A UFOAttention-YOLOv11-Based Method for Surface Defect Detection of Wind Turbine Blades in Complex Inspection Scenarios

Abstract

In wind turbine blade inspection images, defects such as cracks, dirt and peeling often have small scales, weak boundaries, low contrast and obvious background texture interference. The original YOLOv11 is still unstable in recognizing such weak-feature defects. In this paper, a UFOAttention module is introduced after the initial convolution layer of YOLOv11 to construct a UFOAttention-YOLOv11 wind turbine blade surface defect detection model, and shallow spatial correlation enhancement is used to improve the feature representation of defect regions. Experiments are carried out on a dataset containing seven types of blade defects. The results show that UFOAttention-YOLOv11 achieves an mAP@0.5 of 0.895, which is 0.017 higher than the original YOLOv11, and an mAP@0.5:0.95 of 0.682, which is 0.057 higher. The model improves recognition accuracy for multiple defect categories, including crack, deformity, dirt and peeling, indicating that the method can effectively enhance the representation of weak-boundary and irregular defects in complex blade inspection scenarios.