Defect detection in greige fabric based on improved YOLO11

To address the challenges of low detection accuracy and high miss rate in defect recognition of greige fabric, this study systematically optimized and improved the YOLO11 model to enhance detection precision and speed in complex textile inspection scenarios. Firstly, a C3K2-DCNv2 module was introduced into the backbone network, leveraging deformable convolutions to improve the model’s adaptability to target deformation and complex backgrounds. Secondly, in the neck network, the DySample and ASF modules were integrated to construct ASF-DySample module, achieved dynamic sampling and spatially adaptive fusion of multi-scale features, effectively preserving detail information, significantly improving small object detection accuracy and feature representation capability. Finally, the MultiSEAMHead module was adopted in the detection head, which further strengthened the target representation capability through multi-branch semantic decoupling and contextual aggregation strategies. Experimental results demonstrated that the improved DAM-YOLO model achieved a 2.3% increase in mAP@0.5 compared to the original YOLO11, validating the effectiveness and practical value of the proposed method in object detection tasks.