Transmission line insulator condition detection based on improved lightweight YOLOv8n algorithm

Abstract

Insulator stability plays an important role in ensuring the stability of transmission lines. With the rapid development of artificial intelligence, deep learning is increasingly used in transmission line detection. At present, many insulator operation state detection models inevitably have problems such as large number of network parameters, slow transmission image speed and large network computation. In order to solve the problem of insulator fault detection difficulty in complex background, this paper proposes a lightweight insulator fault detection algorithm with improved YOLOv8n. In this paper, a new C2f-DWR-DRB module is designed to replace the C2f module in the original backbone network, which achieves the specific task of selecting the appropriate convolutional kernel size to extract feature information. And the SegNeXt Attention Mechanism module is added at the bottom of the backbone network to prevent the network from extracting redundant low-level information during the first stage of information extraction. Auxiliary detection header DetectAux are also added in the middle of the network, which can extract the missed features of different scales and improve the generalization ability of the network. Finally, the computational complexity of the network is also greatly reduced by the knowledge distillation operation, which improves the FPS(Frames Per Second) value. The experimental results show that, compared to the original YOLOv8n network, the improved model proposed in this paper increases the mAP(Mean Average Precision) value from 88.2–91.6%.The number of parameters is only 77% of the original. At the same time, the FPS of the network decreased by only 12.0% compared to the original YOLOv8n network, and the goal of real-time detection can still be achieved.