The detection of damage in wind turbine blades is of paramount importance in mitigating safety risks and economic losses. To address the detection challenges posed by intricate backgrounds and subtle damage, this study introduces the YOLO-Fan model, based on YOLOv10. Initially, the RELAN4 module is employed as the core feature extraction component of the model, leveraging multi-scale feature fusion to effectively reduce background interference. Next, to overcome the challenge of detecting small-target damage, the C2f-SENetV2 module is designed, incorporating an adaptive channel attention mechanism that dynamically adjusts feature weights, thereby enhancing the model's ability to detect minute defects. Finally, the GSConv module is introduced to optimize the boundary localization precision for multi-scale damage by utilizing channel mixing. Experimental results demonstrate that the YOLO-Fan model achieves a detection time of only 3.8 ms for fan blade damage, significantly outperforming existing models, with a mAP[50:95] score of 77.4%, reflecting a 3.6% improvement over the original model. This satisfies the stringent requirements for both high accuracy and real-time performance in damage detection in wind turbine blades.