Abstract: The identification of risk sources along urban rail transit is an important method to prevent safety accidents caused by illegal operations. In order to solve the problems of low efficiency, high miss rate and high cost of traditional identification methods, this paper takes the UAV metro protection area inspection system as a platform, and uses convolutional neural network to identify the risk source of the image data collected by UAV. Firstly, the process of UAV image acquisition is introduced, and based on the original image data, multi-angle and multi-scale of risk source datasets are produced through data enhancement. Then, a method of convolutional neural network is used to establish risk source identification models to automatically identify and locate the risk source in the images collected by the UAV. Experiments show that the establishment of multi-angle and multi-scale risk source data sets further improves the recognition accuracy of the model, and this method has the advantages of high efficiency and low cost compared with traditional methods.

Key words: UAV, image data, risk source identification, convolutional neural network