Research On License Plate Area Detection And License Plate Character Recognition Based On Convolution Neural Network

Target detection and image recognition technology has been a hot topic in the field of computer vision both here and abroad.In the research process,it is the key to improve the accuracy of target detection and image recognition and improve the efficiency of the model.Deep learning,as a new research field of machine learning,is a research direction attracting a great deal of attention in recent years by constructing deep learning models(such as VGGNet,GoogLeNet,ResNet)to extract features to solve the problem of object detection and image recognition.Convolutional Neural Network(CNN)is a new neural network method that based on artificial neural network and deep learning technology.Because of its features of local reception,weight sharing,space and time sub-sampling,the CNN performances more stable in image scaling and translation.Based on the existing research results of CNN,this paper use the SSD model for target detection and the LeNet-5model,which has excellent performance on handwritten character recognition.Base on that,modify those two models for the license plate detection and license plate character recognition.This article mainly focus on the following two aspects:(1)License plate regional detection based on CNNConsidering the shape and physical structure of the license plate area,this paper presents a license plate area detection method based on SSD model.First of all,we introduce the network structure of SSD,and then introduce the dataset used in this paper.Then the data set used in this article is introduced,and the real tag is marked by BBox-Label-Tool tool.Then the length width ratio and other related parameters of the default frame are designed according to the license plate characteristics of our country.Then we use training set to train and test the improved network model.Experimental results show that the proposed method can achieve a correct detection rate of 85.4% and has a certain detection performance.(2)Research on license plate character recognition based on CNNAiming at the characteristics of license plate characters,complex background,simple shape and structure,this paper improves on the basis of LeNet-5 model.Firstly,the pre-treatment to the split characters are: size normalization,denoising,binarization,refinement,centering of the character area and so on,to remove the complex background;and then use the preprocessed license plate character set to improve the network model's performence.At last examine the model.Experimental results show that the proposed method can achieve the correct recognition rate of99.96% and has a good recognition performance.