Research And Application Of Line Ontology Recognition Of Inspection Robot Based On Convolutional Neural Network

As the main components of the transmission line,the working status of the line components has a significant impact on the operation of the power system.Therefore,regular inspection of the line is required.During the inspection process,the staff will use the equipment to take a large number of pictures of the line parts.When the picture is judged by the staff,it is inefficient and costly.At the same time,traditional detection methods are affected by the limitation of man-made features,and the recognition effect is average.With the development of deep learning,in the field of target detection,convolutional neural networks have outstanding performance in feature learning and processing complex scenes.So it is highly regarded by the public.This paper proposes a method for line component detection based on convolutional neural network.The purpose is to efficiently complete the detection and recognition tasks of the line component pictures during inspection.The main research work is as follows:First,it is proposed to use convolutional neural network to extract the features of transmission line components to solve the problem of insufficient artificial design feature expression ability.After comparing various convolutional networks,VGGNet is selected as the feature extraction network.It also visualizes the network.This paper analyzes the characteristic information of each layer of the network.The network can continue to learn the characteristics of component.The result proved the feasibility of the scheme.Second,Faster R-CNN detection model is selected by comparing and analyzing different detection algorithms.And line components detection model is built under the Tensorflow framework.At the same time,a picture part of line components including 3types of component,1 type of component failure and 2 types of hidden trouble is constructed for model training.On the basis of the model,an improved detection model is proposed through a deeper convolutional neural network,improvement of model structure and increase of anchor size.Finally,the influence of different hyper-parameters on the model was analyzed,and the optimal parameters were selected,so that detection model reached an accuracy rate of82.39% on the test set.Comparing influence of the feature extraction network depth on the model,it was found that the deeper network recognition effect was better,and accuracy rate was increased to 84.93%.The improved detection model improved the recognition accuracy of components,and the average accuracy rate reached 87.67% in the test experiment.The experimental results showed that the line components detection model could accurately complete certain inspection work.