基于改进YOLOv5s的矿用输送带异物检测算法OA

Given the detection challenges posed by the monochrome conveyor-belt scene,the high length-to-width ratio of slender anchor-rod foreign objects,and the occlusion of large coal fragments,together with the real-time demands of industrial applications,an enhanced YOLOv5s algorithm for foreign object detection on mining conveyors was proposed.The C3_Faster network was incorporated to substitute the original C3 backbone network to compress model size,Triplet attention mechanism was embedded into the core feature extraction stage to reweight features along three directions of the characteristic pattern,and a Focaler-IoU loss function with linear interval mapping was adopted to boost accuracy.Comparative experiments show that,relative to the baseline,the improved YOLOv5s raises mean Average Precision(mAP)by 3.2%to 91.4%,trims model volume by 17.2%and parameter count by 17.5%,with a detection speed of 109.89 FPS.The improved YOLOv5s model delivers higher accuracy in foreign object detection and a lighter footprint,meeting the edge-deployment requirements for foreign object detection on coal mine conveyor belts.