Research on Object Detection Method of Four-Electric Equipment in Rail Transit Based on Yolo

Rail transit is a complex system, there will always be a variety of errors in the actual project to bring unnecessary losses. This research aims to increase users’ understanding of equipment by detecting the types of four-electric equipment in rail transit, so as to reduce the occurrence of misuse and omission of equipment by equipment users in the project. In this paper, we introduce dilated convolution in the Yolov3 network structure and use different dilation rates for different scales of network output, which can effectively improve the prediction effect of the network; in addition, we improve the loss function of the network to obtain better detection effect. Experiments show that the network detection speed is 53f/s, and the detection accuracy reaches 83.18%, which is 3.23% higher than the original Yolov3 network, which effectively improves the detection effect.