Insulator Defect Detection in Power Grids Based on Improved YOLOv8

With the advancement of smart grid construction and the acceleration of digital transformation, the inspection of power grid line insulators is facing increasingly severe challenges. Traditional manual inspection methods, due to their large work-load, low efficiency, and high susceptibility to environmental fac-tors, are gradually revealing their limitations. The introduction of drone technology has brought revolutionary changes to power inspection. However, drone-based inspection also faces many difficulties and challenges, such as the complex environment of distribution lines, the difficulty in obtaining sample data, and the insufficient generalization ability and poor robustness of existing intelligent recognition algorithms. This paper proposes a power grid insulator defect detection model based on the YOLOv8 model. This model integrates multi-scale feature fusion and a detection head specifically designed for small insulator defect tar-gets on the basis of the YOLOv8 model. The main contributions include: replacing the original backbone network CSPDarknet53 of YOLOv8 with ConvNeXt, adding a dedicated small target detection head, and performing appropriate data augmentation operations on the power grid insulator defect dataset during the training process. Experimental results demonstrate that the improved YOLOv8 model outperforms the original YOLOv8 model in terms of recall, precision, and mAP50, exhibiting better detection performance and generalization ability.