Abstract: Aiming at the problem that existing RFID tags were easy to fall off and damage, resulting in the loss of vehicle number, this paper proposed an urban rail train number recognition system based on image processing. The paper used industrial cameras to take pictures of train numbers on the side of urban rail trains, and then used Speed Up Robust Features(SURF) algorithm and Transform Invariant Low-rank Textures(TILT) method to locate, correct and segment the captured pictures, used the VGG-16 network model to recognize the segmented train number characters. The test results show that the system has good robustness and high recognition accuracy, and can meet the requirements of urban rail train number acquisition.