Wind turbine blades endure persistent operational stresses, including aerodynamic loads, cyclic fatigue, and environmental corrosion, leading to structural defects such as cracks, perforations, and surface delamination. These defects impair aerodynamic performance, reduce energy output, and may propagate over time, increasing the risk of blade fracture. To find these defects in time, a novel defect detection method for wind turbine blades based on improved YOLOv8 is proposed in this paper. First, standard convolutional layers are systematically replaced with spatial pyramid depthwise convolution modules in the backbone network to improve the recognition of slender and microscale crack defects, and the proposed method enhances defect detection accuracy through the mixed local channel attention mechanism. Additionally, the conventional spatial pyramid pooling (SPP) structure is further redesigned as the Spatial Pyramid Pooling Cross Stage Partial Connections (SPPCSPC) architecture by integrating SPP with cross-stage partial connection (CSPC), thereby enhancing multi-scale feature representation. Finally, the complete intersection over union (IoU) loss is optimized into Pixels-IoU formulation for bounding box regression, which significantly improves detection performance for small targets while ensuring regression accuracy and stability. Experimental results on a wind turbine blade crack defect dataset demonstrate that the proposed method achieves improvements in precision of 9.3% and in mean average precision of 7.3% compared with the baseline YOLOv8 model. These findings validate that the enhanced You Only Look Once v8 (YOLOv8) detection method exhibits superior detection effectiveness, accuracy, and reliability.