Research And Implementation Of License Plate Location And Recognition Based On Deep Learning

With the continuous increase of urban traffic flow and the rapidly development of cities,license plate location and recognition system has been widely used in civil and military fields such as vehicle monitoring,crime fighting and traffic control.The traditional license plate location and recognition algorithm is difficult to adapt to the requirements of license plate location and recognition in different application scenarios because of the artificial design of image features and the adjustment of classifier parameters.In recent years,with the development of deep learning technology,the target detection and recognition algorithm based on deep learning has the advantages of strong anti-interference ability and can simultaneously adapt to the needs of a variety of scenarios.In order to achieve accurate and efficient license plate location and recognition,this paper conducts the following research on license plate location and recognition based on deep learning technology.The specific work is as follows:(1)In order to accurately locate the vehicle license plate in the image,this paper designs a depthwise separable convolution multi-scale vehicle license plate location model.Specifically,in order to improve the detection speed of the model,a depthwise separable convolution feature extraction network is designed to extract the features of the image.In addition,in order to make the model specific to the license plate target,12 anchor point frames with different widths and heights were designed according to the characteristics of the license plate,and the license plate target was located on 4 feature maps of different scales.The experimental results show that the model can not only accurately locate the license plate target in the image,but also solve the problems such as poor real-time performance of the existing depth target detection model and small target error detection to some extent.(2)In order to realize the accurate recognition of license plate characters,this paper designs a CBCNet license plate character recognition model.Firstly,a Convolutional Neural Network(CNN)is designed to extract the initial features of license plate characters.Then,a Bidirectional Gated Recurrent Unit(BGRU)is designed to further process the initial characteristics of license plate characters,and the temporal characteristics that can express the context relation of license plate characters are obtained.Finally,in order to obtain license plate character recognition results,a Connectionist Temporal Classification(CTC)module was used to decode the Temporal characteristics of license plate characters.The experimental results show that the model can not only recognize the license plate string accurately,but also solve the problem that the existing deep license plate character recognition model cannot express the relation between the license plate character context and can only recognize the fixed-length license plate to some extent.(3)The depth separable convolution multi-scale license plate location model and CBCNet license plate character recognition model are integrated to build a license plate location and recognition system,and the system performance is tested on the test set.