一种采用动态子空间的小样本图像分类算法

doi:10.19665/j.issn1001-2400.2022.05.019

摘要/Abstract

摘要：

针对当前基于度量学习的小样本图像分类算法中普遍存在的图像分类精度不高与泛化性能一般的问题,提出了一种采用动态子空间的小样本图像分类算法。首先,使用残差神经网络提取小样本图像特征,对各类特征向量进行分解后,动态生成表征图像类别的正交化投影子空间,增强类间特征的差异性;其次,通过融合子空间损失函数与交叉熵损失函数,增强同类样本的特征相似性,构建基于小样本学习的动态子空间分类器,随采样量与样本相似度的变化动态更新子空间的类间距离;最后,将目标图像的特征向量输入动态子空间分类器,使用平方欧氏距离与softmax函数计算类别概率,预测其所属类别。在mini-ImageNet、CIFAR-100和Pascal VOC2007小样本数据集上进行性能测试,并与当前主流的小样本图像分类算法进行了比较,所提出算法的图像分类精度高于对比算法。在5-way 5-shot的条件下,分类精度比当前性能较好的深度子空间分类网络提高了2.3%。实验结果表明,所提出的算法具有较强的泛化性能与抗干扰能力。

关键词: 动态子空间, 小样本数据, 正交投影, 图像分类, 残差神经网络

Abstract:

The existing few-shot image classification algorithms based on metric learning have a low precision of image categorization and weak generalization performance.A few-shot image classification algorithm by dynamic subspace is proposed in this paper.First,a residual neural network is used to extract few-shot image features.The dynamically orthogonalized projection subspaces representing image categories are generated with decomposed image features of various categorizations so as to enhance the difference of features among categories in orthogonalized projection subspaces.Second,a dynamic subspace classifier based on few-shot learning is constructed by fusing the subspace loss function and the cross-entropy loss function so as to enhance the similarity of samples in the same category.The inter-class distance of the subspace is dynamically updated with the change of sampling amount and sample similarity.Finally,the feature vector of the target image is input into the dynamic subspace classifier,and the Euclidean distance square and the soft max function are used to calculate the category probability of the target feature and predict its category.Performance testing is performed on the few-shot data sets such as mini-ImageNet,CIFAR-100 and Pascal VOC2007.The proposed algorithm is superior to the current mainstream few-shot image classification algorithm,and the average classification precision of the proposed algorithm is 2.3% higher than that of the current DSN with good performance under 5-way 5-shot.Experiments show that the proposed algorithm has a strong generalization performance and an anti-interference ability.

Key words: dynamic subspace, few-shot data, orthogonal projection, image classification, residual network

中图分类号:

TP391

任佳兴, 曹玉东, 曹睿, 闫佳. 一种采用动态子空间的小样本图像分类算法[J]. 西安电子科技大学学报, 2022, 49(5): 166-174.

REN Jiaxing, CAO Yudong, CAO Rui, YAN Jia. Algorithm for classification of few-shot images by dynamic subspace[J]. Journal of Xidian University, 2022, 49(5): 166-174.

图/表 8

图1

表1

ResNet-18网络结构"

卷积层名称	输出尺寸	网络层结构
Conv1	112×112	7×7,64,stride2,3×3maxpool,stride2
Conv2	56×56	$3 × 3,64 3 × 3,64$ ×2
Conv3	28×28	$3 × 3,128 3 × 3,128$ ×2
Conv4	14×14	$3 × 3,256 3 × 3,256$ ×2
Conv5	7×7	$3 × 3,512 3 × 3,512$ ×2
Conv6	1×1	average pool,1000-d fc,softmax
FLOPs		1.8×10⁹

