Visual Detection Algorithm of Foreign Object Intrusion in High-Speed Railway Traction Substation Based on Patch Clustering Learning

Since high-speed railway traction substation is usually built in an open area, the intrusion of foreign objects will cause hidden trouble to the operation safety of the substation, so it is of great significance and practical value to study the foreign object intrusion detection method in traction substation. Therefore, this paper proposes a patch-based clustering learning foreign invasion of visual detection algorithm. Firstly, the global region image of the high-speed railway traction substation is divided into patches, and then features are extracted from the segmented image patches based on the MobileNetV2 network. Then, the image patches are clustered according to these features by the K-means method and the classification results are obtained. Finally, the Patch-SVDD method is used to train the encoder and classifier to detect and locate foreign object intrusion. Based on the real traction substation data, the optimal input size and sampling step size of the image patch were obtained by selecting segmentation image patches of different sizes and sampling step sizes, and the validity and accuracy of the proposed method were verified. The detection accuracy of foreign object intrusion was 96.6%, and the positioning accuracy was 98.8%.