Please wait a minute...
浙江大学学报（理学版）
浙江大学学报（理学版）  2019, Vol. 46 Issue (3): 261-269    DOI: 10.3785/j.issn.1008-9497.2019.03.001
文化计算     
一种基于深度学习的古彝文识别方法
陈善雄1, 王小龙1, 韩旭1, 刘云2, 王明贵2
1.西南大学 计算机与信息科学学院，重庆 400715
2.贵州工程应用技术学院 彝文研究院，贵州 毕节 551700
A recognition method of Ancient Yi character based on deep learning
Shanxiong CHEN1, Xiaolong WANG1, Xu HAN1, Yun LIU2, Minggui WANG2
1.Southwest University, Chongqing 400715, China
2.Department of Information Engineering, Guizhou University of Engineering Science,Bijie 551700, Guizhou Province, China
 全文: PDF(1878 KB)   HTML  
摘要： 作为世界六大古文字之一的古彝文记录了几千年来人类的发展历史。通过对古彝文的识别能够将这些珍贵文献资料转换为电子文档，便于保存和传播。由于历史发展、区域限制等原因，针对古彝文识别的研究鲜有成果。本文将当前新颖的深度学习技术应用于古老的文字识别。在四层卷积神经网络（convolutional neural network, CNN）基础上扩展出5个模型，然后利用Alpha-Beta散度作为惩罚项,对5个模型的输出神经元重新进行自编码，接着用2个全连接层完成特征压缩，最后在softmax层对古彝文字符特征进行重新评分，得到其概率分布，选择对应的最高概率作为识别的字符。实验表明，相对于传统CNN模型，本文方法对古彝文手写体的识别精度更高。
关键词： 古彝文;  深度学习;  卷积神经网络;  散度    
Abstract: Ancient Yi is one of six kinds of ancient word in the world and it recorded human development history in thousands of years. The recognition technology of Ancient Yi character will enable us to transform lots of precious Ancient Yi literature into electronic documents which is convenient for storage and spread. Due to imbalance of historical development and territorial limitation, the research on recognition of ancient Yi is rare. In this article, we apply novel deep learning to ancient character recognition. Our framework consists of five models which are extended based on four-layer convolutional neural network (CNN).Then we take Alpha-Beta divergence as penalty term to implement the coding for output neurons of five models. Next, two fully connected layers finish characteristics compression. Finally, we use softmax layer to reevaluate characteristics of ancient Yi character and get the probability distribution. The character owning the highest probability is identified as the target character. Experiments show that our method has higher precision compared with the traditional CNN model for handwriting recognition of the ancient Yi.
Key words: Ancient Yi    deep learning    convolutional neural network    divergence
收稿日期: 2019-01-11 出版日期: 2019-05-25
CLC:  TP 391  
基金资助: 国家自然科学基金资助项目(61603310)；中国博士后基金项目(2015M580765)；重庆市博士后科研项目(Xm2016041)；中央高校基本科研业务费项目(XDJK2018B020,XDJK2018B019).
作者简介: 陈善雄(1981—)，ORCID:http://orcid.org/0000-0002-3053-7824，男，博士，副教授，主要从事模式识别研究,E-mail:csxpml@163.com.
服务　　
把本文推荐给朋友
加入引用管理器
E-mail Alert
RSS
作者相关文章　　
陈善雄
王小龙
韩旭
刘云
王明贵

引用本文:

陈善雄, 王小龙, 韩旭, 刘云, 王明贵. 一种基于深度学习的古彝文识别方法[J]. 浙江大学学报（理学版）, 2019, 46(3): 261-269.

Shanxiong CHEN, Xiaolong WANG, Xu HAN, Yun LIU, Minggui WANG. A recognition method of Ancient Yi character based on deep learning. Journal of Zhejiang University (Science Edition), 2019, 46(3): 261-269.