表1

图2

图3

表2

表3

图4

图5

参考文献 23

[1]	刘颖, 雷研博, 范九伦, 等. 基于小样本学习的图像分类技术综述[J]. 自动化学报, 2021, 47(2):297-315.
	LIU Ying, LEI Yanbo, FAN Jiulun, et al. Survey on Image Classification Technology Based on Small Sample Learning[J]. Acta Automatica Sinica, 2021, 47(2):297-315
[2]	赵凯琳, 靳小龙, 王元卓. 小样本学习研究综述[J]. 软件学报, 2021, 32(2):349-369.
	ZHAO Kailin, JIN Xiaolon, WANG Yuanzhuo. Survey on Few-shot Learning[J]. Journal of Software, 2021, 32(2):349-369.
[3]	宋闯, 赵佳佳, 王康, 等. 面向智能感知的小样本学习研究综述[J]. 航空学报, 2020, 41(z1):12-25.
	SONG Chuang, ZHAO Jiajia, WANG Kang, et al. A Survey of Few Shot Learning Based on Intelligent Perception[J]. Acta Aeronautica et Astronautica Sinica, 2020, 41(z1):12-25.
[4]	VINYALS O, BLUNDELL C, LILLICRAP T, et al. Matching Networks for One Shot Learning[J]. Advances in Neural Information Processing Systems, 2016, 29:3630-3638.
[5]	KOCH G, ZEMEL R, SALAKHUTDINOV R. Siamese Neural Networks for One-Shot Image Recognition[C]// Proceedings of the ICML Deep Learning Workshop. Piscataway: IEEE, 2015:1-30.
[6]	SNELL J, SWERSKY K, ZEMEL R S. Prototypical Networks for Few-Shot Learning[J]. Advances in Neural Information Processing Systems, 2017, 30:197-209.
[7]	LI H, EIGEN D, DODGE S, et al. Finding Task-Relevant Features for Few-Shot Learning by Category Traversal[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2019:1-10.
[8]	PEREZ-RUA J M, ZHU X, HOSPEDALES T, et al. Incremental Few-Shot Object Detection[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020:13846-13855.
[9]	SIMON C, KONIUSZ P, NOCK R, et al. Adaptive Subspaces for Few-Shot Learning[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020:4136-4145.
[10]	LIANG X, WANG X, ZHEN L, et al. Soft-Margin Softmax for Deep Classification[C]// Proceedings of the International Conference on Neural Information Processing. Piscataway: IEEE, 2017:413-421.
[11]	JIANG Q, YAN X. Parallel PCA-KPCA for Nonlinear Process Monitoring[J]. Control Engineering Practice, 2018, 80:17-25. doi: 10.1016/j.conengprac.2018.07.012
[12]	HE K, ZHANG X, REN S, et al. Deep Residual Learning for Image Recognition[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2016:770-778.
[13]	KANG X, XIANG X, LI S, et al. PCA-Based Edge-Preserving Features for Hyperspectral Image Classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2017, 55(12):7140-7151. doi: 10.1109/TGRS.2017.2743102
[14]	HARANDI M, HARTLEY R, SHEN C, et al. Extrinsic Methods for Coding and Dictionary Learning on Grassmann Manifolds[J]. International Journal of Computer Vision, 2015, 114(2):113-136. doi: 10.1007/s11263-015-0833-x
[15]	HE K, GIRSHICK R, DOLLÁR P. Rethinking ImageNet Pre-Training[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE, 2019:4918-4927.
[16]	KRIZHEVSKY A, SUTSKEVER I, HINTON G. ImageNet Classification with Deep Convolutional Neural Networks[C]// Proceedings of the Advances in Neural Information Processing Systems. Cambridge: MIT Press, 2012:1097-1105.
[17]	EVERINGHAM M, ESLAMI S M A, VAN GOOL L, et al. The Pascal Visual Object Classes Challenge:A Retrospective[J]. International Journal of Computer Vision, 2015, 111(1):98-136. doi: 10.1007/s11263-014-0733-5
[18]	CHERRY J M, ADLER C, BALL C A. SGD:Saccharomyces Genome Database[J]. Nucleic Acids Research, 1998, 26(1):73-79. doi: 10.1093/nar/26.1.73
[19]	FINN C, ABBEEL P, LEVINE S. Model-Agnostic Meta-learning for Fast Adaptation of Deep Networks[C]// Proceedings of the International Conference on Machine Learning. Sydney: ACM, 2017:1126-1135.
[20]	MUNKHDALAI T, YU H. Meta Networks[C]// Proceedings of the International Conference on Machine Learning. New York: ACM, 2017:2554-2563.
[21]	KARLINSKY L, SHTOK J, HARARY S, et al. RepMet:Representative-Based Metric Learning for Classification and Few-Shot Object Detection[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2019:5197-5206.
[22]	GIDARIS S, KOMODAKIS N. Dynamic Few-Shot Visual Learning Without Forgetting[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018:4367-4375.
[23]	LEE K, MAJI S, RAVICHANDRAN A, et al. Meta-Learning with Differentiable Convex Optimization[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2019:10657-10665.

数据集	ResNet-12	ResNet-18	ResNet-34
mini-ImageNet	0.831	0.839	0.783
CIFAR-100	0.854	0.866	0.735
Pascal VOC2007	0.863	0.881	0.822

算法名称	特征提取网络	5-way 1-shot	5-way 5-shot
MAML^[19]	Conv-4	0.487	0.631
Meta-Nets^[20]	ResNet-12	0.571	0.701
RepMet^[21]	Deformable-FPN	0.592	0.739
LwoF^[22]	WRN-28-10	0.601	0.764
CTM^[7]	ResNet-18	0.620	0.786
Meta-Opt^[23]	ResNet-12	0.641	0.800
DSN^[9]	ResNet-12	0.671	0.816
文中算法	ResNet-18	0.685	0.839