which can be best handled by the authorities. Sufficient support from the government in the form of legislation, laws and acts favoring and encouraging the adoption of digital platform in the construction holds immense potential in upgrading the efficiency of the sector. The objective of the paper was to establish a comprehensive framework for estimating the preparedness of the management of the construction industry to embark on the digital phase by adopting Industry 4.0. The framework proposed in this paper should be used to increase the possibility of accepting a new technology by addressing the challenges in a timely and orderly manner. The challenges that are most severe should be dealt with on the priority basis. This would diminish the chances of other related factors or obstacles to intervene in the smooth implementation process. Authors foresee that there is ample room for additional challenges that might show up for a different set-up, which can be included in the framework conveniently and evaluated on the degree of severity. Digitization of construction is inevitable to continue to stay in competition. Constant vigil is of paramount importance in the business to gain an instant indication of anything going out of order. As competition escalates and profit plummets, the search for incremental productivity and gains will be the factors that will be the most sought for. Industry 4.0 is thus a need of the hour that should be included in the operations and activities to help raise the growth and productivity of the construction sector.

6 Limitations The major limitations of the study are listed below which can be considered as future scope of the research: • The generalizability of the results can be increased by including more decision makers in the study. Besides, the number of perspectives can also be increased from 3 which will be a better representation of the reality. • There can be number of other challenges depending upon the industrial set-up, people employed and host of other factors such as location of the plant and geographical factors that should be incorporated in the study to lend validity. • The current study deals with fuzzy TOPSIS to analyze the data which can be substituted with other decision-making tools that are supposedly considered to be advanced versions of dealing with uncertainty and vagueness.

Identifying Challenges in the Adoption of Industry 4.0

397

• Since the analysis relies heavily on the subjective inputs from the decision makers in the form of criteria weights, the results ultimately come under the purview of scrutiny as the weights can be altered according to the perception of the managers appropriate to the setting. • The technical challenges such as complexity of the programs required to run the equipment, etc., are not considered in the study which if included might improve the appropriateness and validity of the results even more.

Acknowledgements. We thank our colleagues from Indian Institute of Technology, Kanpur, who provided insight and expertise that greatly assisted the research.

References 1. MEP (2019) Middle East. Accessed on July 15, 2019. https://www.mepmiddleeast.com/ events/conferences/72305-construction-sector-still-backward-in-tech-adoption-says-voltasuae-head 2. Trading Economics. Accessed on July 15, 2019. https://tradingeconomics.com/india/gdpfrom-construction 3. Soham M, Rajiv B (2013) Critical factors affecting labour productivity in construction projects: case study of south Gujarat region of India. Int J Eng Adv Technol 2(4):583–591 4. World Economic Forum. Accessed on July 15, 2019. https://www.weforum.org/agenda/2018/ 03/how-construction-industry-can-build-its-future/ 5. BCG. Accessed on July 15, 2019. https://www.bcg.com/en-ch/publications/2016/engineeredproducts-infrastructure-digital-transformative-power-building-information-modeling.aspx 6. EPICOR. Accessed on July 16, 2019. https://www.epicor.com/en-in/resource-center/articles/ what-is-industry-4-0/ 7. Schumacher A, Erol S, Sihn W (2016) A maturity model for assessing Industry 4.0 readiness and maturity of manufacturing enterprises. Procedia Cirp 52:161–166 8. Khan A, Turowski K (2016) A survey of current challenges in manufacturing industry and preparation for industry 4.0. In Proceedings of the first international scientific conference “intelligent information technologies for industry” (IITI’16, pp. 15–26. Springer, Cham 9. Stock T, Seliger G (2016) Opportunities of sustainable manufacturing in industry 4.0. Procedia Cirp 40:536–541 10. Li L (2018) China’s manufacturing locus in 2025: with a comparison of “Made-in-China 2025” and “Industry 4.0”. Tech Forecast Social Change 135:66–74 11. Sommer L (2015) Industrial revolution-industry 4.0: are German manufacturing SMEs the first victims of this revolution. J Industr Eng Manag 8(5):1512–1532 12. Iyer A (2018) Moving from Industry 2.0 to Industry 4.0: a case study from India on leapfrogging in smart manufacturing. Proced Manuf 21:663–670 13. Zhong RY, Xu X, Klotz E, Newman ST (2017) Intelligent manufacturing in the context of industry 4.0: a review. Engineering 3(5):616–630 14. Zheng P, Sang Z, Zhong RY, Liu Y, Liu C, Mubarok K, Xu X (2018) Smart manufacturing systems for Industry 4.0: conceptual framework, scenarios, and future perspectives. Front Mech Eng 13(2):137–150 15. Rojko A (2017) Industry 4.0: concept background and overview. Int J Inter Mobile Technol (IJIM) 11(5):77–90

398

A. Singh and S. C. Misra

16. BENCHMARK REPORT 2017—INDIA (PDF) World travel and tourism council. Retrieved 11 April 2018 17. Indo-Italian chamber of commerce. Accessed on July 18, 2019. http://www.indiaitaly.com/ IndoItalianSite/index.aspx? 18. Jha KN, Iyer KC (2006) Critical factors affecting quality performance in construction projects. Total Qual Manag Bus Excell 17(9):1155–1170 19. Wired. The construction industry needs a robot revolution. Accessed on 18 July 2019. https:// www.wired.com/story/the-construction-industry-needs-a-robot-revolution/ ´ 20. Slusarczyk B (2018) Industry 4.0: are we ready? Polish J Manag Stud 17 21. Industry 4.0: The new Industrial revolution. Accessed on 18 July 2019. https://bridgr.co/wpcontent/uploads/2017/06/bdc-etude-manufacturing-en.pdf 22. Bellman RE, Zadeh LA (1970) Decision-making in a fuzzy environment. Manag Sci 17(4):141 23. Hwang CL, Yoon K (1981) Methods for multiple attribute decision making. In: Multiple attribute decision making, pp 58–191. Springer, Berlin, Heidelberg 24. Awasthi A, Chauhan SS, Omrani H, Panahi A (2011) A hybrid approach based on SERVQUAL and fuzzy TOPSIS for evaluating transportation service quality. Comput Ind Eng 61(3):637– 646. https://doi.org/10.1016/j.cie.2011.04.019

Human Action Recognition Using STIP Evaluation Techniques H. S. Mohana and U. Mahanthesha(B) Department of Electronics and Instrumentation Engineering, Malnad College of Engineering, Hassan, Karnataka, India [email protected], [email protected]

Abstract. The activities of human can be classified into human actions, interactions, object–human interactions and group actions. The recognition of actions in the input video is very much useful in computer vision technology. This system gives application to develop a model that can detect and recognize the actions. The variety of HAR applications are surveillance environment systems, healthcare systems, military, patient monitoring system (PMS), etc., that involve interactions between electronic devices such as human–computer interfaces with persons. Initially, collecting the videos containing actions or interactions was performed by the humans. The given input videos were converted into number of frames, and then these frames were undergone preprocessing stage using by applying median filter. The noise of the given input frame is reduced by applying the median filter of the neighboring pixels. Through frames, desired features were extracted. The actions of the person which is recognised from the system is going to extract further. There are three spatial–temporal interest point (STIP) techniques such as Harris SPIT, Gabor SPIT and HOG SPIT used for feature extraction from video frames. SVM algorithm is applied for classifying the extracted feature. The action recognition is based on the colored label identified by classifier. The system performance is measured by calculating the classifier performance which is the accuracy, sensitivity and specificity. The accuracy represents the classifier reliability. The specificity and sensitivity represent how exactly the classifier categorizes its features to each correct category and how the classifier rejects the features that are not belonging to the particular correct category. Keywords: Action recognition · STIP · Harris filter · Gabor filter · Histogram of oriented gradients (HOG)

1 Introduction The recognition of human action from the input video is important for video indexing, retrieval, health care, sports visual systems and security surveillance purposes. Recognition of any type of actions in the video is real challenging to analyze visual architecture of a person. The perfect badge of all actions of person by the custom requires regular techniques. The action detection with recognition methods is very much useful in computer visual perception. © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_38

