A GAN-Based Anomaly Detection Method for Isoelectric Line in High-Speed Railway
Authors:
- Yang LyuYLYang Lyu
- Zhiwei HanZHZhiwei Han
- Junping ZhongJZJunping Zhong
- Changjiang LiCLChangjiang Li
- Zhigang LiuZLZhigang Liu
- 1 May 2019
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
Abstract
The isoelectric line is an important component of the catenary support device in high-speed railway. Isoelectric line is prone to failure during the long-term operation of vehicles. Traditional image processing methods and deep convolution neural networks (DCNNs) classifier cannot obtain good anomaly detection results due to the diverse shapes of isoelectric lines and the lack of fault samples. In this paper, a novel approach is proposed to detect the faulty states of isoelectric lines based on generative adversarial networks (GANs). First, the faster R-CNN (ResNet101) is adopted to extract the isoelectric line features and localize the image areas of isoelectric lines accurately. Second, the generative adversarial representation of isoelectric lines from the first stage is acquired by a deep convolutional generative adversarial network (DCGAN). Finally, an anomaly rating criterion is proposed to distinguish the faulty states of isoelectric line. Experiments and comparisons show that the proposed method can judge faulty isoelectric line correctly. Furthermore, compared with the current typical anomaly detection methods, the proposed method has better performance and robustness for the isoelectric line anomaly detection.Keywords
This publication has 12 references indexed in Scilit:
- Automatic Defect Detection of Fasteners on the Catenary Support Device Using Deep Convolutional Neural NetworkIEEE Transactions on Instrumentation and Measurement, 2017
- Detection and monitoring system of the pantograph-catenary in high-speed railway (6C)2017
- Unpaired Image-to-Image Translation Using Cycle-Consistent Adversarial Networks2017
- Image-to-Image Translation with Conditional Adversarial Networks2017
- Unsupervised Anomaly Detection with Generative Adversarial Networks to Guide Marker Discovery2017
- Deep Residual Learning for Image Recognition2016
- You Only Look Once: Unified, Real-Time Object Detection2016
- Automatic Fastener Classification and Defect Detection in Vision-Based Railway Inspection SystemsIEEE Transactions on Instrumentation and Measurement, 2013
- Theoretical foundation of the intersecting cortical model and its use for change detection of aircraft, cars, and nuclear explosion testsSignal Processing, 2004
- The EM/MPM algorithm for segmentation of textured images: analysis and further experimental resultsIEEE Transactions on Image Processing, 2000