Research On Real-time State Detection Of Pantograph Based On Image Processing

In today’s rapid development of railway industry,intelligent and digital real-time detection is becoming more and more important.Pantograph,as an important part of railway traction power supply system,plays the role of receiving current from the catenary,and its condition is directly related to the safety of railway operation.Research on the pantograph state detection algorithm with high detection accuracy and good real-time performance is of great significance to guarantee the safety of railway operation.In view of the disadvantages of current pantograph state detection methods,such as high cost,narrow application range,low detection accuracy and strong dependence on abnormal data,this thesis designs a real-time pantograph state detection method with wide versatility and no dependence on abnormal data based on image processing technology.Firstly,based on the keypoints object detection technology,a real-time detection method based on improved Center Net is designed to solve the problem of pantograph positioning.By adopting the Mobile Net backbone network,this method effectively reduces the amount of model calculation and speeds up the detection speed.At the same time,Feature Pyramid Networks(FPN)is used to effectively fuse contextual features,which makes up for the lack of feature extraction ability of Mobile Net and improves the detection accuracy.Comparative experiments show that the improved positioning method achieves the best trade-off between accuracy and speed.Subsequently,considering that the abnormal state of the pantograph is often reflected in structural changes,based on the semantic segmentation technology,a U-Net-based segmentation method for the pantograph structure was designed,and compared with the traditional segmentation technology.The experimental results show that,U-Net-based semantic segmentation method can accurately segment the pantograph structure,effectively eliminating the interference of background and other components on subsequent anomaly detection.Finally,in order to solve the shortcomings of the current pantograph state detection algorithm of narrow application range and strong dependence on abnormal data,based on unsupervised learning technology,this thesis designs a pantograph anomaly detection method based on Variational Auto-Encoder(VAE).This method only uses a variational autoencoder to model the normal data distribution,and then uses similarity calculation and labeling algorithms to detect anomalies.Experimental results show that the algorithm can detect the state of the pantograph with high accuracy.