Rail wheel tread defect detection using improved YOLOv3

Abstract

The rail wheel is the key device of the rail vehicle. When the vehicle running conditions worsen or emergency braking, the rail train will have severe sliding friction and collision on the track, resulting in scratches, peeling, and scratches on the wheel tread. The traditional wheel defect detection technology can not realize automatic classification and accurate defect positioning. Thus, an improved YOLOv3 framework for rail wheel surface defect detection is developed. The classification detection of four types of wheel tread defects is realized, and the average accuracy of mean average precision (mAP) is 0.92. Compared with the other detection methods, the experimental results show that the improved YOLOv3 model not only maintains the detection speed but also improves the detection accuracy and meets the real-time requirement.