Research on Self-explosion Defect Detection of Lightweight Glass Insulators Based on Improved YOLOv5

In order to maintain the reliability, safety and sustainability of power transmission, the fault diagnosis of aerial insulators has become an important task in power inspection. Therefore, a lightweight defect detection model L-YOLOv5 is proposed in this paper. First, the residual module in the backbone network is improved for light weight by adding depthwise separable convolution and 1×1 group convolution, and by designing the backbone network L-CSPDarknet53. This network can greatly improve the detection speed without sacrificing a small amount of accuracy. In terms of feature extraction, the DC-SPP module is designed. The convolution and dilated convolution in the module can increase the receptive field and improve the detection performance of the network without losing detailed information. Finally, in accordance with the problem that the self explosion defect area is small and difficult to detect, a method of adding a small target detection layer is proposed. The small target detection layer contains more defect details, which is more conducive to the detection of self explosion defects. The implementation results show that L-YOLOv5 can quickly and accurately detect self explosion defects, with an accuracy of 96.7% and a detection speed of 37.4 frames/s. Compared with YOLOv5 network, the accuracy and speed are improved by 3.5% and 49%, respectively. Compared with other common detection networks such as Faster R-CNN and SSD, L-YOLOv5 is more competitive in insulator defect identification and location. © 2023 Science Press. All rights reserved.