400

H. S. Mohana and U. Mahanthesha

1.1 Classifications of HAR Mainly, there are three types of person action recognition, and they are explained here (a) Single-User, Sensor-Based Action Recognition With machine learning and new data processing, we develop a variety range of human actions by consolidating the rising space detector networks using the technique sensor-based action recognition. Mobile devices offer sufficient detector information and measuring power to allow accurate action recognition to get an energy consumption estimation throughout day-today life. The sensor-based activity recognition researchers feel that computers are better desirable to act on our behalf to observe the behavior of agents. (b) Multi-user, Sensor-Based Action Recognition On-body sensor action was recognized for multiple users in the early 1990s. Throughout workplace scenarios, detector technology like acceleration sensors recognizes the cluster activity patterns. With this, they question the basic problem of identifying actions of many users from detector measurements. Each single-user and multi-user action in a unified answer is known by proposing a novel pattern sound approach. (c) Visual-Based Action Recognition In the method of naming image, current sequences in particular action labels are the vision-based human activity recognition. The videos taken by a number of cameras help us to track and clearly understand the agents’ behavior which are very important. HCI, robot learning, interface design and security surveillance are some of the applications of visual-based action recognition system. Different techniques like hidden Markov model (HMM), Kalman filtering, optical flow, etc., have been tried by analysts under completely different modalities like single camera, stereo and infrared. Motivation: In order to break down video, indexing, repossess and reliability purpose, the actions recognized from the input video are absolutely necessary. Howbeit, we should incorporate unique mechanisms in order to spot all the actions of human by computer system. Process of HAR is to be categorized into two types such as (i) Low-level action recognition process and (ii) high-level recognition process. The low-level recognition processes are simple to implement, and they come under recognizing the actions from the extracted feature values. The high-level recognition processes are more reliable and computationally expensive, and they are employed to find actions in the input video using specific hardware. The objective is to spot the actions of 1 or more person from round-the-clock observation on the person’s actions and deviation within the environmental conditions. The low-level action recognition process identifies the actions utilizing the feature points extracted from the video of specific size. These specific process implementations are easy, and they are unreliable in all the time. The high-level action recognition process requires some special hardware

Human Action Recognition Using STIP Evaluation Techniques

401

such as high-resolution cameras to detect the actions in the video. These processes are more reliable, and they were very much computationally expensive [1, 2]. Organization of the paper: (a) Abstract, (b) Introduction, (c) Related Works, (e) Existing System, (f) Proposed System, (g) Proposed System Methodology, (h) Feature Extraction Techniques, (i) Action Recognition System, (j) Performance Measures, (k) Conclusion, (l) Future Work, (m) References, (n) Author’s Biography.

2 Related Works The automation in every area comes by rapid growing of technology. The human face and facial expression recognition are heavily needed to specific applications in real life of person. It has many specific applications, which are data privacy, image or video security surveillance, information security, biometric identification, human–computer interface (HCI), human behavior interpretation (HBI), etc [3]. As mentioned in first section, the human action recognition using STIP method ignores the spatial–temporal (ST) interrelationships between all types of person visual features. To improve the activity recognition, there are many works been presented to capture STIP information. In the year 2017–18, authors took the challenge in the field of leveraging vision of computer techniques in order to enrich HRI techniques, and this concept explores the systems which can expand the capabilities of action [4]. In the year 2018, authors analyzed to detect and recognize activities using wearable sensor or mobile data which are collected with appropriate sensors. They have presented that feature extraction is a very important stage in order to help for reducing time of execution and improvement of accuracy of all person action [5]. The authors Van and Tran have proposed a technique which exhibits both optical flow and RGB for HAR. They have analyzed the techniques and application of convolutional neural network, and this CNN is very much suitable for the task of person visual activity recognition from various input videos [6].

3 Existing System HAR mechanism provides description, interpretation or comprehension of the scene by bringing out vital options from image. The flawless process cannot be outlined as, recasting the present image in an exceedingly needed manner, and the output of the positioning action and speed is obtained at an equivalent time and real-time aspects [7]. Innumerable SIFT variants were projected in order to spot the actions of the person in the video. SIFT-based sampling and local descriptors are often extracted on the motion trajectories [8].

4 Proposed System The most well-built image processing system which consists of human eye together with the brain is the human visual system. With this resource, we try to develop a computer

402

H. S. Mohana and U. Mahanthesha

vision system. The video is fed to the system which divides it into each different frame and preprocesses it, to reinforce the image frames by removing the unwanted pixels from the frames. By this technique, we can reduce the noise and store those derived pictures for later usage. Options were extracted using the SIFT descriptors of various sort from the preprocessed video frames.

5 Proposed System Methodology The detailed structure of the human activity recognition (HAR) is as shown in Fig. 1.

Fig. 1. Structure of human activity recognition flow

The input video frames were preprocessed to remove the noise from the video. By reducing the noise, we can improve the performance of the process. Variety of noise are present in the frames or images, and the most common is the salt-and-pepper noise and can be seen as white and black pixels in the images. Also, the image is preprocessed to remove the unwanted pixels. We can apply and filter techniques to get rid of noise from the frames [9, 10]. The noised pixel is detected by the use of median filters, and the

Human Action Recognition Using STIP Evaluation Techniques

403

noisy pixel is replaced by average of the neighboring pixels. Using the STIP descriptors of various kinds, the options were extracted from preprocessed video frames. The information in each of the considered descriptors is calculated, and the features like Harris STIP, Gabor STIP and HOG STIP are extracted from the frames. To detect the corners in the video frames, we use Harris STIP and this algorithm is also used to detect the corners in each pixel of the image, by considering the corner localization differential methods with directions and also even considering the sum of squared differences (SSD). The process of input video loading is as shown in Fig. 2. Which clearly shows that the size of the input videos and frames counts per second for image analysis.

Fig. 2. Illustration of loading the input video to the HAR system and illustration of converting video into frames

The frames and its counts of the given input video to the human action recognition system are illustrated in Fig. 2. These frames are very much useful for analyzing motion of an action in the STEP analysis. After the loading input video in the HAR Graphic User Interface need to check for preprocessing of the video which is shown in Fig. 3.

6 Feature Extraction Techniques 6.1 Harris STIP The algorithm is used to spot the corner present in each pixel of a picture using the corner score differentiation into account w.r.t direction. A grip is the sudden modification in the brightness of a picture. Corner is the junction of two edges. The resemblance is computed by locating the sum of squared differences (SSD) between the two patches. If the pixels within the image are of uniform intensity, then the nearby edges will look similar, and if not the edges will look relatively different. To abstract some varieties of options and deduce the contents of a picture in computer vision systems, corner identification is a worthy appeal. Corner identification is applied many times in motion or movement’s detection, image mosaicking, image registration, tracking of videos, panorama sewing, and 3D modeling and various types of object recognition.

404

H. S. Mohana and U. Mahanthesha

Fig. 3. Human activity recognition GUI window representation

Detection Process of Harris Corner Intensity variation mechanism is used to detect all points through a local neighborhood we make use of Harris mechanism, and a very small region of the feature could be showing the maximum change in intensity levels when comparing with the shift of windows in any direction [11, 12]. This concept is explained using the autocorrelation functions illustrated below: Let us consider P as a scalar function which is represented by function P → R and small increment among any position in the domain as represented by h, a ∈ . Corners are defined as the points x that gives large values of the below illustrating functional for very small h, E(h) = Σw(a)P(a + h) − P(a)

(1)

That is the large variation in any other direction. The function w(a) gives permission for selecting the region of support, which is clearly called as a Gaussian function. Taylor expansions will be used to get linearization of the expression P(a + q) as P(a + q)  P(a) + ∇(a)Tq Hence, the right hand of (i) gives E(q)  Σ w(a)(∇P(a) q)2 da = Σw(a)(qT ∇P(a)∇P(a)Tq)

(2)

The last equation 2 depends on the image gradient through the matrix of autocorrelation, or tensor structure, which is represented as Z = Σw(a)(∇P(a)∇P(a)T )

(3)

The largest eigenvalue of Z corresponds to maximum intensity variation direction, and also the second one corresponds to orthogonal direction of the intensity variation.

Human Action Recognition Using STIP Evaluation Techniques

405

6.2 Gabor STIP Gabor function caters to an energy density of local spectral values located around initially represented position and also certain direction of frequency. A two dimensional convolution of a Gabor function in circular domain is separately able to one-dimensional ones series. To detect the corners, we go with Gabor wavelet, and these wavelets serve as second-order PD operator. A linear filter for detecting edges, which is named after Dennis Gabor, is the Gabor filter. Orientation description and frequency and Gabor filter are identical to human component analysis method. They are awfully suitable for texture description and differentiation. A plane wave of sinusoidal signal is controlled using Gaussian kernel function which is also a 2D Gabor filter in a spatial domain. Gabor filter has been used extensively in pattern analysis, optical character recognition, finger print recognition, facial expression recognition, etc. The Gabor filter feature samples of cycling action are represented in Table 1. Table 1. Gabor filter feature samples of cycling action 6.6617e+04 5.6961e+04 3.7978e+04 3.3827e+04 3.3368e+04 6.8448e+04 5.5650e+04 4.0233e+04 3.4343e+04 3.1905e+04 7.1175e+04 6.0219e+04 4.1516e+04 3.4421e+04 3.2473e+04

Features of Gabor filter: The basic feature extraction of Gabor filter in the twodimensional function is as illustrated in expression 1. The Gabor features referred to multiple resolution Gabor feature are generated from outputs of Gabor filters by using multiple filters on many frequencies fa and orientations. Frequency representations are illustrated in Eq. 4 fa = h − afmax a = {0, . . . , A − 1}

(4)

where fa is the ath frequency, f max = 0 is the maximum frequency generated and h > 1 is the scaling factor of frequency. Let us consider θ n as filter orientations drawn as θ n = 2π n/N = 0, . . . , N − 1

(5)

where n is the nth orientation and N is maximum orientations. 6.3 HOG STIP Histogram of oriented gradients (HOG) is the best object detection in computer vision technology and image processing which uses the applications of feature descriptors. Basically the split of the image into very small connected regions which are called cells, and for each cell we compute a HOG directions or edge orientations for the all pixels within the cell. Each pixel of cell provides gradient weights to its respective angular bin. We can take blocks as spatial regions, which are the neighboring cell group [13]. The base for classification and normalization of histograms is assembling of cells as blocks.

