Research And Realization Of License Plate Recognition Algorithm Based On Improved Convolution Neural Network

Due to the large population and rapid economy development in China,there has been a continuous growth in the demand of motor vehicle monitoring,supervision,and control.In this thesis,the license plate character recognition algorithm is deeply researched for the needs of security enterprises.The convolution neural network-based recognition method and the improved LeNet-5 convolution neural network model are used to identify the license plate characters.The improved recognition algorithm is programmed,and the accuracy of the recognition algorithm of license plate recognition software and the recognition time-consuming are identified.In this thesis,the license plate recognition algorithm is studied based on the premise that the location of the license plate has been accurately positioned.First of all,using the method of gray-scale transformation,perspective transformation,binarization and morphological processing,the license plate image is processed to obtain the specific position of the character in the image,and the character image is segmented.Secondly,collecting license plate image samples,build convolution neural network training environment.Select the network model suitable for engineering application,through the actual test analysis network model and training process problems.Thirdly,the network structure and initialization of training parameters are improved,and the improved classifier is trained to obtain the classifier needed for character recognition.Finally,write independent license plate character recognition software,including the license plate image,character segmentation and character recognition,image processing and data visualization of image processing and character recognition process and relevant test validation.Compared with the traditional recognition algorithm,the improved license plate character recognition algorithm not only has the advantages of fast network convergence in classifier training,but also has the advantages of high recognition accuracy and fast recognition speed.After the recognition algorithm is implemented,the average recognition rate of license plate is up to 96.9%,and the average number of single license plate recognition is 64.7ms.Compared with the original BP algorithm,the accuracy is improved by 2.4% and the identification time is reduced by 22.1%.And in multi-imaging conditions,the recognition effect is good.