SDW-YOLO: An Improved Steel Surface Defect Detection Algorithm Based on YOLOv11

The current methods for steel surface defect detection have the disadvantages of low accuracy, poor real-time performance, and limited detection of small targets. To address the above problems, an improved detection model SDW-YOLO is proposed.Firstly, we introduce SKANet (Selective Kernel Attention Network), which enhances feature extraction by multiscale attention. Secondly, the C3k2-DBB module is designed, which is an enhancement of the original C3k2 module, and the incorporation of DBB (Diverse Branch Block) reduces the number of parameters and enhances the feature extraction capability. Finally, the WIoU loss function is introduced in order to dynamically optimise the loss function and improve the positioning accuracy. The experimental results show that SDWYOLO achieves $\mathbf{7 9. 8 \%}$ mean average precision mean (mAP) on the NEU-DET dataset and improves $3 \%$ compared to the baseline YOLOv11 while reducing the number of parameters by 0.4 m. Ablation and comparison experiments validate the effectiveness and accuracy of the model for the detection of steel surface defects.