406

H. S. Mohana and U. Mahanthesha

The block diagram represents the normalized group of histograms. This process yields better invariance to changes in brightness or shadowing. Calculation of Histogram of Oriented Gradients The initial step of generating the descriptor in HOG is to measure the one-dimensional derivative points such as Ga and Gb in a and b directions by the convolution of gradient masks Ma and Mb with original image I:   Ga = Ma ∗ I Ma = −1 0 1 (6)  T Gb = Mb ∗ I Mb = −1 0 1

(7)

With the help of derivatives, basis functions Ga and Gb calculate the degree of HOG [|G(a, b)|] and angle in direction F(a, b) for each one of pixel. The degree of HOG shows its strength at a pixel as shown in Eq. 8:  |G(a, b)| = Ga(a, b)2 + Gb(a, b)2 (8) The feature of 3D HOG extraction is as shown in Fig. 4. All three feature extraction techniques in HAR system of STIP algorithm are shown in Fig. 5.

Fig. 4. Illustration of 3D HOG feature extraction in HAR system

7 Action Recognition System Using multi-SVM classifier, the actions in the video are recognized. Support vector machines are utilized for the classification purpose [2]. Vector machines and regression

Human Action Recognition Using STIP Evaluation Techniques

407

Fig. 5. Illustration of feature extraction techniques in HAR system

analysis are the supervised learning models along with learning algorithms associated, which are capable of analyzing the data and hence recognize the patterns [14–16]. Support vector machine (SVM) is a non-probabilistic binary linear classifier. Expression for hyperplane is represented as (a.h) + t = 0 where t is the set of training vectors, a are vectors perpendicular to the separating hyperplane and h is the offset parameter which permits to raise the margin. The output showing “surfing” and “cycling” is one of the actions identified from the input video processing as illustrated in Fig. 6.

Fig. 6. Output showing “surfing” and “cycling” is one of the actions identified from the input video processing

408

H. S. Mohana and U. Mahanthesha

7.1 HAR This module detects and recognizes the following actions and interactions based on the given videos in their specific actions of various persons of color or gray-scaled videos, and this system exactly recognizes human actions like boxing, surfing, walking, running, clapping, hand waving, jogging, cycling.

8 Performance Measures To calculate the implementation of HAR mechanism, accuracy, sensitivity and specificity of the classifier are estimated. Accuracy of the classifier is the rate at which the classifier is able to identify the image based on the given label. Sensitivity of the classifier is calculated based on how exactly the classifier is able to classify the data to the defined categories [17, 18]. Sensitivity is also recognized as rate of true positive or rate of recall. Specificity of the classifier is calculated based on how exactly the classifier is able to reject the data from each category. Specificity is also known as true-negative rate. The following equations describe the exact calculations of the measuring parameters like sensitivity, specificity and accuracy of the given input action video. Sensitivity =

True Positive True Positive + False Negative

(9)

Specificity =

True Negative False positive + True Negative

(10)

Accuracy =

(True Positive(TP) + False Negative(FN)) (False Positive(FP) + True Negative(TN) + True Positive(TP) + False Negative(FN))

(11)

The performance measurements in HAR system and its calculated results are illustrated in Fig. 7. The accuracy, sensitivity and specificity of the given existing and proposed system are as shown in Fig. 8. The HAR performance measurement bar graph representations are shown in Fig. 9. The confusion matrix result of “surfing” action in HAR system is as shown in Fig. 10. The diagonal elements of the confusion matrix give higher values than comparing upper and lower triangular matrix which shows the 100% of accuracy.

9 Conclusion The action performed by the person in the video is identified by the planned mechanism on the basis of options extracted exploitation color STIPs. Action recognition is created using the kernel function of the SVM classifier. The exactness of the designed system gives high accuracy than existing techniques as the different changes in classifications are minimized to a larger extent. STIP detectors and descriptors were redeveloped so that multiple photometric channels are incorporated additionally with image intensities, leading to color STIPs.

Human Action Recognition Using STIP Evaluation Techniques

409

Fig. 7. Illustration of performance measurements in HAR system

Fig. 8. Bar graph outputs for representation of accuracy sensitivity and specificity of the given activity

Fig. 9. HAR performance measurements

The action performed by the person in the video is recognized accurately by the proposed method based on the extracted options. The results are obtained with the exact correctness even though there have been challenges such as illumination variations,

410

H. S. Mohana and U. Mahanthesha

37 3 0 0 0 0 0 0

0 38 2 0 0 0 0 0

0 0 39 1 0 0 0 0

0 0 0 39 1 0 0 0

0 0 0 0 37 3 0 0

0 0 0 0 0 40 0 0

0 0 0 0 0 0 36 4

0 0 0 0 0 0 1 39

Fig. 10. Confusion matrix of cycling action of HAR system

contrast variations, abrupt motions and scaling of the person in the video. To boost the performance of the system, it is used by automation of supervised learning classifiers. The supervised learning framework classifiers require manual label, and therefore, the system must be trained for the classification purpose. The system performance is improved, and also, some of the feature extraction algorithms are deduced. These algorithms describe the classifier. The additional options that are to be extracted must overcome the problems of real-time implementation of the system.

10 Future Work The human action recognition techniques can also be applied using data fusion techniques. The data fusion techniques are speech action data can add to the human facial expressions or any action recognition to achieve better performance output to the real time given videos. Acknowledgements. The authors would like to thank all who are directly or indirectly involved, advised and coordinated in collecting data and software information for the support of this paper.

References 1. Mahanthesha U, Mohana HS (2016) Identification of human facial expression signal classification using spatial temporal algorithm. Int J Eng Res Electr Electron Eng (IJEREEE) 2(5) May 2016 2. Lalitha K, Deepika TV, Sowjanya MN, Michahial S (2016) Human identification based on iris recognition using support vector machines. Int J Eng Res Electr Electron Eng (IJEREEE) 2(5) May 2016 3. Mahanthesh U, Mohana HS (2016) Identification of human facial expression signal classification using spatial temporal algorithm. Int. J. Eng. Res. Electr. Electron. Eng. (IJEREEE) 2(5) 4. Efthymiou N, Koutras P, Filntisis PP, Potamianos G, Maragos P (2018) Multi-view fusion for action recognition in child-robot interaction. 978-1-4799-7061-2/18/$31.00 ©2018 IEEE

Human Action Recognition Using STIP Evaluation Techniques

411

5. Friday NH, Mujtaba G, Al-garadi MA, Alo UR (2018) Deep learning fusion conceptual frameworks for complex human activity recognition using mobile and wearable sensors: 978-1-5386-1370-2/18/$31.00 ©2018 IEEE 6. Khong V, Tran T (2018) Improving human action recognition with two-stream 3D convolutional neural network. 978-1-5386-4180-4/18/$31.00 ©2018 IEEE 7. El Din Elmadany N (2018) Student Member, IEEE, Yifeng He, Member, IEEE, and Ling Guan, Fellow, IEEE. Information fusion for human action recognition via biset/multiset globality locality preserving canonical correlation analysis. IEEE Trans Image Process 27(11) Nov 2018 8. Pavithra S, Mahanthesh U, Michahial S, Shivakumar M (2016) Human motion detection and tracking for real-time security system. Int J Adv Res Comput Commun Eng ISO 3297:2007 Certified 5(12) Dec 2016 9. Ryoo MS, Aggarwal JK Stochastic representation and recognition of high-level group activities. Robot Research Department, Electronics and Telecommunications Research Institute, Korea, e-mail: [email protected] 10. Xia L, Aggarwal JK Spatio-temporal depth cuboid similarity feature for activity recognition using depth camera. Computer and Vision Research Center/Department of ECE, The University of Texas at Austin, [email protected] 11. Holte MB (2012) Human pose estimation and activity recognition from multi-view videos: comparative explorations of recent developments. IEEE J Sel Top Sign Process 6(5) Sept 2012 12. Aggarwal JK, Ryoo MS Human motion: modeling and recognition of actions and interactions. In Proceedings of the 2nd international symposium on 3D data processing, visualization, and transmission (3DPVT’04) 0-7695-2223-8/04 $ 20.00 IEEE 13. Rasheed MB, Raja HB, Alghamdi TA (2015) Evolution of human activity recognition and fall detection using android phone. In: 2015 IEEE 29th International Conference on Advanced Information Networking and Applications 14. Ambeth Kumar VD, Ashok Kumar VD, Malathi S, Vengatesan K, Ramakrishnan M (2018) Facial recognition system for suspect identification using a surveillance camera. ISSN 10546618, Pattern Recognition and Image Analysis 28(3):410–420. © Pleiades Publishing, Ltd. 15. Chernbumroong S, Cang S, Yu H (2015) Member, IEEE Genetic algorithm-based classifiers fusion for multisensor activity recognition of elderly people. IEEE J Biomed Health Inform 19(1), Jan 2015 16. Turchet L, Bresin R (2015) Effects of interactive sonification on emotionally expressive walking styles. IEEE Trans Affect Comput 6(2) April-June-2015 17. Jafari R, Kehtarnavaz N (2018) A survey of depth and inertial sensor fusion for human action recognition. https://link.springer.com/article/10.1007/s11042-015-3177-1. 07/12/2018 18. Al-Akam R, Paulus D (2018) Local feature extraction from rgb and depth videos for human action recognition. Int J Mach Learn Comput 8(3) June 2018 19. Xia L, Chen CC, Aggarwal JK Human detection using depth information by kinect. The University of Texas at Austin Department of Electrical and Computer Engineering

