Surface Coverage Classification of UAV High Resolution Image Transmission Lines Corridor based on Total Convolution Neural Network

Abstract

In recent years, UAV imagery has been applied more and more to the work of power line inspection. Surface coverage classification is a key step. Because of the size and amount of data of UAV high-resolution images, the accuracy requirements of traditional classification algorithms can no longer meet the practical needs. In this paper, a pixel level classification algorithm based on full convolution neural network is proposed. The total convolution neural network model reduces the error caused by other forms of deformation such as image translation, scaling, tilting and so on. In the application of UAV image, the total convolution neural network is used to classify the image, and the high dimensional information extracted from the coiling layer is used to learn the feature of the image fully, and the accuracy of the classification is improved. By comparing with the random forest algorithm, the advantages of the full convolution neural network in the classification of the surface coverage of the UAV transmission line corridor are verified, and the reference value is provided for the electric patrol line and the image classification of unmanned aerial vehicles.