A Lightweight and Enhanced YOLO11-Based Method for Small Object Surface Defect Detection

To address the challenges of high false detection, missed detection, and inefficiency in small target defect detection in complex backgrounds, this paper proposes an improved surface defect detection method based on the YOLO11 algorithm. The method introduces the Efficient Multi-scale Attention and CMUNeXt Block module into the YOLO11 backbone to enhance the model's ability to capture global context and improve feature expression in complex backgrounds. A Large Separable Kernel Attention module is incorporated at the end of the backbone network to optimize feature extraction for small objects and reduce computational complexity. Additionally, a Ghost Convolution operation is applied during the feature extraction phase to replace standard convolution and improve both computational efficiency and feature representation. Finally, a new regression loss function is proposed to solve the problem of gradient disappearance in small object detection and the problem of insufficient optimization weight in the detection process, and optimize the performance of the model in small object detection. Experimental results show a 3% improvement in mean Average Precision and a 17.7% reduction in parameters, achieving a better balance between detection accuracy and computational efficiency.