Fuzzy Cognitive Map-Based Genetic Algorithm for Community Detection K. Haritha(B) and M. V. Judy Department of Computer Applications, Cochin University of Science and Technology, Cochin, Kerala 682023, India {haritha.kaladharan68,judy.nair}@gmail.com http://dca.cusat.ac.in/

Abstract. One of the most elemental operations concerning the analysis of properties of a network is community detection. It is the process of decomposition of a given network into groups of densely connected nodes that tend to share some similar properties. A wide variety of algorithms to identify the communities in complex networks exists. In this paper, an intelligent genetic algorithm (GA)-based approach to identify communities has been proposed. The efficiency of the solution that resulted from the genetic algorithm depends on the setting appropriate values for the various parameters involved. As a means to reduce the convergence time of the genetic algorithm, a fuzzy cognitive map (FCM) is used. The knowledge derived from the FCM is used to populate the initial population reducing the randomness of the algorithm. The potency of the algorithm is evaluated on various weighted and unweighted benchmark networks. Keywords: Fuzzy cognitive maps · Genetic algorithm detection · Focal nodes · Social networks

1

· Community

Introduction

The different networks around us are highly structured in nature, and exploring these structures gives us insights into the functional properties of a network. The decomposition of a network into different groups that share some similar features and have tightly packed nodes is known as community detection problem [1]. Community detection is the most active area of research in complex networks. A collection of nodes in a network that has high-density associations within the group and low-density associations with nodes in other groups is called a community. The communities form the individual functional units of a system which is encompassed by the entire network, and community detection helps in understanding these individual units better. Extensive research has taken place in this area to discover various methods that can uncover the underlying communities in a network. This work discusses c Springer Nature Singapore Pte Ltd. 2021  C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_39

Fuzzy Cognitive Map-Based Genetic

413

a genetic algorithm-based approach [2]. Inspired by the natural selection mechanisms, genetic algorithms use crossovers and mutations to find the optimum solution to a given problem. Various parameters are taken into consideration when developing a genetic algorithm to ensure the quality of the solution obtained such as the initial population, the crossover probability, the mutation probability, the stopping criteria, the type of selection operator, and the fitness function which is used to solve the problem. This work proposes the use of fuzzy cognitive maps to select the initial population of the genetic algorithm [3]. FCM is a model for the representation of knowledge of a domain. It is a weighted fuzzy graph, wherein the nodes depict the concepts of the domain, and the edges depict the connections between the concepts. The central nodes in the network are identified using the fuzzy cognitive maps, and these nodes give the value of the initial cluster size, which is then adopted with genetic algorithm to build the initial population. This paper is organized as follows. In Sect. 2, we discuss the current works in the areas of genetic algorithm for community detection and fuzzy cognitive maps. Section 3 gives a technical overview of the methods being used. In Sect. 4, a description of the proposed method is given. In Sect. 5, we compare and analyse the different results obtained, and in Sect. 6, the conclusion is provided.

2

Related Work

Girvan and Newman proposed the most prevalent algorithm for community detection. It is based on divisive clustering method where the edges in a cluster that lies between two different communities are removed based on betweenness centrality measure, proposed by Freeman [4], such that in the end only distinct communities are left in the network [1,5]. Other metrics used to assess the efficiency of the detected communities are modularity, normalized mutual information, accuracy, density, etc. Among these, modularity is the most commonly used measure [6–8]. Girvan and Newman introduced it in 2003 [5]. It is the measure of the strength of partition of the nodes into different communities. Modularity values usually lie between 0 and 1. The networks with high modularity have communities which have densely connected nodes within them, and the connections with other nodes in different communities are sparse. An enhancement of the modularity that takes into consideration the weight and direction of the network was proposed by Arenas et al. [9]. Community detection using extremal optimization is also a divisive clustering method that adopts network modularity as the objective function [8]. An agglomerative clustering technique that uses the network modularity to cluster the network was proposed by Girvan and Newman[5]. Genetic algorithms are a set of optimization techniques inspired by the biological evolution process [2]. Genetic algorithms have been successfully used to detect communities in networks [10–14]. Encoding of the chromosome is an important aspect of using genetic algorithm for optimizing a problem. Different encoding schemes have been used to encode solutions for GA in solving

414

K. Haritha and M. V. Judy

the community detection problem. The common encoding schemes used are label-based encoding [15–21] and locus-based encoding [12,21–26]. Most of the implementations of the genetic algorithm use modularity as the fitness function [10,16,17,20,21,23–26]. Other fitness functions being used are community score [12,22], modularity density [18,19], etc. The initial population is determined randomly. The idea of fuzzy cognitive maps was put forward by Kosko in 1986 [3] as an extension of the cognitive map proposed by Axelrod in the year 1976 [27]. It proposes a representation method of highly complex systems that describes the components of the system and the causal interconnections between its entities. The difference in fuzzy cognitive maps from traditional cognitive maps is that FCM incorporates the strength fuzzy logic along with the cognitive maps. Due to its ability to represent any complex system irrespective of its domain, fuzzy cognitive maps have been able to capture research interests of many, and it has been successfully enforced in a vast expanse of scientific fields. Some of the areas where FCM has been used are social and political sciences, engineering, information technology, military sciences, robotics, expert systems, medical sciences, education, prediction systems, and so on. Giles [28] used FCM to study the different causal factors affecting “diabetes”. FCM was used to diagnose obesity based on psychological behaviour by Giabbanelli [29]. Andreou et al. [30] proposed the application of FCM as a technique for modelling political and strategic issues to aid in the decision-making process for crisis management. Credit risk evaluation was done using FCM by Zhai et al. [31]. FCMs were also used to study the influence of family disease history on the possibility of developing breast cancer [32]. Several extensions to FCMs have been proposed. An extension to FCM that included mechanisms to deal with feedback known as rule-based fuzzy cognitive map was proposed by Carvalho and Tome [33]. Salmeron proposed fuzzy grey cognitive maps that deal with multiple meaning environments [34]. Iakovidis and Papageorgiou created the intuitionistic fuzzy cognitive maps (iFCMs) that handle the experts’ hesitancy in decision-making [35]. A variation of FCM that enables defining of dynamic causal relationships between the concepts was proposed by Liu et al. [36]. Dynamic random fuzzy cognitive map (DRFCM) was proposed by Aguilar to model dynamic systems [37]. Fuzzy cognitive network (FCN) was proposed by Kottas et al., a system that always reaches equilibrium points [38]. Evolutionary fuzzy cognitive map (E-FCM) that simulates the realtime variable state was proposed by Cai et al. [39]. An FCM that considers the time relations is the fuzzy time cognitive map (FTCM) [40]. Song et al. proposed a model that makes use of fuzzy IF-THEN rules along with fuzzy cognitive maps [41]. An FCM where the causal relationships are represented by a belief structure was developed by Ruan et al. known as the belief-degree-distributed fuzzy cognitive maps (BDD-FCMs) [42]. A fuzzy cognitive map based on the rough set theory known as rough cognitive maps was proposed by Chunying [43].

