Defect diagnosis technology of typical components on transmission line based on Fully Convolutional Network

Abstract

This paper presents an intelligent defect diagnosis algorithm based on fully convolutional neural network for typical transmission line components. Based on the region-based fully convolutional neural network algorithm and combined with deformable convolution, feature context fusion, clustering, the feature expression ability of neural network is improved. Through the improvement, the algorithm adapts to the object deformation and scale difference. In addition, the training effect is improved by improving the sample labeling strategy. In this paper, the defect diagnosis of four kinds of transmission line components(insulator, vibration damper, grading ring, wire clamp) is realized.