Research on license plate location and recognition in complex environment

Aiming at the problems of license plate location difficulty and low character recognition accuracy in complex environments, such as a small number of license plate samples, illumination transformation, changeable weather and motion blur, this paper proposes an end-to-end license plate recognition method to improve the location and recognition accuracy in complex environments. First, the cyclic generative adversarial network is used to synthesize the approximate real license plate image to enrich the training set and solve the problem of data imbalance to facilitate subsequent model training. Second, a MF-RepUnet license plate location method is proposed, which integrates the improved VGG structure and feature pyramid into the U-Net model to improve the feature extraction capability of the network, and effectively solve the problem of missing detection of inclined license plate and small-scale license plate. Finally, the convolutional recurrent neural network is improved to accurately predict the feature sequence through the way of attention mechanism weighting, which solves the problem of blurred semantic structure sequence features caused by image degradation and further improves the accuracy of license plate character recognition. Experiments show that the proposed method can effectively improve the accuracy and efficiency of license plate location and character recognition, and can be applied to license plate recognition in various complex environments.