Improved YOLOv8-Based Foreign Object Detection Algorithm for Coal Mine Conveyor Belts

Abstract

To address the poor detection performance of existing models for coal mine conveyor belts under low illumination, an improved algorithm based on YOLOv8n (You Only Look Once version 8 nano) is proposed to reduce false and missed detections of slender metallic objects, small targets, and objects with background-similar textures. The Coordinate Attention (CA) mechanism is integrated into the backbone to enhance positional and channel feature extraction; the Efficient Channel Attention (ECA) mechanism is embedded into the neck to improve sensitivity to textural features; and the Complete Intersection over Union (CIoU) loss is replaced with the Focal-EIoU (Focal Efficient Intersection over Union) loss to improve regression and alleviate sample imbalance. Compared with baseline YOLOv8n, the proposed algorithm improves Recall by 4.6% and mAP@50 (mean Average Precision at IoU 0.5) by 1.7%; ablation experiments verify the effectiveness and complementarity of each module. This work provides a practical solution for intelligent safety monitoring in coal mine conveyor systems, advancing industrial vision-based foreign object detection.