改进胶囊网络的小样本图像分类算法

doi:10.3778/j.issn.1673-9418.2102026

计算机科学与探索 ›› 2022, Vol. 16 ›› Issue (10): 2387-2394.DOI: 10.3778/j.issn.1673-9418.2102026

改进胶囊网络的小样本图像分类算法

王飞龙¹, 刘萍¹, 张玲², 李钢²^,⁺()

1.太原理工大学大数据学院,山西晋中 030600
2.太原理工大学软件学院,山西晋中 030600

收稿日期:2021-02-06 修回日期:2021-04-12 出版日期:2022-10-01 发布日期:2021-04-26
通讯作者: + E-mail: ligang@tyut.edu.cn
作者简介:王飞龙（1996—）,男,山西运城人,硕士研究生,CCF学生会员,主要研究方向为图像处理。
刘萍（1976—）,女,山西忻州人,博士研究生,副教授,主要研究方向为水资源大数据。
张玲（1985—）,女,山西吕梁人,博士研究生,讲师,CCF会员,主要研究方向为机器学习、图像处理。
李钢（1980—）,男,内蒙古包头人,博士研究生,副教授,CCF 会员,主要研究方向为人工智能、视觉信息处理。
基金资助:
国家自然科学基金(61976150);山西省自然科学基金(201901D111091);山西省自然科学基金(201801D21135);山西省高校科技创新项目(JYTKJCX201943)

Modified Algorithm of Capsule Network for Classifying Small Sample Image

WANG Feilong¹, LIU Ping¹, ZHANG Ling², LI Gang²^,⁺()

1. College of Data Science, Taiyuan University of Technology, Jinzhong, Shanxi 030600, China
2. College of Software, Taiyuan University of Technology, Jinzhong, Shanxi 030600, China

Received:2021-02-06 Revised:2021-04-12 Online:2022-10-01 Published:2021-04-26
About author:WANG Feilong, born in 1996, M.S. candidate, student member of CCF. His research interest is image processing.
LIU Ping, born in 1976, Ph.D. candidate, asso-ciate professor. Her research interest is big data of water resources.
ZHANG Ling, born in 1985, Ph.D. candidate, lecturer, member of CCF. Her research interests include machine learning and image processing.
LI Gang, born in 1980, Ph.D. candidate, asso-ciate professor, member of CCF. His research interests include artificial intelligence and vi-sual information processing.
Supported by:
National Natural Science Foundation of China(61976150);Natural Science Foundation of Shanxi Province(201901D111091);Natural Science Foundation of Shanxi Province(201801D21135);University Science and Technology Innovation Project of Shanxi Province(JYTKJCX201943)

摘要/Abstract

摘要：

为了解决胶囊网络不能对复杂的小样本图像进行有效分类的问题,提出一种将Darknet进行改进融入胶囊网络的分类模型。首先将Darknet改进为同时包含浅层与深层特征提取器的模型,浅层特征提取器采用5×5的卷积核以捕捉长距离的边缘轮廓特征,深层特征提取器采用3×3的卷积核以捕捉更深层的语义特征,再将图像的浅层边缘特征与深层语义特征进行融合,以保留图像的有效特征;接着利用胶囊网络对图像有效特征进行向量化处理,解决特征空间表征能力缺失的问题;最后在损失函数中加入L₂正则化项,避免模型的过拟合问题。实验结果表明,在小样本数据集上,该模型相比胶囊网络、DCaps模型分类准确率分别提升28.51个百分点和24.40个百分点,相比ResNet50、Xception等卷积神经网络分别提升21.57个百分点和18.02个百分点,显示该方法对复杂小样本图像分类性能提升明显;同时在大样本数据集上,该模型的分类性能也获得了一定程度的提升。

关键词: 小样本图像, 胶囊网络, Darknet, L₂正则化项, 图像分类

Abstract:

In order to address the problem that the capsule network can not classify complex small sample images effectively, a classification model is proposed on the basis of fusing the improved Darknet with the capsule network. Firstly, the Darknet is upgraded containing both the shallow level extractor and the deep level extractor. The shallow level extractor adopts a 5×5 convolution kernel to capture long-distance edge contour features and the deep level extractor uses a 3×3 convolution kernel to capture deeper semantic features. Then, the extracted edge features and semantic features are fused to preserve effective features of images. Next, the capsule network is used to vectorize these effective features to work out the loss of spatial representation. Finally, L₂ regularization is added in the loss function to avoid the over-fitting. Experimental results show that, on the small sample dataset, the classification accuracy of the proposed model is 28.51 percentage points and 24.40 percentage points higher than that of the models of the capsule network and the DCaps respectively, 21.57 percentage points and 18.02 percentage points higher than that of the ResNet50 and the Xception respectively. Hence it suggests that the method proposed in this paper gains a better performance in classifying complex small sample images. Meanwhile, on the large sample dataset, the classification accuracy of the proposed model has also been improved to a certain extent.

