基于改进CRNN算法的电动自行车车牌识别研究

Electric license plate recognition based on improved CRNN algorithm

基于成熟的机动车违章牌照识别系统及电动自行车车牌和机动车车牌的相似性,提出了基于深度学习的电动自行车车牌识别算法:对数据集使用了数据增强以及增加仿造车牌的方法进行扩充,有效提升了数据质量;将所有车牌图像输入改进的卷积神经网络并提取特征,通过双向长短期记忆(bidirectional long short-term memory,BiLSTM)循环神经网络和时序分类网络进行字符序列的识别;对于卷积循环神经网络(convolutional recurrent neural network,CRNN),本文在卷积网络部分减少了层数,调整了隐藏层的神经元个数。在广西柳州电动自行车车牌数据集上,本文提出的识别算法精准率达到98.40%,比CRNN模型提升2.80%,且对于污垢、形变、模糊等复杂自然场景下的车牌,本文的方法也能够实现有效识别,具有广泛的应用场景。

This paper proposes an electric license plate recognition algorithm based on deep learning in view of the fact that the vehicle license plate recognition system has become mature, and the similarity between electric license plates and motor vehicle license plates. Firstly, the data set is expanded by data augmentation and the method of adding counterfeit license plates, which effectively improves the data quality. Secondly, all license plate images are input into the improved convolutional neural network to extract features, and then the bidirectional long short-term memory(BiLSTM) recurrent neural network and the time series classification network are used to recognize the character sequence. Then, for convolutional recurrent neural network(CRNN), this paper reduces the number of layers in the convolutional network and adjusts the number of neurons in the hidden layer. On the Liuzhou electric license plate data set in Guangxi, the accuracy rate of the recognition algorithm proposed in this paper reaches98.40%, which is a 2.80% improvement compared to the CRNN model. And for license plates in complex natural scenes such as dirt, deformation, and blur, the method in this paper can also achieve effective recognition and has a wide range of application scenarios.