Abstract: Rail damage refers to various states that occur during the use of rails, such as breakage, cracks, and other conditions that affect and limit the performance of rail use. Conducting damage detection on railway tracks is a necessary method for maintaining railway transportation safety. This paper proposes a rail crack detection method based on an improved YOLOv5 model to address the problems of current railway maintenance relying mostly on manual visual inspection and poor real-time fault detection. Meanwhile, in response to the shortcomings of existing neural network models in identifying rail damage and inspired by image change detection, a rail change detection method based on scale invariant feature transformation (SIFT) is proposed to achieve rail damage recognition. Experiments have shown that the improved YOLO model has a 4.6% increase in average mean accuracy compared to the original model, and has good application prospects. The SIFT based algorithm has high detection accuracy and can effectively identify rail damage areas, meeting practical engineering needs.

Key words: rail damage detection, YOLOv5, image change detection, SIFT