基于数据增强的小样本字符识别模型

Small Sample Character Recognition Model Based on Data Augmentation

小样本学习是机器学习的重要研究方向,过去的几十年里,针对小样本学习已经展开了许多研究。针对一些银行支票原始数据积累不足、支票磁条码字符识别准确率低的问题,提出了一种基于数据增强的小样本字符识别方法。通过原始少量样本设计了定制的数据增强方案,有效解决了小样本冷启动的问题,提高模型泛化能力,实现磁条码的准确识别。经测试,该方法的识别准确率达到了95%以上。

Small sample learning is an important research direction of machine learning. In the past decades, many researches have been carried out on small sample learning. Aiming at the problem of insufficient original data accumulation and low character recognition accuracy of magnetic bar code of check in some banks, a small sample character recognition method based on data enhancement is proposed. A customized data enhancement scheme is designed through a small number of original samples, which effectively solves the problem of cold start of small samples, improves the model generalization ability, and realizes the accurate recognition of magnetic bar codes. Through testing, the recognition accuracy of this method is more than 95%.