Defect Detection Method Based on Improved YOLOv5s Model

Due to the problems of low detection accuracy and inaccurate positioning in the process of rail surface defect detection, this study introduced YOLOv5s into the SE module (Squeeze-and-Excitation) to increase the small amount of calculation to improve the sensitivity of the YOLOv5s model to the rail surface defect detection channel features, and then effectively improve the defect detection recognition ability and improve the problem of small defects missed detection. In addition, we used the $\alpha$-CIOU loss function instead of the GIOU loss function as the bounding box loss function to improve the accuracy of bounding box localization. The results showed that the proposed model has an average accuracy AP of 92.8%, a detection speed of 32 FPS on GPU for rail surface defect images, and a model memory occupation of 34.8 Mb. In this paper, the improved algorithm based on YOLOv5s can realize fast and accurate identification on the railroad. Compared with the YOLOv5s algorithm for rail surface defect detection, it has obvious advantages.