Lightweight building crack segmentation based on an improved YOLOv11-seg

To address the issues of insufficient accuracy and high model complexity in YOLOv11-seg for building crack segmentation, this paper introduces a lightweight crack segmentation algorithm based on an improved YOLOv11-seg. First, GhostConv modules are introduced into the backbone network to replace standard convolutions, to reduce parameters and computational cost. Then, DWConv (Depthwise Convolution) modules are introduced into the downsampling part of the neck network, which reduces computational cost while improving segmentation accuracy. Finally, the CA (Coordinate Attention) mechanism is embedded after multi-scale feature fusion to enhance the model’s ability to perceive crack characteristics. Experimental results show that, compared with YOLOv11n-seg on the Crack Segmentation dataset, the proposed model reduces Weight, Params, and FLOPs by 13.3%, 14.5%, and 9.4% respectively, while precision, mAP50, and mAP50:95 are improved by 2.5%, 3.4%, and 1.3% respectively. The proposed model achieves lightweighting while effectively improving the accuracy of building crack segmentation.