Latest News

 

The 9th International Symposium on Computational Intelligence and Industrial Applications (ISCIIA 2020) will be held in Beijing, P. R. China. The ISCIIA 2020 is an international conference that provides a forum for scientists and engineers over the world to present their own theoretical results and techniques in the field of computational intelligence, information technology, and industrial/control applications. The past conference proceedings have been indexed in EI since 2004. Selected excellent papers will be recommended for publication in a special issue in JACIII (Journal of Advanced Computational Intelligence and Intelligent Informatics, ESCI/SCOPUS/EI indexed) after revision. Taking this opportunity, we sincerely welcome our colleagues worldwide to join us for this conference.Technical topics of the conference include but are not limited to: 

Control Theory and Applications                        Data Mining
Network-Based Control                                       Intelligent Control
Data Fusion                                                          Fuzzy Systems
Fuzzy and Neural Systems                                  Neural Networks
Human-Computer Interface                                 Evolutionary Computing
Multi-Agent Systems                                           Learning Systems
Network Security                                                 Mechatronics
Intelligent Transportation Systems                      Motion Control
Robotics                                                               Signal and Image Processing

 

Organizer:

Beijing Institute of Technology (BIT)

 

Sponsors:

Beijing Association of Automation (BAA)

Japan Society for the Promotion of Science (JSPS)

Japan Society for Fuzzy Theory and Intelligent Informatics (SOFT)

International Fuzzy Systems Association (IFSA)

IEEE Systems, Man, and Cybernetics Society (Beijing Capital Region Chapter)

IEEE Computational Intelligence Society (Beijing Chapter)

IEEE-IES Technical Committee of Human Factors

 

Co-Sponsors:

Fuji Technology Press Ltd

KTI Semiconductor Manufacturing Machine Co. Ltd

 

----------------------------------------------------------------------------------------------------------------------------------------------------------------

Publication Ethics and Publication Malpractice Statement

It is necessary to agree upon standards of expected ethical behavior for all parties involved in the act of publishing (authors, editors, peer reviewers, publisher). ISCIIA 2020 ethic statements are based on the guidelines and standards developed and published by the Committee on Publication Ethics (COPE). 

 

Editors

1. ISCIIA 2020 editor is responsible for deciding which of the articles submitted to the conference should be published.

2. ISCIIA 2020 editor will accept and evaluate manuscripts for their intellectual content regardless of race, gender, nationality, ethics origin, political philosophy, religious belief or sexual orientation of the authors.

3. Editors must not disclose any information about a submitted manuscript to anyone other than stakeholders.

4. Unpublished manuscripts or materials must not be used in an editors own research.


Authors

1. Authors attest that paper is their own, original and unpublished work.

2. Authors guarantee that the manuscript has not been copied or plagiarized (in whole or in part); a paper should contain sufficient details and references to permit others to replicate the work; any used information resources (phrases, data, images…) must be appropriately cited or quoted.

3. Authorship should be limited to those who have made a significant contribution to the manuscript content. All those who made significant contribution should be listed as co-authors.

4. Affiliation (all authors/co-authors) and corresponding address (main author) must be clearly stated.

5. When authors discover a significant error or inaccuracy in theirown published work, it is the authors' obligation to promptly notify the Publisher and cooperate with the editor to retract or correct the paper.


Reviewers

1. All manuscripts received for review must be treated as confidential documents.

2. Privileged information/ideas obtained by reviewers through peer review process must be kept confidential and not used for personal advantage.

3. Personal criticism of the author is inappropriate; review should be conducted objectively.

4. All suggestions and comments should be followed by supporting arguments. 

 

Call for Papers

Part 1: General Session ⇒  

Part 2: Organized Session ⇒

---------------------------------------------------------------------------------------------------------------------------------------------------------------

PART 1: General Session

A PDF version of CFP can be downloaded by the link "/pub/isciia2016/docs/CPF/" (Pls. click it).

The International Conference of ISCIIA 2020 will be held in Beijing, P. R. China. ISCIIA 2020 is an international conference that provides a forum for scientists and engineers over the world to present their own theoretical results and techniques in the field of Computional Intelligence and Industrial Applications. 
 The past conferences are Ei indexed and excellent selected papers will be recommended to be included in the special issue of JACIII(Journal of Advance Computational Intelligence and Intelligent Informatics, ESCI/SCOPUS/EI indexed). Technical topics of the conference include but are not limited to:

Control Theory and Applications Data Mining
Network-Based Control Intelligent Control
Mechatronics Intelligent Transportation Systems
Data Fusion Fuzzy and Neural Systems
Neural Networks Motion Control 
Human-Computer Interface Fuzzy Systems
Evolutionary Computing Multi-Agent Systems
Learning Systems Signal and Image Processing
Robotics Network Security

 

 

 

----------------------------------------------------------------------------------------------------------------------------------------------------------------

PART 2: Call for Organized Session of ISCIIA 2020

Welcome to participate in the session "Advances in Signal Processing and Control of Biomedical Systems".  For more information, please follow the link  "/pub/isciia2020/docs/CPOS/" .

 

 

 

 

 

Paper Submission Deadline           ----------------   July 15, 2020     July 31, 2020(No more extension!)

Notification of Acceptance                   -----------------  Aug. 15, 2020   

Final Camera-Ready Paper Due           ----------------    Aug. 25, 2020

Conference Period                                ----------------   Oct. 31-Nov. 3, 2020

      Honorary Chair

Kaoru Hirota (BIT, China; JSPS, Japan)

 

General Chairs                                                                  Advisory Board

Bin Xin (BIT, China)                                                           Yaping Dai (BIT, China)

Shinichi Yoshida (KUT, Japan)                                            Toshio Fukuda (Meijyo Univ., Japan)

 

Program Committee Chairs                                             Organized Committee Chairs

Hongbin Ma (BIT, China)                                                    Dawei Shi (BIT, China)

Jinhua She (TUT, Japan)                                                      Xiangyuan Zeng (BIT, China)

                                                                                              Syoji Kobashi ( Univ. of Hyogo, Japan)

 

Publication/Publicity Chairs                                              Local Organizing Chairs

Huifang Li (BIT, China)                                                        Yuan Li (BIT, China)

Zhiyang Jia (BIT, China)                                                       Zhen Li (BIT, China)

Qing Wang (BIT, China)

 

Registration Chairs                                                              General Affairs

Ru Lai (BIT, China)                                                                Rongli Li (BIT, China)

Zhuoyue Song (BIT, China)                                                              

 

Y.Q. Bai                 T. Cai                    C. Chen                     J. Chen                  K. Chen                L. Chen                S. Chen 

W. J. Chen            X. Chen               E. Dadios                 Y. Dai                    J. Dan                   F. Dong                H. Dong  

H. Fang                 Z. Feng                K. Fujimoto             E. F. Fukushima   T. Furuhashi        K.Galkowski        M. G. Gan 

Z. Geng                 T. Hashimoto      Y. Hatakeyama        X. He                    Y. He                  Y. Horiguchi       Y. Hoshino  

N. Ikoma               A. M. Iliyasu       M. Inuiguchi            H. Ishibuchi          H. Iyatomi          Z. Y. Jia              J. Kacprzyk  

K. Kawamoto        S. Kawata            A. Kecskemethy      D. Kim                  S. Kobashi          I. Kobayashi       L. T. Koczy 

N. Kubota                K. Kurashige         R. Lai                        C. Li                       H. Li                    X. Li                    Y. Li    

Z. Li                        X. Liao                B. Liu                        G. Liu                    X. Liu                   Z. Liu                  M.B. Lu    

P. Lu                         Z. Luo                 H. Ma                       H. Masuta             Y. Matsuo             M. Moniwa          Y. Nakagawa  

Y. Nakanishi           K. Nitta               H. Nobuhara             Y. Nojima            C. N. Nyirenda     T. Ohkubo           K. Ohnishi   

S. Ohno                   H. Ohtake            K. Okamoto               I. Ono                  F. Pan                  G. Park               W. Pedrycz      

Q. Pan                       Z. Peng                N. H. Phuong              A. Ralescu            D. Ralescu           X. Ren                I. J. Rudas    

H. Sakaniwa             H. Seki               J. She                         D. W. Shi             A. Shibata            T. Shibata           E. Shimokawara     

 Z. Y. Song                J. W. Spencer       W. Su                           J. Sun                     K. Tachibana        H. Takahashi      T. Terano      

Y. Takama               K. Uehara            Y. Ueno                     G. Vachkov           J. Wang                 L. Wang             Q. Wang       

Y. Wang                   K. Wong              M. Wu                       Q. Wu                    Y. Wu                    Y. Xia                 B. Xin      

X. Xin                      Y. Xiong              B. Xu                          L. Xu                    Y. Xu                     T. Yamaguchi    T. Yamanoi  

F. Yan                        Q. K. Yang             J. Yi                             R. Yokoyama        K. Yoshida            S. Yoshida        T. Yoshikawa    

C.P. Yu                       L. Yu                     X.L. Zeng                    X. Y. Zeng             C. Zhang             X. Zhao              D. Zhao          

J. Zhou                       G. Zhu                  Q. Zhu                          X. Zuo

 

 

Keynote Speech 1: Synthesis of Minimum Noise Non-Reciprocal and Phase-Insensitive Quantum Amplifiers 

By Ian R. Petersen ( Australian National University)

 

Keynote Speech 2: Multiview Granular Rule-Based Modeling: Design and Analysis 

By Witold Pedrycz (University of Alberta, Canada)

 

Keynote Speech 3: Intelligent Alarm Monitoring of Complex Industrial Processes 

By Tongwen Chen (University of Alberta, Canada)

 

Keynote Speech 4: Seeing through human body with near-infrared light - Attempt for new noninvasive medical imaging 

By Koichi Shimizu (Waseda University, Japan)

 

Keynote Speech 5: Modelling and Prediction of COVID-19 Spreading 

By C. K. Michael Tse (City University of Hong Kong)

 

Invited Talk 1: Topological Mapping based on Computational Intelligence in Robotics 

By Naoyuki Kubota (Tokyo Metropolitan University, Japan)

 

Invited Talk 2: Decision analytics using data-driven modelling and evidential reasoning 

By Yu-Wang Chen (The University of Manchester)

 

Invited Talk 3: New Trend of Engineering Medicine and Our Related Work 

By Hongbin Ma (Beijing Institute of Technology)

 

Invited Talk 4: Artificial Pancreas: Current Situation and Future Directions 

By Dawei Shi (Beijing Institute of Technology)

 

Keynote Speaker 1: Dr.  Ian R. Petersen

Title: Synthesis of Minimum Noise Non-Reciprocal and Phase-Insensitive Quantum Amplifiers

Abstract: We present a systems theory approach to finding the minimum required level of added quantum noise in phase-insensitive and non-reciprocal  quantum amplifiers. We also present a  synthesis procedure for constructing  quantum optical phase-insensitive and non-reciprocal quantum amplifiers which adds the minimum level of quantum noise and achieves a required gain and bandwidth. This synthesis procedure is based on a singularly perturbed quantum system involving the broadband approximation of a Bogoliubov transformation. In the case of a phase-insensitive amplifier it requires two squeezers and two beamsplitters. In the case of a  non-reciprocal and phase-insensitive quantum amplifier  it requires three squeezers and two beamsplitters.

Bio: 

Dr. Ian R. Petersen was born in Victoria, Australia. He received a Ph.D in Electrical Engineering in 1984 from the University of  Rochester. From 1983 to 1985 he was a Postdoctoral Fellow at the  Australian National University. From 1985 until 2016 he was with UNSW Canberra where he was most recently a Scientia Professor and an Australian Research Council Laureate Fellow in the  School of Engineering and Information Technology.  He has previously been ARC Executive Director for Mathematics Information and Communications, Acting Deputy Vice-Chancellor Research for UNSW and an Australian Federation Fellow. From 2017 he has been a Professor at the Australian National University. He was the Interim Director of  the Research School of Electrical, Energy and Materials Engineering at the Australian National University from 2018-2019. He has served as an Associate Editor for the IEEE Transactions on Automatic  Control, Systems and Control Letters, Automatica, IEEE Transactions on Control Systems Technology and SIAM Journal on  Control and Optimization. Currently he is an Editor for Automatica. He is a fellow of IFAC, the IEEE and the Australian Academy of Science.  His main  research interests are in robust control theory, quantum control theory and stochastic control theory.

