2.765

2022影响因子

(CJCR)

  • 中文核心
  • EI
  • 中国科技核心
  • Scopus
  • CSCD
  • 英国科学文摘

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

基于半监督编码生成对抗网络的图像分类模型

付晓 沈远彤 李宏伟 程晓梅

付晓, 沈远彤, 李宏伟, 程晓梅. 基于半监督编码生成对抗网络的图像分类模型. 自动化学报, 2020, 46(3): 531-539. doi: 10.16383/j.aas.c180212
引用本文: 付晓, 沈远彤, 李宏伟, 程晓梅. 基于半监督编码生成对抗网络的图像分类模型. 自动化学报, 2020, 46(3): 531-539. doi: 10.16383/j.aas.c180212
FU Xiao, SHEN Yuan-Tong, LI Hong-Wei, CHENG Xiao-Mei. A Semi-supervised Encoder Generative Adversarial Networks Model for Image Classification. ACTA AUTOMATICA SINICA, 2020, 46(3): 531-539. doi: 10.16383/j.aas.c180212
Citation: FU Xiao, SHEN Yuan-Tong, LI Hong-Wei, CHENG Xiao-Mei. A Semi-supervised Encoder Generative Adversarial Networks Model for Image Classification. ACTA AUTOMATICA SINICA, 2020, 46(3): 531-539. doi: 10.16383/j.aas.c180212

基于半监督编码生成对抗网络的图像分类模型

doi: 10.16383/j.aas.c180212
基金项目: 

国家自然科学基金 61601417

详细信息
    作者简介:

    付晓  中国地质大学(武汉)数学与物理学院硕士研究生. 2015年获得中国地质大学(武汉)数学与物理学院学士学位.主要研究方向为深度学习与图像处理. E-mail: cugfuxiao@163.com

    李宏伟  中国地质大学(武汉)数学与物理学院教授.主要研究方向为信息处理与智能计算. E-mail: hwli@cug.edu.cn

    程晓梅  中国地质大学(武汉)数学与物理学院硕士研究生. 2016年获得山东大学(威海)数学与统计学院统计系学士学位.主要研究方向为深度学习与图像处理. E-mail: 13016471716@163.com

    通讯作者:

    沈远彤  中国地质大学(武汉)数学与物理学院教授.主要研究方向为小波分析理论与应用, 数字图像处理.本文通信作者. E-mail: whsyt@163.com

A Semi-supervised Encoder Generative Adversarial Networks Model for Image Classification

Funds: 

National Natural Science Foundation of China 61601417

More Information
    Author Bio:

    FU Xiao   Master student at the College of Mathematics and Physics, China University of Geosciences. She received her bachelor degree from China University of Geosciences in 2015. Her research interest covers deep learning and image processing.)

    LI Hong-Wei   Professor at the College of Mathematics and Physics, China University of Geosciences. His research interest covers information processing and intelligent computing.)

    CHENG Xiao-Mei   Master student at the College of Mathematics and Physics, China University of Geosciences. She received her bachelor degree from Shandong University in 2016. Her research interest covers deep learning and image processing.)

    Corresponding author: SHEN Yuan-Tong   Professor at the College of Mathematics and Physics, China University of Geosciences. His research interest covers theory and application of wavelet analysis
  • 摘要: 在实际应用中, 为分类模型提供大量的人工标签越来越困难, 因此, 近几年基于半监督的图像分类问题获得了越来越多的关注.而大量实验表明, 在生成对抗网络(Generative adversarial network, GANs)的训练过程中, 引入少量的标签数据能获得更好的分类效果, 但在该类模型的框架中并没有考虑用于提取图像特征的结构, 为了进一步利用其模型的学习能力, 本文提出一种新的半监督分类模型.该模型在原生成对抗网络模型中添加了一个编码器结构, 用于直接提取图像特征, 并构造了一种新的半监督训练方式, 获得了突出的分类效果.本模型分别在标准的手写体识别数据库MNIST、街牌号数据库SVHN和自然图像数据库CIFAR-10上完成了数值实验, 并与其他半监督模型进行了对比, 结果表明本文所提模型在使用少量带标数据情况下得到了更高的分类精度.
    Recommended by Associate Editor JIN Lian-Wen
    1)  本文责任编委 金连文
  • 图  1  SSE-GAN模型中流形一致结合方式

