Multi-Modal Data Fusion Using Deep Neural Network for Condition Monitoring of High Voltage Insulator

Abstract

This research proposes a novel Fusion Convolutional Network (FCN) combining a CNN with a binary multilayer neural network (MNN) sub-classifier, forming a multi-modal information fusion system (MMIF) for real-time monitoring of high-voltage insulator surface condition via UAV-captured imagery and leakage current data. The fusion of image-based classification and leakage current readings allowed classification accuracy to increase from 92 % to 99.76 %. Traditional classifiers based on wavelet transform and PCA, as well as various CNN architectures, were benchmarked, and the hardware implementation potential on UAV edge devices was discussed.,