To address the issues of low sensitivity to multi-scale targets, severe complex background interference, and insufficient localization accuracy in wind turbine blade defect detection, an improved YOLOv10n defect detection algorithm is proposed. In the SPPF module, a large-kernel attention mechanism (LSKA) is embedded to construct a multi-scale feature enhancement network, effectively solving the problem of false detections caused by reflections on the blade surface. A channel-priority attention mechanism (CPCA) is designed to enhance the response to small defect features by establishing a cross-dimensional interaction channel selection model. The original loss function of the model is replaced with the Wise-MPDIoU loss function, which integrates dynamic weight adjustment strategies and shape-aware constraints. Experimental results show that on a dataset of wind turbine blades collected by drones, the improved algorithm achieves an mAP@0.5 of 84.1%, representing a 2.9 percentage point improvement over the baseline model. This provides a reliable technical solution for the intelligent operation and maintenance of wind power equipment.