Key words: small sample image, capsule network, Darknet, L₂ regularization, image classification

中图分类号:

TP391

王飞龙, 刘萍, 张玲, 李钢. 改进胶囊网络的小样本图像分类算法[J]. 计算机科学与探索, 2022, 16(10): 2387-2394.

WANG Feilong, LIU Ping, ZHANG Ling, LI Gang. Modified Algorithm of Capsule Network for Classifying Small Sample Image[J]. Journal of Frontiers of Computer Science and Technology, 2022, 16(10): 2387-2394.

图/表 11

图1 胶囊网络与Darknet融合模型结构

Fig.1 Fusion model structure of capsule network and Darknet

图2 未经改进的Darknet模型

Fig.2 Unimproved Darknet model

图3 改进后的Darknet模型

Fig.3 Improved Darknet model

图4 未经改进的Darknet模型提取的特征图谱

Fig.4 Feature map extracted through unimproved Darknet model

图5 改进后的Darknet模型提取的特征图谱

Fig.5 Feature map extracted through improved Darknet model

图6 CIFAR数据示例

Fig.6 Example of CIFAR data

表1 Cap-Dark不同参数的分类准确率单位：%

Table 1 Classification accuracy of Cap-Dark with different parameters

网络参数	CIFAR-10	CIFAR-100
seperable_conv2d+relu+BN+L₂	62.78	39.60
conv2d+selu+BN+L₂	76.29	43.24
conv2d+relu+GN+L₂	80.68	35.45
conv2d+relu+BN+L₁	76.93	30.07
conv2d+relu+BN+L₂₁	76.75	30.56
conv2d+relu+batch_norm+L₂	82.42	47.83

表2 Bi-Cap-Dark不同参数的分类准确率单位：%

Table 2 Classfication accuracy of Bi-Cap-Dark with different parameters

网络参数	CIFAR-10	CIFAR-100
seperable_conv2d+relu+BN+L₂	71.92	46.39
conv2d+selu+BN+L₂	79.39	49.86
conv2d+relu+GN+L₂	82.73	43.52
conv2d+relu+BN+L₁	80.21	42.24
conv2d+relu+BN+L₂₁	79.85	40.63
conv2d+relu+batch_norm+L₂	85.16	54.69

表3 不同模型分类准确率单位：%

Table 3 Classification accuracy of different models

网络模型	CIFAR-10	CIFAR-100
CapsuleNet	55.13	26.18
DCaps	62.65	30.29
ResNet50	82.87	33.12
Xception	83.25	36.67
Cap-Dark-NR	80.66	47.78
Cap-Dark	82.42	47.83
Bi-Cap-Dark	85.16	54.69