 

Keynote Speaker 2: Dr. Witold Pedrycz

Title: Multiview Granular Rule-Based Modeling: Design and Analysis

Abstract: Multiview models are models describing a real-world system being perceived from different points of view. In establishing such individual perspectives, we typically engage locally available features (attributes, input variables). When considered together, a collection of multiview models has to be aggregated. Multiview models also arise as a consequence of system modeling carried out in the presence of data characterized by a large number of variables. Under such circumstances, building a monolithic model involving all attributes is neither feasible nor computationally sound. We formulate and discuss these two categories of scenarios by focusing on fuzzy rule-based architectures. An important task when building an aggregate of multiview models is to endow the overall global model with a sound measure of quality using which one can efficiently assess the relevance of the individual results produced by the rule-based models as well as establish the quality of the overall fusion. We advocate that the quality of the results can be quantified in terms of some information granule. In the two scenarios outlined above, the family of multiview models is aggregated with the use of the augmented principle of justifiable granularity -one of the fundamentals of Granular Computing. The related optimization criteria of coverage and specificity are discussed along with the associated optimization process. The emergence of type-2 information granules is also motivated and analyzed in depth.

Bio: 

Dr. Witold Pedrycz (IEEE Fellow, 1998) is Professor and Canada Research Chair (CRC) in Computational Intelligence in the Department of Electrical and Computer Engineering, University of Alberta, Edmonton, Canada. He is also with the Systems Research Institute of the Polish Academy of Sciences, Warsaw, Poland. In 2009 Dr. Pedrycz was elected a foreign member of the Polish Academy of Sciences. In 2012 he was elected a Fellow of the Royal Society of Canada. In 2007 he received a prestigious Norbert Wiener award from the IEEE Systems, Man, and Cybernetics Society. He is a recipient of the IEEE Canada Computer Engineering Medal, a Cajastur Prize for Soft Computing from the European Centre for Soft Computing, a Killam Prize, a Fuzzy Pioneer Award from the IEEE Computational Intelligence Society, and 2019 Meritorious Service Award from the IEEE Systems Man and Cybernetics Society. 

His main research directions involve Computational Intelligence, fuzzy modeling and Granular Computing, knowledge discovery and data science, pattern recognition, data science, knowledge-based neural networks, and control engineering. He has published numerous papers in these areas; the current h-index is 114 (Google Scholar) and 87 on the list top-h scientists for computer science and electronics  http://www.guide2research.com/scientists/. He is also an author of 21 research monographs and edited volumes covering various aspects of Computational Intelligence, data mining, and Software Engineering. 

Dr. Pedrycz is vigorously involved in editorial activities. He is an Editor-in-Chief of Information Sciences, Editor-in-Chief of WIREs Data Mining and Knowledge Discovery (Wiley), and Co-editor-in-Chief of Int. J. of Granular Computing (Springer) and J. of Data Information and Management (Springer).  He serves on an Advisory Board of IEEE Transactions on Fuzzy Systems and is a member of a number of editorial boards of international journals. 

 

 

Keynote Speaker 3: Dr. Tongwen Chen

Title: Intelligent Alarm Monitoring of Complex Industrial Processes

Abstract: In operating industrial facilities, alarm systems are configured to notify operators about any abnormal situation. The industrial standards (EEMUA and ISA) suggest that on average an operator should not receive more than six alarms per hour. This is, however, rarely the case in practice as the number of alarms each operator receives is far more than the standard. There exist strong industrial needs and economic benefits for better interpreting and managing the alarms, and redesigning the alarm systems to reduce false and nuisance alarms, and increase the alarm accuracy. In this talk, we plan to summarize our recent work in this new area, targeting an intelligent and data-based approach, called “alarm analytics,” and presenting a new set of advanced tools for alarm visualization, performance evaluation and analysis, alarm rationalization design, alarm flood classification, and root cause analysis, thereby to help industrial processes to comply with the new standards. The tools have been tested with real industrial data and used by process engineers in Canada and elsewhere.

Bio: 

Dr. Tongwen Chen is currently a Professor and Tier 1 Canada Research Chair in Intelligent Monitoring and Control at the University of Alberta, Canada. He received the BEng degree in Automation and Instrumentation from Tsinghua University (Beijing) in 1984, and the MASc and PhD degrees in Electrical Engineering from the University of Toronto in 1988 and 1991, respectively. His research interests include computer and network based control systems, event triggered control, process safety and alarm systems, and their applications to the process and power industries. He is a Fellow of IEEE, IFAC, as well as Canadian Academy of Engineering.

 

Keynote Speaker 4: Dr. Koichi Shimizu

Title: Seeing through human body with near-infrared light - Attempt for new noninvasive medical imaging

Abstract: Medical imaging is indispensable in modern medicine. To overcome various problems of current imaging techniques such as X-ray, MRI, PET, etc., we are developing an optical transillumination imaging technique. With visible light, we can see only the surface of our body. In contrast, near-infrared light with 700-1200 nm wavelength has relatively high transmission through animal tissue. However, the strong optical scattering in the tissue makes transillumination imaging difficult. If we can appropriately handle this scattering effect, we can see through our body without using hazardous X-ray. Moreover, using the rich legacy of optical spectroscopy, we can realize various functional imaging and molecular imaging for the three-dimensional internal structure of an animal body.

Bio: 

Dr. Koichi Shimizu received M.S.(1976) and Ph.D.(1979) degrees, from University of Washington(UW), Seattle, USA. He was Research Associate in UW 1974-79. He was an Assistant-, an Associate- Professors, and a Professor in Hokkaido University, Sapporo, Japan in 1979-2016. He is currently a Professor Emeritus of Hokkaido University and a Professor of Waseda University, Kitakyushu, Japan. He served as an associate editor of IEEE Trans. ITB in 1999–2007. He has been a Fellow of the Electromagnetics Academy, and an editorial board member of Scientific Reports, Nature.

 

Keynote Speaker 5: Dr. C. K. Michael Tse

Title: Modelling and Prediction of COVID-19 Spreading

Abstract: The COVID-19 outbreak began to occur and escalate in a special holiday period in China (about 20 days surrounding the Lunar New Year), during which a huge volume of intercity travel took place, resulting in outbreaks in multiple regions in China connected by an active transportation network. Thus, in order to understand the COVID-19 spreading process in China, it is essential to examine the human migration dynamics, especially between the epicentre Wuhan and other Chinese cities. A new Susceptible-Exposed-Infected-Confirmed-Removed (SEICR) model with consideration of intercity travel and active intervention is proposed for predicting the spreading progression of COVID-19. The model takes into account the known or reported number of infected cases being fewer than the actual number of infected individuals due to insufficient testing. The model integrates intercity travel data to track the movement of exposed and infected individuals among cities, and allows different levels of active intervention to be considered so that realistic prediction of the number of infected individuals can be performed. Data from over 500 cities/regions around the globe are included in the study.

Bio: 

Dr. C. K. Michael Tse received the BEng degree with first class honors and the PhD degree from the University of Melbourne, Australia, in 1988 and 1991, respectively. He is presently Chair Professor of Electrical Engineering at City University of Hong Kong. Prior to joining City University in October 2019, he was with the Hong Kong Polytechnic University, where he assumed Head of the Department of Electronic and Information Engineering (2005-2012), member of University Council (2013-2015), and chairman of the culture promotion committee (2000-2019) which organised cultural programmes in history, art, theatre and music. His research interests include network applications, power electronics and nonlinear systems. He was recipient of a number of research and invention prizes including a few Best Paper Prizes from IEEE and other journals, as well as a Grand Prize in Silicon Valley International Invention Festival (2019). In 2005, 2010 and 2018, he was selected and appointed as IEEE Distinguished Lecturer. In 2006 he chaired the IEEE CAS Technical Committee on Nonlinear Circuits and Systems. He serves and has served as Editor-in-Chief of IEEE Transactions on Circuits and Systems II (2016-2019), IEEE Circuits and Systems Magazine (2013-2016), IEICE Nonlinear Theory and Applications (since 2013); as Editor of IJCTA (2014-2020) and associate editor of a few other IEEE journals. He has served on a number of IEEE committees including the IEEE Fellows Committee and the IEEE Awards Committee, and chaired the Steering Committee for IEEE Transactions on Network Science and Engineering. He has been appointed to honorary professorship and distinguished fellowship by a few Australian, Canadian and Chinese universities, including the Chang Jiang Scholar Chair Professor with Huazhong University of Science and Technology, Honorary Professor of Melbourne University, and Distinguished Professor-at-Large with the University of Western Australia. He currently serves or has served on panels of Hong Kong Research Grants Council, Innovation Technology Fund, and on the Quality Education Fund Steering Committee. While with Hong Kong Polytechnic University, he received the President’s Award for Outstanding Research twice in 1996 and 2000. He is an IEEE Fellow and an IEAust Fellow.

 

Invited Speaker 1: Dr. Naoyuki Kubota

Title: Topological Mapping based on Computational Intelligence in Robotics

Abstract: Recently, we often face unexpected disasters all over the world. Various types of mobile or multi-legged robots have been developed for exploration and monitoring in such a disaster. Especially, environmental sensing, environmental monitoring and environmental modelling, are required as the important tasks in the remote control and monitoring. The environmental modelling is used to conduct task planning through map building. Simultaneous localization and mapping (SLAM) is one of essential technologies to deal with environmental information. In general, there are two main approaches of grid mapping and topological mapping in SLAM. Basically, we have to reduce the computational cost while keeping the accuracy of localization and mapping in real time monitoring and control. 


In this talk, we focus on topological mapping methods to extract environmental features from a 3D point cloud. The applicability of unsupervised learning methods based on topological mapping has been discussed to deal with environmental features in unknown and dynamic environments. One of them is Growing Neural Gas (GNG) that can dynamically change the topological structure composed of nodes and edges. One important advantage of GNG is in the incremental learning capability of nodes and edges according to a target data distribution. We have proposed several different types of topological mapping methods based on GNG to extract the environmental features from a 3D point cloud until now. First, we explain the learning algorithm of GNG and several experimental results of GNG for SLAM in various environments. Next, we discuss the advantage and disadvantage of standard GNG by comparing with other topological mapping methods such as self-organizing map and growing cell structure. Next, we explain multi-layer GNG to extract hierarchical features in environmental maps as a multi-scale approach, and batch learning algorithm for GNG (GL-GNG) to improve the convergence property. Furthermore, we explain the modified method of GNG-utility (GNG-U), that is called GNG-U2. GNG-U2 can improve the real-time adaptability of extracting topological structure in non-stationary data distribution. Next, we show several experimental results of mobile robots and multi-legged robots. Finally we discuss the applicability and future direction of topological approaches in robotics.

Bio: 

Dr. Naoyuki Kubota graduated from Osaka Kyoiku University in 1992, received the M.E. degree from Hokkaido University in 1994, and received the D.E. from Nagoya University, Japan, in 1997. He was an Assistant Professor and Lecturer at the Department of Mechanical Engineering, Osaka Institute of Technology, Japan, from 1997 to 2000. He joined the Department of Human and Artificial Intelligence Systems, the School of Engineering, Fukui University, as an Associate Professor in 2000. He joined the Department of Mechanical Engineering, the Graduate School of Engineering, Tokyo Metropolitan University, Japan, as an Associate Professor in 2004. He was an Associate Professor from 2005 to 2012, and a Professor from 2012 to 2017 at the Department of System Design, the Graduate School of System Design, Tokyo Metropolitan University, Japan. He is currently a Professor at the Department of Mechanical Systems Engineering, the Graduate School of Systems Design, Tokyo Metropolitan University, Japan. His current interests are in the fields of cognitive robotics, robot partners, coevolutionary computation, fuzzy control, spiking neural networks, and informationally structured space. He was a Visiting Professor at University of Portsmouth, UK, in 2007 and 2009, and was an Invited Visiting Professor at Seoul National University from 2009 to 2012. His current interests are in the fields of coevolutionary computation, fuzzy control, spiking neural networks, perception-based robotics, robot partners, and informationally structured space. He has published more than 200 refereed journal and conference papers in the above research fields. He received the Best Paper Award of IEEE IECON'96, the Best Paper Award of IEEE CIRA'97, and so on. He was an associate editor of the IEEE Transactions on Fuzzy Systems from 1999 to 2010, the IEEE CIS Intelligent Systems Applications Technical Committee, Robotics Task Force Chair from 2007 to 2014, IEEE Systems, Man, and Cybernetics Society, Japan Chapter Chair since 2018, and others.

