A Multi-feature Fusion-based Deep Learning for Insulator Image Identification and Fault Detection

Abstract

Insulators are used as faulty multiple components in transmission lines. Problems such as contamination, cracks and damage will seriously affect the normal operation of transmission lines. Therefore, it is necessary to perform Insulator Image Identification and Fault Detection on the insulator. The traditional method is always affected by the complex background and multi-size of the massive aerial image, causing the image segmentation is difficult, the model calculation is complex, and the fault detection type is single. This paper proposes to combine the insulator multi-fault target detection algorithm (Fast R-CNN) and the deep learning to automatically learn the advanced features of the insulator image of multiple different surface faults, adding the traditional low-level visual features (color feature and texture feature) to more fully extract the effective features of the image, thus improving the accuracy of the recognition. In this paper, the model is trained by the multi-type insulator aerial image, such as normal, dropped, damaged, contamination, cracked and noisy. The multi-class neural network is used to classify the surface fault types of insulators in the end. The results show that the method can detect multiple-faults of Insulator image simultaneously, which improves the accuracy of insulator fault identification.