Fuzzy Cognitive Map-Based Genetic

3

415

Technical Background

In this section, the theoretical background of how genetic algorithm is used for detecting communities in a network and the working of a fuzzy cognitive map model for detection of focal nodes is provided. 3.1

Genetic Algorithm

Darwin’s theory of “survival of the fittest” is the foundation of genetic algorithms. In GA, a population of individual solutions is repeatedly enhanced until an optimum solution is obtained that is best in resolving the problem than all the other possible solutions. In each generation, individual solutions may be combined to get a solution that increases its level of fitness in the problem. The generated offspring might undergo mutation so that the fitness of the solution can be further enhanced. A fitness function is used to evaluate the level of fitness of individual solution to a particular problem. The newly generated population would have individuals that provide a more optimum solution to the problem than the population they were created from. Figure 1 depicts the chronological order of operations in a genetic algorithm. 3.2

Fuzzy Cognitive Map

FCM represents a system in a way that is similar to how humans perceive it. The concepts of the system are nodes of the FCM, and the causal relationships are depicted by the connections between the nodes. Causal weighted diagraphs are used to represent the system in an FCM [3]. Figure 2 represents a simple fuzzy cognitive map with Ci as the concept of the system and Wij as the degree of the causal relationships between the concepts. Fuzzy cognitive maps are basically matrix-vector calculations, and the different elements in an FCM model are as follows: • Concepts (Ci ) : Concepts are the critical components in a system which play an essential role in the problem under consideration. • State Vector (A): State vector is a vector that comprises all the concepts in the system with values usually between 0 and 1. • Weight Matrix (W): Weight matrix consists of the weights of the connections between different concepts (Wij = weight between Ci Cj ), and the diagonal is always zero. The positive value of Wij represents excitatory or positive causality, and the negative value represents the inhibitory or negative causality. Fuzzy Cognitive Map Learning The initial state vector is determined, taking into consideration the properties of the fuzzy cognitive map. FCM learning is performed on the state vector using the formula: ⎞ ⎛ N  (k+1) (k) (k) (1) = f ⎝Ai + Aj .Wij ⎠ Ai j=i,j=1

416

K. Haritha and M. V. Judy

Fig. 1. Genetic algorithm workflow

Fig. 2. A simple fuzzy cognitive map

Fuzzy Cognitive Map-Based Genetic

417

(k+1)

where Ai represents the value of concept Ci at step k + 1 and Wij is the weight of the connection between Ci and Cj . The threshold function f (x) used is the sigmoid function. 1 (2) 1 + eλx The state vector values are computed iteratively until epsilon which is a residual value that gives the minimum error difference between subsequent concepts. The values obtained from state vector are filtered based on a threshold value (mean of the concept values in the final state vector), and the result gives the total number of focal nodes which is then passed on as input to the genetic algorithm as the initial number of clusters. f (x) =

4 4.1

Proposed Methodology Overall Framework

In this paper, we use an intelligent genetic algorithm that detects the communities while preserving the static as well as the dynamic properties of the network. Figure 3 gives the framework of the proposed community detection algorithm using fuzzy cognitive map. Genetic algorithm helps in optimizing the network modularity to identify the best community structure in a network. Existing community detection algorithms based on genetic algorithm randomly choose the initial number of modules in the network, and the nodes are partitioned based on this value. In this paper, a more intelligent method is proposed using fuzzy cognitive maps to detect the initial number of communities that reduces the randomness factor in the algorithm. We use fuzzy cognitive maps to identify the focal nodes in the network, which will be used as the initial number of clusters for the genetic algorithm. 4.2

Identification of Focal Nodes Using Fuzzy Cognitive Maps

Initializing State Vector The values of the state vector must take into consideration the properties of the network and the problem statement. Since the problem which is to be solved using FCMs is finding the focal nodes of the network, the value of the different concepts is set such that it takes into consideration how well the nodes are connected. Hence to compute the initial state vector, the given betweenness centrality [4] measure is used: g(v) =

 σst (v) σst

(3)

s=v=t

In Eq. 3, σst is the total number of shortest paths from s to t and σst (v) gives the number of shortest paths from s to t that passes through the vertex v. High centrality scores indicate that a vertex lies on a considerable fraction of shortest paths connecting pairs of vertices.

418

K. Haritha and M. V. Judy

Fig. 3. Framework of the community detection algorithm based on genetic algorithm using fuzzy cognitive maps

Fig. 4. A sample network

FCM Learning Consider the sample network in Fig. 4. The initial state vector of the FCM obtained using the betweenness centrality measure is

Fuzzy Cognitive Map-Based Genetic

419

A = [0.0, 0.06944444444444445, 0.29166666666666663, 0.0416666666666666 64, 0.5972222222222222, 0.1388888888888889, 0.5, 0.38888888888888884, 0.0, 0.0] The weight matrix of the sample network is given by ⎤ ⎡ 0.0

⎢4.0 ⎢4.0 ⎢ ⎢0.0 ⎢ ⎢0.0 W =⎢ ⎢0.0 ⎢ ⎢0.0 ⎢ ⎢0.0 ⎣0.0 0.0

4.0 0.0 2.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0

4.0 2.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0

0.0 1.0 0.0 0.0 0.0 4.0 0.0 0.0 0.0 0.0

0.0 0.0 1.0 0.0 0.0 2.0 1.0 0.0 0.0 0.0

0.0 0.0 0.0 4.0 2.0 0.0 0.0 0.0 0.0 0.0

0.0 0.0 0.0 0.0 1.0 0.0 0.0 3.0 0.0 0.0

0.0 0.0 0.0 0.0 0.0 0.0 3.0 0.0 3.0 2.0

0.0 0.0 0.0 0.0 0.0 0.0 0.0 3.0 0.0 4.0

0.0 0.0⎥ ⎥ 0.0⎥ 0.0⎥ ⎥ 0.0⎥ ⎥ 0.0⎥ ⎥ 0.0⎥ ⎥ 2.0⎥ 4.0⎦ 0.0

FCM learning is applied to the state vector and the weight matrix. After all the iterations, the FCM converges, and the final state vector when epsilon condition is achieved is A = [0.80914196, 0.66695489, 0.73648484, 0.66075637, 0.8411309, 0.81757448 0.90584185, 0.86862879, 0.76254197, 0.68520098] Identifying Focal Nodes The values obtained from state vector are filtered based on a threshold value (mean of the concept values in the final state vector), and the result gives us the total number of focal nodes which is given as input to the genetic algorithm as the initial number of clusters. For the final vector obtained, the mean value is 0.7754257012345555. Hence, total of 5 focal nodes are identified in the network and the focal nodes identified are depicted in Fig. 5.

Fig. 5. Focal nodes identified in the sample network

4.3

Genetic Algorithm to Identify the Best Community Structure

Creating Initial Population The first step in GA is to initialize the population. The initial number of communities (n) obtained from the FCM aids in

420

K. Haritha and M. V. Judy

creating the initial population for the GA by randomly partitioning the network into n communities. Using FCM to find the probable count of communities in the network helps the genetic algorithm to converge much faster and avoid unnecessary iterations. Label-based encoding strategy is used to represent the solution where the chromosome holds the community configuration information and each gene in the chromosome has values representing the community it corresponds to (Fig. 6).

Fig. 6. Chromosome representation of the communities in a network

Fitness Function After the initial population construction, the set of genetic operations are applied to the population iteratively. The fitness score for each solution in the population is determined by the algorithm. Modularity [5] is used as a fitness function to optimize the solutions until the best possible partition of the network is obtained. Modularity compares a different partitioning of a network, and the best partitioning gives the maximum modularity value. Modularity lies in range [−1, 1]. Modularity can be computed using Eq. 4. Q=

N N wi wj  1  wij − δ(Ci Cj ) 2w i=1 j 2w

(4)

where wi is the degree of node i and the Kronecker delta function δ(Ci Cj ) is 1 if a connection exists between vertex i and j and 0 otherwise. The equation for modularity works best with unweighted, undirected graphs. For weighted, directed graphs, an enhanced version of the modularity equation is used. Weighted modularity [9] equation is given by Q=

N N wiout wjin  1  wij − δ(Ci Cj ) 2w i=1 j 2w

where



wiout =

(5)

wij

(6)

wij

(7)

j

wjin =

 i

2w =

 i

wiout =

 j

wjin =

N  N  i=1

wij

(8)

j

Crossover and Mutation The crossover operation is applied to a subset of the chromosomes based on a crossover probability value. Crossover operation

Fuzzy Cognitive Map-Based Genetic

421

is done by selecting any two chromosomes based on their fitness values and recombining them to produce a new genetically improved chromosome that has a better fitness value than its parents. In this paper, the single-point crossover (Fig. 7) is used to generate the offspring.

Fig. 7. Single-point crossover used in GA

After applying a series of crossovers, the mutation is performed on randomly selected chromosomes based on a probability value. Random resetting is used to mutate the population, and a node in the chromosome is reassigned to a random community in the network. Selection Based on Ranking After computing the fitness value for each chromosome in the population, the solutions are ranked based on their fitness values. Ranking ensures that the best chromosomes are always at the top of the list. In this paper, ranking is used as the objective function to select the chromosomes for the next generation. The selection, crossover, and mutation operations are applied iteratively for a predefined number of generations. The solution with maximum modularity is preserved across all generations. This ensures that the optimum solution is never lost irrespective of the number of generations in the algorithm.

5

Experimental Results

In this paper, benchmark networks have been used to test the efficiency of the algorithm. The fuzzy cognitive map identified the focal nodes in different networks taking into consideration the centrality properties of the network (Fig. 8). The hyperparameters for the GA are common for all the data sets (mutation probability = 0.01, crossover probability = 0.5, population size = 200, number of generations = 500). And the algorithm was tested on an Intel i7 system with 32 GB RAM and Nvidia GeForce GTX 1050 GPU.

