Insulator Recognition Method for Distribution Network Overhead Transmission Lines Based on Modified YOLOv3
Zhihao Chen, Yewei Xiao, Yan Zhou, Zhiqiang Li, Yang Liu · 2020
As an important part of distribution network overhead transmission lines, accurate recognition of insulator is an important prerequisite for intelligent detection and fault diagnosis of insulator. In order to accurately identify insulators, a method of insulator recognition for distribution network overhead transmission lines based on modified YOLOv3 is proposed. Firstly, K-means is used to cluster the target box in the dataset to obtain the appropriate anchor. Then, DenseNet is introduced in DarkNet53 to enhance the reuse and fusion of network characteristic information. At the same time, an additional scale is added to predict the insulator. Finally, combining Cross Entropy function and Focal Loss function is adopted to replace the original loss function. Building insulator of distribution network overhead transmission lines dataset for training and testing, the experimental results show that this method can accurately classify the insulator of distribution network overhead lines, accuracy rate increased by 10% than the original algorithm, which has higher detection accuracy and stronger robustness. It basically meets the inspection task of power network and has high engineering application value.