An American license plate detection and recognition technology based on deep learning

There are many problems in American license plates, such as varied background patterns and complex text information. Traditional license plate recognition methods are difficult to meet the demand for different types of license plate recognition problems. Through the research of the connectionist text proposal network (CTPN) and convolutional recurrent neural network (CRNN), an American license plate detection and recognition method was proposed in this study. The SE-MobileNetV2 fast feature extraction model was proposed to greatly improve the speed of feature extraction. Aiming at the situation that many text boxes were produced in American license plate detection, and that some license plate numbers were not in the same text box because of the pattern partition, an anchor box method for screening license plate numbers was designed, which used Adam optimization algorithm to train CRNN. We built an American license plate data set and carried out verification experiments. The recognition rates of state name, license plate number and common recognition rate reached 92%, 84% and 82% respectively. Moreover, the network model proposed in this study was less than 60 MB in size, having strong real-time performance. © 2023 Editorial Board of Journal of Harbin Engineering. All rights reserved.