Research on Subway Axle Box Bolt Localization Method Based on Improved YOLOv8s Algorithm

To address the issues of low positioning accuracy, missed detections, and false alarms in traditional network models for subway axle box bolt localization, a subway axle box bolt localization detection model based on the improved YOLOv8s algorithm is proposed. In the YOLOv8s model, the Slice-based Efficient Inference (SAHI) module is incorporated, and the SimAM attention mechanism is introduced into the C2f module of the YOLOv8s model. Experimental results show that the improved model achieved a 5.5% increase in mean average precision (mAP) and a 6.0% increase in recall compared to the original YOLOv8s model on a self-built dataset. Particularly, the precision of BOLT-E, which experiences significant occlusions and reflections in the dataset, improved by 9.2% compared to the original model. In conclusion, the improved model demonstrates significantly enhanced detection accuracy for subway axle box bolts compared to the original YOLOv8s model, showcasing practical value for real-world applications.