A programmable UAV-based measurement system for wind turbine surface defect detection: the WB-YOLO method

Wind turbine surface defect detection is critical for ensuring operational safety and efficient maintenance of wind energy systems. Although unmanned aerial vehicle (UAV)-based inspection offers a promising alternative to manual methods, it faces challenges such as environmental interference and poor recognition of small-scale defects. To address these issues, this paper proposes WB-YOLO, an efficient detection model based on an improved YOLO architecture. An additional P2 detection head is introduced to enhance small-target feature representation, and an Inner-SIoU loss function is adopted to improve bounding box regression accuracy. The model is integrated into a programmable UAV-based vision measurement system with a metamaterial-inspired structural design, forming a flexible and low-cost inspection platform. Experiments on a self-constructed wind turbine surface defect dataset demonstrate that WB-YOLO outperforms representative YOLO- and DETR-based methods, achieving improvements of 1.9% in Precision, 2.5% in Recall, 2.1% in mAP@50, and 2.3% in mAP@50–95, while maintaining an inference time of 3.2 ms (312 FPS). Validation in simulated wind farm environments further confirms stable detection performance under challenging lighting conditions.