Research On Algorithms Of License Plate Detection And Recognition In Complex Scenes Based On Deep Learning

In today’s society,the utilization rate of vehicles is increasing,and the pressure of traffic control is also increasing.License plate detection and recognition can help traffic control departments to quickly query the driver information and track suspected vehicles in real time.With the popularity of artificial intelligence and Internet of things,intelligent transportation gradually begins to play an important role,and license plate detection and recognition algorithm is further developed through the use of deep learning.However,due to the bad weather,uneven illumination,license plate tilt,license plate fouling and other interference factors in the real environment,the performance of license plate detection and recognition network still has a lot of room to improve.This paper focuses on the problem of license plate detection and recognition in complex scenes based on deep learning:Firstly,from the perspective of license plate data set expansion,this paper proposes a license plate generation method in complex scenes based on generation confrontation network.Based on the generative countermeasure network,a data set expansion method for license plate is designed.Generating confrontation network can change the style of license plate while keeping the important information of license plate(character,contour,etc.)unchanged,and get more realistic license plate image,which can make up for the deficiency of the original license plate data set in quantity and environment complexity.This improves the robustness of license plate detection and recognition network in different scenes.Secondly,for the license plate detection task,this paper proposes a license plate detection method based on lightweight feature fusion network.This method uses deep separable convolution instead of ordinary convolution,builds a lightweight backbone network,and performs lightweight processing on bidirectional feature fusion pyramid module and detector head.These designs not only reduce the complexity of the network,but also enhance the ability of feature representation,making the model more balanced in accuracy and speed.Finally,for the license plate recognition task,this paper proposes a license plate recognition method based on attention.This method uses the improved residual network as the backbone network,adds the license plate correction module and the sequence feature enhancement module,and applies the self-attention mechanism to the text recognition module.These designs give full play to the excellent characteristics of attention mechanism,making the model more accurate and stable.