采煤工作面隐患目标检测中YOLO11与Faster R-CNN的对比研究

Comparative Study of YOLO11 and Faster R-CNN for Hidden Danger Detection in Coal Mining Face

摘要: 针对煤矿井下采煤工作面裂缝、鼓包、坍塌等隐患检测中样本有限、特征微弱的问题，构建500张实景图像数据集，对比YOLO11与Faster R-CNN。实验表明，YOLO11最优mAP50达94.80%，精确率96.30%，召回率89.60%，参数量仅2.59 M，推理速度72 FPS，对细长裂缝等小目标检测完整精准；Faster R-CNN的mAP50为81.91%，对大尺度坍塌较好，但小目标漏检明显。YOLO11在精度与实时性上更具优势，适合井下部署。

Abstract: To address the issues of limited samples and weak features in detecting cracks, bulges, and collapses on underground coal mining faces, a dataset of 500 real-scene images is constructed to compare YOLO11 and Faster R-CNN. Results show that YOLO11 achieves a best mAP50 of 94.80%, precision 96.30%, recall 89.60%, with only 2.59 M parameters and 72 FPS, accurately detecting slender cracks and small hazards. Faster R-CNN obtains mAP50 of 81.91%, performing well on large collapses but missing small targets. YOLO11 is superior in accuracy and real-time performance, making it more suitable for underground deployment.