Research And Application Of Civil Aviation Vehicle License Plate Recognition Algorithm In Smog Environment

With the development of China's economy,the number of vehicles has exploded,which puts forward higher requirements for license plate recognition technology.At present,many domestic license plate recognition systems have been put into commercial use,but the current license plate recognition technology has two problems: low accuracy of license plate recognition in the smog environment and failure to recognize the license plates of civil aviation vehicles.On the one hand,with the increasingly serious problem of smog,the low accuracy of license plate recognition in the smog environment has become an urgent problem that needs to be solved by license plate recognition technology.On the other hand,in recent years,relevant policies have allowed some vehicles with civil aviation license plates to drive on roads outside the airport.However,because the specifications of the license plates of civil aviation special vehicles are different from the specifications of ordinary civilian license plates,after investigation,it is found that there is currently no license plate recognition method capable of recognizing civil aviation special license plates.Based on the above situation,this thesis proposes a license plate recognition method that can recognize common license plates and civil aviation license plates,and improves the accuracy of license plate recognition in a haze environment.The main research contents of the thesis are as follows:Firstly,a defogging algorithm is used to defog the picture to improve the accuracy of license plate recognition in the haze environment.In this thesis,an image defogging algorithm based on gated context aggregation is used to defog the pictures.The defogging algorithm has the characteristics of automatically adapting to different fog and haze concentrations,and will not negatively affect the pictures without fog and haze.Secondly,a license plate detection algorithm based on deep learning is designed and trained.This license plate detection algorithm is based on the characteristics of license plate detection in this thesis,after the target detection algorithm YOLOv3 has been modified and optimized,the network suitable for license plate detection is obtained.There are two main improvements: an anchor box parameter calculation method based on hierarchical clustering algorithm is proposed to make the calculation of anchor box parameters more stable and reasonable.on the other hand,for the characteristics of the larger license plate target,the network The multi-scale feature fusion has been optimized to improve the calculation speed.The license plate detection algorithm in this thesis can detect various license plates including civil aviation license plates.Thirdly,a character recognition algorithm capable of recognizing the characters of civil aviation license plates was trained.The character recognition algorithm can recognize the character content of license plates with different character lengths.It solves the problem that the traditional license plate recognition algorithm can only recognize fixed-length character strings.The character recognition algorithm can recognize the image of the license plate area detected by the license plate detection module as a character string without character segmentation.Finally,a set of visual civil license plate recognition system was developed.The license plate recognition system can call the above algorithm to realize the demonstration and verification of the license plate recognition algorithm,which is convenient for the improvement of the algorithm.And by setting up a test environment and test cases,the system was tested for functions and performance.