关键词: 无人机影像　, 浅埋深煤层开采　, 地表裂缝检测　, YOLO模型

Abstract: Shallow-buried coal seam mining causes severe surface subsidence and deformation,readily induces ground cracks,and increases the risks of spontaneous combustion of residual coal in goafs and water inrush during the rainy season,po sing serious threats to mine production safety.Currently,surface crack detection in mining areas mainly relies on manual in spection,which suffers from high labor intensity,poor accessibility,and low detection efficiency,making it difficult to meet the requirements of rapid monitoring in complex terrains.Based on UAV low-altitude remote sensing imagery,this study proposes an automated detection workflow for mining-induced ground cracks and achieves efficient identification using deep learning ob ject detection models.Taking the 12407 and 22603 working faces of the Huangyuchuan Coal Mine in Inner Mongolia as exam ples,high-resolution UAV images were collected.Four representative object detection models,including YOLOv8,YOLOv10, and the YOLOv9-series YOLOv9 and GELAN models,were selected for systematic comparison and performance evaluation. Through training and validation on a unified dataset,the performance of different models in terms of detection accuracy,recall, and inference speed was analyzed,and their applicability to mining-induced crack detection in shallow-buried coal seams was assessed.The results show that the YOLO series models perform well in detecting mining-induced surface cracks in this mining area,with mAP_0.5 exceeding 70.4%.Among them,the YOLOv10 model achieves the best balance between detection accura cy and speed for surface cracks,making it the optimal choice for automated detection of surface cracks in shallow-buried coal seam mining.These findings provide a scientific basis for UAV remote sensing-based automated detection of surface cracks in shallow-buried coal seam mining in arid and semi-arid regions such as Inner Mongolia,Shaanxi,and Xinjiang,and offer a practical foundationfor intelligentmining-induceddisastermonitoringandprevention.