YOLOv8-WTDD: multi-scale defect detection algorithm for wind turbines

In addressing the challenges of wind turbine defect detection, such as different defect scales in UAV aerial photography, interference from different lighting conditions, and small-sized target defects leading to low detection accuracy and inaccurate localization, a YOLOv8-WTBB model based on YOLOv8 is proposed. Firstly, the Diverse Branch Block is designed to enhance multi-scale feature fusion capabilities. Next, the Receptive-Field Attention Convolution is introduced to focus on the spatial features of the receptive field, increasing the distinction between target features and the surrounding environment. Finally, introducing the Minimum Point Distance Intersection over the Union bounding box regression loss function notably improves localization accuracy in object detection and accelerates model convergence. Experimental results demonstrate that the proposed algorithm significantly outperforms the baseline network, with a 4.3% improvement in mean average precision, achieving 89.1%, and a 7.4% increase in mean average recall, reaching 84.8%.