Deep Learning-Based Detection of Catenary Support Component Defect and Fault in High-Speed Railways

Description

This book focuses on the deep learning technologies and their applications in the catenary detection of high-speed railways. As the only source of power for high-speed trains, the catenary's service performance directly affects the safe operation of high-speed railways. This book systematically shows the latest research results of catenary detection in high-speed railways, especially the detection of catenary support component defect and fault. Some methods or algorithms have been adopted in practical engineering. These methods or algorithms provide important references and help the researcher, scholar, and engineer on pantograph and catenary technology in high-speed railways. Unlike traditional detection methods of catenary support component based on image processing, some advanced methods in the deep learning field, including convolutional neural network, reinforcement learning, generative adversarial network, etc., are adopted and improved in this book. The main contents include the overview of catenary detection of electrified railways, the introduction of some advance of deep learning theories, catenary support components and their characteristics in high-speed railways, the image reprocessing of catenary support components, the positioning of catenary support components, the detection of defect and fault, the detection based on 3D point cloud, etc. Preface Contents Abbreviations 1 Overview of Catenary Detection of Electrified Railways 1.1 Introduction 1.2 Catenary Detection and Monitoring System 1.3 Catenary Component Detection Technologies 1.4 Catenary Parameter Detection Technologies 1.5 Summary References 2 Advance of Deep Learning 2.1 Introduction 2.2 Image Feature 2.2.1 Handcrafted Features 2.2.2 Deep CNN Feature 2.3 Deep Learning Frameworks in Computer Vision 2.3.1 Image Object Detection 2.3.2 Image Object Segmentation 2.3.3 Image Anomaly Detection 2.4 Other Deep Learning Approaches 2.4.1 Deep Autoencoder 2.4.2 Generative Adversarial Networks 2.4.3 Deep Reinforcement Learning 2.5 Summary References 3 Catenary Support Components and Their Characteristics in High-Speed Railways 3.1 Introduction 3.2 Catenary System 3.3 Catenary Characteristics 3.4 Summary References 4 Preprocessing of Catenary Support Components’ Images 4.1 Introduction 4.2 Catenary Image Denoise with LWBCTCS 4.2.1 Denoise Method LWBCTCS 4.2.2 Catenary Image Denoising Experiment and Analysis 4.3 Catenary Image Enhancement with Deep Learning 4.3.1 Catenary Image Enhancement Based on Zero-DCE Model 4.3.2 Catenary Image Enhancement Based on Retinex Theory and Generative Adversarial Networks 4.3.3 Catenary Image Enhancement Based on Reinforcement Learning 4.4 Summary References 5 Positioning of Catenary Support Components 5.1 Introduction 5.2 Simultaneous Positioning of CSCs with Variant Deep CNN Object Detection Frameworks 5.2.1 CSCDNET—CSC Positioning with Multiple Scale Feature Prediction 5.2.2 CSCNET—CSC Positioning with Unsupervised Coarse Image Classification 5.2.3 CSCSIN—CSCs Positioning with Position Relationship Learning 5.3 Positioning Refinement of CSCs with Deep Learning Techniques 5.3.1 Horizontal Positioning Box Refinement with Deep Reinforcement Learning 5.3.2 Obliqued Positioning Box Refinement with Generative Adversarial Networks 5.3.3 Segmentation Boundary Refinement with the CascadePSP Network 5.4 Summary References 6 Detection of Catenary Support Component Defect and Fault 6.1 Introduction 6.2 Messenger Wire Base Defect Detection and Evaluation Using the Wavelet Transform 6.2.1 Suspected Crack Region Extraction 6.2.2 Crack Detection of Messenger Wire Bases 6.2.3 Analysis of Experimental Results 6.2.4 Conclusions 6.3 Insulator Defect Detection and Evaluation Based on Autoencoder Networks (AE) 6.3.1 Insulator Piece Separation 6.3.2 Insulator Defect Extraction 6.3.3 Insulator Defect Evaluation 6.3.4 Analysis of Experimental Results 6.3.5 Conclusions 6.4 Isoelectric Line Defect Detection and Evaluation Based on Generative Adversarial Networks (GANs) 6.4.1 Generative Adversarial Representation and Fault Diagnosis of Isoelectric_Lines 6.4.2 Analysis of Experimental Results 6.4.3 Conclusions 6.5 Summary References 7 Detection of the Parameters of Catenary Support Devices Based on 3D Point Clouds 7.1 Introduction 7.2 Geometry Parameter Detection of Contact Wires with RANSAC 7.2.1 Acquisition and Preprocessing of the Catenary 3D Point Cloud Data 7.2.2 Detection and Extraction of Contact Wires 7.2.3 Detection of the Catenary Conductor Height and Stagger 7.2.4 Analysis of Experimental Results 7.2.5 Conclusions 7.3 Structure Parameter Detection with 3D Deep Segmentation Neural Networks 7.3.1 Cantilever Component Separation 7.3.2 Cantilever Structure Parameter Measurement 7.3.3 Analysis of Experimental Results 7.3.4 Conclusions 7.4 Summary References