 

Invited Speaker 2: Dr. Yu-Wang Chen

Title: Decision analytics using data-driven modelling and evidential reasoning

Abstract: Decision analytics allow individuals and organizations to transform data and combine evidence to support informed decision making. However, most real-world decision making problems are often characterized by multiple sources of data and different types of information. In this talk, I will briefly introduce data-driven modelling and evidential reasoning in the context of decision analytics under uncertainties, where the evidential reasoning methodology constitutes a conjunctive probabilistic reasoning process generalised from Dempster’s rule and Bayesian inference. A set of examples and real-world applications will be presented for the illustration of leveraging data-driven modelling and evidential reasoning for business decision making.

Bio: 

Dr. Yu-Wang Chen is currently Senior Lecturer in Decision Sciences at Alliance Manchester Business School (AMBS), The University of Manchester. Prior to his current appointment, he was a Postdoctoral Research Associate, and then appointed as a Lecturer in 2011 at the Decision and Cognitive Sciences (DCS) research centre of AMBS, and a Postdoctoral Research Fellow at the Department of Computer Science, Hong Kong Baptist University. He serves as the Programme Director for MSc Business Analytics at AMBS, which is ranked by QS the 8th in the world. He received the PhD degree in Control and System Engineering from Shanghai Jiao Tong University in 2008. His research interests are mainly in the areas of Decision and System Sciences, Operational Research and Data Analytics. He has published over 50 research articles in leading journals, such as EJOR, IS, C&OR, KBS, IEEE T-FS and IEEE T-SMC, 5 books or book chapters and 30 publications in conference proceedings. He has completed as PI/Co-I several research projects funded by ERC, EPSRC, Innovate UK, NSFC, etc. He acts as Associate Editor or Editorial Board Member of several international journals.

 

Invited Speaker 3: Dr. Hongbin Ma

Title: New Trend of Engineering Medicine and Our Related Work

Abstract: Engineering Medicine is an emerging trend of science and technology. In this talk, a brief summary on how engineering has changed and is changing the medicine science will be given first. Among various engineering technologies which have been applied in medicine, imaging technology as well as related image processing and artificial intelligence have been proved to be crucial to current medicine. Besides, this talk will illustrate several applications of engineering medicine, which cover areas of medicine manufacturing, medicine circulation, medical instruments and healthcare information system. Some new algorithms of computer vision and artificial intelligence such as real-time light-weight learning have been applied towards the intelligentization of medical systems, which exhibit promising potential of adaptation, learning and intelligence rooted from cybernetics.

Bio: 

Dr. Hongbin Ma has been a Professor at the School of Automation, Beijing Institute of Technology since he joined Beijing Institute of Technology in 2009. He received his bachelor degree from Zhengzhou University in 2001 and doctoral degree from the Academy of Mathematics and Systems Science, Chinese Academy of Sciences in 2006. He joined Bell Labs Research Center at Beijing in March of 2006 and later joined Temasek Laboratories, National University of Singapore in August 2006 as a Research Scientist. His research focuses on adaptation, learning and recognition, especially adaptive estimation and control, as well as their applications in robots, smart cars, and UAVs. He is also a member of IEEE, SIAM, CSIAM, ACM and Automation Society of China. Dr. Ma is the principle investigator of several scientific projects supported by National Natural Science Foundation (NSFC), the programme of New Century Excellent Talents in University by Ministry of Education, the programme of Beijing Outstanding Talents Project, and so on. He also won several awards, such as Wu Wenjun Artificial Intelligence Award, Outstanding Research Work Award, ICIRA Recognition Award, ICMA best paper award, etc. Dr. Ma has published more than 100 academic papers in reputable journals or conferences. And under his supervision, a dozen of students have performed well in scientific research, innovation contests, and are establishing themselves.

 

Invited Speaker 4: Dr. Dawei Shi

Title: Artificial Pancreas: Current Situation and Future Directions

Abstract: The artificial pancreas (AP) is an intelligent drug delivery system that aims at achieving closed-loop glucose control for patients with type 1 diabetes mellitus, based on real-time glucose sensing, dose decision and infusion technologies. In this talk, we will first briefly review the history and basics of AP systems. We will then discuss the state-of-the-art dose decision technologies in AP and results in recent clinical studies, focusing on the major considerations in technological and clinical design. We will conclude with challenges in AP, which point out current/future directions for this technology.

Bio: 

Dr. Dawei Shi received the B.Eng. degree in electrical engineering and its automation from the Beijing Institute of Technology, Beijing, China, in 2008, and the Ph.D. degree in control systems from the University of Alberta, Edmonton, AB, Canada, in 2014. In December 2014, he was appointed as an Associate Professor at the School of Automation, Beijing Institute of Technology. From February 2017 to July 2018, he was with the Harvard John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, MA, USA, as a Postdoctoral Fellow in Bioengineering. He is currently a professor at the School of Automation and the Institute of Engineering Medicine, both at the Beijing Institute of Technology. His research focuses on analysis and design of complex sampled-data control systems with applications to biomedical engineering, robotics, and motion systems. He serves as an associate editor for IET Control Theory and Applications and IET Cyber-Systems and Robotics. He also served as a guest editor for European Journal of Control. He served as an associate editor for IFAC World Congress and is a member of the IEEE Control Systems Society Conference Editorial Board. He is a senior member of IEEE.

All manuscripts must be submitted electronically in PDF format before the deadline. Only English manuscripts are acceptable. The initial version of each paper cannot exceed 8 pages. The camera-ready version of each accepted paper is also up to 8 pages in total, but each page in excess of 6 will incur an extra charge. Manuscripts should be on A4 sized paper (210 mm x 297 mm), in two-column format, with top and bottom margins of 25 mm, left and right margins of 20mm, column width of 80 mm, and inter-column gap of 5 mm. All figures and tables should be clearly visible, and the paper format should be referenced to https://www.fujipress.jp/jaciii/jc-authors/#materials. A complete paper should be submitted via https://conf.fujipress.com/isciia2020/. Selected papers will also be published in the special issue of JACIII (currently scheduled in 2021 January issue).

Download Paper Template here! (Word, Latex)

 

Registration Form:

Please fill in Registration Form here and send it to lirongli@bit.edu.cn with the subject Registration before August 25, 2020.

Registration Payment:

For Chinese

For Chinese participants, please transfer the registration fee to the following account; please also pay special attention to the subsequent important reminder notice:

账户信息:

单位名称

纳税人识别号

北京理工大学

12100000400009127B

单位开户行及账号

中国工商银行北京紫竹院支行   0200007609014435495

开户行联行号

预算代码

上级部门

102100000763

124202

工业和信息化部

单位地址

单位电话

北京市海淀区中关村南大街5号

010-68918558

 

特别提醒:

1. 请务必在汇款或转账时填写附言,格式为:isciia2020 +文章编号+作者姓名(中文)

2. 作者将汇款/转账回执单扫描件、付款人手机号、填写北京理工大学增值税发票开具信息确认表并在甲方一栏签字盖章,扫描件(表格请在会议网站“Registration”条目内下载)、营业执照复印本扫描件发送至邮箱lirongli@bit.edu.cn;邮件标题格式:文章编号+回执信息;

3. 请在转账/汇款时留出银行转账时间,以保证在2020年8月25号前转账汇款到账;

 

For Japanese
Since it may bring some difficulties to transfer registration fee from Japan to Beijing by using international bank transfer, we would like to recommend participants living in Japan to pay the registration fee (in Japanese yen) to the following account

    みずほ銀行 大岡山支店(支店番号145)
 普通口座 1825115
    名義  ヒロタケンキュウシツダイヒョウヒロタカオル

Please note that

 1) Money transfer charge should be paid by participants,

 2) Remitter's name should be the participants name (in the case if other name is used, please inform the name and the date to conference secretary by e-mail).

 

For Other Foreigners

If you can attend the conference ISCIIA 2020 during Oct 31-Nov 3, 2020 at Beijing(北京), China, you can pay by Chinese money cash only (no credit card service) on site.

 

If you have any further questions about the registration and payment, please contact Miss Rongli Li  (Email: lirongli@bit.edu.cn ; Office Tel.: 86-10-68917913) .

 

 

1A1-1-1 
Research on Dynamic Positioning of Competitive Skiing Based on UWB

Rui YAN, Xiaoyan ZHAO, Wenjing ZHANG, Zhaohui ZHANG, and Ke DENG

Dynamic positioning algorithm of skiers based on Ultra Wide Band (UWB) is discussed in this paper. Firstly, a Non-Line of Sight (NLOS) base station screening method combining with Geometric Dilution Precision (GDOP) is proposed to suppress the degradation of positioning accuracy caused by athletes’ body occlusion. Secondly, the CTRUM (constant turn rate and Uniform motion) model is constructed according to the skiing state of athletes; then, the Unscented Kalman Filter (UKF) is used to fuse the UWB positioning information and the angular velocity and acceleration of the Inertial Measurement Unit (IMU). According to the information of the fusion, the CTRUM model is dynamically updated to improve the positioning deviation of the skiers during the curve skiing. Finally, the algorithm is deployed in the upper computer system and experiments are carried out to verify that the method can screen out the NLOS base station and improve positioning accuracy of the athletes’ curve movement.

1A1-1-2 
Indoor Positioning System Based on Fusion of UWB and IMU with Strong Tracking Kalman Filter

Tianrui LIAO, Kaoru HIROTA, Zelong ZHANG, Xiangdong WU, Yaping DAI, and Zhiyang JIA

A Loosely-Coupled Fusion Positioning (LCFP) system is designed, which integrates Ultra Wide Band (UWB) and Inertial Measurement Unit (IMU) sensor information to locate, and is conducive to suppress the electromagnetic interference to UWB and the error accumulation of IMU. Traditional fusion positioning framework uses Kalman filter for data fusion, which is difficult to process non-Gaussian observation noise. To improve the problem, this system uses Strong tracking Kalman Filter (SKF) for UWB and IMU data fusion, which is conducive to suppress the influence of non-Gaussian noise and improves the positioning accuracy of the system. In the dynamic positioning experiments with non-Gaussian observation noise, the maximum positioning error of this system is about 2 cm lower than that of the traditional fusion positioning system, and the accuracy is improved by about 14%. This means that data fusion using SKF really helps to suppress the effects of non-Gaussian observation noise on the system and enables the system to achieve higher accuracy. In the future, this system framework will not only be applied to the fusion positioning system of UWB and IMU, but also to other fusion positioning systems susceptible to non-Gaussian observation noise, such as UWB and lidar fusion positioning system.

1A1-1-3 
Intelligent Error Compensation

Yiping SONG and Ning LIU

A temperature error compensation method based on RBF neural network is proposed to solve the problem that the navigation accuracy of MEMS gyroscope is reduced due to the influence of temperature. The temperature related output data of MEMS gyroscope are collected through the designed experimental scheme. The conventional temperature compensation model is a polynomial model, and the polynomial model as the temperature compensation model of MEMS gyroscope has shortcomings, that is, it cannot show the deviation in a small temperature interval. Therefore, the temperature error compensation method of RBF neural network is suggested as an improvement of polynomial fitting.

1A1-1-4 
A Speaker Localization Method Based on Voice and Image Multimodal Fusion

Hao-Ran JIN, Chen-Ning LU, Ying-Zuo LONG, Bao-Han WU, and Zhen-Tao LIU

Single-modal localization technology based on computer hearing or computer vision has the deficiencies of accuracy and stability. Considering that the comprehensive utilization of multimodal information can effectively improve the accuracy and anti-interference of the positioning system, a speaker localization method based on voice and image multimodal fusion is proposed. Firstly, we use the method based on TDOA using microphone array for voice localization and AdaBoost algorithm for face detection separately. Secondly, a multimodal fusion method based on temporal and spatial fusion between voice and image is proposed. After developing the frame rate tracker for temporal fusion, the pixel coordinate system and the world coordinate system are fused for the fusion of the features from face image and voice localization. The proposed method was tested by positioning the speaker stand at 15 different points, and each point was tested for 50 times, from which the experimental results show that there is a high accuracy when speaker standing in front of the positioning system with a distance between 0.5 m to 3.5 m, and an azimuth between −45° to 45°.

1A1-1-5 
Research on 3D Surveying Based on Binocular Stereo Vision

Haoran WEI and Xiangyang XU

Surveying technology is a technology to obtain the spatial information of the target. With the development of computer technology, the use of computer technology for surveying can effectively improve the accuracy and speed of surveying. As for surveying equipment, compared with some distance sensors, the binocular camera has the advantages of simple structure, low price and easy operation. This research will study the method of surveying using binocular vision technology and perform the surveying on a specific object. Binocular stereo vision surveying includes five steps: camera calibration, stereo rectification, stereo matching, calculating coordinates and three-dimensional reconstruction. This work uses OpenCV and OpenGL libraries to implement each step on the VS2016 platform to complete the surveying of a specific object. Then this research proposes an evaluation criterion to evaluate the surveying results. Evaluation results show that, method using binocular stereo vision technology is able to carry out the surveying successfully, its accuracy is close to method using laser rangefinder and its resolution is higher under the same price.

