基于注意力机制的免分割车牌字符识别

Segmentation-Free License Plate Character Recognition Based on Attention Mechanism

针对车牌数据量小的情况下对车牌样本图像进行了数据增强,扩充数据解决小样本数据的均衡性问题。对增强后的车牌图像采用一种基于注意力机制的轻量级车牌字符识别算法,实现了端对端免分割字符的深度学习算法。在算法中引入了SE注意力机制模块,可实现不同通道的权重不同。算法的优点模型结构比较精简,提高识别速度,对车牌的倾斜和光线有一定的鲁棒性。

In the case of small amount of license plate data,the sample image of license plate is enhanced,and the small sample data is extended to solve the problem of data equalization.A light-weight license plate character recognition algorithm based on attention mechanism is applied to the enhanced license plate image to realize the end-to-end segmentation-free character deep learning algorithm.SE attention mechanism module is introduced into the algorithm,which can realize different channel weights.The model structure of the algorithm is simplified,the recognition speed is improved,and the algorithm is robust to the slant of the license plate and the light.