Research On License Plate Recognition Based On Improved RBF And CNN Neural Network

With the rapid increase of the number of private cars,the problems of road congestion and traffic accidents are becoming more and more serious.In order to alleviate the above problems,we should try to improve the modern intelligent traffic system.License plate recognition system as the basis of intelligent transportation system,not only can automatically extract license plate image from complex background,but also can automatically extract the license plate image segmentation of the characters,and accurately recognize the extracted characters.In this paper,the traditional neural network and the convolution neural network in depth learning were improved,and the license plate recognition model was constructed.The main work is as follows.Based on the principle of digital image processing and neural network,this paper studies the character extraction and character recognition in the early stage of license plate image,improves the traditional neural network,and designs a hybrid optimized RBF neural network model for license plate character recognition.The simulation experiments are carried out under the MATLAB environment.Based on the latest research results of deep learning,the structure and algorithm of convolution neural network(CNN)model in depth learning are analyzed and improved,and applied to license plate character recognition.The CNN neural network model is deeply studied,and the convolution neural network model is carried out to construct and train the convolution neural network model in the MatConvNet framework.On the basis of the LeNet-5 network model,the improved CNN network model is applied to the license plate character recognition problem.The license plate character recognition model and classifier based on CNN are designed,and the recognition of license plate characters is completed.By comparing the CNN network model in the simulation and the shallow network model,based on deep learning study of license plate recognition accuracy has been significantly improved.The license plate recognition GUI interface is designed with MATLAB software.The result of model recognition is represented intuitively.Finally,in the Keras,on the basis of deep learning framework,a license plate character recognition model based on the combination of Inception network and ResNet residual network is designed.The simulation results show that the method of license plate recognition has a good accuracy,and on the training time has the very great improvement.