Locking Latch Loosening Detection of Train Based on YOLO Framework

This study introduces a novel YOLO-based method enhanced with the Convolutional Block Attention Module (CBAM) for detecting loosening in electrical box cover latches, a crucial safety component in train operations. Utilizing a dataset of dynamic train images captured by side-mounted cameras, the proposed two-stage detection process effectively addresses challenges such as small target detection and high background noise. Our findings indicate that CBAM integration significantly improves the model's fault detection capabilities, achieving a recall rate of 0.992 for loosened latches and maintaining perfect precision (0.956) for normal latches, thereby substantially reducing false positives. This approach marks a significant advancement in computer vision applications for railway maintenance, enhancing the safety and reliability of rail transport.