422

K. Haritha and M. V. Judy

Fig. 8. Focal nodes identified using FCM in few benchmark networks. a karate club network, b animal social network, c network on books about US politics, d American college football network, e Jazz musician network, f C. elegans network

Fig. 9. Communities in Zachary’s karate club

The karate club data set [44] is one of the most commonly used data sets to identify the community structures in a network. The network consists of 34 vertices and 78 edges. Fuzzy cognitive map identified 13 focal nodes in the network. Finally, two communities were identified using FCM with GA (Fig. 9). Another network taken into consideration is the animal social network [45]. It is a weighted network that depicts the interactions among group-living European badgers. The network data set was used to analyse the impact of social

Fuzzy Cognitive Map-Based Genetic

423

Fig. 10. Clusters identified by GA in animal social network data set

interactions among the badgers on disease spreading. The network consists of 51 nodes and 494 edges. FCM identified 21 focal nodes in the network, and after applying the GA for community detection, the network was partitioned into eight communities (Fig. 10). Many other networks were taken into consideration. Table 1 portrays the summary of results obtained after applying FCM-based GA on all the networks considered. Table 1. Summarization of results Input network

No. of nodes No. of edges No. of focal nodes Modularity

Zachary’s karate club [44]

34

78

13

0.37179

Animal social network [45]

51

494

21

0.75

Dolphin social network [46]

62

159

32

0.38

Books about US politics [7]

105

441

41

0.521

American college football [1]

115

613

50

0.55

Les Miserables [6]

77

254

37

0.50

Jazz musician network [47]

198

2742

96

0.44

C. Elegans metabolic network [8] 453

2025

244

0.4001

A comparative analysis of the results obtained by applying fuzzy cognitive map-based genetic algorithm and simple genetic algorithm is depicted in Fig. 11.

6

Conclusion

This paper presents a genetic algorithm-based approach to identify the communities in a network and optimize the network modularity in order to obtain the

424

K. Haritha and M. V. Judy

Fig. 11. Comparison of execution times of FCM-based GA and simple GA for community detection

best solution. Fuzzy cognitive map is used to obtain the initial community size taking into account the centrality properties of the network rather than randomly choosing the total count of communities in the network. The algorithm was tested on various networks. And in all the cases using the fuzzy cognitive map to detect the initial community size greatly improved the overall computational time of the algorithm and helped the algorithm to converge faster.

References 1. Girvan M, Newman MEJ (2002) Community structure in social and biological networks. Proc Natl Acad Sci 99(12):7821–7826 2. Holland JH (1975) Adaptation in natural and artificial systems. University of Michigan Press, Ann Arbor, MI 3. Kosko B (1986) Cognitive fuzzy maps 4. Freeman LC (1977) A set of measures of centrality based on betweenness. Sociometry 5. Newman MEJ, Girvan M (2003) Finding and evaluating community structure in networks, pp 1–16 6. Newman MEJ (2004) Fast algorithm for detecting community structure in networks. Phys Rev E - Stat Nonlinear Soft Matter Phys 69(6):5 7. Clauset A, Newman MEJ, Moore C (2004) Finding community structure in very large networks. Phys Rev E 70(6):6 8. Duch J, Arenas A (2005) Community detection in complex networks using extremal optimization. Phys Rev E 72(2) 9. Arenas A, Duch J, Fern´ andez A, G´ omez S (2007) Size reduction of complex networks preserving modularity. New J Phys 9 10. Tasgin M, Herdagdelen A, Bingol H (2007) Community detection in complex networks using genetic algorithms, pp 1–6 11. Mazur P, ZmarzLowski K, OrLowski AJ (2010) Genetic algorithms approach to community detection. Acta Phys Pol A 117(4):703–705

Fuzzy Cognitive Map-Based Genetic

425

12. Pizzuti C (2008) GA-Net: a genetic algorithm for community detection in social networks. Lect Notes Comput Sci (including Subser Lect Notes Artif Intell Lect Notes Bioinformatics 5199 LNCS:1081–1090 13. Guerrero Manuel, Montoya Francisco G, Ba˜ nos Ra´ ul, Alcayde Alfredo, Gil Consolaci´ on (2017) Adaptive community detection in complex networks using genetic algorithms. Neurocomputing 266:101–113 14. Pizzuti C (2018) Evolutionary computation for community detection in networks: a review. IEEE Trans Evol Comput 22(3):464–483 15. Tasgin M, Bingol H (2006) Community detection in complex networks using genetic algorithm. arXiv preprint, p 6 16. Gog A, umitrescu D, Hirsbrunner B (2007) Community detection in complex networks using collaborative evolutionary algorithms. In: Advances in artificial life SE - 89 17. He D, Wang Z, Yang B, Zhou C (2009) Genetic algorithm with ensemble learning for detecting community structure in complex networks. In: ICCIT 2009 - 4th international conference on computer sciences and convergence information technology, pp 702–707 18. Gong M, Fu B, Jiao L, Du H (2011) Memetic algorithm for community detection in networks. Phys Rev E - Stat Nonlinear Soft Matter Phys 19. Gong M, Cai Q, Li Y, Ma J, An improved memetic algorithm for community detection in complex networks. In: IEEE Congress on Evolutionary Computation (CEC) 20. Jia G, A multimodal optimization and surprise based consensus community detection algorithm, pp 1407–1408 21. Shang R, Bai J, Jiao L, Jin C (2013) Community detection based on modularity and an improved genetic algorithm. Phys A Stat Mech its Appl 22. Pizzuti C (2009) Overlapped community detection in complex networks. In: Proceedings of the 11th annual conference on genetic and evolutionary 23. Shi C, Wang Y, Wu B, Zhong C (2009) A new genetic algorithm for community detection. Part II LNICST 24. Shi C, Cai Y, Fu D, Dong Y, Wu B (2013) A link clustering based overlapping community detection algorithm. In: Data and knowledge engineering 25. Jin D, He D, Liu D, Baquero C (2010) Genetic algorithm with local search for community mining in complex networks. In: 2010 22nd IEEE international conference on tools with artificial intelligence 26. Liu D, Jin D, Baquero C, He D, Yang B, Yu Q (2013) Genetic algorithm with a local search strategy for discovering communities in complex networks. Int J Comput Intell Syst 27. Axelrod R (1976) Structure of decisions: the cognitive maps of political elites 28. Giles BG, Scott Findlay C, Haas G, LaFrance B, Laughing W, Pembleton S (2007) Integrating conventional science and aboriginal perspectives on diabetes using fuzzy cognitive maps. Soc Sci Med 29. Giabbanelli PJ, Torsney-Weir T, Mago VK (2012) A fuzzy cognitive map of the psychosocial determinants of obesity. Appl Soft Comput J 30. Andreou AS, Mateou NH, Zombanakis GA (2005) Soft computing for crisis management and political decision making: the use of genetically evolved fuzzy cognitive maps. Soft Comput 31. Zhai DS, Chang YN, Zhang J (2009) An application of fuzzy cognitive map based on active Hebbian learning algorithm in credit risk evaluation of listed companies. In: 2009 international conference on artificial intelligence and computational intelligence, AICI 2009

426

K. Haritha and M. V. Judy

32. Papageorgiou EI, Subramanian J, Karmegam A, Papandrianos N (2015) A risk management model for familial breast cancer: a new application using fuzzy cognitive map method. Comput Methods Programs Biomed 33. Carvalho JP, Tome JAB (2001) Rule based fuzzy cognitive maps expressing time in qualitative system dynamics. In: 10th IEEE international conference on fuzzy systems (Cat. No.01CH37297) 34. Salmeron JL (2010) Modelling grey uncertainty with fuzzy grey cognitive maps. Expert Syst Appl 35. Iakovidis DK, Papageorgiou E (2011) Intuitionistic fuzzy cognitive maps for medical decision making. IEEE Trans Inf Technol Biomed 36. Miao Y, Liu ZQ, Slew CK, Miao CY (2001) Dynamical cognitive network-an extension of fuzzy cognitive map. IEEE Trans Fuzzy Syst 37. Aguilar J (2004) Dynamic random fuzzy cognitive maps. Comput y sist 38. Kottas Theodoros L, Boutalis Yiannis S, Christodoulou Manolis A (2007) Fuzzy cognitive network: a general framework. Intell Decis Technol 1(4):183–196 39. Cai Y, Miao C, Tan AH, Shen Z, Li B (2010) Creating an immersive game world with evolutionary fuzzy cognitive maps. IEEE Comput Graph Appl 40. Park KS, Kim SH (1995) Fuzzy cognitive maps considering time relationships. Int J Hum - Comput Stud 41. Song HJ, Miao CY, Wuyts R, Shen ZQ, D’Hondt M, Catthoor F (2011) An extension to fuzzy cognitive maps for classification and prediction. IEEE Trans Fuzzy Syst 19(1):116–135 42. Ruan D, Mkrtchyan L (2011) Using belief degree-distributed fuzzy cognitive maps for safety culture assessment. In: Advances in intelligent and soft computing 43. Chunying Z, Lu L, Dong O, Ruitao L (2011) Research of rough cognitive map model. In: Communications in computer and information science 44. Zachary WW (1977) An information flow model for conflict and fission in small groups. J Anthropol Res 45. Weber N, Carter SP, Dall SRX, Delahay RJ, McDonald JL, Bearhop S, McDonald RA (2013) Badger social networks correlate with tuberculosis infection 46. Lusseau D, Schneider K, Boisseau OJ, Haase P, Slooten E, Dawson SM (2003) The bottlenose dolphin community of doubtful sound features a large proportion of long-lasting associations: can geographic isolation explain this unique trait? Behav Ecol Sociobiol 47. Gleiser P, Danon L (2003) Community Structure in Jazz 6(4):565–573

