Design And Implementation Of Automotive Circuit Diagram Character Recognition Method Based On Artificial Neural Networks

The arrival of a new era of artificial intelligence on people's way of life has brought the huge change,deep learning as a branch of artificial intelligence has been widely used in various fields,it expanded the field of artificial intelligence,which in certain circumstances the characters of detection and character recognition has been applied successfully.As car with combination of artificial intelligence,automobile circuit diagram is also more subtle and complex,and because the production models of cars in our circuit diagram with foreign little difference,this makes maintenance personnel in the vehicle maintenance and repair increased difficulty and reduce the efficiency of artificial identification.Through deep learning of the character recognition technology to automobile circuit diagram for information extraction,image information can be converted to text information,which will greatly improve the efficiency of reading circuit diagram of automobile maintenance personnel of the repair work at the same time provides important assistance by automotive circuit maintenance assist system.Usually in practice the use of traditional algorithm for image recognition and extraction of the character,on account of non-universality of artificial formulation characteristics make application has some limitations.Commonly used in target detection algorithm in real-time and accuracy is a certain degree to improve but still have plenty of advancement,character recognition algorithm not only produces in the process of transmitting information loss and improves the computational cost,in distinguishing between different postures and other aspects of the effect of the difference is pool.So the innovation uses in this paper is relatively mature development of artificial neural network to come true automobile circuit diagram of character extraction from two aspects of character detection and character recognition system design.A network model of artificial neural network training to get strong generalization,improve the speed and accuracy of character recognition.Combined with for the purpose of this article to achieve a specific scenario analysis of automobile circuit diagram and through determine the research directionfor the development of the deep learning task to clear aims,convolution neural network and capsule neural network was adopted to realize recognition task,which simplified into two parts to complete character detection and character classification,according to the characteristics of the character images information on network design respectively.Part of the character detection,Yolo v3 network model based on improved positioning and detection for character areas,by BN layer merged into convolution layer,lightweight is made for the network model,under the premise of without changing the result and performance,improves the speed of the network by the method of multiple scale integration do forecast and improving the detecting precision of the algorithm.Part of the character classification,the neural network model based on the capsule,which keep the details of testing the object's position and posture,at the same time only need very small relative to the convolutional neural network training data set can achieve the same level,which will be more suitable for practical application.Finally,this paper collected as much as possible the comprehensive automobile circuit diagram and manual annotation.Finally,this paper collect more comprehensive automobile circuit diagram and manual annotation,input data set into to the character to detect network to extract characters graphical information,then the effective split image after preprocessing of the input data sets to character recognition in the network are classified,reached for the automobile circuit diagram of the character information for identification purposes.