Abstract: At present, the missing detection of steel structure canopy bolts in high-speed railway passenger stations in China relies too much on manual visual inspection, which has great risk coefficient, high cost, low efficiency and high false detection rate. In order to solve this problem, this paper proposed a bolt missing detection system based on YOLO algorithm. The system used YOLO v4 convolution neural network to mark the steel structure canopy and catenary bolts collected on site, determined the number and size of anchor frames by K-means clustering algorithm, and used cutmix, mosaic and other data to enhance the operation, increase the diversity of training data, and avoid training over fitting. The trial results show that the category recognition accuracy of the system can reach more than 85%, and the recognition effect is good, which meets the requirements of real-time detection.