Research And Implementation Of License Plate Recognition System Based On Convolutional Neural Network

The license plate recognition system is an important part of the vehicle management system.Accurate positioning and recognition of the license plate has been a hot spot in the research of intelligent traffic management systems.However,the existing license plate recognition technology is often in specific scenarios,and the requirements for the environment are normally strict.In addition,with the continuous increase of new energy vehicles,there are differences in color and structure between new energy vehicle license plates and ordinary license plates,which makes the identification of new energy license plates in nat ural scenes still challenging.In response to this situation,this thesis has carried out extensive research the positioning and recognition of new energy license plates and ordinary license plates in complex natural environments.The specific research content is as follows:First of all,to solve the problem of license plate location in complex natural scenes,this thesis studies the license plate location algorithm based on Advanced EAST.Using the transfer learning method,the text detection network Advanced EAST is migrated to the license plate location task,and the paramet ers in the Advanced EAST network are continuously adjusted to improve the performance of the model.In the convolution part,the sparse connection in the Inception module is combined to improve the feature extraction ability of the convolution structure for the license plate image.The combination of non-maximum suppression and the license plate aspect ratio is used to screen the license plate area to achieve the accurate position of the license plate.Experimental results show that the positioning algorithm has a better positioning effect in complex natural scenes with uneven illumination and long shooting distances.Then,the Le Net-5 model structure is improved.Research on the license plate recognition algorithm based on improved Le Net-5 and CRNN.The license plate is divided proportionally into two parts: Chinese characters,numbers and letters.For the Chinese part of the license plate,the improved Le Net-5 network is used for recognition,and the other part of the license plate is recognized by the CRNN network.Experiments have proved that compared with using CRNN to recognize the entire license plate characters,the once segmentation strategy improves the accuracy of license plate character recognition.In the character recognition experiment,the recognition accuracy of Chinese characters was 95.2%,and the recognition accuracy of numbers and letters was 96.3%.Finally,combine the license plate location module and the recognition module to construct a complete license plate recognition system.Experiments have proved that the algorithm in this thesis can realize the positioning and recognition of license plates in complex scenes,and has good practical significance.