Improved Fabric Defect Detection Algorithm of YOLOv11n

In the field of fabric defect detection, to address the challenges of complex textile textures and target multiplicity that lead to low detection accuracy, an improved fabric defect detection algorithm based on the YOLOv11n framework is proposed. A Residual Hybrid Attention Mechanism is implemented in the backbone network to enhance feature extraction capability in defect regions. The ADown downsampling method replaces conventional downsampling operations in both the backbone and neck layers, effectively reducing model parameters. A Multi-Head Fusion Dilated Detection Head is designed to strengthen feature information integration and resolve target multiplicity issues. The enhanced method achieves a 4.7% improvement in mean Average Precision while maintaining lightweight characteristics, with a detection speed reaching 106 frames per second.