图7 CIFAR-10消融实验分类准确率

Fig.7 Classification accuracy of CIFAR-10 in ablation experiments

图8 CIFAR-100消融实验分类准确率

Fig.8 Classification accuracy of CIFAR-100 in ablation experiments

参考文献 15

[1]	LECUN Y, BOTTOU L. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324. DOI URL
[2]	KRIZHEVSKY A, SUTSKEVER I, HINTON G E. Image-Net classification with deep convolutional neural networks[C]// Advances in Neural Information Processing Systems 25, Lake Tahoe, Dec 3-6, 2012. Red Hook: Curran Associa-tes, 2012: 1106-1114.
[3]	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[J]. arXiv:1409.1556, 2014.
[4]	HE K M, ZHANG X Y, REN S Q, et al. Deep residual lear-ning for image recognition[C]// Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recog-nition, Las Vegas, Jun 27-30, 2016. Washington: IEEE Com-puter Society, 2016: 770-778.
[5]	SZEGEDY C, VANHOUCKE V, IOFFE S, et al. Rethin-king the inception architecture for computer vision[C]// Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, Jun 27-30, 2016. Washington: IEEE Computer Society, 2016: 2818-2826.
[6]	HU J, SHEN L, SUN G, et al. Squeeze-and-excitation net-works[C]// Proceedings of the 2018 IEEE Conference on Computer Vision and Pattern Recognition, Jun 18-23, 2018. Washington: IEEE Computer Society, 2018: 7132-7141.
[7]	REDMON J, FARHADI A. YOLOv3: an incremental improve-ment[J]. arXiv:1804.02767, 2018.
[8]	RATNER A J, EHRENBERG H R, HUSSAIN Z, et al. Lear-ning to compose domain-specific transformations for data augmentation[C]// Advances in Neural Information Proces-sing Systems 30, Long Beach, Dec 4-9, 2017. Red Hook: Curran Associates, 2017: 3236-3246.
[9]	HOU P, WANG Y H, LIU Q J. A part-based and feature fu-sion method for clothing classification[C]// LNCS 9916: Pro-ceedings of the 17th Pacific-Rim Conference on Multimedia, Xi’an, Sep 15-16, 2016. Cham: Springer, 2016: 231-241.
[10]	SABOUR S, FROSST N, HINTON G E. Dynamic routing between capsules[C]// Advances in Neural Information Proces-sing Systems 30, Long Beach, Dec 4-9, 2017. Red Hook: Curran Associates, 2017: 3859-3869.
[11]	HINTON G E, SABOUR S, FROSST N. Matrix capsules with EM routing[C]// Proceedings of the 6th International Conference on Learning Representations, Vancouver, Apr 30-May 3, 2018: 1-15.
[12]	ZHANG X K, SUN Y, WANG Y, et al. A novel effective and efficient capsule network via bottleneck residual block and automated gradual pruning[J]. Computers & Electrical Engineering, 2019, 80: 106481.
[13]	CHOLLET F. Xception: deep learning with depthwise sepa-rable convolutions[J]. arXiv:1610.02357, 2016.
[14]	WU Y X, HE K M. Group normalization[C]// LNCS 11217:Proceedings of the 15th European Conference on Computer Vision, Munich, Sep 8-14, 2018. Cham: Springer, 2018: 3-19.
[15]	KLAMBAUER G, UNTERTHINER T, MAYR A, et al. Self-normalizing neural networks[C]// Advances in Neural Infor-mation Processing Systems 30, Long Beach, Dec 4-9, 2017. Red Hook: Curran Associates, 2017: 971-980.

改进胶囊网络的小样本图像分类算法

Modified Algorithm of Capsule Network for Classifying Small Sample Image

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 11

参考文献 15

相关文章 15

编辑推荐

Metrics

[1]	安凤平, 李晓薇, 曹翔. 权重初始化-滑动窗口CNN的医学图像分类[J]. 计算机科学与探索, 2022, 16(8): 1885-1897.
[2]	李宽宽, 刘立波. 双线性聚合残差注意力的细粒度图像分类模型[J]. 计算机科学与探索, 2022, 16(4): 938-949.
[3]	张海涛, 柴思敏. 改进双分支胶囊网络的高光谱图像分类[J]. 计算机科学与探索, 2022, 16(10): 2405-2414.
[4]	张梦倩, 张莉. 粗-细两阶段卷积神经网络算法[J]. 计算机科学与探索, 2021, 15(8): 1501-1510.
[5]	刘靖祎, 史彩娟, 涂冬景, 刘帅. 零样本图像分类综述[J]. 计算机科学与探索, 2021, 15(5): 812-824.
[6]	张利, 邱存月, 张凯鑫, 张大波, 罗浩. 改进胶囊网络优化分层卷积的亚健康识别算法[J]. 计算机科学与探索, 2021, 15(4): 712-722.
[7]	刘晓龙, 王士同. 面向开放集图像分类的模糊域自适应方法[J]. 计算机科学与探索, 2021, 15(3): 515-523.
[8]	杨章静, 王文博, 黄璞, 张凡龙. 基于潜子空间去噪的子空间学习图像分类方法[J]. 计算机科学与探索, 2021, 15(12): 2374-2389.
[9]	安平，冀中，刘西瑶. 任务感知双原型网络的人物交互少样本识别[J]. 计算机科学与探索, 2021, 15(11): 2184-2192.
[10]	李祥霞, 吉晓慧, 李彬. 细粒度图像分类的深度学习方法[J]. 计算机科学与探索, 2021, 15(10): 1830-1842.
[11]	马翔, 邓赵红, 王士同. 多粒度融合的模糊规则系统图像特征学习[J]. 计算机科学与探索, 2021, 15(1): 173-184.
[12]	杨萌林，张文生. 分类激活图增强的图像分类算法[J]. 计算机科学与探索, 2020, 14(1): 149-158.
[13]	陈德运，付立军，张学松，于梁，陈海龙，李骜. 多种表示的图像分类方法[J]. 计算机科学与探索, 2019, 13(12): 2138-2148.
[14]	焦志成，李洁，王颖，高新波. 浅层模糊K均值图像分类网络[J]. 计算机科学与探索, 2015, 9(8): 1018-1024.
[15]	田昊, 李国辉, 廉蔺, 贾立. 用于遥感图像建筑物目标分类的层次匹配核[J]. 计算机科学与探索, 2011, 5(7): 588-594.