As a clean and renewable energy source, wind energy is increasingly receiving attention from countries worldwide. However, with the gradual increase of wind turbine assembly capacity, the operation and maintenance problems of wind turbines have become increasingly prominent. Wind turbine blades are one of the core components of wind turbines, accounting for about 15%-20% of the total cost of wind turbine operation and maintenance, which is directly related to the performance and benefit of wind turbines. To solve the problem of fan blade defect detection, we propose an improved object detection model, RSD-YOLO, based on YOLOv9 architecture. RSD-YOLO introduces the 4iRMB attention mechanism and extracts key information from the wind turbine blade images combined with the feature extraction module. In addition, we design SDI_Ghost feature fusion layer and DySample upsampling module instead of the traditional pyramid pooling layer and nearest neighbor interpolation module, respectively. The experimental results show that the proposed RSD-YOLO defect detection model shows better performance with \(94.52\%\) precision, \(93.55\%\) recall, \(93.73\%\) \(mAP\_0.5\) , \(74.63\%\) \(mAP\_0.5:0.95\) , and 0.94 F1-Confidence.