With the development of the economy and the advancement of science and technology, the number of vehicles has gradually increased, which has led to an increase in the difficulty of vehicle management in society. How to effectively manage vehicles has become a popular direction at present. As a key component of intelligent transportation systems, license plate recognition is of great significance for application scenarios such as traffic monitoring and violation management. However, traditional license plate recognition methods face challenges from complex backgrounds, different lighting conditions, and various license plate fonts and colors. In order to solve these problems, this study proposes a license plate character recognition method based on improved CNN. In the process, the improved U-Net is first introduced to build the license plate positioning (LPL) model; then the license plate image is effectively preprocessed to improve the accuracy of character recognition. The LPCR model based on the improved LeNet is fused to construct the character recognition model with the proposed improved CNN. The results show that the improved U-Net model can accurately locate the LP region in the license plate image in various scenarios. The average IOU values of the improved U-Net model, ATPA model and MK-means model are 0.964, 0.939 and 0.931, respectively. In the comparison of the license plate image localization precision and recall, the improved U-Net achieves a localization precision of 99.64%, which is 0.65% and 1.06% higher than that of ATPA and MK-means, respectively. The Recall value is 99.35%, which is 0.29% and 0.68% higher than ATPA and MK-means, respectively. In the comparison of ACRR and CCRR values, the improved LeNet model has an ACRR of 99.25%, a CCRR of 99.29%, an RA of 99.18% and a CRA of 99.56%. This indicates that the improved LeNet model can effectively recognize license plate characters in different scenarios. Meanwhile, it has good application effect in vehicle identity detection, which can effectively improve the detection efficiency of license plate recognition.