The operating mileage of high-speed railroad is growing rapidly, and the detection of electric trackside equipment still mainly relies on manual visual inspection, which has the problems of shortage of personnel, low efficiency, low accuracy, and great influence by the environment. This study proposes a high-speed railroad trackside equipment identification and automatic detection method based on multi-scale convolutional neural network, which aims to solve the problems of low efficiency and poor accuracy of traditional manual inspection. The study adopts a modular-designed high-speed industrial camera for data acquisition, and constructs a dataset containing a total of 3274 pictures of five types of equipment, namely, choke transformer voltage box, cable diverter box, cable terminal box, transformer box and signaling machine. Based on the Faster-RCNN framework, ResNet101 is selected as the backbone network, and the trackside equipment detection model is designed by feature pyramid network, Rol pooling and improved loss function. The experimental results show that the model achieves an average accuracy of 97.65% in the detection of five types of trackside equipment, and the processing speed is 21.42 frames/second. Compared with other detection algorithms, this model improves the recognition accuracy, and the introduction of the feature pyramid network improves the average accuracy of the model by 4.17%. In addition, the detection accuracy is significantly improved by increasing the candidate region size to {128,256,512}. The proposed multi-scale convolutional neural network method provides an effective solution for the automated detection of trackside equipment in high-speed railroads, and provides technical support to ensure the safety of railroad operation.
