Research And Realization Of Automatic Character Recognition Of Electric Energy Meter Based On Neural Network Technology

With the continuous development of information technology,people's life is becoming more and more automated and intelligent,and the traditional meter reading mode is gradually eliminated.The research of smart energy meter character recognition technology has become a hot topic.In view of the current situation of electric power measurement in China,this paper studies the automatic recognition technology of characters at home and abroad,and implements three different models of electric energy meter automatic recognition system,and carries out the experiment analysis.The main work of this article is as follows:1.The traditional electric energy meter character automatic recognition system,before the image recognition,the system adopted image gray,two value and image enhancement measures to preprocess the image.In the later recognition process,the manual segmentation and template matching algorithm are applied to complete the recognition of the electric energy meter.MATLAB software is used in the experimental link.In the experimental comparison analysis,the recognition results after the image grayscale are better than the recognition results after the two value of the image,and the recognition rate is 83.6%.2.A multi-layer convolution neural network Lenet-5 is constructed.There are 7 layers of the network model.The convolution neural network structure is used to classify the character images of the electric energy meter.Four methods of data enhancement are adopted in the training process.By analyzing the results of the experiment,the confusion matrix is constructed,and the average classification accuracy of 91.4%is realized.3.A classification model of convolution neural network is realized by using transfer learning technology.The depth model based on VGG-16 migration learning is used to extract the features of the power meter characters and images,and then the Softmax classifier is used to classify the features.From the analysis of the experimental results,we can see that the accuracy of the test has reached 96%when the character preprocessing of the meter characters is not ideal.