Research And Implementation Of Driver License Identification Technology Based On Neural Network In Complex Background

In recent years,OCR(Optical Character Recognition)teclhnology has been developed more maturely,and it is widely used in all aspects of life.This is the technology used in the driver's license recognition system at the school entrance control.However,the accuracy of the name recognition text recognition system of the school's driving license recognition system is not very high,about 80%.In order to further improve the recognition rate of the driver's name,the research team will redesign the driver's license name recognition system by means of deep learning.The algorithm is optimized and improved in the preprocessing part and the text recognition part respectively.In view of the low recognition rate of the school driver's license identification system,this topic mainly improves from two aspects:First,the algorithm is improved in the preprocessing stage,and the algorithm of red chapter positioning is adopted to make the system more accurate positioning and extraction.The driver's license name area;the second is the LeNet-5 training model using the improved algorithm of the convolutional neural network in the identification stage.In the preprocessing stage,this topic will perform some necessary image processing operations on the image,such as non-uniform light illumination,Gaussian filtering,tilt correction,grayscale,binarization,and adding noise.In the stage of positioning and extracting the name area,the algorithm of red chapter positioning is adopted,that is,the area of the red chapter is first positioned,and then the driver license name area is located according to the position proportional relationship between the red chapter and the name area.After testing,the accuracy of this method to extract the complete name area is about 99%.In the process of identifying Chinese characters,the most important thing is to choose neural network and training network model.Due to the limitations of experimental data and experimental equipment,this topic can not complete the recognition of all Chinese character categories in the huge Chinese character library for the time being.Therefore,20 Chinese characters are selected as the recognition categories in the selection of Chinese character categories,and each Chinese character category is about 1500 pieces of data are generated for model training.In addition,this method adopts the method of single Chinese character recognition.The single Chinese character recognition is relatively simple.Even if the neural network is not used for more detailed feature learning,it can achieve a good recognition effect.In summary,this topic selects the LeNet-5 classical neural network with shallow network layer and improves it.The improved method is to add the BN(Batch Normalization)algorithm before the second layer convolution network layer.After comparing the improved and improved network models,it is found that the training time of the model is reduced from 30min to 15min,the recognition time of a single Chinese character is within 100ms,and the accuracy of the model is above 99%.