License Plate Recognition Method based on Convolutional Neural Network

Automatic license plate detection and recognition is very important for intelligent vehicle management. At present, the license plate recognition technology mainly obtains the better picture under the good illumination condition to achieve the high precision recognition effect. However, there are still great challenges to achieve better recognition effect by using the existing technology in long distance and large Angle license plate recognition. In order to solve the problem of license plate recognition under complex conditions, this paper constructs a suitable network structure and training parameters to improve the accuracy of license plate recognition. The license plate contains Chinese characters, whose structure is more complex than alphanumeric ones. Therefore, the paper uses the existing convolutional neural network LeNet-5 for transfer learning and improves it into a network suitable for license plate character recognition. Finally, 31 kinds of Chinese character data and 34 kinds of alphanumeric data are trained and tested to get the best parameter combination in the network. Meanwhile, the influence of parameter and function selection on recognition accuracy is analyzed and illustrated.