    Fig.  1  The manifold agreement combination method in SSE-GAN

    图  2  SSE-GAN框架图

    Fig.  2  The framework of SSE-GAN

    图  3  模型收敛后生成图像与原MNIST数据库图像对比

    Fig.  3  The generated image and the image from MNIST database after model converges

    图  4  模型收敛后生成图像与原SVHN数据库图像对比

    Fig.  4  The generated image and the image from SVHN database after model converges

    图  5  模型收敛后生成图像与原CIFAR-10数据库图像对比

    Fig.  5  The generated image and the image from CIFAR-10 database after model converges

    表  1  MNIST数据库上不同数量带标数据的半监督训练分类准确率

    Table  1  Using different number of labeled data when semi-supervised training on MNIST

    模型 带标数据个数及对应分类准确率(%)
    100 1 000 全部数据
    Ladder-network[6] 98.14 99.06 -
    Cat-GAN[9] 98.09 99.11 $99.40\pm0.03$
    Improved-GAN[10] 98.58 99.15 $99.40\pm0.02$
    ALI21] 98.77 99.16 $99.45\pm0.01$
    GAR[22] 98.92 99.21 $99.55\pm0.03$
    SSE-GAN 99.10 99.23 99.61±0.03
    下载: 导出CSV

    表  2  SVHN数据库上不同数量带标数据的半监督训练分类准确率

    Table  2  Using different number of labeled data when semi-supervised training on SVHN

    模型 带标数据个数及对应分类准确率(%)
    100 1 000
    Ladder-network[6] 75.50 87.06
    Cat-GAN[9] 77.68 88.90
    Improved-GAN[10] - 90.78
    Virtual Adversarial[23] 79.71 90.99
    Adversarial Training[7] 79.99 91.11
    Bayesian GAN[11] 80.53 92.01
    GAR[22] 80.87 92.08
    SSE-GAN 81.08 92.92
    下载: 导出CSV

    表  3  CIFAR-10数据库上不同数量带标数据的半监督训练分类准确率

    Table  3  Using different number of labeled data when semi-supervised training on CIFAR-10

    模型 带标数据个数及对应分类准确率(%)
    1 000 2 000 4 000
    Ladder-network[6] - 76.52 79.31
    Cat-GAN[9] - 78.83 80.42
    improved-GAN[10] 77.17 79.39 81.37
    ALI[21] 80.02 80.91 81.48
    Adversarial training[7] 81.25 82.88 83.61
    Bayesian GAN[11] 81.89 83.13 84.20
    GAR[[22] 82.10 83.35 84.94
    SSE-GAN 82.34 83.66 85.14
    下载: 导出CSV
  • [1] 张号逵, 李映, 姜晔楠.深度学习在高光谱图像分类领域的研究现状与展望.自动化学报, 2018, 44(6): 961-977 doi: 10.16383/j.aas.2018.c170190

    Zhang Hao-Kui, Li Ying, Jiang Ye-Nan. Deep learning for hyperspectral imagery classification: the state of the art and prospects. Acta Automatica Sinica, 2018, 44(6): 961-977 doi: 10.16383/j.aas.2018.c170190
    [2] Suddarth S C, Kergosien Y L. Rule-injection hints as a means of improving network performance and learning time. Neural Networks. EURASIP 1990, 1990. 120-129
    [3] 李敏, 禹龙, 田生伟, 吐尔根·依布拉音, 赵建国.基于深度学习的维吾尔语名词短语指代消解.自动化学报, 2017, 43(11): 1984- 1992 doi: 10.16383/j.aas.2017.c160330

    Li Min, Yu Long, Tian Sheng-Wei, Ibrahim T, Zhao Jian-Guo. Coreference resolution of uyghur noun phrases based on deep learning. Acta Automatica Sinica, 2017, 43(11): 1984-1992 doi: 10.16383/j.aas.2017.c160330
    [4] 王坤峰, 左旺孟, 谭营, 秦涛, 李力, 王飞跃.生成式对抗网络:从生成数据到创造智能.自动化学报, 2018, 44(5): 769-774 doi: 10.16383/j.aas.2018.y000001

