Research Of License Plate Recognition Based On Deep Convolutional Neural Network

In the process of urbanization in China,traffic congestion,poor parking planning has become increasingly prominent,which makes people's work,study and life inconvenient.In order to build a harmonious society,traffic management has become the trend of intelligence.License plate recognition technology as an important part of intelligent transportation,has been widely used in the road information monitoring,parking management,highway and bridge toll station management.DCNN(Deep Convolution Neural Network),as an important branch of artificial intelligence,exhibits higher accuracy than the general algorithm in image processing.Therefore,this thesis mainly studies the license plate recognition technology based on DCNN,Identify the three processes to start the analysis.(1)Aiming at the problem that the rough location algorithm based on edge information and color information can't get the target area when the environment is complex,the body texture and the color are close to the license plate.In this thesis,the algorithm is improved based on the MSER text extraction algorithm.The experimental results show that the improved coarse localization algorithm is more robust than the license plate area in the image area.(2)In the screening phase,the existing methods use feature extractors(such as SIFT,HOG)plus a combination of classifiers(such as SVM,shallow BP neural network)to separate the real license plate.However,this method does not analyze all the pixels of the picture to be detected,so some features of the license plate will be discarded.After designing DCNN,this thesis designs the license plate screening network LPC-Net,which receives the full pixel input and obtains license plate characteristics from more dimensions,the screening effect is better than the existing method.(3)In order to solve the problem of character sticking or Chinese character disconnection problem when segmenting license plate characters,this thesis proposes a character segmentation algorithm based on connected domain analysis,which uses Chinese license plate characters as a priori information.The algorithm makes the segmentation faster after avoiding the defects of the projection method.(4)The algorithm of the feature extractor plus the classifier used in the character recognition of the license plate usually ignores the difference between the similar characters such as "5" and "S".After studying the LeNet handwritten numeral recognition DCNN,this thesis improves it into the license plate character recognition network Res-LeNet,which uses a large number of samples to learn the difference between these similar characters,the experimental results show that the recognition accuracy is superior to the LeNet.Finally,a simulation experiment is designed.The experimental results show that the license plate recognition technology integrated into DCNN has improved the accuracy rate compared with the existing identification technology.