To address the challenges posed by the complex background of wind turbine blade damage images and the multi-scale distribution of features—which often result in the loss of small target features and difficulties in precise localization—a novel wind turbine blade damage detection model, YOLO-Nan, is proposed. This model is based on an improved version of YOLOv11. First, the model enhances its ability to extract small target features in complex scenarios by integrating the innovative (Softmax-based Attention Module) SBAM into the backbone network. Second, Use the (Receptive-Field Attention Convolution) RFAConv enhances the network's performance with large convolutional kernels while incurring minimal increases in computational cost and parameter count. Third, Feature fusion through the Generalized Feature Pyramid Network (GFPN), enabling a greater focus on small target features within deep features during multi-scale feature fusion. Finally, a lightweight upsampling module, DySample, is incorporated to enhance the model's ability to capture fine details. The experimental findings indicate that, compared to the YOLOv11 model, the YOLO-Nan model achieves improvements of 1.4%, 3.5%, 3.1%, and 1.8% in precision, recall, mAP0.5, and mAP [0.5:0.95], respectively. These results confirm the feasibility and effectiveness of YOLO-Nan.