Insulator Fault Recognizing via Modified Faster R-CNN Using UAV Data

Abstract

Insulator fault recognition is a common hidden danger that affects the normal operation of the power system. In recent years, UAV aerial images have been widely used to identify defects in insulators. In order to solve the problem of low efficiency caused by manual inspection and traditional image defect detection methods, this paper proposed a modified Faster R-CNN model to improve the accuracy of model detection and reduce the amount of model parameters. Based on the traditional Faster R-CNN detection framework, the proposed method selected ResNet- 50 to replace VGGNet-16 as the feature extraction networks. The database used 781 labeled UAV aerial insulator images, and they are divided into training set, validation set and test set according to the ratio of 6:2:2. The mAP of the modified feature network model reached 62.41%, and the network model parameter size was 25.26M. Compared with the original framework with VGGNet-16 as the feature network, the mAP increased by 3.43% and the parameter amount was compressed by 4.41 times. The results show that the improved algorithm reduces the missed detection rate and false detection rate, and on the basis of improving the recognition accuracy, it can greatly reduce the amount of network parameters. And it can better meet the needs of lightweight network structure in actual scenarios.