Abstract
Wind turbine blades are critical components of wind power systems, and early accurate detection of surface defects is essential for ensuring operational safety. However, automated detection faces significant challenges due to drastic illumination variations, complex background interference, and wide-ranging defect scales in real wind farm environments. This paper proposes RCS-YOLOv8, a multi-module collaborative optimization model for wind turbine blade defect detection. The model introduces RepViT as the backbone network, integrating local convolutional modeling with global context perception to enhance robustness against complex backgrounds. A Channel-Spatial Cooperative Attention Module (CAFM) is designed and embedded in the Neck layer, which performs dual-dimensional adaptive feature recalibration to focus on critical defect regions. SlimNeck with VoVGSCSP is adopted to reconstruct the feature pyramid network, strengthening multi-scale fusion between shallow localization information and deep semantic features. WIoU replaces CIoU as the bounding box regression loss, employing a dynamic non-monotonic focusing mechanism to improve recall for hard-to-detect defects. On a wind turbine blade dataset containing seven defect categories, RCS-YOLOv8 achieves 90.3%, outperforming the baseline YOLOv8n by 5.8% and the second-best method RT-DETR by 3.0% The recall reaches 87.5%, the highest among all compared models. Ablation studies validate the effectiveness of each module, and visualization analysis confirms that the model accurately focuses on defect regions while effectively suppressing background interference. The proposed method provides a reliable technical solution for automated wind turbine blade defect detection.