Research on the License Plate Recognition Method Based on Convolutional Neural Network

A method based on convolutional neural network for license plate recognition with character segmentation mode is proposed, which aims to solve the problems of low recognition accuracy and long response time caused by abnormal weather such as rain and snow. A dataset for highway license plate character recognition under abnormal weather conditions, which is constructed by analyzing the proportional segmentation method and the segmentation method based on connected domain, is optimized. Finally, a large number of model training and comparison experiments are carried out based on the optimized data set. The results show that the ResNeXt-LPCR model, which has the highest accuracy, has the accuracy rates of 99.89 %, 98.13 % and 97.74 % on the training set, verification set and test set respectively. The MNASNet-LPCR, whose recognition time is the least with a single character recognition time of 0.0061 s. The ResNeXt-LPCR model, which has the highest accuracy on complete license plate recognition reaching 95.52 %, demonstrates that this character segmentation mode license plate recognition method surpasses the accuracy of similar algorithms and has significant recognition efficiency.