A novel approach for detecting small targets of wind turbine blade surface damage, particularly minute defects within intricate scenes depicted in low-resolution images, is introduced through the utilization of an enhanced YOLO-v7 algorithm. The method incorporates the Ghost-Shuffle Convolution (GSConv) to improve detection accuracy without compromising inference speed, tailored to the specific characteristics of wind turbine blade defects. Furthermore, the integration of the Simple Attention Mechanism (SimAM) within the ELAN structure aims to enhance the identification of smaller wind turbine blades. To expedite model convergence, the Edge Intersection over Union (EIoU) is adopted as the edge loss function. Experimental findings demonstrate that the refined algorithm attains an average accuracy of 78.7%, surpassing the original YOLO-v7 algorithm by 4.2%. Moreover, it achieves a recognition speed of 105.1 frames per second, facilitating more effective detection of wind turbine blade defects while meeting real-time target detection requisites.
