Segmentation of Concrete Cracks based on Improved YOLO11 Lightweight Model

Cracks are one of the typical diseases of concrete structures, and it is of great engineering significance to achieve efficient detection and accurate segmentation of concrete surface cracks for structural health monitoring and preventive maintenance. Although deep learning technology has pushed crack detection into the intelligent stage in recent years, the existing methods have the problems of high computational complexity, insufficient multi-scale feature extraction capability, and poor segmentation integrity in complex backgrounds. In this study, a lightweight model named YOLO11-DMS is proposed to achieve simultaneous identification and segmentation of concrete cracks. The model enhances the multi-scale feature extraction capability of the backbone network by fusing the cavity window residual module with the multi-scale cavity attention mechanism; the Slim-neck neck network is introduced to optimise the feature transfer efficiency. To evaluate the method and test its generalisation ability, experiments are conducted on two public datasets, and the results show that the improved model improves the mAP by $2.4\%$ in the detection task and $5.4\%$ in the segmentation task, which are both better than the mainstream lightweight models, effectively balancing the accuracy with the efficiency and improving the performance of crack detection and segmentation.