链接本文:

https://www.zjujournals.com/sci/CN/10.3785/j.issn.1008-9497.2019.03.001    或    https://www.zjujournals.com/sci/CN/Y2019/V46/I3/261

1 ZHUC X. Collection and Study of Ancient Books in Yi Language[M].Beijing:Ethnic Press,2008
2 GAOJ,LIUJ Z. Problems and countermeasures of digitization of ancient books in mainland China[J]. Journal of Chinese Library,2013,39(4):110-119.
3 WANGJ M,WENY H,LIY Q,et al. The recognition system of old-Yi character based on the image segmentation[J].Journal of Yunnan Nationalities University (Natural Science Edition),2018,17(1):76-79.
4 ZHUL H,WANGJ M. Off-line handwritten Yi character recognition based on the multi-classifier ensemble with combination features [J]. Journal of Yunnan Nationalities University (Natural Science Edition), 2010,19(5):329-333.
5 朱宗晓,吴显礼.脱机印刷体彝族文字识别系统的原理与实现[J].计算机技术与发展,2012,22(2):85-88. ZHU Z X,WU X L. Principle and implementation of an off-line printed Yi character recognition system[J].Computer Technology and Development,2012,22(2):85-88.doi:10.3969/j.issn.1673-629X.2012.02.022
6 LIUS,LIY D. Design and realization on character segmentation method for Yi language[J]. Journal of South-Central University for Nationalities（Natural Science Edition）,2007,26(3):74-76.
7 WUB. Analysis and research of standard Yi characters based on the angel of character recognition[J].Journal of Southwest Minzu University（Humanities and Social Science Edition）,2018(9):47-53.
8 Yi Literature Cooperation Group.Collection of Yi Chracters[M]. Kunming:Yunnan Nationalities Press,2004.
9 RENX H, ZHOUY, HEJ H, et al. A convolutional neural network-based Chinese text detection algorithm via text structure modeling[J]. IEEE Transactions on Multimedia, 2017,19(3):506-518.doi:10.1109/tmm.2016.2625259
10 AKHANDM A H, AHMEDM, RAHMANM M H,et al. Convolutional neural network training incorporating rotation-based generated patterns and handwritten numeral recognition of major Indian scripts[J]. IETE Journal of Research, 2018,64(2):176-194.doi:10.1080/03772063.2017.1351322
11 NASEEA, ZAFARK. Comparative analysis of raw images and meta feature based Urdu OCR using CNN and LSTM[J]. International Journal of Advanced Computer Science and Applications,2018,9(1):419-424.doi:10.14569/ijacsa.2018.090157
12 SINDAGIV A, PATELV M. A survey of recent advances in CNN-based single image crowd counting and density estimation[J].Pattern Recognition Letters,2018,107:3-16doi:10.1016/j.patrec.2017.07.007
13 DENGX M, ZHANGY D,YANGS, et al. Joint hand detection and rotation estimation using CNN[J].IEEE Transactions on Image Processing, 2018,27(4):1888-1900.doi:10.1109/tip.2017.2779600
14 CICHOCKIA, AMARIS I. Families of Alpha-Beta-and Gamma- divergences: Flexible and robust measures of similarities[J]. Entropy, 2010,12(6):1532-1568.
