Design And Implementation Of License Plate Recognition System Based On Deep Learning

Currently,artificial intelligence,the Internet of Things,5G and other technologies are developing at a rapid rate.Smart transportation makes a great difference in smart cities,and license plate recognition technology plays an important role in the smart transportation system.The traditional Chinese license plate recognition method has low accuracy and speed,and recognition rate is greatly affected by the environment.which can no longer meet the application requirements.Based on deep learning theory,this paper designs an end-to-end deep learning model for license plate localization and recognition in natural traffic scenarios.In terms of detection,optimize the existing object detection network,the convolutional neural network is used to extend the depth of detection,which refine the localization accuracy,and improve the original detection network of YOLOv3 and multi-scale detection with different fine-grained features at high and low layers was used to enhance the model's detection accuracy for small targets.In terms of recognition,the joint structure of BGRU and CTC is used to optimize the character recognition network so as to capture the interdependence between features in a sequence,complete the character segmentation-free recognition task of the license plate,Obviously promote the traditional schema relying on segmentation and recognition,significantly shorten the training time and improve the convergence speed and recognition accuracy of the network,and good results can still be obtained when processing low-quality images.During training of detection and recognition models,system parameters was optimized,multi-source data was collected,and high-quality data sets were produced through pre-processing such as manual annotation and data cleaning and by using affine transformation to simulate license plate images obtained in complex scenes.which enhance the training effect of network model a lot.The subject combines the detection module and recognition module based on deep learning technology,further integrates into a complete end-to-end license plate recognition system,and test the system performance in common scenarios and complex scenarios.The experimental results show that compared with the traditional existing license plate recognition technology,the improved method greatly improves the speed and accuracy of license plate recognition,and has better robustness and reliability.