Aiming at the technical challenges of complex background interference and multi-scale defect detection in wind turbine industrial scenes, this study proposes an improved defect detection algorithm SMLA-YOLO based on YOLOv11. By adding the SWS attention mechanism to the EMO backbone network, the computational efficiency is balanced and the feature response of small defects is enhanced. By topologically fusing the C3K2 module with the attention network LANet, a feature extraction unit with cross-dimensional interaction ability is constructed. MSFE-PConv is proposed to improve the effect of sparse feature processing and improve the efficiency of spatial feature extraction in fan blade defect detection. Experiments show that on the self-built wind turbine blade defect dataset, the mAP0.5 of SMLA-YOLO reaches 96%, which is 3.1% higher than that of the benchmark model YOLOv11, and the mAP0.95 is 1.5% higher than that of YOLOv11, which effectively solves the balance between multi-scale defect detection accuracy and computational overhead in industrial scenes.