Research on Foreign Object Intrusion Detection for Railway Tracks Utilizing Risk Assessment and YOLO Detection

Addressing the limitation of existing object detection methods in accurately distinguishing the hazard level posed by foreign objects on railway tracks, this study proposes an intrusion warning approach for foreign object detection and risk assessment on railway surfaces. This method integrates MobileNetv3 with Transformer to detect foreign objects on railway tracks, constructing a novel backbone feature extraction network, MobileNetV3-CATr, aimed at reducing model complexity. Furthermore, a BiFPN-Lite module is introduced to fuse more target features without increasing complexity, with YOLO Head utilized to output the type information of foreign objects on the track surface. An improved track detection method is adopted toextact the railway tracks, using least squares to establish a track linear equation and delineate risk level zones. Key points of foreign objects within the image positions are selected to evaluate the intrusion risk level. Verification of detection accuracy for foreign object intrusions on railway tracks is conducted on a self-constructed image dataset. The results demonstrate that the proposed detection model can effectively identify foreign object intrusions on tracks, mitigating issues such as missed and false alarms. It achieves a 3.7% improvement in mean average precision (mAP) compared to the baseline model. Additionally, the refined Unet semantic segmentation network $A_{MIoU}$ attains an accuracy of 90.53%, representing a 2.65% enhancement over its predecessor, with better extracted track edge integrity. The proposed risk assessment method determines different intrusion risk levels based on the location of the intruding objects, thereby assisting relevant personnel in adopting appropriate actions to enhance the safe operation of trains.