Evaluation of Digital Forensic Tools in MongoDB Database Forensics Rupali Chopade(B) and Vinod Pachghare College of Engineering Pune, Savitribai Phule Pune University (SPPU), Wellesely Road, Shivajinagar, Pune 411005, India {rmc18.comp,vkp.comp}@coep.ac.in

Abstract. Wide usage of online applications has increased the risk of misuse of data by affecting privacy and security policies. Digital forensics is a process of solving criminal cases related to digital devices. Technical growth in this area is the expansion of forensic tools to collect the pieces of evidence. Database forensics is one of the categories of digital forensics. Database forensics covers the scanning of various parts of it for data recovery or finding data tampering. Forensic tools are available for most of the relational databases. Very few tools are available in the market for NoSQL databases. This paper is an attempt to present available digital forensic tools and to experiment with relevant free tools on the MongoDB database to check the usefulness. Keywords: Digital forensic tools · Database forensics · MongoDB · NoSQL · Recovery

1 Introduction The numbers of digital forensic tools are available for the forensic investigation process. These tools cover forensic areas like computer, memory, file, network, database, etc. Digital forensics involves the detail investigation of such devices to find out pieces of evidence [1]. Database forensics is not that much-explored area as compared to digital forensics [2]. Database forensics involves the analysis of database artifacts [3] including log, function/procedure, index, trigger, etc. Most of the forensic tools available are for relational databases like Oracle, MySQL, PostgreSQL, SQLite, etc. Due to huge storage capability and open-source functionality, NoSQL databases are becoming popular. From database forensics perspective, NoSQL databases have not received research consideration, and also, the tools available for the same are very limited [4]. This paper is an attempt to enlist most of the digital forensics tools available as of now and show the applicability of few tools for MongoDB database forensics. These tools are separated as freeware and proprietary along with their functionality [5] and use, as shown through Table 1 [6–11]. As a relational database, forensic tools [3] are designed considering the specific format of that database, so these tools are not relevant and not enlisted in Table 1. MongoDB is a document-based NoSQL database. The reason behind MongoDB © Springer Nature Singapore Pte Ltd. 2021 C. R. Panigrahi et al. (eds.), Progress in Advanced Computing and Intelligent Engineering, Advances in Intelligent Systems and Computing 1198, https://doi.org/10.1007/978-981-15-6584-7_40

428

R. Chopade and V. Pachghare

database selection is its top ranking in the NoSQL category [12]. The database forensic investigation process includes data recovery, finding tamper detection, log recovery, etc. [13]. Any random tool is not applicable to the database as its functionality is intended for different purposes. In databases, different files like log, other database files are available at memory and disk locations. Considering this, few memory analysis tools, disk image creation and hash calculating tools are selected for experimentation purposes. Organization of Paper The existing research conducted so far has discussed in Sect. 2 and the tools namely FTK Imager, Autopsy, Hex Workshop and Compare it experimented on the MongoDB database are described in Sect. 3. Finally, the paper is concluded by enlisting available digital forensic tools with their functionality and future work of the study.

2 Related Work Cankaya et al. [14] enlisted ten different digital forensics tools that are useful for database extraction. These tools are oxygen forensics detective, Xplico, digital detective blade, kernel database recovery, SQL log analyzer, Winhex, Netcat, Windows Forensic Toolchest, SQL CMD and Forensic Toolkit (FTK). Due to the different properties of tools, all tools are not executed on the same database. The three tools which have experimented on the same database are compared with each other based on their execution time. The less execution time is shown by FTK, then by digital detective blade, and finally, highest execution time is by kernel database recovery. Analysis of open-source and proprietary forensic tools is presented by Maurya et al. [15]. As per experimentation, open-source tools are also giving equivalent results compared to proprietary tools. This is explained by Maurya et al. using a comparison matrix. Digital evidence is acceptable in a court of law following Daubert’s guidelines of testing, error rate, publication and acceptable. Comparison among six digital forensic tools is presented by Yadav et al. [16]. The six tools namely Encase, DFF, FTK, TSK, Helix and Liveview are compared with respect to cost, user interface, platform, quality, language interface, feature support, image creation and hash calculation. From the above-mentioned list, few tools are open source, while few are proprietary. Waziri et al. presented [17] analysis of Linux-based digital forensic tools Fiwalk, Autopsy, Bulk Extractor, Sleuth Kit and Foremost. One single tool is not effective for the forensic investigation process. Based on functionality and features, a combination of forensic tools can be utilized. Ghazinour et al. discussed [18] digital forensic tools with respect to features like disk imaging, data carving, memory dump analysis, data recovery, password recovery, file system support, real-time alert, slack space, email analysis, static and live analysis, decryption and incident response. This is helpful for the selection of appropriate tools.

Useful for

HELIX3

LastActivityView

Linux DD

ExifTool

CrowdStrike CrowdResponse

DSi USB Write Blocker

3

4

5

6

7

8

Computer forensics

Digital forensics

Disk forensics

Disk forensics

Computer forensics

Computer forensics

Memory forensics

The Coroner’s Toolkit Digital forensics

Mandiant RedLine

Freeware

1

Name of tool

2

Open source/proprietary

S. No.

Table 1. Digital forensic tools

The command-line tool, used for disk imaging. Available by default in most of the Linux system today

It views the actions performed by the user, events occurred on the machine. Can be exported in CSV or XML format

Data carving

Memory and file analysis

Data recovery

Functionality

Windows

Windows

(continued)

Collects logs, metadata, allows connecting in read-only mode to restrict changes

Data collection tool

Windows, Mac, Unix similar To extract metadata from the OS file system. Supports a large number of file formats.

Linux

Windows

Linux-based

Windows

UNIX-related operating systems

Operating system (OS)

Evaluation of Digital Forensic Tools in MongoDB 429

Computer-aided investigative environment (CAINE)

The Sleuth Kit (Autopsy)

15

16

Digital Forensics Framework

13

Open computer forensics architecture (OCFA)

HxD

12

14

P2 eXplorer

11

Open Source

USB Historian

10

Name of tool

FireEye RedLine

Open source/proprietary

9

S. No.

Computer forensics

Tools supported for disk forensics/memory forensics/network forensics

Computer forensics

Digital chain of custody/digital forensics

Computer forensics

Computer forensics

Computer forensics

Computer forensics

Useful for

Table 1. (continued)

Unix/Windows-based

Built on Linux platform

Built on Linux platform

Windows, Linux

Windows

Windows

Windows

Windows

Operating system (OS)

(continued)

Disk imaging, analysis of file systems

Useful for the forensic investigation process

Data storage in PostgreSQL

Forensic analysis of Windows/Linux, recovery of deleted files, file metadata search

Hex editor and disk editor

A forensic image of physical and logical disk

Scans windows registry to get the details about when USB mounted, user account, etc.

Memory/file analysis, registry data, file metadata

Functionality

430 R. Chopade and V. Pachghare

Xplico

FTK Imager

PlainSight

Paladin Forensic Suite Computer forensics

19

20

21

22

Computer forensics

Disk forensics/memory forensics

Network forensics

Disk forensics/memory forensics

Bulk Extractor

Computer forensics

Useful for

18

Name of tool

Volatility

Open source/proprietary

17

S. No.

Table 1. (continued)

Based on Ubuntu

Windows

Windows

Linux

Windows, Linux

Windows, Linux

Operating system (OS)

(continued)

Tools available for hashing, malware analysis, imaging

File carving/recovery, extracts USB information, Internet history, memory dump

Forensic analysis of file/folders from hard drive, network drive, CD/DVDs, Hash computation

Analysis of network traffic, extraction of an email message from POP/SMTP/IMAP

Extracts of useful information by scanning file, file directory, disk image

Malware analysis, incident response, network connection/socket analysis, Extracts important information from RAM dump

Functionality

Evaluation of Digital Forensic Tools in MongoDB 431

Wireshark

DEFT

Supports Free/Open-Source tools

Shareware

25

26

27

Hex Workshop

WinHex

Belkasoft Evidence Center

28

29

30

Free Hex Editor Neo

Scalpel

24

Name of tool

EVTXtract

Open source/proprietary

23

S. No.

Windows

Windows, Linux, Mac

Windows, Linux, Mac

Operating system (OS)

Digital forensics

Disk forensics/memory forensics

File forensics

File forensics

Windows

Windows

Windows

Windows

Computer Linux-based Forensics/network forensics/mobile forensics

Network forensics

Digital forensics

Digital forensics

Useful for

Table 1. (continued)

(continued)

Forensic acquisition including analysis of disk image, mobile image, RAM image, drive, folder, etc.

