Research On Image Classification Of Power Equipment Based On Convolutional Neural Network

With the large-scale expansion of power grid,image acquisition technology has been widely used in power system.The power equipment in the substation is operating at high intensity every day,so it is essential to detect and classify the power equipment every day for better management of power equipment.In order to solve the above problems,this paper adopts convolutional neural network for classification of electric equipment images.The convolutional neural network-based power equipment image classification method is studied,which solves the problem of low efficiency of manual long-term detection image analysis and processing,and provides guarantee for power equipment operation and fault monitoring.The dataset is obtained by downloading online and shooting on site to obtain the power equipment dataset.The obtained power equipment dataset was pre-processed and expanded to 6 times the original size by data augmentation.ACDSee software was used to batch annotate the expanded dataset.In order to improve the accuracy of power equipment image classification,several improvements are proposed in this paper based on the DenseNet network model.1.Adding the residual attention mechanism to the DenseNet network model,the residual attention mechanism can enhance the effective features and suppress the ineffective ones.The structure of the DenseNet network model can be optimized by adding the residual attention mechanism.2.By modifying the dense block in the DenseNet network model and adding the multi-resolution module,the integrity of feature map extraction can be improved by modifying the growth rate in the dense block so that the number of feature map outputs in each layer is not fixed and the number of feature map outputs in each layer keeps increasing.In this paper,the improved DenseNet network model is tested on a public data set.The experimental results show that the test accuracy of the improved DenseNet network model has been significantly improved,which shows the effectiveness of the method proposed in this paper.The improved DenseNet network is applied to the power equipment data set for testing.The experimental results show that the test accuracy of the improved DenseNet network model has been significantly improved,indicating that the method proposed in this paper has strong generalization ability.In order to further verify the residual attention mechanism,two networks,Xception and mobileNet V2,are also added for comparative testing.Compared with the original network,the improved DenseNet network is tested on the power equipment data set,and the accuracy is increased by 8.89%,5.14%,and 8.06% respectively.The improved Xception and the improved MobileNet V2 network are tested on the power equipment data set compared with the original network,and the accuracy is increased by 1.39% and 4.31%respectively.From the above experimental comparison,the following conclusions can be drawn: The method proposed in this paper has a good improvement effect on image classification of power equipment.