15 CICHOCKIA, CRUCESS, AMARIS. Generalized Alpha-Beta divergences and their application to robust nonnegative matrix factorization[J].Entropy, 2011,13(1):134-170.doi:10.3390/e13010134
16 SHIW W, GONGY H, TAOX Y,et al. Training DCNN by combining max-margin, max-correlation objectives, and correntropy loss for multilabel image classification[J].IEEE Transactions on Neural Networks and Learning Systems,2017,29(7):2896-2908.doi:10.1109/tnnls.2017.2705222
17 SENGUPTAA , SHIMY, ROY K. Proposal for an all-spin artificial neural network: Emulating neural and synaptic functionalities through domain wall motion in ferromagnets[J].IEEE Transactions on Biomedical Circuits and Systems, 2016,10(6):1152-1160.doi:10.1109/tbcas.2016.2525823
18 KNAGP, KIM J K, CHENT, et al. A sparse coding neural network ASIC with on-chip learning for feature extraction and encoding[J]. IEEE Journal of Solid-State Circuits, 2015,50(4):1070-1079.doi:10.1109/jssc.2014.2386892
19 QIANS,LIUH,LIUC,et al. Adaptive activation functions in convolutional neural networks[J]. Neurocomputing, 2018,272:204-212.doi:10.1016/j.neucom.2017.06.070
20 áARCOS-GARCíA, álVAREZ-GARCíAJ A, SORIA-MORILLOL M. Deep neural network for traffic sign recognition systems: An analysis of spatial transformers and stochastic optimisation methods[J]. Neural Networks, 2018,99:158-165.doi:10.1016/j.neunet.2018.01.005
21 GOPALAKRISHNANK, KHAITANS K, CHOUDHARYA, et al. Deep convolutional neural networks with transfer learning for computer vision-based data-driven pavement distress detection[J]. Construction and Building Materials, 2017,157:322-330.doi:10.1016/j.conbuildmat.2017.09.110
22 Guizhou Institute of Yi Studies. Southwest Yi Chi[M].Guiyang: Guizhou Nationalities Press,2015.
[1] 徐圣嘉,苏程,朱孔阳,章孝灿. 基于深度学习的岩石薄片矿物自动识别方法[J]. 浙江大学学报（理学版）, 2022, 49(6): 743-752.
[2] 刘华玲,张国祥,马俊. 图嵌入算法研究进展[J]. 浙江大学学报（理学版）, 2022, 49(4): 443-456.
[3] 李丰庆, 何林李. 链刚性和剪切场对纳米棒在纳米复合体系中分布和取向的影响[J]. 浙江大学学报（理学版）, 2021, 48(6): 694-699.
[4] 钱立辉, 王斌, 郑云飞, 章佳杰, 李马丁, 于冰. 基于图像深度预测的景深视频分类算法[J]. 浙江大学学报（理学版）, 2021, 48(3): 282-288.
[5] 陈园琼, 邹北骥, 张美华, 廖望旻, 黄嘉儿, 朱承璋. 医学影像处理的深度学习可解释性研究进展[J]. 浙江大学学报（理学版）, 2021, 48(1): 18-29.
[6] 傅颖颖, 张丰, 杜震洪, 刘仁义. 融合图卷积神经网络和注意力机制的PM2.5小时浓度多步预测[J]. 浙江大学学报（理学版）, 2021, 48(1): 74-83.
[7] 王协, 章孝灿, 苏程. 基于多尺度学习与深度卷积神经网络的遥感图像土地利用分类[J]. 浙江大学学报（理学版）, 2020, 47(6): 715-723.
[8] 李君轶, 任涛, 陆路正. 游客情感计算的文本大数据挖掘方法比较研究[J]. 浙江大学学报（理学版）, 2020, 47(4): 507-520.
[9] 郑锐, 钱文华, 徐丹, 普园媛. 基于卷积神经网络的刺绣风格数字合成[J]. 浙江大学学报（理学版）, 2019, 46(3): 270-278.
[10] 黄婕, 张丰, 杜震洪, 刘仁义, 曹晓裴. 基于RNN-CNN集成深度学习模型的PM2.5小时浓度预测[J]. 浙江大学学报（理学版）, 2019, 46(3): 370-379.
[11] 胡伟俭, 陈为, 冯浩哲, 张天平, 朱正茂, 潘巧明. 应用于平扫CT图像肺结节检测的深度学习方法综述[J]. 浙江大学学报（理学版）, 2017, 44(4): 379-384.
[12] 刘玉杰, 庞芸萍, 李宗民, 李华. 融合抽象层级变换和卷积神经网络的手绘图像检索方法[J]. 浙江大学学报（理学版）, 2016, 43(6): 657-663.
[13] 贾厚玉. 有界凸区域上的Dirichlet问题的正则性[J]. 浙江大学学报（理学版）, 1999, 26(2): 1-7.