Research on algorithm for high-speed train bolt loosening detection based on improved YOLOv8n

To tackle the safety hazards posed by bolt loosening during high-speed train operations, alongside addressing the high computational cost and slow inference speed associated with existing detection algorithms, a lightweight bolt detection system based on YOLOv8n has been proposed. This system integrates several enhancements: SimAM attention to boost feature representation while mitigating overfitting, DSConv for superior memory efficiency and accelerated processing, and MPDIoU loss to improve bounding box accuracy and model stability. Experimental evaluations conducted on a simulated undercarriage platform have demonstrated significant improvements, including a 34.34% increase in inference speed, a 27.75% reduction in inference time, and a 3.7% enhancement in mean Average Precision (mAP). Notably, the model showcases effective identification and localization of three distinct types of bolts, achieving this with low rates of false positives and missed detections. These results highlight the system's robust performance and generalizability, making it a promising solution for enhancing the safety and maintenance efficiency of high-speed trains.