1A1-2-1 
Probabilistic Auto-Encoder Matrix Factorization Model for Public Digital Cultural Resources Recommendation

Lingjun MENG, Feng JIN, Yuqing HOU, and Wenjuan ZHANG

The results of the public digital cultural resource recommendation model will directly affect the user experience and the popularity of traditional culture. Public digital cultural resources have the problems of information overload and data sparsity. In order to achieve more accurate recommendation, we propose a Probabilistic Auto-encoder Matrix Factorization Model. Firstly, we make a data pre-filling to alleviate the missing data, then we build a new deep collaborative network based on the double hidden layer edge noise auto-encoder to train. The recommendation results integrate the Bayes prior information and the recommendation performance of deep learning. The experimental results show that our model has a good effect on the accuracy and recall rate.

1A1-2-2 
An Automatic Scoring System for Factual Subjective Questions Based on Language Element Extraction

Xudong GUO, Yuan LI, and Qinglin WANG

In the field of education, examinations are used more and more frequently as a mean to measure learning outcomes and summarize knowledge for both teachers and students. But the workload of scoring is also increasingly heavy, especially for subjective questions. In this paper, we designed an automatic question scoring system based on language element extraction, combining the deep learning related technology and the characteristics of objective fact-based subjective question. The system is mainly composed of entity recognition module and automatic scoring module. Attention mechanism is applied in entity recognition module to handle the different influences of word in a sentence. A double-layer network is designed considering the complexity of the entity. Experimental results verified the method, and the system achieved the expected design target.

1A1-2-3 
Proposal of Treemap-Based Cluster Visualization and its Application to News Article Data

Yasufumi TAKAMA, Yuna TANAKA, and Hiroki SHIBATA

This paper proposes Treemap-based visualization for supporting cluster analysis of news article dataset. It is important to grasp data distribution in a target dataset for such tasks as machine learning and cluster analysis. When dealing with multi-dimensional data such as statistical data and document set, dimensionality reduction algorithms are usually applied to project original data to lower-dimensional space. However, dimensionality reduction tends to lose the characteristics of data in the original space. In particular, the border between different data groups could not be represented correctly on lower-dimensional space. To overcome this problem, the proposed visualization method applies Fuzzy c-Means to target data and visualizes the result on the basis of the highest membership values with Treemap. The membership values to the second closest clusters are also visualized, which is expected to be useful for identifying instances around the border between different clusters, as well as the relation between different clusters. A prototype interface is implemented to handle news article dataset, of which the effectiveness is investigated with a user experiment.

1A1-2-4 
A Recommendation Algorithm Focusing on Time Bias via Neural Graph Collaborative Filtering

Simin LI, Yaping DAI, Kaoru HIROTA, and Wei DAI

To reduce the error caused by time bias to the accuracy of recommendation algorithm, based on neural graph collaborative filtering (NGCF), a NGCF-Shallow tower recommendation algorithm is proposed. By inputting the users’ and items’ interaction social networks and time features, the recommended objects for a target user is obtained. Compared with other recommendation structures, this algorithm can make up for the influence of users’ and items’ time characteristics on users’ selection. The effect of the algorithm is tested on the Gowalla dataset and the Hetrec2011-delicious-2k dataset. Compared with NGCF algorithm, the accuracy of the proposed algorithm is improved by about 1.5% on Gowalla dataset. In the experiment under the Hetrec2011-delicious-2k dataset, the accuracy rate can be increased up to 4.42%.

1A1-2-5 
K-Means Clustering Gan Based Fault Diagnosis Approach for Imbalanced Dataset

Huifang LI, Rui FAN, Qisong SHI, and Zijian DU

Due to the different occurring frequencies of various faults, class distribution of fault data is often imbalanced. However, most existing machine learning based diagnosis methods, which didn’t take this imbalance into consideration, tend to be biased toward the majority classes and result in poor accuracy for minority ones. To solve this problem, we propose a K-Means clustering GAN (KM-GAN) based fault diagnosis approach which can reduce the imbalance of fault data and improve the diagnostic accuracy for minority classes. First, we design a new oversampling method based on K-Means clustering algorithm and GAN, to generate diverse minority-class samples which have similar distribution with the original minority data. The K-Means clustering algorithm is adopted to divide minority-class samples into K-clusters, while GAN is applied to learn the distribution of the resulting clusters and then generate a certain number of minority samples as a supplement to the original dataset. Second, we construct a Deep Neural Network (DNN) and Deep Belief Network (DBN) based heterogeneous ensemble model as a fault classifier to improve generalization, while DNN and DBN models are trained separately on the resulting dataset, and then the outputs from DNN and DBN are averaged as the final diagnostic results. A series of comparative experiments are conducted to testify the effectiveness of our proposed method, and the experimental results show that our method can improve the diagnostic accuracy of minority samples.

1A1-3-1 
Study on Fire Early Warning System Based on Sparse Coding Visual Bag-of-Words Model

Jiajun WANG and Yahui WANG

Based on the early design and implementation of wireless multimedia sensor networks (WMSNs), the mobile network platform for fire early warning was established for verifying the problem of using video monitoring to directly identify fire to avoid the misjudgment of traditional fire smoke and temperature detector. Aiming at the contradiction between time-delay and real-time of massive data transmission in WMSNs, an algorithm for image information big data processing was studied and applied to fire warning. Simulation and experimental results have shown that this strategy was effective. The research focus on the image preprocessing at the acquisition end of wireless network sensors, and an algorithm based on sparse coding visual word package model multi-attribute feature fusion was proposed. First, the fire target was detected, and the compressed image information was transmitted according to the detection results and decision-making requirements. This new processing strategy of fire image based on WMSNs has greatly reduced the transmission burden of WMSNs (Wireless Multimedia Sensor Networks), improved the data transmission rate and enhanced the real-time performance of information. The experimental verification results have also shown that the model was effective for target recognition. The classification decision learning algorithm of support vector machine (SVM) was trained with positive and negative samples, and the experimental verifications of the decision function were conducted, and the results were accurate.

1A1-3-2 
A Skeleton-Based Method for Recognizing the Campus Violence

Yi XING, Yaping DAI, Kaoru HIROTA, and Zhiyang JIA

With the surveillance video acquired by campus surveillance devices as input, this paper adopts an action recognition method to determine whether or not campus violence has occurred. We incorporate attention module into the two-stream adaptive graph convolutional network. This method can process skeletal information extracted from the video with graph convolution network. And then it determines the presence of school violence through action recognition based on the human skeleton. To verify the validity of this method, we do experiments on the filtered NTU RGB+D dataset. The accuracy of our method is 96.07%, which is 1.31% higher than the conventional method.

1A1-3-3 
IoT-Based Precision Irrigation System for Eggplant and Tomato

Maria Gemel B. PALCONIT, Edgar B. MACACHOR, Markneil P. NOTARTE, Wenel L. MOLEJON, Arwin Z. VISITACION, Marife A. ROSALES, and Elmer P. DADIOS

Water management, specifically for areas with scarce water, is essential. Hence, the sensors and smartphone camera were deployed to gather data the needed parameters and to access these data through an application remotely. The data gathered from the soil moisture sensor was used to set up an automatic irrigation system that would irrigate the precise water needed for the tomato and eggplant. In 4 weeks, the irrigation system watered the tomato with a total volume of 4009.875 mL and 4543.08 mL for the eggplant while in traditional watering, the tomato consumed a 6650 mL and 8550 mL for the eggplant. Overall, this resulted in a total of 44% water consumption savings, while the plants were visually healthier than the traditional watering method.

1A1-3-4 
Object-Action Interaction Region Detection in Egocentric Videos

Shinobu TAKAHASHI and Kazuhiko KAWAMOTO

We propose a deep model for detecting object-action interaction regions in egocentric videos. This task includes both object detection and action recognition simultaneously, and we need to detect the only object involved in the action. We design a two-stream deep architecture that enables learning by annotating a single frame in a video clip so that we can avoid the time-consuming annotation that assigning bounding boxes and object classes to every frame of the video clip. In this paper, we report the results of a comparative evaluation of four possible structures of the output layer of the two-stream architecture for multitask learning. Experimental results on the EPIC-KITCHENS dataset shows that the structure of fusing object detection and action recognition provides better performance than the other structures.

1A1-3-5 
DQN Based Reinforcement Learning Algorithm for Scheduling Workflows in the Cloud

Huifang LI, Jianghang HUANG, Yizhu WANG, Binyang WANG, and Chenghao GU

With more and more scientific and social media applications, the amount of data is growing exponentially. Any type of computing applications, such as data calculation or analysis, can be described as workflows. Cloud computing provides an effective platform for executing large and complex workflow applications conveniently and cheaply through its delivering internet-based services as a pay-as-you-go model. However, the performance of workflow scheduler directly affects the Quality of Service (QoS) of the cloud users, and how to efficiently allocate the heterogeneous cloud resources to execute workflows still faces big challenges. In this work, an improved Deep Q Network (DQN)-based reinforcement learning (RL) algorithm for workflow scheduling is developed to optimize dual objectives like makespan and cost simultaneously. First, we test the performance of DQN and Actor-critic (AC) based RL algorithm in scheduling workflows respectively, then modify the reward function for the DQN algorithm to improve its convergence and universality for optimization problems. Extensive experiments are conducted to verify our approach and the simulation results show that the proposed algorithm can minimize both makespan and cost, as well as adjust user preference for the specific optimization objective and accordingly increase the diversity of generated scheduling schemes.

1A1-4-1
A Short-Term Traffic Flow Prediction Framework Based on Deep Learning

Yuqing HOU, Feng JIN, and Lingjun MENG

Accurate transportation prediction in real time affects schedule strategy of transportation system directly, which makes short-term traffic flow prediction module play an important role in intelligent transportation system. To improve the prediction accuracy of short-term traffic flow, a short-term traffic prediction framework based on deep learning is proposed, combining with the basic property of short-term traffic flow and road network topology. The framework can mine complex short-term traffic flow patterns automatically with Short-Term processing module, Long-Term processing module and Feature Embedding module. The experimental results show that the framework gets higher prediction accuracy in short-term traffic prediction tasks of three different time intervals of 5 min, 10 min and 15 min.

1A1-4-2 
Vehicle Detection System on Dynamic Camera Sources Using CNN and SORT Tracker

Fae Nicole C. SERRANO, Elmer P. DADIOS, Argel A. BANDALA, Robert Kerwin C. BILLONES, John Carlo V. PUNO, Allysa Kate M. BRILLANTES, and Jay Robert B. del ROSARIO

With the increasing popularity of dashcam usage, there are other applications of these devices that can contribute to Intelligent Transport Systems such as road surveillance, law violation detection, and autonomous driving. This study mainly tackles detecting vehicles in a dynamic background, tracking vehicles despite varying angle views (front, side, and back views) and extracting vehicular information (vehicle type or license plate presence). Several CNN models were compared based on accuracy and speed in order to identify the architecture best suited for the set-up. The result shows that FRCNN Inception, is the choice for accuracy-based implementation with 85% mAP and SSD MobileNet for speed-based implementation with a frame rate of 7.81 fps on a GPU.

1A1-4-3 
Semi-Supervised Real-Time Roadway Detection Based on ML-ELM

Wenjuan ZHANG, Feng JIN, and Lingjun MENG

To meet the autopilot in non-standard highway of real-time and robustness requirement, this paper proposes a half supervision based on ML-ELM road driving regional rapid segmentation methods. This method contains image driving regional clustering and the pavement decision-making in two stages. In the clustering stage, the first to drop pavement grayscale sample build multi-scale image pyramid, decision-making based on scale factor in each layer to find the best contrast images, and then to clustering of grey value, to search connected domain in the output image of clustering, inhibit the abnormal characteristics of the connected domain and within its domain sampling according to the rules for decision-making sub-graph. In road area of decision-making, using multilayer ELM in the process of the training as a classifier, treat decision-making sub-graph for binary classification, clustering diagram decision will result feedback to the connected domain category attributes, namely road may exercise area pixel level segmentation. Experimental results show this method is compared with the traditional FCM algorithm speed improved, compared with the semantic network FCN segmentation based on deep learning, Multi-net has similar accuracy, effectively ruled out the lane line, pavement crack and other non-standard interference pixels of interference, it has better robustness and real-time performance on curves and multi-mode pavement.

1A1-4-4 
Wireless Video Monitoring Intelligent Car Based on Single Chip Microcomputer

Qichen DONG and Jia ZHANG

Aimed at the management of the enterprise warehouse, a wireless video monitoring intelligent car is designed and implemented to replace manual inspection and recording. The intelligent car takes STM32 Micro Control Unit (MCU) as the main controller and uses the PWM signal to control the motor and realize movement. The camera, temperature and humidity sensor, smoke sensor, and other devices mounted on the car can collect video data and environmental parameter information. The host computer communicates with the intelligent car through Wi-Fi to transmit data and control instructions, and monitors the environment in real time to ensure safety. Two control modes including autonomous obstacle avoidance control and non-contact remote control are designed. The operator can choose the control mode of the car according to different work needs, so as to manage the warehouse more effectively.

1A2-1-1 
An Improved Dynamic Window Approach for Intelligent Pedestrian Avoidance of Mobile Robot

Bijun TANG, Kaoru HIROTA, Junkui WANG, Yaping DAI, and Zhiyang JIA

To ensure the safety of pedestrian, an improved dynamic window approach (IDWA) based on the pedestrian’s motion information is applied to the mobile robot to navigate in the complex environment with dynamic pedestrian. The robot navigates by the IDWA not only preventing collisions with pedestrian and static obstacle but also not affecting the movement of the pedestrian. Through the Robot Operating System (ROS) with Gazebo simulator, the experiments are carried out to compare the classic DWA with the IDWA. It is shown that the IDWA keeps the distance between the robot and the pedestrian above 0.4 meters, and the robot bypasses the pedestrian along the safe path generated by the IDWA behind the pedestrian. When the pedestrian is close to the static obstacle, the IDWA can still generate a smooth and safe path between the pedestrian and the static obstacle. Applying the IDWA to the mobile robot can solve the problem of automatic robot navigation in public places.

1A2-1-2 
Bionic Robot Motion Control Based on Visual Track and Gesture Measurement Glove Oriented to Rotation-Traction Manipulation Training

Haiyu HE, Zhen CHEN, Wenliang LIN, Liguo ZHU, Minshan FENG, and Liwei SHAO

Rotation-Traction Manipulation (RTM) has been proved that is an effective treatment for cervical spondylosis. However, the manipulation is hard to grasp for beginners. In previous study, a prototype training robot has been proposed and obtained good training effect. During the manipulation process, we find that patients need to follow the doctor’s hand gesture. In order to achieve better training experience, an algorithm based on computer vision and gesture measurement glove are proposed to simulate patients tracking hand trajectory in this paper. Firstly, the platform of RTM training robot is simply described. Then, a visual tracking and gesture measurement system are introduced to track the clinicians’ gesture in the first step of manipulation. An improved Camshift algorithm is proposed to enhance the robustness of the tracking trajectory. Meanwhile, a gesture measurement glove is adopted to guaranteed the real-time performance and accuracy of gesture recognition. In this paper, experiments of visual track and gesture measurement carried out. The results show that the robot can track gestures well and satisfy the clinical application requirement.

1A2-1-3 
A Novel Control Scheme Combining Wave Variable and Generalized Predictive Control for Teleoperation

Ru LAI, Shaodong QIN, Jian LI, Shiming CHEN, and Yujun CHEN

A control scheme for teleoperation system is developed in this paper, which combines wave variables and predictive control methods to reduce the impact of time delay between local and remote. Wave variable transmission technology is adopted to ensure the passivity of the communication and the stability of the teleoperation system under any time delay. The velocity information transmitted by the wave variable is used to generate the predicted trajectory. Then a novel predictive control structure and generalized controller (GPC) is proposed. Unlike traditional method, the proportional and derivative terms were added to the dynamic compensation to facilitate the design of predictive controllers, which allow the control system has better tracking ability. Finally, the superiority of the proposed control scheme is proved by comparing the traditional predictive control simulation.

1A2-1-4 
Design Methodology of an All-Terrain Autonomous Quadruped Robot

Zelong ZHANG, Kaoru HIROTA, Junyi YUAN, Yujie WANG, Yixin PENG, Jiaming HE, Yaping DAI, and Zhiyang JIA

The design and control method of a quadruped robot is presented for autonomous cruise over complex terrain. The unique leg design paired with specially modified motors allows the robot to accomplish various tasks with simpler control method and minimum cost. According to a series of real-life tests including hill-climbing, rope-crossing and obstaclecrossing, performance of the robot is proved to be able to meet its design objectives. By verifying performance on real-world prototypes, a new design method of quadruped robot is proposed as an alternative, cost-effective way of achieving certain types of goals.

1A2-1-5 
One-Dimensional Model of Cubic Hopping Rover

Jianxun JIANG, Xiangyuan ZENG, and Shuquan WANG

Due to the irregular gravity field and complex surface terrain, asteroid surface exploration is challenging. Hopping is considered a more promising asteroid surface motion scheme. This paper designs and presents one-dimensional wheel pendulum model that can jump up, roll and keep one-side balance. The reaction wheel mounted on the protective plate is the power source of the system. The sudden braking of the reaction wheel with high speed causes one-dimensional wheel pendulum model to jump up. When the pendulum body approaches the central axis of the system with weak momentum, the Single Chip Micyoco exerts control signal on the motor to keep the system in one-side balance. In this paper, numerical simulations of the system are carried out. The experimental study indicates that the system has the characteristics of quick start, strong stability and excellent immunity.

1A2-2-1 
Particle Swarm Optimization-Based Dark Channel Prior Parameters Selection for Single Underwater Image Dehazing

Vincent Jan D. ALMERO, Ronnie S. CONCEPCION II, Jonnel D. ALEJANDRINO, Argel A. BANDALA, and Elmer P. DADIOS

Underwater images are confronted with blurriness and poor color consistency due to the haze produced by the absorption and scattering effects of the turbid water. Dark Channel Prior (DCP) is the state-of-the-art and the algorithmic basis to solve underwater image restoration. However, the default parameters of DCP may not be applicable to underwater images with different levels of degradation. The selection of the appropriate DCP parameters for each underwater image is considered as an optimization problem and can be solved using Particle Swarm Optimization (PSO). The proposed PSO-based selection algorithm is defined by its operators: objective function, swarm size, inertial weights and acceleration coefficients. Obtaining appropriate combination of these operators are elaborated. The qualitative and quantitative evaluations observed acceptable visual improvements and measurements to underwater images applied with DCP at optimally selected parameters, in comparison to underwater images applied with DCP at default parameters. Hence, the proposed algorithm provides good adaptability and effectivity to the exhaustive search of appropriate DCP parameters.

1A2-2-2 
Tabu-Model-Based Estimation of Distribution Algorithm Framework for Permutation Optimization Problems

Sai LU and Bin XIN

The estimation of distribution algorithm (EDA) is a common and effective method to solve the permutation optimization problems. EDA describes the distribution of the superior samples by establishing a probability model and samples the model to find better solutions in search space. Although EDA has a strong ability for global search, it lacks a mechanism to jump out of the local optima. In this paper, a tabu probability model is designed to depict the distribution of the solutions which needs to be taboo. A similarity between a solution and the tabu model is defined to judge whether the solution should be taboo. A local search operator is conducted for the taboo solutions to improve the sample diversity, and a modification strategy of sampling model is proposed to regain the sample diversity. Based on the above concepts and operators, a tabu-model-based EDA framework is proposed. In this paper, two calssic permutation optimization problems, quadratic assignment problem (QAP) and travelling salesman problem (TSP), are selected to test the proposed EDA framework. The node histogram model (NHM) and the edge histogram model (EHM) are used to design the tabu-model-based EDAs, T-EHM and T-NHM, to customize for the two problems. Finally, compared with the traditional EDAs, T-NHM and T-EHM have much stronger abilities to relieve the premature convergence, and can find the better solutions for QAP and TSP respectively.

1A2-2-3 
PMSM Direct Torque Control Based on Genetic Algorithm and Neural Network

Xiangzhou DENG, Zhida LIU, and Dongpo QU

To reduce the bigger torque ripple in the system of direct torque control (DTC) of permanent magnet synchronous motors (PMSM), genetic algorithm (GA) is introduced in the speed PI controller to realize self-tuning of PI parameters, neural network (NN) is introduced in space vector modulation (SVM) to achieve the desired space voltage vector. Starting from the dynamic mathematical models of PMSM, the system of DTC of PMSM based on GA and NN is proposed. Then, concrete implementation methods of self-tuning PI controllers and NN SVM are analyzed in detail. Finally, modeling and simulation are performed by MATLAB/Simulink. The results show that the proposed control method has the advantages of fast speed response, small speed overshoot and lower torque ripple. The system has better static and dynamic performances.

1A2-2-4 
Independent Evaluations of Each Fuzzy Rule for Derivative-Free Optimization of Fuzzy Systems: Toward Fast Fuzzy-Rules Learning for Fuzzy Inputs

Kiyohiko UEHARA and Kaoru HIROTA

A method is proposed for evaluating fuzzy rules independently of each other in their optimization. It is derived by extending the conventional method called α-FUZZI-ES so as to cope with facts (inputs) given by fuzzy sets (non-singletons). A method is further proposed for fuzzy rules learning based on the evaluation method. It attains fast fuzzy-rules learning by optimizing fuzzy rules independently of each other in parallel. The proposed method is effective especially when evaluation functions for fuzzy rules learning are not differentiable and then derivative-free optimization is required. Numerical results indicate that the learning method achieves proper convergence with derivativefree optimization.

1A2-2-5 
Using Granular Representation of Time Series to Spread Risk of Portfolio Selection

Haoran REN, Benyu ZHANG, and Xuemei YANG

In the process of portfolio selection decisionmaking, investors always want to achieve a portfolio selection result that is as evenly distributed as possible to spread the risk from financial market. In this paper, a granulation method for time series data is proposed to improve the quality of portfolio decision results. In the method, the time series data corresponding to the investment objects are first constructed to information granules which are represented by fuzzy numbers as their mathematical representation, and then the portfolio decision model is employed to give a corresponding decision result after a comprehensive analysis of these information granules. In the process of constructing information granules, in order to make information granules which associate with the corresponding investment objects as informative as possible, the principle of justifiable granularity is used to achieve a compromise between justifiability and specificity of the information granules. The experiments using data from China Shanghai Stock Exchange clearly show that the proposed method improves the dispersion of portfolio investment.

1A2-3-1 
An Incremental Multi-Model Learning Architecture for Emotion Recognition

Akihiro MATSUFUJI, Erina KASANO, Eri SATO-SHIMOKAWARA, and Toru YAMAGUCHI

Emotion strongly influences our work performance, decision making, mental health, and relationship satisfaction in our daily life. To provide the supports and feedbacks that correspond to human emotion appropriately, emotion recognition technologies have been developing. Recent emotion recognition methods are developed with sophisticated machine learning models using a large amount of labeled data. However, the accuracy of these models has been insufficient even of the sophisticated models. According to the field of psychology, the important features of emotion recognition/expression are different among people. We assumed that these models that discover a relationship pattern between features and emotion state using labeled data are suffered from individual differences. Therefore, we integrate the multiple models learned from each person and each feature as knowledge and use appropriate models to infer the emotion of a new user. In this study, we focus on training a model by incrementally using new data obtained from the new user. In the practice of dealing with various individual differences or domains, our architecture flexibly increments and updates with new data. We evaluate the performances of the previous integration method and our architecture in the scenario of training a new user model incrementally and stocking multi-models learned from each user and each modal.

1A2-3-2 
Dynamic Facial Expression Recognition Based on Optical Flow and Geometric Features

Da MA, Kaoru HIROTA, Yaping DAI, and Zhiyang JIA

For the recognition of facial expressions in video images, we propose to use optical flow method to improve geometric features for dynamic expression recognition in this paper. The optical flow method can better capture facial motion and record dynamic features compared with traditional geometric methods. After feature extraction, support vector machines are used to classify the two types of features separately. Finally, a weighted fusion strategy is used for decision-level fusion of the two classifiers. The proposed method is verified on the extended Cohn-Kanade database and the overall recognition accuracy is 95%.

