摘 要:当前越来越多的场景需要将手写体的文字转换为电子格式,手写体识别成为人机交互最便捷的手段之一,拥有广泛的应用前景。文章提出了一种基于 TensorFlow 框架的深度学习手写识别方法,包含手写数字识别和手写汉字识别。以TensorFlow 为框架,采用 CNN 神经网络模型建立训练集以降低识别错误率。实验结果最终表明,对手写数字的识别率达到95%,对手写汉字的识别率达到 90%。
关键词:TensorFlow;手写字体识别;深度学习;人工智能
DOI:10.19850/j.cnki.2096-4706.2021.19.022
中图分类号:TP18;TP391.4 文献标识码:A 文章编号:2096-4706(2021)19-0089-04
Handwritten Word Recognition Based on Deep Learning
WAN Ruyue, HAI Ling, GU Zheng, LIU Wen
(School of Control Engineering, Xinjiang Institute of Engineering, Urumqi 830023, China)
Abstract: At present, more and more scenes need to convert handwritten words into electronic format. Handwritten recognition has become one of the most convenient means of human-computer interaction and has a wide application prospect. This paper proposes a deep learning handwriting recognition method based on TensorFlow framework, including handwritten numeral recognition and handwritten Chinese character recognition. Taking TensorFlow as the framework, the CNN neural network model is used to establish a training set to reduce the recognition error rate. The experimental results finally show that the recognition rate of handwritten numeral reaches 95%, and the recognition rate of handwritten Chinese characters reaches 90%.
Keywords: TensorFlow; handwritten word recognition; deep learning; artificial intelligence
参考文献:
[1] 杨佶 . 基于深度学习的手写汉字识别技术研究 [D]. 沈阳:沈阳师范大学,2019.
[2] 张茹玉 . 基于卷积神经网络和度量学习的脱机手写汉字识别 [D]. 上海:华东师范大学,2018.
[3] 张达峰 . 基于深度卷积神经网络的文字识别算法研究 [D]. 贵阳:贵州大学,2019.
[4] 孙巍巍 . 基于深度学习的手写汉字识别技术研究 [D]. 哈尔滨:哈尔滨理工大学,2017.
[5] 苏日娅 . 基于深度学习的手写汉字识别的研究 [D]. 呼和浩特:内蒙古大学,2019.
[6] 齐照辉 . 基于 TensorFlow 的卷积神经网络应用 [D]. 武汉:武汉大学,2018.
[7] 周甲甲 . 基于深度学习的汉字识别方法研究 [D]. 武汉:华中师范大学,2019.
作者简介:万茹月(1993.06—),女,汉族,新疆乌鲁木齐人,讲师,硕士,研究方向:嵌入式系统;通讯作者:刘文 (1982.02—),男,汉族,新疆乌鲁木齐人,教师,博士,研 究方向:人工智能、大数据。