《计算机应用》唯一官方网站 ›› 2023, Vol. 43 ›› Issue (5): 1330-1335.DOI: 10.11772/j.issn.1001-9081.2022030367

• 第九届中国数据挖掘会议 • 上一篇    

改进的基于多路径特征的胶囊网络

徐清海1, 丁世飞1,2(), 孙统风1,2, 张健1,2, 郭丽丽1,2   

  1. 1.中国矿业大学 计算机科学与技术学院, 江苏 徐州 221116
    2.矿山数字化教育部工程技术研究中心(中国矿业大学), 江苏 徐州 221116
  • 收稿日期:2022-03-22 修回日期:2022-04-20 接受日期:2022-04-28 发布日期:2023-05-08 出版日期:2023-05-10
  • 通讯作者: 丁世飞
  • 作者简介:徐清海(1998—),男,江西景德镇人,硕士研究生,主要研究方向:深度学习、神经网络
    丁世飞(1963—),男,山东青岛人,教授,博士,CCF会员,主要研究方向:人工智能、模式识别、机器学习、数据挖掘 dingsf@cumt.edu.cn
    孙统风(1977—),男,江苏徐州人,副教授,博士,CCF会员,主要研究方向:机器学习、神经网络、深度学习
    张健(1990—),男,山东泰安人,讲师,博士,CCF会员,主要研究方向:机器学习、深度学习
    郭丽丽(1990—),女,山东临沂人,讲师,博士,CCF会员,主要研究方向:深度学习、多模态情感识别。
  • 基金资助:
    国家自然科学基金资助项目(61976216)

Improved capsule network based on multipath feature

Qinghai XU1, Shifei DING1,2(), Tongfeng SUN1,2, Jian ZHANG1,2, Lili GUO1,2   

  1. 1.School of Computer Science and Technology,China University of Mining and Technology,Xuzhou Jiangsu 221116,China
    2.Engineering Research Center of Mine Digitization,Ministry of Education (China University of Mining and Technology),Xuzhou Jiangsu 221116,China
  • Received:2022-03-22 Revised:2022-04-20 Accepted:2022-04-28 Online:2023-05-08 Published:2023-05-10
  • Contact: Shifei DING
  • About author:XU Qinghai, born in 1998, M. S. candidate. His research interests include deep learning, neural network.
    DING Shifei, born in 1963, Ph. D., professor. His research interests include artificial intelligence, pattern recognition, machine learning, data mining.
    SUN Tongfeng, born in 1977, Ph. D., associate professor. His research interests include machine learning, neural network, deep learning.
    ZHANG Jian, born in 1990, Ph. D., lecturer. His research interests include machine learning, deep learning.
    GUO Lili, born in 1990, Ph. D., lecturer. Her research interests include deep learning, multimodal emotion recognition.
  • Supported by:
    National Natural Science Foundation of China(61976216)

摘要:

针对胶囊网络(CapsNet)在复杂数据集上的分类效果差,而且在路由过程中参数数量过大等问题,提出一种基于多路径特征的胶囊网络(MCNet),包含新的胶囊特征提取器和新的胶囊池化方法。该胶囊特征提取器从多个不同路径中并行地提取不同层次、不同位置的特征,然后将特征编码为包含更多语义信息的胶囊特征;胶囊池化方法则在胶囊特征图的每个位置选取最活跃的胶囊,用少量的胶囊表示有效的胶囊特征。在4个数据集(CIFAR-10、SVHN、Fashion-MNIST、MNIST)上与CapsNet等模型进行了对比。实验结果显示,MCNet在CIFAR-10数据集上的分类准确率为79.27%,可训练的参数数量为6.25×106,与CapsNet相比,MCNet的分类准确率提升了8.7%,参数数量减少了46.8%。MCNet能够有效提升分类准确率,同时减少可训练的参数数量。

关键词: 胶囊网络, 深度学习, 动态路由, 胶囊池化, 反卷积重构

Abstract:

Concerning the problems of poor classification of Capsule Network (CapsNet) on complex datasets and large number of parameters in the routing process, a Capsule Network based on Multipath feature (MCNet) was proposed, including a novel capsule feature extractor and a novel capsule pooling method. By the capsule feature extractor, the features of different layers and locations were extracted in parallel from multiple paths, and then the features were encoded into capsule features containing more semantic information. In the capsule pooling method, the most active capsules at each position of the capsule feature map were selected, and the effective capsule features were represented by a small number of capsules. Comparisons were performed on four datasets (CIFAR-10, SVHN, Fashion-MNIST, MNIST) with models such as CapsNet. Experimental results show that MCNet has the classification accuracy of 79.27% on CIFAR-10 dataset and the number of trainable parameters of 6.25×106; compared with CapsNet, MCNet has the classification accuracy improved by 8.7%, and the number of parameters reduced by 46.8%. MCNet can effectively improve the classification accuracy while reducing the number of trainable parameters.

Key words: Capsule Network (CapsNet), deep learning, dynamic routing, capsule pooling, deconvolutional reconstruction

中图分类号: