Artificial Intelligence and Robotics: 7th International Symposium, ISAIR 2022, Shanghai, China, October 21-23, 2022, Proceedings, Part II

Description

This two-volume set (CCIS 1700-1701) constitutes the refereed proceedings from the 7th International Symposium on Artificial Intelligence, ISAIR 2022, held in Shanghai, China, in October 2022. The 67 presented papers were thoroughly reviewed and selected from 285 submissions. The volumes present the state-of-the-art contributions on the cognitive intelligence, computer vision, multimedia, Internet of Things, robotics, and related applications. Preface Organization Contents – Part II Contents – Part I Brain Modeling for Surgical Training on the Basis of Unity 3D 1 Introduction 2 Tissue Deformation Modeling and Kinetic Equations 2.1 Biomechanical Properties of Brain Tissue 2.2 Mass-Spring Model 2.3 Physical Modeling of Brain Tissue 2.4 Kinetic Equations 3 Experimental Environment and Experimental Results 3.1 Soft and Hardware Environment 3.2 Experimental Platform 3.3 Experimental Simulation 4 Conclusion References Motion Saliency Detection Based on Drosophila Vision-Inspired Model 1 Introduction 2 Related Work 2.1 Drosophila Vision 2.2 Saliency Detection 2.3 Our Contributions 3 Methodology: Drosophila Vision-Inspired Model 3.1 Motion Pathway 3.2 Color Pathway 3.3 Central Brain 4 Experiment Results 4.1 Evaluation Metrics 4.2 Results on FBMS and DAVIS 4.3 Performance Discussion 5 Conclusions References Research on Matching Mechanism and Route Planning of Intercity Carpool 1 Overview 1.1 Introduction 1.2 Related Work 1.3 Study Content 1.4 Thesis Organization 2 Problem Definition and Model 3 Algorithm Design 4 Theorem Proof 5 Experiments 6 Conclusion Annex 1 Source Code (code available for review only) References Image Undistortion and Stereo Rectification Based on Central Ray-Pixel Models 1 Introduction 2 Related Work 2.1 Corner Detection 2.2 Pinhole Models 2.3 Ray-Pixel Models 2.4 Image Undistortion and Stereo Rectification 3 Algorithms and Pipeline 3.1 Projection and Unprojection 3.2 Bundle Adjustment 3.3 Image Undistortion 3.4 Stereo Rectification 3.5 Calibration Pipeline 4 Evaluation 4.1 Image Undistortion Results 4.2 Stereo Rectification Results 4.3 Sector Grid Results 4.4 3D Reconstruction Results 5 Evaluation References GGM-Net: Gradient Constraint on Multi-category Brain MRI Segmentation 1 Introduction 2 Methods 2.1 Image Acquisition and Preprocessing 2.2 Gradient-Guided Multi-category Segmentation Network 2.3 Training 3 Experiment Result 3.1 Evaluation Metric 3.2 Analysis and Presentation of Results 4 Conclusion References Linear Split Attention for Pavement Crack Detection 1 Introduction 2 Related Work 2.1 Traditional Methods 2.2 Deep Learning Based Methods 3 Proposed Method 3.1 Model Architecture 3.2 Linear Split Attention Module 3.3 Multi-scale Feature Fusion Module 4 Experimental Result and Analysis 4.1 Experimental Setting 5 Dataset Introduction 6 Evaluation Metrics 7 Evaluation on Image Datasets 8 Ablation Studies 9 Conclusions References Information Acquisition and Feature Extraction of Motor Imagery EEG 1 Introduction 2 EEG Signals and Signals Acquisition 3 Preprocessing of MI-EEG Signals 4 Features Extraction and Analyses 4.1 Time Domain Methods 4.2 Frequency Domain Methods 4.3 Time-Frequency Domain Methods 4.4 Spatial Domain Methods 4.5 Spatio-Temporal Domain Methods 4.6 Spatio-Spectral Domain Methods 5 Classification of MI EEG Signals 6 Conclusions References Lightweight 3D Point Cloud Classification Network 1 Introduction 2 Related Work 2.1 Hand Crafted Features 2.2 Point Cloud Classification 2.3 Deep Learning on Point Cloud 3 Lightweight 3D Point Cloud Classification Network 3.1 Background 3.2 Position Encoding 3.3 Framework of the Network 4 Experiment 4.1 Detail Experimental Setting 4.2 Results in ModelNet40 4.3 Results in ScanObjectNN 4.4 Ablation Study 5 Conclusion References Unsupervised Domain Adaptive Image Semantic Segmentation Based on Convolutional Fine-Grained Discriminant and Entropy Minimization 1 Introduction 2 Related Works 2.1 Semantic Segmentation 2.2 Domain Adaptation 2.3 Unsupervised Domain Adaptive Semantic Segmentation 3 Method 3.1 Semantic Segmentation 3.2 Entropy Minimization 3.3 Fusion and Adversarial Learning 4 Experiments 4.1 Datasets 4.2 Network Architecture 4.3 Implementation Details 4.4 Experimental Results 5 Conclusion References Ensemble of Classification and Matching Models with Alpha-Refine for UAV Tracking 1 Introduction 2 Related Work 2.1 Correlation-Filter-Based Methods 2.2 Deep-Learning-Based Methods 3 Proposed Algorithm 3.1 Overview 3.2 Discriminator 3.3 Redetection Module 4 Experiment 4.1 Dataset 4.2 Implementation Details 4.3 Results and Analysis 5 Conclusion References LayoutLM-Critic: Multimodal Language Model for Text Error Correction of Optical Character Recognition 1 Introduction 2 Approach 2.1 Local Optimum Criterion of OCR Text 2.2 Implementation of LayoutLM-Critic 3 Experiment 3.1 LayoutLM-Critic 3.2 OEC 4 Conclusion References Blind Image Deblurring Via Fast Local Extreme Intensity Prior 1 Introduction 2 LEP: Local Extreme Intensity Prior 2.1 Building Local Extreme Intensity Collect Model 2.2 Fast Local Extreme Intensity Prior 3 Proposed Blind Deblurring Model 3.1 Estimating Latent Image 3.2 Estimating Blur Kernel k 4 Experiment 4.1 Deblurring Images 4.2 Analysis and Discussion 5 Conclusions References STRDD: Scene Text Removal with Diffusion Probabilistic Models 1 Introduction 2 Related Work 2.1 Scene Text Removal 3 Background: SDE 4 Method 4.1 Overview 4.2 Training 4.3 Inference 5 Experiment 5.1 Datasets 5.2 Evaluation Metrics 5.3 Comparison with Other Scene Text Removal Methods 6 Conclusion References Geometry-Aware Network for Table Structure Recognition in Wild 1 Introduction 2 Related Work 2.1 Table Detection 2.2 Table Structure Recognition 2.3 Transformation Learning Networks 3 Proposal Method 3.1 Framework 3.2 Geometry-Aware Representation 3.3 Loss Function 4 Experiments 4.1 Datasets 4.2 Evaluation Metric 4.3 Evaluation Result 5 Conclusion References A Differential Evolution Algorithm with Adaptive Population Size Reduction Strategy 1 Introduction 2 The APRDE Algorithm 2.1 Variation Strategy 2.2 Parameter Adaptive Adjustment 2.3 Population Reduction Strategy 3 Numerical Experiment and Analysis 3.1 Experiments Setup 3.2 Comparison of Solution Accuracy and Stability 3.3 Comparison of Convergence Speed 4 Conclusion References A Semi-supervised Road Segmentation Method for Remote Sensing Image Based on SegFormer 1 Introduction 2 Our Method 2.1 Training Strategy 2.2 Unsupervised Module Design 3 Super Pixel Segmentation 3.1 SegFormer Network Structure 4 Experiment 4.1 Datasets 4.2 Experimental Setup 4.3 Unsupervised Module Experiment 4.4 Comparative Test 5 Conclusion References Cross-Layer Feature Attention Module for Multi-scale Object Detection 1 Introduction 2 The Proposed Method 2.1 Cross-Layer Feature Fusion 2.2 Feature Refinement 3 Experiments 3.1 Datasets and Settings 3.2 Comparison with State-of-the-Arts 3.3 Ablation Studies 4 Conclusion and Future Work References Interact-Pose Datasets for 2D Human Pose Estimation in Multi-person Interaction Scene 1 Introduction 2 Related Works 2.1 Datasets for 2D Human Pose Estimation 2.2 Multi-person Human Pose Estimation 3 Method 3.1 Interact-Pose Dataset 3.2 Multi-dataset Fusion Training 3.3 Data Augmentation Scheme 4 Experiments 4.1 Datasets 4.2 Training 5 Conclusion References Multimodal Breast Cancer Diagnosis Based on Multi-level Fusion Network 1 Introduction 2 Related Work 2.1 Disease Diagnosis Based on Manually Extracted Feature 2.2 Disease Diagnosis Based on Convolutional Neural Network (CNN) 2.3 Disease Diagnosis Based on Electronic Medical Records 2.4 Disease Diagnosis Based on Multimodality 3 Methods 3.1 Image Feature Representation 3.2 Text Feature Representation 3.3 Multimodal Fusion 3.4 Classification Prediction 4 Experiments 4.1 Dataset 4.2 Experimental Setup and Evaluation Metrics 4.3 Comparative Experiment 4.4 Model Analysis 5 Conclusion References Behavior Control of Cooperative Vehicle Infrastructure System in Container Terminals Based on Q-learning 1 Introduction 2 Design of Model 3 Design of Cooperative Vehicle Infrastructure System 4 Test Verification 4.1 Simulation Experiment 4.2 Outdoor Experiment 5 Conclusion References Backdoor Attack Against Deep Learning-Based Autonomous Driving with Fogging 1 Introduction 2 Related Work 3 Fogging Backdoor Attack 3.1 Problem Definition 3.2 Mathematical Modeling of Fogging 3.3 The Fogging Backdoor Attack Pipeline 4 Experiment 4.1 Experiment Setup 4.2 Effectiveness of Our Fogging Attack 5 Conclusion References A Study on Japanese Text Multi-classification with ALBERT-TextCNN 1 Introduction 2 Related Work 2.1 Solutions for Text Classification 2.2 Text Classification for Different Languages 3 Methodology 3.1 ALBERT-TextCNN Model 3.2 Tokenization 4 Experiments 4.1 Test Environment 4.2 Experiment Data 4.3 Evaluation Strategy 4.4 Experiment Design 4.5 Experiment Results 5 Conclusions and Future Work References Part Based Face Stylization via Multiple Generative Adversarial Networks 1 Introduction 2 Methodology 2.1 Improved Structure of Generative Adversarial Network CycleGAN 2.2 Network Design 3 Detailed Design Scheme 3.1 Data Preprocessing Module 3.2 Portrait Segmentation Module 3.3 Style Transfer Module 4 Experiment Results 5 Conclusion References Network Intrusion Detection Method Based on Optimized Multiclass Support Vector Machine 1 Introduction 2 Optimized Feature Selection Method 2.1 Eliminating Irrelevant Feature Based on ReleifF Algorithm 2.2 Eliminating Redundant Feature Based on Heuristic Search Rules Using Variable Similarity 3 Optimized Training Based on Data Block Method 4 Multiclass SVM for Intrusion Detection 5 Experiment and Results Analysis 5.1 Experiment Dataset 5.2 Experimental Results and Analysis 6 Conclusion References Face Recognition Based on Inverted Residual Network in Complex Environment of Mine 1 Introduction 2 Inverted Residual Face Recognition Algorithm 2.1 Analysis of Typical Face Recognition Framework 2.2 Activation Function Optimization 2.3 Fusion of Residual Networks 3 Experiment and Result Analysis 3.1 Experimental Setup 3.2 Performance Analysis 4 Conclusion References Automated Cobb Angle Measurement Using MVIE-Net Combined with Vertebral Segmentation and Landmarks Detection 1 Introduction 2 Method 2.1 Confidence Map of Landmarks 2.2 The Proposed MVIE-Net 2.3 Cobb Angle Measurement 3 Experimental Details 3.1 Dataset 3.2 Implement 3.3 Evaluation Metrics 4 Results and Analysis 4.1 Segmentation Results of MVIE-Net 4.2 Cobb Angle Measurement Results 5 Conclusion References Emotion-Sentence-DistilBERT: A Sentence-BERT-Based Distillation Model for Text Emotion Classification 1 Introduction 2 Related Works 3 Emotion-Sentence-DistilBERT 3.1 Model Structure 3.2 ESDBERT for Sentence Pair Emotion Classification Tasks 3.3 Distillation Method 4 Experiments and Results 4.1 Dataset 4.2 Evaluation Method 4.3 Results and Discussion 5 Concludes References Gait Recognition for Laboratory Safety Management Based on Human Body Pose Model 1 Introduction 2 Related Work 3 Human Body Pose (HBP) Model for Gait Recognition 3.1 Human Body Pose Estimation 3.2 Feature Extraction 3.3 Gait Recognition Network 4 Experiment 4.1 Dataset 4.2 Experimental Results 4.3 Discussion 5 Conclusion and Future Work References Use Active Learning to Construct Japanese Emoji Emotion Database 1 Introduction 2 Related Work 3 Constructing an Emotional Database Through Active Learning 3.1 Data Collection 3.2 Text Sentiment Classification Based on BERT 3.3 Introduction to Active Learning Algorithm 4 Experiment 5 Conclusions References Target-Driven Autonomous Robot Exploration in Mappless Indoor Environments Through Deep Reinforcement Learning 1 Introduction 2 Autonomous Exploration 2.1 Global Navigation 2.2 The Dynamics Model of Wheeled Robot 2.3 Local Navigation 3 Simulation and Evaluation 4 Conclusions References Background Subtraction Based on Visual Saliency 1 Introduction 2 Relationship Between Visual Saliency and Potential Movement Targets 2.1 Potential Moving Target Attracting Visual Saliency 2.2 Visual Saliency Region in the Background Frame 3 The Proposed Algorithm 3.1 Moving Target Perception 3.2 Moving Object Extraction 4 Experimental Results 4.1 Comparison on Background 4.2 Performance on CDnet2014 5 Conclusion References Image Rectification of Industrial Equipment Nameplate Based on Progressive Probabilistic Hough Transform 1 Introduction 2 Related Works 3 Methodology 3.1 Feature Transformation Based on Hough Transform 3.2 Corner Detection Based on Double Line Segment Interception 3.3 Nameplate Correction Based on Perspective Transformation 4 Experiment 4.1 Mining Equipment Nameplate Dataset 4.2 Comparative Experiment 5 Conclusions References Author Index