A novel edge crop method and enhanced YOLOv5 for efficient wind turbine blade damage detection.

Accurately and rapidly detecting damage to wind turbine blades is critical for ensuring the safe operation of wind turbines. Current deep learning-based detection methods predominantly employ the gathered blade images directly for damage detection. However, due to the slender geometry of wind turbine blades, non-blade background information accounts for a considerable proportion of the captured images with complex background features, affecting the detection of blade damage. To address this challenge, we propose a novel edge cropping method combined with an enhanced YOLOv5s network for detecting damage in wind turbine blades, termed Edge Crop and Enhanced YOLOv5 (EC-EY). The edge cropping method adaptively modifies the cropping stride by the edge features of both sides of the blade, thereby procuring image content that predominantly encompasses the blade region. This procedure effectively mitigates the interference from complex background features and augments the utilization of image pixels. Furthermore, the enhanced YOLOv5 network incorporates the global attention mechanism into the head section of the network and substitutes the original SPPF module with an attention-based intra-scale feature interaction module. The EC-EY aims to improve the detection accuracy for small and variable-shape damages in wind turbine blades. EC-EY achieved excellent performance on a dataset of wind turbine blade damage collected in western Inner Mongolia. Notably, the edge cropping method significantly improves the accuracy of wind turbine blade damage detection.