Research On License Plate Recognition System Based On Neural Network

With the social and economic development of our country,the number of cars is also increasing year by year.In the process of urbanization,people have been paying more attention to the urban traffic conditions.Especially with the increasing pressure of traffic management departments,how to carry out more effective traffic management has gradually become the focus of management departments and people.In particular,the types of vehicle license plates in China are characterized by complexity and diversity,such as nonstandard hanging positions and weak awareness of traffic rules.Due to the lack of human resources of the traffic police,there are serious pollution of vehicle license plates caused by illegal listing,illegal travel and other factors.At present,there are many reasons why the recognition rate of vehicle license plates cannot be broken,such as man-made intentional damage of license plates,cover of license plates The problem of shooting angle,license plate recognition in special scenes such as nights,etc.At present,many mature methods and systems have been launched for license plate character recognition at home and abroad,but from the current point of view,the effect of domestic license plate recognition has not fully met the needs of the public and relevant departments,so we need to further research to improve the accuracy of recognition.In order to solve the problem that the accuracy of traditional license plate character recognition and detection methods is not high in special environments and the prediction cannot fully meet the existing traffic management needs,the research work done in this paper mainly includes:(1)Compared with traditional license plate location and character recognition algorithms and artificial intelligence neural network algorithms,the robustness of license plate recognition in complex scenes is improved.Compare and study the license plate region extraction method based on color information,the license plate region extraction method based on morphological edge feature analysis,the license plate location algorithm based on texture feature analysis,the license plate character recognition algorithm based on template matching,the character recognition algorithm based on support vector machine(SVM),and the license plate location and recognition algorithm based on neural network,and propose the problem of low recognition accuracy in special environments,In combination with YOLOv4 algorithm and mathematical tools,the data enhancement method is used to improve the recognition accuracy of license plates in common special scenes and improve the speed,accuracy and prediction effect of the license plate system.(2)Based on the existing model structure of YOLOv4 version,a target detection and recognition algorithm based on YOLOv4-SFA is proposed.We added the self attention mechanism based object detection method to the YOLOv4 model,combined with license plate features,trained these features using an improved Soft Attention algorithm,and replaced them with the K-means++algorithm,solving the limitation of K-means algorithm randomly selecting clustering centers.By self collecting data and downloading data,combined with the CCPD license plate dataset,an experimental dataset was constructed.The improved YOLOv4-SFA model and the original YOLOv4 model were tested and compared using evaluation methods such as plotting PR curves,Recall recall,Precision accuracy,and m AP to verify the implementation effect of YOLOv4-SFA.(3)The design and research of license plate cloud recognition system were carried out.Aiming at the challenges faced by the traffic management department in terms of vehicle management such as illegal listing,illegal travel and accident escape,the deployment and architecture design of license plate cloud recognition system were studied.The cloud deployment of license plate recognition system was realized by building private cloud and public cloud.With the help of big data,cloud computing and other technologies,the predictive analysis of license plates was carried out,It enables the management personnel to check,manage and control the status of vehicles,realize the real-time control,scientific control and man-machine joint control of vehicle management,and thus improve the intelligent control ability of vehicles.(4)The design and research of license plate cloud recognition system were carried out.Aiming at the challenges faced by the traffic management department in terms of vehicle management such as illegal listing,illegal travel and accident escape,the deployment and architecture design of license plate cloud recognition system were studied.The cloud deployment of license plate recognition system was realized by building private cloud and public cloud.With the help of big data,cloud computing and other technologies,the predictive analysis of license plates was carried out,It enables the management personnel to check,manage and control the status of vehicles,realize the real-time control,scientific control and man-machine joint control of vehicle management,and thus improve the intelligent control ability of vehicles.