License Plate Recognition Based On Deep Convolutional Neural Network

With the development of modern society,people’s living standards are improving,and the number of’vehicles is increasing,which will bring great pressure to traffic management.How to quickly and accurately identify and locate vehicles in front of massive vehicle traffic data is particularly important and urgent.The license plate recognition system plays a huge role in the transportation field and which is an important part of the intelligent transportation system.It is an important part of the urban intelligent traffic management system,and it is an intelligent transportation system for real-time monitoring of motor vehicles on the road.In recent years.How to accurately locate the license plate and how to accurately identify characters has been the focus of the field of machine vision,With the systematic research of neural network science,you can use the powerful feature extraction capabilities of deep learning to make the system more stable and adaptable to various complex scenes.The main research contents of the license plate recognition algorithm in this article include the follo.wing three aspects:1)In the license plate location part,through the detailed explanation of the traditional license plate location scheme based on digital image processing.aiming at the shortcomings of the positioning part,the improvement scheme of the vehicle license plate positioning by using the Principal Component Analysis algorithm to reduce the dimension of the data and then combining the Support Vector Machine algorithm is proposed.In this article.100 samples to be tested are randomly selected as test data.Through experimental comparison,the correct positioning rate of the recognition scheme based on digital image processing is 94%.and the correct scheme of the new scheme proposed in this article is 98%.2)In the character recognition part,the problem that the recognition rate of the template matching method is greatly blurred,the license plate pollution is greatly reduced,and the problem that the single character recognition method is excessively dependent on the character segmentation has a low recognition rate.This article uses a new character recognition method-end-to-end recognition.The end-to-end recognition method used in this article is to train a recognition model for each character of the license plate.This method no lonser depends on the segmentation of the license plate characters and does not rely too much on the license plate pixel information.This solution better overcomes the shortcomings of the above two identification schemes.Randomly selected 500 blocks containing license plate information were used for experiments,and the overall recognition accuracy of the improved recognition method was 95.6%.3)At the end of the article,two license plate location schemes and three license plate recognition character recognition schemes are combined to form six different systems.Test the above system with a database containing 500 panoramic license plate images.The experimental results show that the improved recognition scheme and the end-to-end character recognition scheme combined with the license plate recognition system have a character recognition accuracy rate of 94.6%.