基于A-CapsNet的西夏文字识别研究  

TANGUT CHARACTER RECOGNITION BASED ON A-CAPSNET

作　　者：杨丽娟 孟一飞 王葭 毛威 孟斌 Yang Lijuan;Meng Yifei;Wang Jia;Mao Wei;Meng Bin(School of Physics and Electronic-electrical Engineering,Ningxia University,Yinchuan 750021,Ningxia,China;Zhengzhou Branch of China United Network Communication Co.,Ltd.,Zhengzhou 450000,Henan,China)

机构地区：[1]宁夏大学物理与电子电气工程学院,宁夏银川750021 [2]中国联合网络通信有限公司郑州分公司,河南郑州450000

出　　处：《计算机应用与软件》2024年第8期219-224,239,共7页Computer Applications and Software

基　　金：宁夏重点研发计划项目(2020BFG02013)。

摘　　要：针对西夏文字结构复杂、笔画繁多、类别之间相似度较高以及各类别样本数量分布不均衡等问题,论证了将CapsNet网络架构应用于西夏文识别的可行性和优越性,进而提出A-CapsNet网络,运用AlexNet网络在深层次上对图像信息进行提取的优越性能,来弥补CapsNet高层胶囊所接收的缺失特征信息,从AlexNet模块、Capsule模块进行实验分析,实验结果表明,A-CapsNet网络对西夏文字的识别率可以达到94%,比原始的胶囊网络提高了3百分点,并且都优于深度学习卷积神经网络,具有很好的适用性,为研究西夏文字做了一定的贡献。Tangut characters have the characteristics of complex structure,numerous strokes,high similarity between categories,and uneven distribution of samples in each category.In view of these characteristics,we demonstrated the feasibility and superiority of applying the CapsNet network architecture to Tangut text recognition,and the A-CapsNet network was proposed.The superior performance of the AlexNet network to extract image information at a deep level was used to make up for the missing feature information received by the CapsNet high-level capsule.From the experimental analysis of the AlexNet module and the Capsule module,the experimental results show that the proposed A-CapsNet network can achieve 94% recognition rate of Tangut characters,which is 3 percentage points higher than the original capsule network.And A-CapsNet network is better than deep learning convolutional neural network,and it has good applicability.This research has made a certain contribution to the study of Tangut characters.

关 键 词：AlexNet A-CapsNet 西夏文字识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]