1A2-3-3 
Facial Expression Recognition with Feature Matrix for Low-Resolution Images

Jian WANG, Kaoru HIROTA, Yaping DAI, and Zhiyang JIA

Low-resolution (LR) facial images usually lack enough visual information for feature extraction, which increases the difficulty of expression recognition and reduces the recognition accuracy. In this paper, we propose a new model framework for LR images, which main idea is to extract the emotion feature matrix of the images and add it to the convolutional neural network (CNN). More specifically, the emotion feature matrix determined by analyzing the distribution of salient regions in the dataset can eliminate redundant regions and retain salient regions. The emotion feature matrix is used to construct CNN model to extract limited features from LR images, which can enhance the expression of salient region features. Experimental results on several facial expression datasets, including CK+, JAFFE and FER2013 show the superior performance of the proposed method for LR facial expression recognition, compared with several state-of-the-art methods.

1A2-3-4 
The Multi-Modal Emotion Recognition Based on Text and Image

Wenlong LI, Kaoru HIROTA, Xingwang LIU, Yaping DAI, and Zhiyang JIA

The Multi-modal emotion recognition based on text and image (MMER) is proposed to solve the problem of inaccurate emotion recognition and poor model robustness of a single modality such as text, image or speech. The Multi-modal emotion recognition based on text and image compares the shallow features of text and image by cosine similarity, and inputs the obtained results to the decision-making layer, and participates in the final emotional decision-making together with the respective results of text and image. The experimental data set is made by ourselves, and each row includes an image, a sentence of text and the emotional label. Results of experiments on the dataset show that the Macro-F1 score for the multimodal model based on text and image is 73.54, achieving 6.4% and 11.8% improvement compared with the text emotion recognition model various LSTM and the image emotion recognition model ResNet.

1A2-3-5 
Investigation of Relationship Between Robot Expressions and Human Perception Considering Negative and Anxiety Scale

Youdi LI, Eri SATO-SHIMOKAWARA, and Toru YAMAGUCHI

Perception is one of the most important capabilities in either interpersonal or human-robot interaction. To design the appropriate robot expression styles to gratify different types of people, a better understanding of the relationship between human perception and detectable personal physical and phycological features is essential. In this study, an online video-based questionnaire was utilized to investigate human perception of robot different emotional expressions based on various combinations of voice and motion traits. The negative and anxiety scale were chosen as the phycological indicators. Results have shown that robot’s different expression patterns can elicit participant’s different perceptions towards robot and participants are generally more perceptive about robot’s voice trait than its motion trait. Meanwhile, significant negative correlation between anxiety of the communication capacity and the perception of the robot with negative motion and voice has strengthen the idea that it is necessary to take people attitudes towards robots into consideration in social robots’ expression design. This study provides the first investigations into how to make robots generate appropriate reactions considering individual inner conditions.

1A2-3-6 
Indoor Crowd Posture Recognition and Emotion Perception

Yishan CHEN, Yaping DAI, Kaoru HIROTA, and Zhiyang JIA

In order to detect the abnormal human beings’ emotions in public hall, an emotional awareness alarm (E-alarm) system is proposed to perceive crowd emotions. The E-alarm system should monitor the indoor crowd posture by image analysis of monitoring video and identifies three emotions of “calm,” “nervous” and “angry.” The alarm emails will be sent automatically when getting “nervous” and “angry” emotions. The system is required to achieve rapid alarm adapted to the complex indoor environments in a short working cycle with high accuracy. It uses Microsoft Kinect to extract typical posture features precisely and machine learning models to achieve the stable performance in the continuous working period. Experimental result shows that E-alarm system can effectively identify human emotions and alarm abnormal emotions. The alarm cycle is about 30 s, and the alarm accuracy rate reaches 91.7%.

1A2-4-1 
A Novel Visual-Inertial Navigation System with Yawing Constraints

Jingzhe WANG, Leilei LI, Xunya GUI, Zucheng LI, and Yixiang WANG

Recent years have witnessed the rapid development of Visual-Inertial SLAM amid numerous theories and implementations ever emerging. Notwithstanding VI-SLAM’s high accuracy and low computational cost, its yawing angle is still unobservable all the same, from which it follows that not only is its heading estimation drifts with time but its overall trajectory is bound to deviate, and therein lies the motivation of this paper. We propose a novel Visual-Inertial-Magnetic navigation system comprising an efficient initialization procedure capable of recovering IMU biases, the scalar factor of monocular vision, and the vectors of gravity and Magnetic North; a non-linear optimization module taking visual-inertial-magnetic information as observation. The system has been examined and evaluated on several datasets collected in large-scale outdoor environments. Analysis and comparisons validate its superiority in accuracy over VI-SLAM.

1A2-4-2 
Research on the MINS Navigation Method of the Vehicle Based on Satellite Assistance

Xiaomeng TAN, Ning LIU, and Zhong SU

Aiming at the problem of excessive positioning error caused by Global Navigation Satellite System (GNSS) lock-out in vehicle navigation, this paper proposes a vehicle integrated navigation method based on GNSS-assisted Micro Inertial Navigation System (MINS). This method introduces state feedback and uses the Kalman filter to perform information fusion to obtain the optimal estimation of system navigation information. Simulation results show that the algorithm can effectively suppress the divergence of strapdown solution errors, and at the same time make up for the error problem caused by GNSS loss of lock. It is worth mentioning that among the inertial devices, low-cost, medium-precision Micro Electro Mechanical System (MEMS) devices can be selected. After the integrated navigation algorithm, the navigation results can reach higher accuracy, which has certain engineering application value. This method not only ensures the accuracy of vehicle positioning but also greatly reduces the cost of navigation devices.

1A2-4-3 
An Improved Manta Ray Foraging Optimizer for Mobile Robot Path Planning

Xiangdong WU, Kaoru HIROTA, Bijun TANG, Tianrui LIAO, Yaping DAI, and Zhiyang JIA

A path planner based on improved manta ray foraging optimizer (IMRFO) is proposed for global path planning of mobile robot, in which manta ray foraging optimizer (MRFO) and Cauchy mutation are combined to find the optimal path with the designed cost function. The proposed IMRFO not only is effective to find an optimal path, but also improves the global search accuracy, speed and stability in path planning of MRFO. To compare the performance of mobile robot path planning based on IMRFO, MRFO and Particle Swarm Optimization (PSO), simulation experiments are implemented by MATLAB 2019a in the Windows 10 operating system. With 25 times independent experiments, it is shown that IMRFO acquires an effective path and its performance is superior to MRFO and PSO with the evaluation of statistical results and convergence curves. The IMRFO provides a powerful tool in dealing with mobile robot path planning problem.

1A2-4-4 
An Improved Q-Learning Algorithm for Mobile Robot Path Planning

Junkui WANG, Kaoru HIROTA, Xiangdong WU, Yaping DAI, and Zhiyang JIA

An improved Q-learning (IQL) combined with Q-function initialization method, action selection strategy and reasonable reward function is proposed. The Q-function is initialized using the inverse Euclidean distance to update the entries in the Q-table efficiently and reduce the episodes significantly. And an improved ε-greedy exploration which combined with Boltzmann exploration and heuristic searching strategies is provided to reduce the search space and limit the variation range of orientation angle. In addition, a reasonable reward distribution is designed to reduce the torque generated by the robot in planning. Experiments undertaken on grid environment confirm that the Q-table obtained by the proposed initialization method outperforms both the classical initialization and the method based on neural network. The experimental results show that the proposed algorithm can keep a safe distance from obstacles and requires less torque in comparison to its classical counterpart.

1A2-4-5 
Equilibrium-Breaking Artificial Potential Field for Path Planning of Mobile Robot by Randomly Generating Proximal Repulsion Sources

Zilong YU and Bin XIN

Artificial potential field (APF) is a very attractive method for the path planning of mobile robots, borrowing ideas from physics in the from of introducing attraction and repulsion between robots and their targets/environments. However, APF suffers from the issue of local minimum due to the equilibrium between attraction and repulsion. In this paper, a novel equilibrium-breaking APF is proposed to drive the robot to break away from deadlock caused by the equilibrium. The new APF will detect the equilibrium and randomly generate virtual repulsive sources near any local minimum according to a Gaussian model. The repulsive force changes the resultant potential field and gets rid of the local minimum point. This method can achieve effective path planning for a mobile robot in an unknown environment. Experimental simulations validated the validity and feasibility of the APF-based path planning method.

2M1-1-1 
The Trajectory Generation via Deep Neural Network for Hypersonic Vehicle

Tirui MA, Yongzhi SHENG, and Xiaoping ZHANG

In this paper, we propose a method based on deep neural networks to generate the reentry trajectory of hypersonic vehicle. There are some complex constraints in the reentry trajectory optimization problem, such as dynamic pressure constraint and no-fly zones for threat avoidance. It’s hard to present the real-time controls by traditional optimal control methods, due to the limitation of onboard computer. In order to obtain the training data, we adopt gauss pseudo-spectral method (GPM) to compute the optimal trajectories, which contain the optimal states and controls. Deep neural network (DNN) are trained with the optimal states and controls. The reentry trajectory driven by the trained DNN is efficient, and the trained neural networks can present the real-time optimal controls, according to the reentry states of hypersonic vehicle.

2M1-1-2 
Terminal Angle Constraint Guidance Law Through Time-to-Go Estimation Using Neural Network

Dan WANG, Yongzhi SHENG, and Xiaoping ZHANG

In this paper, a terminal guidance law for the impact-angle and the terminal angle of attack (AoA) constraint is proposed. The main feature of this guidance law is that the neural network is utilized to fit the relationship between flight states and time-to-go, and the information of altitude is used in guidance law to achieve terminal-angle constraint. The algorithm firstly is designed using the SDRE method to constrain the impact-angle and the terminal AoA using the function of altitude and time-to-go respectively. Then, the neural network is trained to achieve the impact-angle constraint and the terminal AoA constraint simultaneously using the information of altitude. The simulation results show that the proposed guidance law is effective for terminal angle constraining and a comparative study is carried out to show that it has shorter computing time, flatter trajectories, and faster convergence rate of AoA than the SDRE guidance law designed by the function of time-to-go.

2M1-1-3 
The Formation Transformation of the UUVs Based on the Leader-Follower Strategy and the Binary-Tree Structure

Junxi ZHANG, Bin XIN, Qing WANG, and Yun QU

This paper addresses the approaches for the formation transformation of the Unmanned Underwater Vehicles (UUVs), which only rely on sensor measurement. Due to the limited measuring range, the leader-follower strategy and the binary-tree structure are employed to design the desired formation and the switching process. In order to achieve the desired formation, the controller for each follower UUV is designed with the integral sliding mode method, which makes the follower surge velocity and heading angle achieve consistent with the leader UUV. The simulation results demonstrate the formation transformation can be achieved, and the approaches adopted in the paper are effective.

2M1-1-4 
Distributed Finite-Time Practical Consensus of Second-Order Multi-Agent System by Event-Triggered Strategy

Wei HAN, Pingli LU, Shaopan ZHANG, Haikuo LIU, Changkun DU, and Chao LI

This paper considers the finite-time and distributed event-triggered consensus control for multi-agent system (MAS) with second-order dynamics under a undirected graph. A novel event-triggered function, which depends only on local information, is adopted. By employing a novel distributed event-triggered controller for each agent, practical finite-time consensus (FTC) of MAS can be achieved without continuous communication. A sufficient condition is obtained through the analysis by Lyapunov theory and graph theory. In addition, the feasibility of the proposed approach is guaranteed by comprehensive theoretical demonstration of practical FTC and analysis of the Zeno behavior. Finally, a numerical simulation is provided to illustrate the correctness of theoretical results.

2M1-1-5 
Distributed Dynamic Event-Triggered Control for Leader-Following Multi-Agent Systems

Shaopan ZHANG, Ning DONG, Haikuo LIU, Changkun DU, and Pingli LU

This paper is devoted to address the distributed dynamic event-triggered consensus problem for leader-following multi-agent systems with general linear dynamics. Compared with the existing event-triggered scheme, the proposed dynamic event-triggered scheme involves internal dynamic variables, which play an important role to reduce the inter-agent communications. Besides, distributed event-triggered protocol is designed, which only uses the information from neighboring agents instead of global information. It is shown that the state of followers converge to the leader, the triggering number of the agents reduce significantly and the feasibility of the proposed control scheme is verified by excluding the Zeno behavior. Finally, a numerical simulation is provided to illustrate the correctness of theoretical results.

