OSCD-YOLO: A Surface Crack Detection Method for Open-pit Mines in Complex Scenes

This study proposes a crack detection model for complex scenes, named OSCD-YOLO, aimed at addressing the challenges of complex backgrounds and multi-scale features in surface crack detection in open-pit mining. The model integrates partial convolution to reduce redundant computations; multidimensional collaborative attention enhances the crack recognition ability in complex scenarios; and content-aware feature re-adjustment during upsampling improves the detection accuracy of small cracks. Experimental results on multiple datasets demonstrate that OSCD-YOLO outperforms existing models such as YOLOv8 and YOLO11 in both detection accuracy and inference speed. Particularly, on a custom-built dataset, OSCD-YOLO achieved an average precision improvement of 7.28%, reaching 91.7%. The experimental results indicate that OSCD-YOLO exhibits exceptional robustness, efficiency, and generalization capability in the complex open-pit mining environment, providing an efficient and real-time applicable crack detection solution.