Research On The Classification Method Of Transmission Line Fault Image Of Lightweight CNN

Traditional transmission line fault inspection methods are mostly manual visual inspection,which consumes a lot of manpower and material resources,and is affected by the environment of the repair site,and the efficiency is low,and real-time detection cannot be realized.With the development of grid intelligence and image processing technology,the use of advanced image classification algorithms for fault image detection of transmission lines is one of the important measures in the development and fault diagnosis of smart grids.Therefore,the use of drones with image processing has been carried out.Conduct research on intelligent inspection of transmission lines.Combining UAV inspection and transmission line fault image detection tasks can realize the classification and processing of fault images and fault types,thereby realizing intelligent inspection of the power grid and greatly improving the efficiency of fault detection.This paper collects fault images of high-voltage transmission lines within a mining group through drone inspections,and finally obtains pictures and video data of transmission lines,analyzes common fault types of transmission lines and their mechanism characteristics after failures,and selects all The required experimental image.Perform operations such as cropping and preprocessing on the filtered fault images,expand the data set through data enhancement,and mark the fault types of the image files in this image library,and finally construct a transmission line fault image that can be used for classification model training data set.In view of the particularity of the transmission line fault diagnosis task,combined with the more widely used convolutional neural network image classification algorithm and model compression technology,In this paper,a lightweight CNN transmission line fault image recognition and classification algorithm is designed,combined with batch normalized BN-Mobile Net,and simulation experiments are carried out on the keras platform.First of all,experiments on the standard data set prove that the model designed in this paper reduces the complexity of the network by changing the network structure of the convolutional layer.The amount of model parameters is reduced to one-sixth of the original,and the training time is also reduced to a certain extent.And the accuracy of the model is improved by more than 3% on each data set than the original model.Experiments were conducted on the fault image data set and the test software was developed.The final experimental results show that the model proposed in this paper is much better than the previous tests developed using artificial visual inspection and original neural network classification algorithms in terms of accuracy and training speed.The software can realize the fault classification of a single picture and multiple pictures,and has a strong generalization ability,which provides a certain experimental basis for the intelligent inspection of the power grid.Figure [39] table [6] reference [80]...