A Novel License Plate Recognition System Based On Memoristor Neural Networks

License plate recognition(LPR)employing neural networks have developed deeply in recent years,but it is still hard to complete the design of hardware circuits for the complexity of its own,making the use of LPR based on neural networks is only be limited in the software simulation,which requires plenty of computer resources and shows a serious limitation in applications.In this paper,a novel memristor neural network was constructed in view of the unique electrical properties of the memristor and the applications of LPR based on memristor neural network were studied to explore a new LPR technology which is easier to implementin hardware struction,demonstrating a great realistic significance in circuit integration of neural networks.In the present work,the LPR system including acquisition of the license plate images,preprocessing,location,character correction,segmentation and recognition,etc.,was studied making use of MATLAB on the basis of researches of image processing technology.Thereinto,a memristor model was set up and simulated in the light of the existing models and electrical properties of memristors.Then a synapse-based neural network was constructed using memristors as the synapses of an artificial neural network,and the performances of such memristors synapse-based neural networks were analyzed in detail,thus enabling a LPR system of a dynamic synapse neural network.Meanwhile,the LPR system performance of memristors synapse-based neural networks was compared with those formed with the traditional BP neural network.The simulation results showed that memoristor neural networks exhibit prominent advantages in training speed although they are weaker in recognition rate.The hardware circuit of this LPR system is easy to realized,solving the problems like higher power consumption and integration difficulty in the design of artificial synapses relying onthe integrated circuits and ultra-large scale integrated circuits,which helps to make up shortcomings in hardware implementation of traditional neural networks.Therefore it provides a theoretical basis and implementing scheme in applications of neural networks,artificial intelligence and machine vision etc..