Insulator self-shattering detection based on YOLOv5 under small sample conditions

Abstract

Insulators are indispensable devices on power transmission lines. The self-shattering of insulators pose a threat to the operation of transmission lines. Insulators where self-shattering occurs are usually small objects in an image, and samples of insulator self-shattering are scarce. This brings challenges to the intelligent detection of insulator self-shattering. The insulator self-shattering is detected by a YOLOv5 under small sample conditions. Considering the scarcity of insulator self-shattering images, normal samples are used to assist in training a YOLOv5 model. Tests on a few public datasets and some simulated self-shattering images show that the recognition rate can reach 97.43% when the ratio of training to test samples is 1:10.4. The results show that our method provides a way to identify self-shattering insulator under small sample conditions.