License Plate Recognition Based On GASA-BP Neural Network In Haze Weather Method Study

Recent years,with the great progress of science and technology,the computing power of computer has been greatly improved.And license plate recognition technology based on artificial neural network has become a research hotspot.With the development of industry,haze weather appears in our life,whether in the South or in the north,which brings new challenges to the license plate recognition technology,so the research on license plate recognition under haze conditions has certain practical significance.In order to recognize the license plate in haze weather,so the first thing is to complete license plate defogging.In the paper,we deeply study the dark channel defogging algorithm and single-scale Retinex defogging algorithm.For the dim brightness of the image after using the dark channel defogging algorithm,we use logarithmic transformation to improve the brightness of the defogging image.In order to solve the problem that the image details are not prominent after single-scale Retinex defogging,in this paper,Marl operator edge enhancement is performed on the image before single-scale Retinex defogging.Finally,a new algorithm combining improved dark channel and improved single scale Retinex dehazing algorithm is proposed.The experimental results show that the fusion defogging algorithm is more effective than the single defogging algorithm.Secondly,in the license plate location link,the dehazing image is preprocessed first,the image is further enhanced,and then the algorithm of edge operator detection and mathematical morphology fusion positioning algorithm is used for license plate location.The experiment verifies that the fusion location is better than the edge detection algorithm alone.The success rate of fusion localization is much higher than that of single edge detection algorithm.Then,in the character segmentation,after doing some research of the traditional projection method,aiming at the problem that the first character is very to over segmented,and then an improved vertical projection algorithm is comed out to segment the first characters.The algorithm designs four character boxes based on the separator.The number of pixels at the interval between the first character and the second character is counted,and the space character box is counted When the pixel width is set,the first two characters are successfully segmented when the preset threshold is reached.And the last experiment shows that the problem of over segmentation of the first character can be solved in this algorithm.And the last five characters are divided according to the traditional vertical projection method.Eventually,BP network is used for license plate recognition.Due to the random initialization of weights and thresholds,so the traditional BP network is prone to fall into local minimum value and the number of iterations during the training.In order to solve the above problems,a license plate recognition Algorithm based on GASA-BP neural network is proposed in this paper,which combines Genetic Algorithm with Simulated Annealing Algorithm to obtain a set of optimal weights and thresholds and then sends them into BP network for further training.And the algorithm has the global search ability of genetic algorithm(GA)and local search ability of simulated annealing algorithm(SA).Experimental results show that the network has advantages over the unoptimized network in character recognition accuracy and the number of network training iterations,the GASA-BP neural network is used for the license plate recognition after single dehazing and fusion dehazing.In terms of accuracy,fusion dehazing algorithm has obvious advantages over single dehazing algorithm.