The wind turbine blade surface defect detection task faces challenges such as large blade sizes, wide detection areas, difficulty in image feature extraction, and image fog. These issues make defect detection extremely difficult. In particular, when the model is deployed on a drone for real-time detection, it needs to optimize both accuracy and efficiency under limited computational resources. To address these problems, we propose a YOLO-based detection model with feature enhancement, fusion, and dual attention mechanisms (FFDA-YOLO). FFDA-YOLO consists of three innovative lightweight modules: the Dual Attention Module (DAM), Multi-scale Feature Fusion Enhancement Module (MFFEM), and Detection Head Module (DHM). These three modules improve the global correlation ability across channels and spaces, as well as the network’s multi-scale feature fusion and enhancement capabilities. They also enhance the model’s adaptability to changes in object shape and scale. These modules strengthen the weak feature representation of defect targets and suppress confusing background elements. Finally, we validated the detection performance of FFDA-YOLO using the public wind turbine blade surface defect detection dataset (wind turbine-v2). The accuracy of FFDA-YOLO reached 0.903 (based on mAP50), surpassing several benchmark models and classical defect detection methods. Additionally, ablation experiments demonstrated the robustness of FFDA-YOLO.