Anomaly Detection of Filter Mesh Based on DifferNet and SSIM Algorithm

The power supply equipment of electric traction provides electricity for subway operation and produces a large amount of heat during operation, and the filter mesh on it plays a vital role in heat dissipation. However, in the long-term operation of the train, the filter mesh is easily blocked and entangled by foreign objects, causing heal accumulation and reducing the heat dissipation effect of the equipment. Therefore, it is very significant to realize the anomaly detection of the filler mesh of traction box. In this paper, We propose a computer vision-based anomaly detection method, which firstly adopts DifferNet based on normalizing flow to realize the qualitative analysis of anomaly detection, then combines with SSIM algorithm to achieve the localization of the anomaly area, and finally marks the anomaly area on the original image to accomplish the whole anomaly detection process.