A UAV Aerial Image-Based Detection Algorithm for Wind Turbine Blade Defects

This study proposes an enhanced YOLOv11 framework for wind turbine blade defect detection, which achieves an mAP@0.5 of 0.905 on a self-constructed wind turbine blade defect dataset, representing a 2.1% improvement over the baseline model. The architecture incorporates three key improvements: replacing the C3k2 module with the C3k2_SCSA module to enhance the ability to perceive subtle defects; replacing all convolutional layers with depthwise separable convolutions (DSConv) except for the first conv, which optimizes computational efficiency while well preserving detection accuracy; and replacing the standard concat with Concat_BiFPN to improve multi-scale feature fusion performance. These improvements result in the model achieving a Precision of 0.941, with the parameter count optimized from 9.4MB to 8.6MB and the FLOPs reduced from 21.6G to 20.5G. The framework demonstrates excellent performance in wind turbine blade defect detection tasks, achieving an effective balance between detection accuracy and model efficiency, and provides a practical computer vision solution for wind turbine blade maintenance and inspection.