2M1-2-1 
Drum Water Level Control System of Sintering HRSG Based on Modified ADRC Controller

Yuheng WANG, Fanwei MENG, and Zhenzhong LUO

In this paper, a modified adaptive disturbance rejection control (ADRC) method was proposed and applied in sintering heat recovery steam generator (HRSG) drum water level control system to overcome the strong disturbance effect. The system adopted three-element cascade control scheme, which employed PID controller in the inner loop to rapidly eliminate the feed-water disturbance and the modified ADRC controller in the outer loop to overcome disturbances from flow rate of steam and perturbations of boiler-drum water level. With the proposed control scheme, the system can keep stable and maintain satisfied control result under the influence of different disturbance signals. Simulation results show that, compared with the traditional PID-PID and ADRC-PID cascade control, the modified ADRC-PID control strategy has better dynamic regulation performance and robust quality.

2M1-2-2 
Prediction Method of Multi-Injection Pressure Fluctuation of Diesel Engine Based on Recurrent Neural Network Model

Zhe ZUO, Yu ZHANG, and Meng DU

In the high-pressure common rail fuel injection system, the pressure fluctuation caused by single injection in one cycle makes the injection pressure unstable, which reduces the control accuracy and makes it difficult to determine the circulating fuel supply. In recent years, with the development of deep learning, it provides a new idea for the prediction of pressure fluctuation. The general time series model and neural network model are difficult to meet the accuracy requirements. This paper proposes an algorithm: using LSTM and Seq2Seq as the infrastructure, introducing the Global Attention Mechanism, and optimizing by using Probabilistic Teacher Forcing, Adam and other strategies. The performance of the new model in different target injection pressure, different injection duration and different pre-main injection time interval is analyzed.

2M1-2-3 
A Pressure Fluctuation Prediction Algorithm for High Pressure Common Rail System Based on CNN

Zhe ZUO, Kuichen QUAN, and Meng DU

In this paper, aiming at the pressure fluctuation problem of high-pressure common rail system, based on the convolution neural network of WaveNet structure, a pressure fluctuation prediction algorithm is proposed. Compared with other pressure fluctuation algorithms, deep learning is applied in this algorithm. The effectiveness of the proposed algorithm is approved in the experiment in different target rated injection pressure, different injection interval and other working conditions. The results of the experiment show that the average accuracy of the prediction algorithm is about 99.1%, and meets the demand. The accuracy of the prediction the pressure fluctuation from the injection interval to the starting point of the main injection is higher, and the accuracy of the prediction of the pressure fluctuation and its attenuation process slightly decreases after the main injection duration.

2M1-2-4 
A Novel Toolbox for Bearing Fault Detection Based on PCC and Residual Blocks

Chengkun LI, Yufan LIN, Yujing LIU, and Qi GAO

Bearing is one of the essential components of mechanical systems, bearing fault detection is of great importance in bearing production and system fault diagnosis. In this paper, a novel toolbox for bearing fault detection using the bearing vibration signals is proposed. Two baseline models are included: 1. Baseline for Feature Engineering Based Method, which consists of three steps: time-frequency feature extraction, Pearson Correlation Coefficient (PCC) reduction and classification. 2. Baseline for Deep Learning Based Methods: a powerful deep neural network model consists of Convolutional Blocks and Residual Blocks. In the paper, the experimental results show that the methods in our toolbox are sufficiently robust to produce results with accuracy between 98% and 100%.

2M1-2-5 
Design of Positive Control of Three-Phase Asynchronous Motor Based on MCGS and PLC

Dawei ZHANG, Zhida LIU, and Qian ZHAO

With the continuous development of industrial construction, PLC, touch screen and other automation technologies are more and more widely used in industrial production. In the production line, the positive and negative rotation control of the motor is also essential. In this paper, we realize the positive rotation control based on the double interlock of ac contactor and button. Our team studied the design of positive and negative rotation control of MCGS and PLC three-phase asynchronous motor. With the help of Siemens S7_200 series, the control function is realized. Touch screen MCGS realizes configuration software function, in the project of positive rotation of three-phase asynchronous motor controlled by MCGS and PLC, hardware configuration and software configuration must be completed simultaneously.

2M1-3-1 
COVID-19 Risk Assessment Through Multiple Face Mask Detection Using MobileNetV2 DNN

Marielet GUILLERMO, Athena Rosz Ann PASCUA, Robert Kerwin C. BILLONES, Edwin SYBINGCO, Alexis FILLONE, and Elmer P. DADIOS

The current pandemic caused by COVID-19 outbreak is continuously putting lives into peril. It has already spread to over 200 countries, some of which are still fighting heavily to survive. This study aims to promote the importance of disease control and preventive measures such as the use of face masks in crowded places. An active face masks detection and monitoring system can help authorities to identify people who might be vulnerable to infectious diseases such as coronavirus. The problem in strategic planning such as allocation of personnel in high risk areas, due to lack of reliable and prompt means to identify COVID-19 cases, is giving much headache to Filipino community especially to the Philippine Government. An artificial neural network-based system capable of detecting if people in the crowd are wearing face masks, will be discussed comprehensively in this paper. The implementation of the study resulted to 99% in all training and testing key parameters.

2M1-3-2 
Two-Stream Graph Convolutional Networks for 2D Skeleton-Based Fall Detection

Yan LIU, Yuelin DENG, Chen JIA, Yanru YANG, Ruonan WANG, and Chi LI

The vision-based fall detection solutions play more and more significant role in the field of elder care. By reducing waiting time for rescue, life is saved. To improve the performance of fall detection, we propose a 2D skeleton-based fall detection method relying on the graph convolutional networks in this paper. The method is designed to a two-steam structure. Both the Cartesian coordinate and the polar coordinate are used to represent the skeleton feature of human body. The detection process to action sequence is accomplished by the fusion of two-stream of spatial temporal graph convolutional networks. To enhance the detection effect, we extend the scale of training dataset by converting the public 3D skeleton to 2D skeleton. The experimental results demonstrate that the performance of our method exceeds baseline method on both the benchmark NTU-RGB dataset and the proposed dataset.

2M1-3-3 
Image Processing for Functional Transillumination Imaging of Animal Body Using Near-Infrared Light

Koichi SHIMIZU

With a near-infrared (NIR) light, one can get a transillumination image of a living body. The transillumination images are seriously blurred by the light scattering at body tissue. A fundamental study has been conducted to visualize the functional change inside a living biological body. In this study, a technique was developed to visualize the attenuation change occurred in a diffuse scattering medium. Transillumination images are obtained before and after the physiological change. By taking the ratio of the transmitted intensities of these two images, one can obtain the spatial distribution of attenuation change while suppressing the effect of scattering. The effectiveness of this principle was verified in experiments. To examine the applicability of this principle to a biological body, localized physiological changes were made in the mouse abdomen and the rat brain. The hypoxia in one of the mouse kidneys was visualized selectively from another normal kidney. The local increase in the blood volume was visualized in the somatosensory area of a rat brain when its forelimb was electrically stimulated. The blood increase was detected in a symmetrical position with respect to the sagittal plane, when the forelimb of the opposite side was stimulated. Through these experiments, it was found that the changes in the tissue oxygenation and the blood volume could be detected noninvasively and that they are visualized in the transillumination images using the NIR light. In transillumination imaging, the physiological analysis has been limited with a few wavelengths. To advance this technique, we applied a hyperspectral camera to achieve the spectroscopy at each pixel of a transillumination image. The distribution of oxygen saturation was successfully visualized in the transillumination image of an adult hand. However, the transmitted light through animal body is generally weak and the image is greatly contaminated with noise. Moreover, the wavelength-scanning mechanism of the hyperspectral camera makes typical streak noise. To solve these problems, we analyzed the nature of the noise in Fourier domain, and constructed a new noise filter for image processing. In combination with the denoising filter using deep neural network, we succeeded to obtain clear transillumination image which can be used for the multi-spectral analysis of physiological functions.

2M1-3-4 
A Novel Virtual Reality Rehabilitation System for Upper-Extremity Amputee by Fusion Analysis of sEMG and Depth Image Data

Naifu JIANG, Jinwei XUE, Yanjuan GENG, Han CUI, and Guanglin LI

Upper-limb myoelectric prosthesis is commonly-used to improve the life quality of upper-extremity amputees. Before the use of this prosthesis, the rehabilitation training program under virtual reality environment can help the amputees to enhance the muscle strength of the residual arm and get used to the prosthesis earlier. Nevertheless, the current programs often focused on the training of hand motion instead of both the hand motion and arm motion. Thus, this study proposed a novel virtual reality rehabilitation system for upper-extremity amputee by fusion analysis of the surface electromyography (sEMG) and depth image data. Firstly, based on four time-domain features extracted from eight-channel sEMG signals, we applied support vector machine (SVM) algorithm to recognize the hand motion classes (hand-open or hand-close). Secondly, based on color and depth images, the arm motion could be recognized and tracked by using image processing algorithm. Finally, by fusion analysis of the recognition results, we designed two kinds of rehabilitation programs. It preliminarily showed good performance results on recognition of participant’s (able-bodied subject’s and amputee’s) hand motion and arm motion. This rehabilitation system will help to improve the efficiency of upper-limb myoelectric prosthesis.

2M1-3-5 
Predicting Dengue Outbreaks Using Local Weather Factors and the North Atlantic Oscillation Index

Guohun ZHU, Ying JIANG, and Zejiao CHEN

Climate is an important contributing factor in the outbreak and spread of dengue fever because it significantly affects the density and distribution of the mosquitoes carrying dengue virus. Dengue forecasting models based solely on local weather factors have had limited success. This paper proposes a novel dengue fever outbreak prediction model based on local weather factors and the multivariate North Atlantic Oscillation (NAO) index. The local weather data includes weekly temperature, precipitation, and humidity in San Juan, Puerto Rico from 1990 to 2013. Pacific/North American pattern (PNA) and NAO combined with the local weather are forwarded into a Support Vector Machine (SVM) to predict dengue outbreak. Statistical analysis shows that the outbreak of dengue cases has a strong negative-correlation with NAO indices and a strong positive-correlation with local temperatures and humidity. Classification results show that the accuracy of the dengue outbreak can only achieve 53.6% using weather data nine weeks in advance. However, combining local weather factors and NAO data in 15-week ahead, the model can predict dengue outbreak with 77.1% accuracy.

2A1-1-1 
Automatic Segmentation of Liver Tumor Focus Region Based on CT Image

Mohan JIA, Zhongjian DAI, Zhiyang JIA, and Yaping DAI

In view of the small proportion of liver tumor in abdominal CT images, an automatic segmentation algorithm of liver tumor CT image based on U-net is proposed in this paper. In order to enlarge the size of the target area, image processing was carried out before image input into the segmentation network. And a hybrid loss function combining Dice Loss and Cross Entropy Loss is used to improve the accuracy and stability of the training network. Finally, three evaluation criteria, Dice similarity coefficient, Recall and Precision, are adopted to evaluate the performance of tumor segmentation. The results of experiments show that this method can use less data to complete the end-to-end training and obtain good training effect.

2A1-1-2 
AHOBO: Frailty Care Robot for Elderly and its Application for Blood Pressure Measurement

Yoichi YAMAZAKI, Masayuki ISHII, and Takahiro ITO

A robot support system for elderly people is proposed to care their own frailty, where a support robot “AHOBO” instructs elderly on the right blood pressure measurement, and gives them advice to improve their health with the visualized measurement data. Two types of evaluation are performed on 16 subjects (from their 20’s to 60’s). It is confirmed that there is no influence of robotic blood pressure measurement support on blood pressure readings.

2A1-1-3 
Ensemble Learning-Enabled Type 1/Type 2 Diabetes Classification Through Flash Glucose Monitoring Data

Yicun LIU, Wei LIU, Xiaoling CAI, Rui ZHANG, Zhe AN, Dawei SHI, and Linong JI

Clinically, diabetes type diagnosis is difficult and inaccurate by relying on biochemical indicators and doctors’ experience. In order to improve the efficiency and accuracy of diabetic diagnosis, we propose a data-driven diabetes type classification algorithm, called ensemble-based linear discriminant analysis (LDA) algorithm. Furthermore, supported by the Department of Endocrinology and Peking University People’s Hospital, a dataset containing blood glucose records of 113 diabetic patients is created for model training and testing. In our algorithm, to reduce the data redundancy and increase the category discrimination, a data preprocessing method composed of data reorganization and downsampling is presented, in which the classification accuracy is significantly improved by 22% on average. The LDA method is adopted to build a generalized base model. Then an ensemble model is formed by integrating the LDA and the data preprocessing methods utilizing stepwise functions. In the experiment, compared with five machine learning algorithms and their combination with the Adaboost algorithm, our proposed algorithm obtains the highest sensitive value (0.8182) in diabetes type classification and the largest F-Measure, Matthews correlation coefficient score (81.82%, 72.73%). The model achieves satisfactory classification performance in terms of the percentage of classification accuracy which is 87.88%.

