Algorithm Of License Plate Detection And Recognition Based On Convolutional Neural Network

License plate detection and identification is the core part of the intelligent transportation system.The accurate detection and identification of the license plate number determines the development speed and technical level of the intelligent transportation system.A large number of researchers have proved that the algorithm in the convolutional neural network can be well adapted to the detection and recognition of license plates in various natural scenes.This paper analyzes the advantages and disadvantages of traditional license plate detection and recognition algorithms and convolutional neural network algorithms.Finally,it is decided to use Faster-RCNN framework and LeNet-5 network to improve and realize license plate detection and license plate character recognition.This paper studies the detection of license plates and the recognition of license plate characters.The main research contents include the following three aspects:An improved Faster-RCNN algorithm is proposed for the problem of license plate detection.The original Faster-RCNN algorithm classifies 22 types of objects including the background in the image.Because there are many types of categories,the ratio to the background is similar,and the aspect ratio of each object is basically in accordance with 1:2,1:1 and 2:1.The license plate usually accounts for a small proportion in the image.At the same time,the license plate detection is to classify the license plate and the background as a two-category problem.In different natural scenes,the aspect ratio of the license plate will not be generated due to the difference in shooting angle and distance.same.Comparing the above differences,this paper improves the original Faster-RCNN algorithm,and finally conducts experiments in CCPD datasets containing multiple natural scenes.The experimental results show that the improved algorithm in this paper detects the license plate in CCPD while ensuring accuracy.Improve the accuracy of the algorithm.Identification of license plate characters In this paper,the license plates located in the first step are binarized and then divided one by one,and the obtained 7-digit license plate characters are divided into three segments.The LeNet-5 algorithm with good handwriting character recognition is improved to obtain the license plate character recognition algorithm.The three-character characters of the license plate characters are 24 city codes(except I and O)and 34 characters and numbers in 31 provinces.The complexity of Chinese characters and characters is usually higher than the number.According to the above differences,this paper has made related improvements to the LeNet-5 algorithm.Experiments show that the accuracy of the proposed algorithm is 3.9% higher than that of the original network.