    Wang Kun-Feng, Zuo Wang-Meng, Tan Ying, Qin Tao, Li Li, Wang Fei-Yue. Generative adversarial networks: from generating data to creating intelligence. Acta Automatica Sinica, 2018, 44(5): 769-774 doi: 10.16383/j.aas.2018.y000001
    [5] Dosovitskiy A, Fischer P, Springenberg J T, Riedmiller M, Brox T. Discriminative unsupervised feature learning with exemplar convolutional neural networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 38(9): 1734-1747 doi: 10.1109/TPAMI.2015.2496141
    [6] Rasmus A, Valpola H, Honkala M, Berglund M, Raiko T. Semi-supervised learning with ladder networks. arXiv: 1507. 02672, 2015.
    [7] Miyato T, Maeda S, Ishii S, Koyama M. Virtual adversarial training: a regularization method for supervised and semi-supervised learning. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, DOI: 10.1109/TPAMI.2018. 2858821
    [8] Kingma D P, Rezende D J, Mohamed S, Welling M. Semi-supervised learning with deep generative models. In: Proceedings of the 2014 Neural Information Processing Systems. Massachusetts, USA: MIT Press, 2014. 3581-3589
    [9] Springenberg J T. Unsupervised and semi-supervised learning with categorical generative adversarial networks. arXiv: 1511.06390, 2015.
    [10] Salimans T, Goodfellow I, Zaremba W, Cheung V, Radford A, Chen X. Improved techniques for training GANs. In: Proceedings of the 2016 Neural Information Processing Systems. Massachusetts, USA: MIT Press, 2016. 1-10
    [11] Saatchi Y, Wilson A G. Bayesian GAN. In: Proceedings of the 2017 Neural Information Processing Systems. Massachusetts, USA: MIT Press, 2017. 1-16
    [12] Radford A, Metz L, Chintala S. Unsupervised representation learning with deep convolutional generative adversarial networks. In: Proceedings of the 2016 International Conference on Learning Representations. Piscataway, USA: IEEE, 2016. 1-16
    [13] Donahue J, Krähenbühl P, Darrell T. Adversarial feature learning. In: Proceedings of the 2017 International Conference on Learning Representations. Piscataway, USA: IEEE, 2017. 111-128
    [14] Tenenbaum J B, de Silva V, Langford J C. A global geometric framework for nonlinear dimensionality reduction. Science, 2000, 290(5500): 2319-2323 doi: 10.1126/science.290.5500.2319
    [15] Ioffe S, Szegedy C. Batch normalization: accelerating deep network training by reducing internal covariate shift. In: Proceedings of the 2015 International Conference on Machine Learning. Piscataway, USA: IEEE, 2015. 11-21
    [16] Zheng L, Wang S J, Tian L, He F, Liu Z Q, Tian Q. Query-adaptive late fusion for image search and person re-identification. In: Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway, USA: IEEE, 2015. 1741-1750
    [17] LeCun Y, Bottou L, Bengio Y, Haffner P. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 1998, 86(11): 2278-2324 doi: 10.1109/5.726791
    [18] Netzer Y, Wang T, Coates A, Bissacco A, Wu B, Na A Y. Reading digits in natural images with unsupervised feature learning. In: Proceedings of the 2011 Neural Information Processing Systems. Massachusetts, USA: MIT Press, 2011. 5-16
    [19] Krizhevsky A. Learning Multiple Layers of Features from Tiny Images [Ph. D. dissertation], University of Toronto, Toronto, Canada, 2009.
    [20] Krizhevsky A, Sutskever I, Hinton G E. Imagenet classification with deep convolutional neural networks. In: Proceedings of the 2012 Neural Information Processing Systems. Massachusetts, USA: MIT Press, 2012. 1106-1114
    [21] Dumoulin V, Belghazi I, Poole B, Mastropietro O, Lamb A, Arjovsky M, et al. Adversarially Learned Inference. In: Proceedings of the 2017 International Conference on Learning Representations. Piscataway, USA: IEEE, 2017. 111-128
    [22] Kilinc O, Uysal I. GAR: an efficient and scalable graph-based activity regularization for semi-supervised learning. Neurocomputing, 2018, 296: 46-54 doi: 10.1016/j.neucom.2018.03.028
    [23] Miyato T, Maeda S, Koyama M, Nakae K, Ishii S. Distributional smoothing with virtual adversarial training. In: Proceedings of the 2016 International Conference on Learning Representations. Piscataway, USA: IEEE, 2016. 1-12
  • 加载中
图(5) / 表(3)
计量
  • 文章访问数:  2777
  • HTML全文浏览量:  1246
  • PDF下载量:  390
  • 被引次数: 0
出版历程
  • 收稿日期:  2018-04-12
  • 录用日期:  2018-08-30
  • 刊出日期:  2020-03-30

目录

    /

    返回文章
    返回