An Edge Algorithm for Assessing the Severity of Insulator Discharges Using a Lightweight Improved YOLOv8

Insulators are crucial for power transmission lines, and issues related to discharge from insulators are one of the leading causes of faults in these lines. Therefore, an algorithm that can accurately assess the severity of insulator discharge quickly and that can provide real-time monitoring at the edge is needed. In this paper, these issues are addressed by making lightweight improvements to the YOLOv8 object detection algorithm. First, the input side is enhanced by introducing the Mosaic-9 data augmentation method, which improves the algorithm’s robustness and versatility. Next, the backbone network is replaced with the GhostNet network, achieving model lightweighting. The RELU activation function is replaced with GELU to enhance convergence speed and detection accuracy. Finally, the SIoU loss function is introduced to optimize the network, resulting in the Lightweight Improved YOLOv8 algorithm for assessing the severity of insulator discharge at the edge. Experimental validation shows that this algorithm achieves an 87.6% mean average precision (mAP) and 58 frames per second inference speed on edge devices, which meets the requirements for assessing the severity of insulator discharge at the edge.