Research on Track Fasteners and Rails Surface Detection using Image Recognition and Data Management System Technology

This research uses image recognition to conduct track fastener and rail surface defect detection. To counteract the dataset imbalance, design an image manager system to change the image's size, brightness, flip, and rotation. Comparing original and enhanced datasets by the precision of three different image recognition, results indicate the improvement of YOLO V8 Nano and optimized YOLO V4. However, the YOLO V7 model has a lower precision rate but a more promising recall rate.