Blade-YOLOv8:Improved YOLOv8 for Wind Turbine Blade Defect Detection

In the task of using drones to detect defects in wind turbine blades, there are challenges such as missed detections due to the small size of the defects and their random, varying scales. However, the existing model's detection speed is slow, rendering it unsuitable for deployment on edge devices. To address these problems, this paper introduces Blade-YOLOv8, a lightweight edge detection algorithm based on the YOLOv8 model. Firstly, the Global Multi-scale Fusion-Space Pooling Pyramid-Fast (GMF-SPPF) module is introduced, which captures global perspective information and mitigates the impact of varying scales on detection results. Secondly, a new C2f EMA structure, based on Partial Convolution (PConv) and Efficient Multi-Scale Attention (EMA) mechanisms, is employed in the backbone and neck networks to make the network lightweight while enhancing the detection ability for small targets and reducing the number of model parameters. Experimental results show that, compared to the original YOLOv8s model, the improved algorithm reduces the number of parameters by 21.6%, while increasing detection accuracy, specifically mAP50 and mAP50:95, by 4.3% and 7.5%, respectively. And the speed on an NVIDIA Jetson Orin Nano edge computing platform reaches 38 FPS. Consequently, the improved algorithm is well-suited for practical applications and deployment in real-world scenarios.