The Research On License Plate Recognition Based On BP Artificial Neural Network

In recent years,vehicles often have congestion in public places,which have affected public image and delayed people's travel.In order to solve the above problems,many experts and scholars at home and abroad have begun to study the license plate automatic identification,but the existing algorithm is not ideal for identifying characters in complex and harsh conditions.Therefore,there is a need for a fast way to identify the license plate.In this paper,the hardware system of license plate acquisition and recognition is set up.In order to get a clear image,the selection of the corresponding equipment is made.On the building of the system hardware,the distance between the ground coil and the camera,the height of the camera and the angle of the horizontal tilt of the camera are designed.At the same time,the lens resolution,focal length,field angle,depth of field and other parameters are calculated,and the model of the camera in the system is selected.In this paper,the preprocessing methods of image grayscale,image grayscale stretching and image filtering are adopted.Then edge detection and morphological processing are used to locate the parts of the license plate in the image.The character samples in the license plate image are obtained by using the vertical projection and cutting algorithm combined with the prior knowledge.After normalization and feature extraction of characters,we design the scale and train parameters of the network,and input the feature vectors after character processing to the network for training and recognition,and the results can achieve high recognition rate for each character.In this paper,100 samples of each character are extracted from 1000 collected license plates,and 60 of them are used as training samples and 40 as experimental samples.After statistics,the recognition rate of the 65 characters in license plate can be over 95%,and the recognition effect is better than that of the three layer BP neural network without momentum added.By adding a hidden layer and momentum item to the three level BP neural network,the model realizes the function of more accurately identifying the license plate characters,and realizes the fast recognition of license plate characters through the fast transmission of information between the BP networks.