The Research On Plate Recognize System With Deep-learning Algorithm

Intelligent transportation system is one of the main applications of Thing's connection nowadays.It is widely used to monitor and manage cars in the highway,and it is also used to monitor vehicle and control traffic in urban roads,and the primary task of intelligent transportation system is License Plate Recognition.At the same time,since the Google made the introduction of man-made warfare,the deep learning has begun to become one of the main focus in current community researching,it is widely used in automatic driving,face recognition and other fields.This paper designs an improved license plate recognition system based on deep learning algorithm.This paper first analyzes the methods of current license plate recognition and talks about its advantages and disadvantages,then it analyzes the current researching status at home or abroad,analyzes the particularity of domestic license plate and technical difficulties.Secondly,I design a license plate recognition system,and I set up the MATLAB software for license plate's positioning and segmenting,build caffe for making the plate character's recognition network based on deep learning algorithm,analyzes some deep learning algorithms,like CNN,MLP,Pooling and other related algorithms,analyzes network's training and understands the data update process.Thirdly,the paper analyzes the license plate's location and segmenting system,puts forward the improved way to make the picture to be gray,and give the license plate's locating algorithm which is combined with color recognition and edge recognition.Analyze the advantages and disadvantages of three different edge extraction operators like Prewitt,Laplacian and Sobel,then I give the improved way to segment license plate character.Finally,the paper analyzes three network structures like GoogleNet,LeNet-5 and Cifar-10,and analyzes the testing results and finds out if they are good for the license plate character's recognition.At the same time I analyze the activation functions like Sigmoid,tanh and ReLu,and then I make the adopted network.The main innovation of the improved network is combining the convolutional sampling module with the inception module,then I write network's protext code,and test the error rate of the network,and then I get two final identification network for segmenting of Chinese characters and alphanumeric.This paper proposes a identification network for Chinese character based on the deep learning and a identification network for alphanumeric improved recognition network at the same time.The error rate of network for Chinese character is reduced to 2.355%,the error rate of network for alphanumeric character is reduced to 2.091%,and compared with other methods,the recognition error rate is also reduced.The improved recognition system proposed in this paper completes the whole process from the inputting of initial color image to the final recognition of the license plate's character.The results of the design meet the initial requirements.