An Improved YOLOv11-based Algorithm for Foreign Object Detection on Coal Conveyor Belts

Addressing the challenge of insufficient foreign object detection accuracy on mine coal conveyor belts, prevalent in environments characterized by low illumination and high noise, this paper proposes an improved YOLOv11n-based foreign object detection algorithm for coal conveyor belts. The algorithm aims to achieve high precision and high recall in foreign object detection while maintaining real-time performance. Specific improvements include: optimizing the backbone network structure by replacing the original C2PSA module with a Self-Calibrated Spatial Attention (SCSA) module; combining the RFA-Conv module with the Squeeze-and-Excitation (SE) module to enhance the extraction of local target features; and incorporating the Focaler-IOU loss function in the detection head to achieve adaptive fusion of multi-scale features and focus training on hard samples. Experimental results demonstrate that, compared to a high-performance baseline model, the proposed improved model achieved further enhancements across key metrics. Specifically, Precision (P), Recall (R), mean Average Precision (mAP), and F1-score reached 0.910, 0.902, 0.895, and 0.906, respectively. Notably, there was a significant improvement in recall, successfully reducing the miss detection rate to below 9.8%. With a parameter count of 2.67M, the model effectively reduces false and missed detections with almost no increase in parameters, providing a more efficient, accurate, and lightweight technical solution for foreign object monitoring on mine coal conveyor belts.