YOLOv11-ASE: fabric defect detection algorithm based on improved YOLOv11

To improve fabric defect detection under small defect sizes, diverse shapes, and complex backgrounds, the YOLOv11-ASE object detection network based on YOLOv11 is proposed. This detection network retains the original backbone network of YOLOv11n while achieving multi-dimensional performance improvements through three core modules: first, ASFYOLO strengthens small-defect and texture representation through scale sequence feature fusion, triple feature encoder, and channel-position mechanism; second, the Semantic and Detail Infusion module improves cross-level fusion by combining semantic and detailed features through bidirectional sampling and attention weighting; third, the Efficient Up-Convolution Block introduces learnable upsampling with depthwise separable convolution, reducing computational cost while preserving geometric information. On the evaluated fabric defect dataset, YOLOv11-ASE achieves a mAP50 of 91.8%, outperforming the YOLOv11n baseline (89.7%) by 2.1 percentage points, while maintaining an inference speed of 112.0 FPS. These results demonstrate that YOLOv11-ASE offers an effective lightweight solution for real-time, low-cost fabric defect detection.