Application Of The Deep Learning Algorithm In Plate Number Detection

With the explosive growth of the number of vehicles in China,how to obtain the license plate information has become an important subject of traffic management.Therefore,finding an algorithm with strong self-adaptive and high detection precision is an important improvement direction of the license plate detection algorithm.Although the traditional license plate detection has better detection result and put into practical application,however,according to the actual requirements of the project,the effect of the traditional detection methods are poor,the detection speed can not meet the actual requirements,and detection difficult development and long development time.The use of depth learning model can solve the problem of license plate detection,and meet the requirements of detection precision and speed.The paper starts with the traditional license plate detection algorithm,and uses the latest SSD model in the depth learning algorithm to detect the license plate.Experimental results show that the license plate detection algorithm based on deep learning proposed in this paper has strong adaptability,excellent noise robustness and detection accuracy,and shows good performance in practical applications.The paper's innovations and main work are summarized as follows:? The license plate detection system based on the depth learning SSD model is designed.According to the license plate positioning of the biggest difficulty,no longer follow the traditional method of using SSD model too observant of conventional standards.,but deep learning,designed two times using license plate detection system of SSD model,the orientation model of license plate location,character detection model using character detection of license plate region,and to better meet the actual needs of the scene.Under the premise of ensuring the accuracy,improves the speed of operation.? Improved and perfected the license plate character detection module.Aiming at the character recognition rate is low,the similar characters easily confused,with deep learning method,method of convolution neural network is introduced into the classification task of the widely used,and for the three types of characters:Chinese characters,letters and numbers,according to the difference of the structural characteristics,modify the parameters of the network structure,and change the structure and quantity of laminated roll.Under the premise of large sample data sets,the false detection rate is maximally reduced,and the trivial steps such as character tilt correction and segmentation are avoided,which improves the recognition efficiency and improves the robustness of classification.? To complete the construction of data required for model training and testing.The problems in experiment,and then construct the data set when facing rare Chinese characters problems,four solutions are proposed and applied in a limited data set,the biggest expansion of the training sample is augmented,improve the detection effect of rare Chinese characters in license plate detection.? For license plate fuzzy problem,from the two aspects of training set and model parameters,a solution is proposed,and a certain detection effect is obtained,which has great reference significance.