Research On Character Recognition Of Vectorgraph Based On Deep Learning

With the rapid economic development,the automotive industry has developed rapidly,the workload of car maintenance personnel has increased,how to effectively read the circuit diagram into the database,and complete the circuit diagram by performing data extraction,data changes,and data reconstruction on the circuit diagram.The WEB information is reconstructed to facilitate maintenance personnel's tasks such as maintenance guidance and rapid training.Extracting the information in the circuit diagram is a very important part of the research content in the project.The information in the circuit diagram is very complex.The character information is the most important part of the information extraction in the circuit diagram.The characters in the circuit diagram contain a large amount of information,such as the length of the line,device names,connection methods,etc.How to effectively extract the character information in the circuit diagram is the focus of this study.The character recognition in the image belongs to the field of pattern recognition.The traditional character recognition method is also developed in accordance with the pattern recognition method.The original character recognition model has complicated steps,poor recognition effect,and many manual operations.So combining rapid development of science and technology,putting forward a set of effective detection model is the focus of this article.Combined with the process of character recognition model,and by studying the characters of the circuit diagram,we deeply studied the multiple network models of deep learning,especially the typical convolutional neural network.We proposed a model based on Faster RCNN fusion location information + Tesseract character recognition.The main tasks of this article are as follows.First,the traditional character recognition model is analyzed.By studying the traditional character recognition model and combining various network characteristics of deep learning,the character recognition model in the circuit diagram is simplified to the character area positioning and character recognition.Second,jump out of the original character recognition model framework to locate the character area in the circuit.At present,common target detection methods are based on convolutional neural networks.Using convolutional neural networks can reduce a series of pre-processing operations on images,which is very simple and convenient.By studying the convolutional neural network and using the Faster RCNN network model combining RPN and Fast RCNN,the accuracy of the target detection of the network is relatively high.The circuit diagram data set was made and was annotated using the manual labeling method.The data set was trained on the network model.Through experiments,it is found that the detection accuracy of the ResNet network and VGG network as the basic feature extraction network is lower than the accuracy of fusion of the two model position information.Therefore,the position information obtained by the two Faster RCNN models is fused by experiment to obtain the final position information.Third,there are a large number of characters in the circuit diagram.But because of the different meanings of the characters,such as the length of the characters,the direction of indefiniteness,and sometimes multiple lines of characters,the characters in the framed area are cut,and then recognized,in addition to the special circuit diagram.Based on the model,a more effective solution is proposed.The result shows that our model can achieve better results in recognizing character in circuit diagram.