Useful for disk cloning/imaging, Hash computation, data recovery

Checksum generator, comparison, structure viewer, basic editing, binary data analysis

Binary file editing, analyze hex data

Data recovery, hashing

Captures network packets and analyses network traffic

Data carving tool

Recovery of windows events log

Functionality

432 R. Chopade and V. Pachghare

Registry Recon

Oxygen Forensics Suite

Computer Online Forensic Evidence Extractor(COFEE)

Cellebrite UFED (Universal Forensic Extraction Device)

SANS SIFT

33

34

35

36

37

X-ways forensics

EnCase

Proprietary

31

Name of tool

32

Open source/proprietary

S. No.

Computer forensics

Mobile forensics

Computer forensics

Mobile forensics

Computer forensics

Computer forensics

Digital forensics

Useful for

Table 1. (continued)

Windows, Linux, MAC

N/A

Windows

Mobile platform—Android, Sony, iPhone, Blackberry

Windows

Windows

Works on all versions of windows

Operating system (OS)

(continued)

Examines file system, disk

Logical/physical data extraction

Evidence extraction from computer

Extract information from mobile device and cloud services used on a mobile device

Registry analysis tool

Disk imaging/cloning and report generation

Disk imaging and cloning Memory and RAM analysis Supports most of the file systems activity logging, automated registry report, file carving and data recovery

Functionality

Evaluation of Digital Forensic Tools in MongoDB 433

XRY

MailXaminer

40

41.

Perpetual

WindowsSCOPE

39

Name of tool

LogMiner

Open source/proprietary

38

S. No.

Email forensics

Mobile forensics

Memory forensics

Database forensics

Useful for

Table 1. (continued)

Windows

Windows

Windows

Windows

Operating system (OS)

Analysis of Email

Extraction of data by analyzing mobile devices

Used to analyze volatile memory

Forensic analysis of Oracle redo log file to extract useful information.

Functionality

434 R. Chopade and V. Pachghare

Evaluation of Digital Forensic Tools in MongoDB

435

3 Tools Experimented on MongoDB Database Open-source tools are also giving trustful results as compared to proprietary tools [19] which can be used as proof in the court of law. This section presents a few opensource tools working on windows platforms which may be useful for forensic analysis of the MongoDB database. A data recovery tool is developed by [20, 21] for the MongoDB database. But as of now, there is not any open-source tool for forensic process of MongoDB database. 3.1 FTK Imager This tool is used for disk or memory forensics [22]. It is also useful to calculate the hash value. If there is any change in the MongoDB database collection file, that change can be observed through a change in hash value as shown in Fig. 1. Along with hash values, FTK creates a copy of the disk image. For any forensic investigation case, a forensic image is first created, and further investigation is done through that image copy. This image can be analyzed using Autopsy, one more open-source tool. 3.2 Autopsy An Autopsy is an open-source file analysis tool [23]. MongoDB database collection file is created with .wt extension. This file is not readable. Image file created using FTK imager is analyzed using Autopsy [17]. After analysis, it shows the contents as per Fig. 2. It can extract only string contents. So, for string content extraction, this tool is useful. 3.3 Hex Workshop Hex Workshop is a shareware binary data analysis, editing, data interpretation and data visualization tool for windows [24]. It is also useful to compare two files. When MongoDB collection file is analyzed using this tool, it shows the string contents of the file, which can be observed from Fig. 3. When contents are deleted from the database, these changes can be observed using this tool. It compares the contents of two files. It is shown in Fig. 4. 3.4 Compare It Compare it tool is useful to compare two files [25]. Though it is not enlisted under digital forensic tools, it may be useful in the digital forensic process. As the Hex Workshop compares and gives the results with binary data, Compare it compares the two files by clearly showing the different contents. The comparison result of the two files is presented in Fig. 5. The first section shows the original file, and in the second section, contents of the file after deletion of the document are given. Change in these two files is highlighted with a rectangle. The other tools named RAM capturer and Belkasoft evidence center (trial version) are also tested. Initially, the disk image is created where MongoDB files are generated, and then using the evidence center, a disk image is analyzed to find out any relevant information. But these tools are not applicable, though they work on files stored on disk.

436

R. Chopade and V. Pachghare

Fig. 1. Hash calculation using FTK

4 Conclusion Due to the unavailability of relevant open-source forensic tool for the MongoDB database, authors attempted to evaluate the applicability of existing freely available digital forensic tools. Though any tool is not directly useful (even though memory and disk image analysis tools are selected), but tools experimented in this paper might be partially helpful. With the Autopsy, all string contents are extracted from the MongoDB collection file by skipping the unreadable contents. Similarly, change in the collection file can be detected using the FTK imager and Hex Workshop. Future work of this research is to develop a tool for forensic analysis of MongoDB.

Evaluation of Digital Forensic Tools in MongoDB

Fig. 2. Collection contents using Autopsy

Fig. 3. Collection contents using Hex Workshop

437

438

R. Chopade and V. Pachghare

Fig. 4. Comparison of two files using Hex Workshop

Fig. 5. Comparison of two files using Compare it

References 1. Sanap VK, Mane V (2015) Comparative study and simulation of digital forensic tools. Int J Comput Appl 975:8887 2. Hauger WK, Olivier MS (2015) The state of database forensic research. In: 2015 Information Security for South Africa (ISSA), pp 1–8 3. Chopade R, Pachghare VK (2019) Ten years of critical review on database forensics research. Digit Investig 4. Beebe N (2009) Digital forensic research: the good, the bad and the unaddressed. In: IFIP International conference on digital forensics, pp 17–36 5. Patankar M, Bhandari D (2014) Forensic tools used in digital crime investigation. Indian J Appl Res 4:278–283 6. Carrier B (2002) Open source digital forensics tools: the legal argument 7. Prasanthi BV (2016) Cyber forensic tools: a review. Int J Eng Trends Technol 41:266–271 8. Ambhire VR, Meshram BB (2012) Digital forensic tools. IOSR J Eng 2:392–398 9. Roman RFM, Mora NML, Vicuña JPN, Orozco JP (2016) Digital forensics tools. Int J Appl Eng Res 11:9754–9762 10. Altheide C, Carvey H (2011) Digital forensics with open source tools. Elsevier 11. Rafique M, Khan MNA (2013) Exploring static and live digital forensics: methods, practices and tools. Int J Sci Eng Res 4:1048–1056

Evaluation of Digital Forensic Tools in MongoDB

439

12. “DB-Engines Ranking—popularity ranking of database management systems.” [Online]. Available https://db-engines.com/en/ranking 13. Olivier MS (2009) On metadata context in database forensics. Digit Investig 5:115–123 14. Cankaya EC, Kupka B (2016) A survey of digital forensics tools for database extraction. In: Future technologies conference (FTC), pp 1014–1019 15. Maurya N, Awasthi J, Singh RP, Vaish A (2015) Analysis of open source and proprietary source digital forensic tools. Int J Adv Eng Glob Technol 1:3 16. Yadav S, Ahmad K, Shekhar J (2011) Analysis of digital forensic tools and investigation process. In: International conference on high performance architecture and grid computing, pp 435–441 17. Okongwu NO, Adebayo OS et al (2013) Cyber crimes analysis based-on open source digital forensics tools. Int J Comput Sci Inf Secur 11:30 18. Ghazinour K, Vakharia DM, Kannaji KC, Satyakumar R (2017) A study on digital forensic tools. In: 2017 IEEE international conference on power, control, signals and instrumentation engineering (ICPCSI), pp 3136–3142 19. Manson D, Carlin A, Ramos S et al (2007) Is the open way a better way? Digital forensics using open source tools. In: 2007 40th annual hawaii international conference on system sciences (HICSS’07), pp 266b–266b 20. Yoon J, Lee S (2018) A method and tool to recover data deleted from a MongoDB. Digit Investig 24:106–120 21. Yoon J, Jeong D, Kang C, Lee S (2016) Forensic investigation framework for the document store NoSQL DBMS: MongoDB as a case study. Digit Investig 17:53–65 22. “Access Data|FTK Imager version 3.2.0.” [Online]. Available https://accessdata.com/de/pro duct-download/ftk-imager-version-3-2-0 23. “Autopsy.” [Online]. Available https://www.sleuthkit.org/autopsy/ 24. “Hex Workshop: Hex Editor, Sector Editor, Base Converter and Hex Calculator for Windows.” [Online]. Available http://www.hexworkshop.com/ 25. “Compare It! 4.0—file compare and merge tool helps in file comparison.” [Online]. Available https://www.grigsoft.com/wincmp3.htm