Insulator defect detection in severe weather using improved YOLOv8

Abstract

Insulators, as a vital component of the power system, encounter issues such as misdetection, leakage, and low detection accuracy in inclement weather. To address this problem, this paper proposes a YOLOv8-based insulator defect detection algorithm, YOLOv8-SSF. Firstly, SimAM (parameter-free attention mechanism) is included in the algorithm’s backbone network, which improves the ability to focus on critical features while maintaining a lightweight model. Secondly, the SPDConv layer is added to enhance the algorithm’s feature extraction capability for small-size defective targets. Furthermore, the Focal_EIOU loss function, which balances high- and low-quality anchors to increase detection and localization accuracy, replaces the CIOU loss function. According to experimental results, the enhanced algorithm reduces the rate of misdetection and omission of defects on transmission conductors, accomplishes a comprehensive simultaneous improvement, and achieves 87.2% mean average accuracy (mAP@0.5) on the dataset.