2A1-1-4 
Non-Invasive Glycosylated Hemoglobin Monitoring Using Artificial Neural Network and Optimized SVM

Marife A. ROSALES, Krizshel Benaly A. CABRAL, Erika Bianca D. CASTILLO, Mark Paolo L. Dela CRUZ, Aileen F. VILLAMONTE, Maria Gemel B. PALCONIT, and Elmer P. DADIOS

The study aims to develop a non-invasive system using sensors and machine learning algorithms to monitor the Glycosylated Hemoglobin (HbA1c) level. The device incorporates a breath analyzer with sensors to determine the amount of humidity, temperature and acetone that a person has exhaled. Artificial Neural Network (ANN), a supervised machine learning algorithm will be used to evaluate and to correlate the measured acetone level to HbA1c level. Based on the result of the neural network regression, temperature, humidity, sensor voltage and sensor resistance are strongly correlated to glycosylated hemoglobin level. The predicted HbA1c level will be classified further into three (3) categories. To classify the 3 categories of HbA1c levels, hyperparameter optimization using support vector machine (SVM) algorithm was performed. Linear kernel function of SVM was the best kernel function based on the results of training with 98.8% accuracy. The trained model was tested also using new or unseen samples. Based on the testing results, the system is 100% accurate.

2A1-1-5 
A Denoising Method for Ultrasound Images

Min CUI

Compared with other medical image diagnosis, ultrasound diagnosis has the advantages of simplicity and safety. Therefore ultrasound diagnosis has become one of the important methods of medical clinical diagnosis. However, when structural size of human tissue is near to wavelengths of incident ultrasound, ultrasound beam is scattered, and scattered echoes with different phases influence with each other to generate Speckle-Noise, which brings great difficulties to the post-processing of ultrasound image. Aiming at medical ultrasound image of rheumatoid arthritis, one improved algorithm is proposed, and the denoising effect of different filters is compared according to the evaluation index.

2A1-2-1 
Robust Visual Tracking via Hierarchical Representation

Zihao DING, Xuqian REN, Chunlei SONG, and Jianhua XU

Sparse representation has been applied to visual tracking by solving the target templates’ representation coecient accurately. The robust tracking algorithm needs to construct an appropriate object representation model. However, the existing trackers’ representation model is designed without regarding the relationships between templates. In this paper, we propose a novel Hierarchical Visual Tracking (HVT) algorithm, that thoroughly investigated the dictionary’s internal structure. The dictionary of templates with a grouping structure is designed in our HVT tracker to study the relationship between target templates. Furthermore, a novel sparse representation model is constructed based on the Hierarchical Lasso model. Quantitative evaluations on challenging benchmark data sets demonstrate that the proposed HVT algorithm performs favorably against several state-of-the-art methods. Experimental results verify the robustness of the proposed HVT algorithm.

2A1-2-2 
A Flower Classification Approach with MobileNetV2 and Transfer Learning

Wenxin DAI, Yaping DAI, Kaoru HIROTA, and Zhiyang JIA

In order to promote the use of mobile or embedded vision applications to realize flower recognition in daily life, a flower classification model named N-MobileNetV2 is proposed based on MobileNetV2 and transfer learning. The proposed flower classification approach reduces training time and space, which has good robustness and generalization performance. It is a faster way to train a deep convolutional neural network with a smaller dataset and limited computational resources. Experimental results show that the transfer learning model based on the strategy of training all layers performs best, with a classification accuracy of 96%. However, the classification accuracy of the training without transfer learning is only 25% under the same number of iterations.

2A1-2-3 
Lettuce Leaf Necrotic and Chlorotic Surface Defect Assessment Using Recurrent Neural Network Optimized by Electromagnetism-Like Mechanism

Ronnie S. CONCEPCION II and Elmer P. DADIOS

Detection of plant stress is crucial to improve cultivation management. This study presents a non-destructive solution in detecting lettuce crop health and quantitatively assessing the necrotic and chlorotic leaf surface defects due to drought-based senescence. A total of 210 matured lettuce images were collected over a week of water stressed testing using digital camera. Crop quality was classified into healthy and defective based on its canopy visual appearance using deep transfer image networks in which InceptionV3 bested other networks with accuracy of 97.321%. Necrotic and chlorotic regions of defective canopy were separately segmented using CIELab color space thresholding and extracted with color, texture, and morphological properties. Hybrid neighborhood component analysis and ReliefF confirmed that texture features are highly significant than colors for this problem. Artificial neurons on the three hidden layers of recurrent neural network (RNN) were fine-tuned using electromagnetism-like mechanism (EM). Combined EM-RNN exhibited the best R2 performances of 0.9796 and 0.9565 in predicting necrotic and chlorotic surface defect percentages respectively. Necrosis has faster spread factor of 45.3419% than chlorosis in weekly basis per canopy. This developed comprehensive model of InceptionV3-EM-RNN is an objective, reliable and quantitative approach in providing quality assessment on leaf surface defect phenotyping.

2A1-2-4 
Dual ASPP for Lightweight Semantic Segmentation on High-Resolution Image

Dongpeng XIAO, Meiling WANG, Lin ZHAO, and Siyuan CHEN

In recent years, the efficient and lightweight convolutional neural networks (CNNs) such as ShuffleNet and MobileNet, have been widely applied in the field of image classification. But in image semantic segmentation, challenges remain a lot. Although many network models perform well in semantic segmentation tasks, most of them contain large parameters and suffer from high computational complexity. In this paper, we explored the application of lightweight CNNs and atrous spatial pyramid pooling (ASPP) module in semantic segmentation. In the model, MobileNetV2 / MobileNetV3 were chosen as encoders and a new segmentation head named Dual ASPP was proposed as decoder, which was an improved version of DeepLabV3+. By this method, the amount of parameters can be compressed from 47.73M to 1.03M, and the computation amounts can be reduced from 458.5G to 26.1G accordingly. Besides, while testing on the high-resolution (1024 × 2048) Cityscapes datasets, the accuracy of the proposed lightweight model is significantly improved up to 2%.

2A1-2-5 
Vision-Based Automatic Archery Target Reporting System

Jiarong DU, Ru LAI, and Jian LI

Archery is a classic sport with a long history. Most competitions adopt manual target reporting system, which is time consuming, and less efficiency in data processing. This paper presents a vision-based automatic target reporting system for archery through OpenCV. By using the improved corner detection algorithm to eliminate unnecessary parts of the image and combine with the template-matching method, the accuracy of the obtained corner coordinates has been increased. On the other hand, by analyzing various complex cases that might occur in archery competitions, a newly developed judging and calculation method for arrow tip has been employed that makes the system more complete and adaptive. The designed automatic target reporting system has the advantages of convenient installation, low cost and efficient which can effectively overcome the shortcomings of manual target reporting system in the past. The proposed system has been tested by images in different situations, and the results show that the target accuracy rate reaches more than 97%. It could be applied to other sport competitions.

2A1-3-1 
Video Anomaly Detection Using Cycle-Consistent Adversarial Networks

Zheyi FAN, Mengjie CUI, Di WU, Yu SONG, and Zhiwen LIU

Anomaly detection is a challenging work in the area of intelligent video surveillance. It aims to identify abnormal events from monitoring videos. The main challenge of this task is the ambiguity of anomaly definition. In recent years, many researchers exploit hand-crafted features to detect abnormal events, and all these methods follow a two-stage learning strategy, including feature extraction and model establishment. In this paper, we propose an end-to-end anomaly detection framework using cycle-consistent adversarial networks. In the training phase, the representation of regularity is learned from normal video frames and corresponding optical flow images. Our networks are trained with only normal frames, so our model is sensitive to abnormal behavior in abnormal frames. At testing time, the abnormal areas will have larger reconstruction errors than the normal. We can detect abnormal behaviors according to the error between the reconstructed frame and the original frame through a reasonable threshold. Experimental results in challenging datasets show that our method surpasses state-of-the-art methods.

2A1-3-2 
Gait Recognition Based on Dual-Channel Dynamic and Static Fusion Network

Wenjing ZHANG, Xiaoyan ZHAO, Zhaohui ZHANG, and Tianyao ZHANG

As a biometric identification method that can be realized under a distance, gait recognition has a broad application scope. The majority of existing gait recognition methods utilize the Gait Energy Image (GEI) for feature extraction. However, GEI ignores the dynamic information of gait, which causes the recognition effect to be significantly affected by variations in carrying objects and clothes. A novel strategy is proposed in this paper to overcome this limitation based on the dual-channel dynamic and static fusion network. The proposed method trains a neural network to achieve feature extraction. The static features are initially extracted from GEI, followed by the extraction of dynamic features from the image sequence. The static and dynamic features are then fused for classification with the nearest neighbor classifier. The application of the proposed method on the CASIA-B dataset presents higher recognition accuracy than the conventional gait recognition method.

2A1-3-3 
A Two-Stream Bi-Directional LSTM Network for Automatic Counting and Localizing Repetitive Actions

Longsheng WEI, Yuyang YE, Xuefu YU, and Dapeng LUO

Latent information includes geological information and drilling progress can be inferred by how long it takes to drill pipes and the number of pipes which are used to drill in a coal mine well site. Since deep learning has made great achievements in the field of computer vision recent years. This work aims at developing a deep learning model for automatic localization when the drilling process happened and how many times it has happened through surveillance videos. We use frames as appearance stream and pixel trajectories of a frame as motion stream in which the displacement values corresponding to a moving scene point are at the same spatial position across several frames. The two-stream CNN is followed by a bi-directional Long Short-Term Memory (LSTM) layer. Instead of predict a video-level classification, our model output a frame-level classification sequence for temporal localization and counting drilling process. We train and test our model using our own dataset which straightly come from coal mine well site which is relatively small compared to common temporal action detection dataset. Our model performance well at test and opens the possibility to validate an automatic methodology for the automatic localization and counting of repetitive actions in industrial environment based on machine-learning algorithm.

2A1-3-4 
An Abnormal Behavior Detection Method for Elderly People at Home

Zhipeng CHENG, Kaoru HIROTA, Yaping DAI, and Zhiyang JIA

With the aggravation of the aging problem and empty-nester problem, caring for the health problems of the elderly has become an important field. Compared with some available behavior detection technologies, an abnormal behavior detection method for elderly people at home with Kinect v2 is proposed. This method obtains skeleton sequences with Kinect v2 and constructs a graph convolutional network to classify actions. The skeleton sequences are treated as a graph and separated into 4 partition graphs as the input of a spatial-temporal graph convolutional network (ST-GCN). The effectiveness of this method is verified on filtered NTU RGB+D datasets including 10 actions. The accuracy rate of our method achieves 90.33%, and is 4.98% higher than traditional ST-GCN.

2A1-3-5 
Real-Time Hand Detection Based on YOLOv3

Yihui XIE, Zhongjian DAI, Zhiyang JIA, and Yaping DAI

In this paper, a rapid detection method of human hand is proposed, which can determine the position of human hand in real time. Since YOLOv3 shows good accuracy in the field of object detection, a detection method based on YOLOv3 is proposed. Because of the larger model after training and the high demand for the computing power of the platform, the model is difficult to be applied to embedded devices. In order to solve the problem of large model, a new network slimming method is proposed in this paper. The method combines channel pruning and layer pruning to compress the model, and finally the precision is picked up by fine-tuning. The sparsity training step for the model is required before the pruning operation and this step can be applied to any typical CNNs or fully connected networks. The experimental results demonstrate that the mAP (Mean Average Precision) reaches 0.76, and the inference speed reaches 9.0 ms. Therefore, the proposed algorithm is capable of realtime detection.

2A1-3-6 
Cycle Skeleton Structure for Occluded Multi-Person 2D Pose Estimation

Longsheng WEI, Xuefu YU, Yuyang YE, and Dapeng LUO

There are two main pipelines in multi-person pose estimation. Compared to top-down approaches, bottom-up approaches save more computational cost in inference phase, but get lower accuracy for the final prediction result. Openpose is the first bottom-up approach and makes great progress in bottom-up field. However, this approach has room for improvement in both speed and accuracy. In this paper we modify the encoding method that uses only one heatmap to represent one connection to increase speed of inference step. And for tackling the isolated human parts problem caused by occlusion, we propose a new skeleton structure called Cycle Skeleton Structures for assembling step. In network structure, we use Hourglass module to extract multi-scale features at same time. In our experiment, we got accuracy improved on the subset of COCO validation dataset meanwhile speed up the runtime.

Organizers & Co-organizers



Co-sponsors