The Intelligent Recognition of the Disaster Damage of Distribution Network Detected by UAV

Abstract

Unmanned Aerial Vehicle (UAV) is used to reduce the difficulty in the collection of the disaster damage information of distribution network. However, the efficiency still needs to be improved because the judgment of disaster damage by human is inefficient and fallible. So intelligent recognition method is considered to increase the efficiency and accuracy. For the training of intelligent recognition model based on deep learning network requires a large number of samples, a novel method is proposed in this paper. The Self Organizing Map (SOM) is introduced between the convolutional neural network (CNN) and support vector machine (SVM), to reduce the number of training samples. Meanwhile, median filtering detection with Gaussian distribution is taken in the convolutional layer to decrease the influence of stochastic noise in the input picture. The dropout technique and multi winning neurons are introduced in the SOM to improve the clustering robust and reduce time consumption. Experimental results show that the proposed method shows good performance on the recognition of the damage images of the wire poles with complex backgrounds in the distribution network, and it is obviously superior to the traditional ways.