摘要
研究西波手写文字智能识别是保护和传承彝族古老文献资料的一项重要研究工作.西波手写文字收集困难,基于神经网络的智能识别技术难以直接应用于西波手写文字智能识别.针对西波手写文字样本少的特点,提出了MANN(memory augemented neural networks,记忆增强网络)的元学习方法.在小样本条件下,该方法能提高西波手写文字的智能识别精度.实验结果显示,引入元学习方法后,不仅明显缩短了识别算法训练时间,而且显著提高了识别精度.
The research on the intelligent recognition of Xibo handwritten characters is an important work to protect and inherit the ancient documents of the Yi nationality. Due to the difficulty in collecting Xibo handwritten texts, it is no easy to apply directly the intelligent recognition technology based on neural networks to the intelligent recognition of Xibo handwritten characters. With a consideration of the small samples of Xibo handwritten characters in the database, a meta-learning method of MANN(Memory Augmented Neural Networks) is proposed to improve the intelligent recognition accuracy of Xibo handwritten characters. The experimental results show that this meta-learning method not only significantly reduces the training time for the recognition algorithm, but also much improves the recognition accuracy.
作者
何翠玲
杨柱元
唐轶
蒋作
HE Cui-ling;YAN Zhu-yuan;TAN Yi;JIANG Zuo(School of Mathematics and Computer Science,Yunnan Minzu University,Kunming 650500,China)
出处
《云南民族大学学报（自然科学版）》
CAS
2020年第6期592-600,共9页
Journal of Yunnan Minzu University:Natural Sciences Edition
基金
国家自然科学基金(61866040,11361076)。
关键词
西波手写文字识别
小样本
记忆增强网络
元学习
recognition of Xibo handwritten characters
few-shot
memory augmented neural networks
meta-learning
