Insulator Detection in Intelligent Monitoring Based on Yolo Family and Customizing Hyperparameters

Abstract

Monitoring of power transmission lines plays an important task in high voltage transmission systems. The problem of damaged insulator causes bad effects on an electrical power grid. To make sure the power grid worked properly, electrical personnel usually need to be climbed on the electric post to inspect them which consists of risk latent in occupational safety. Therefore, constructing smart monitoring systems plays an important task to monitor and inspect insulator conditions without manual monitoring. Nowadays, UAV (unmanned aerial vehicles) systems have been become popular, which combines with computer vision supporting to implement an autonomous system for detecting and monitoring bad insulators. In this study, we present an approach which apply deep learning based well-known models to automatic detect insulators with hyperparameter optimization. In this experiment, some models are investigated to select the best one for the insulator detection task in the intelligent monitoring system. Image insulator data was collected by using drone system. The experimental image data consists of 972 samples taken under a variety condition such as clutter backgrounds, high contrast… This study has surveyed and experimented some well-known models such as the Yolov5 family and the newer Yolov7. The hyperparameter optimization and augmentation were applied for improving detected performance. Yolov5x with optimal hyperparameters achieves higher performance with the ratio of recall, precision, mAP:0.5, and mAP:0.5:0.95 are 98.5%, 99.3%, 99.0% and 68.6%, respectively. Meanwhile, the default hyperparameters Yolov5 achieved results with the ratio of recall, precision, mAP_0.5 and mAP_0.5:0.95 are 97.3%, 97.1%, 99.2%, 65.7% and Yolov7 are 95.7%, 97.6%, 98.6% and 63.3%.