轴向注意力与尺度感知的小样本细粒度图像分类

doi:10.19678/j.issn.1000-3428.0252234

摘要/Abstract

摘要： 在细粒度图像分类(FGIC)任务中,充足的样本能够提供丰富的局部特征信息。然而,在小样本场景下,数据稀疏性导致模型难以充分捕捉具有判别性的局部信息。为解决该问题,提出一种融合轴向注意力与尺度感知机制的小样本学习(FSL)方法。首先,设计频率自适应特征选择(FAFS)模块,旨在减少背景噪声和非目标区域的干扰,突出判别性局部特征,从而扩大不同类别间的特征区分度;其次,构建轴向尺度联合增强模块,融合全局上下文信息,关注关键区域,并行处理不同感受野的特征,增强模型对不同尺度细节的表征能力;最后,采用双相似度度量模块,通过2种相似度度量方式指导学习,提升特征的泛化性,减少特定特征的偏向性。在公开数据集CUB和Stanford Dog上,相较基准方法,该方法在1-shot和5-shot场景下的分类准确率分别提升1.4、1.45百分点和1.86、3.49百分点。在Stanford Car数据集上,在1-shot场景下该方法达到最优性能,在5-shot场景下也取得了具有竞争力的结果。实验结果表明,所提方法有效提升了小样本细粒度图像分类(FSFGIC)的性能,能够更好地捕捉判别性特征信息。

关键词: 小样本学习, 细粒度图像分类, 轴向注意力, 尺度感知机制, 度量学习

Abstract: In Fine-Grained Image Classification (FGIC) tasks, a sufficient number of samples can provide rich local feature information. However, in few-shot scenarios, data sparsity makes it difficult for models to fully capture discriminative local information. To address this problem, a Few-Shot Learning (FSL) method integrating axial attention and scale-aware mechanisms is proposed. First, a Frequency-Adaptive Feature Selection (FAFS) module is designed to reduce background noise and interference from nontarget regions, highlight discriminative local features, and thereby increase feature separability among different categories. Second, an axial-scale joint enhancement module is constructed to fuse global contextual information, focus on key regions, process features with different receptive fields in parallel, and enhance the representation capability of the model for details at various scales. Finally, a dual-similarity metric module is adopted, which guides learning through two similarity metric methods to improve feature generalization and reduce bias toward specific features. On public datasets CUB and Stanford Dog, compared with the baseline method, the proposed method improves the classification accuracy by 1.4 and 1.45 percentage points under 1-shot and 5-shot settings, respectively, and by 1.86 and 3.49 percentage points on the other dataset. On the Stanford Car dataset, the proposed method achieves the best performance under the 1-shot setting and obtains competitive results under the 5-shot setting. Experimental results demonstrate that the proposed method effectively improves the performance of Few-Shot Fine-Grained Image Classification (FSFGIC) and is capable of better capturing discriminative feature information.

Key words: Few-Shot Learning(FSL), Fine-Grained Image Classification (FGIC), axial attention, scale-aware mechanism, measurement learning

中图分类号:

TP391

高雄, 苟光磊, 周琳杰, 贾朋昊. 轴向注意力与尺度感知的小样本细粒度图像分类[J]. 计算机工程, 2026, 52(10): 252-262.

GAO Xiong, GOU Guanglei, ZHOU Linjie, JIA Penghao. Axial Attention and Scale-Awareness for Few-Shot Fine-Grained Image Classification[J]. Computer Engineering, 2026, 52(10): 252-262.

https://www.ecice06.com/CN/Y2026/V52/I10/252

参考文献

[1] 孙红, 陈玉娟, 宋冬豪. 融合选择性稀疏采样的细粒度图像分类[J]. 小型微型计算机系统, 2024, 45(6): 1460-1465. SUN H, CHEN Y J, SONG D H. Fine-grained image classification with selective sparse sampling[J]. Journal of Chinese Computer Systems, 2024, 45(6): 1460-1465. (in Chinese)
[2] FU J L, ZHENG H L, MEI T. Look closer to see better: recurrent attention convolutional neural network for fine-grained image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C.,USA:IEEE Press,2017: 4476-4484.
[3] ZHANG N, DONAHUE J, GIRSHICK R, et al. Part-based R-CNNs for fine-grained category detection[EB/OL].[2025-01-05].https://www.researchgate.net/publication/319770362_Part-based_R-CNNs_for_Fine-grained_Category_Detection.
[4] WEI X S, WANG P, LIU L Q, et al. Piecewise classifier mappings: learning fine-grained learners for novel categories with few examples[J]. IEEE Transactions on Image Processing, 2019, 28(12): 6116-6125.
[5] TANG H, YUAN C C, LI Z C, et al. Learning attention-guided pyramidal features for few-shot fine-grained recognition[J]. Pattern Recognition, 2022, 130: 108792.
[6] WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[EB/OL].[2025-01-05].https://arxiv.org/abs/1807.06521.
[7] FENG H, WANG S S, GE S S. Fine-grained visual recognition with salient feature detection[EB/OL].[2025-01-05].https://arxiv.org/abs/1808.03935.
[8] SHIH K J, MALLYA A, SINGH S, et al. Part localization using multi-proposal consensus for fine-grained categorization[EB/OL].[2025-01-05].https://arxiv.org/abs/1507.06332.
[9] ZHU Y, LIU C, JIANG S. Multi-attention meta learning for few-shot fine-grained image recognition[EB/OL].[2025-01-05].https://www.ijcai.org/Proceedings/2020/0152.pdf.
[10] LIU C B, XIE H T, ZHA Z J, et al. Filtration and distillation: enhancing region attention for fine-grained visual categorization[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(7): 11555-11562.
[11] WANG C M, FU H Y, MA H D. Learning mutually exclusive part representations for fine-grained image classification[J]. IEEE Transactions on Multimedia, 2024, 26: 3113-3124.
[12] 白尚旺, 王梦瑶, 胡静, 等. 多区域注意力的细粒度图像分类网络[J]. 计算机工程, 2024, 50(1): 271-278. BAI S W, WANG M Y, HU J, et al. Multi-region attention network for fine-grained image classification[J]. Computer Engineering, 2024, 50(1): 271-278. (in Chinese)
[13] LIU H Q, PHILIP C C L, GONG X R, et al. Robust saliency-aware distillation for few-shot fine-grained visual recognition[J]. IEEE Transactions on Multimedia, 2024, 26: 7529-7542.
[14] ALFASSY A, KARLINSKY L, AIDES A, et al. LaSO: label-set operations networks for multi-label few-shot learning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C.,USA:IEEE Press,2020: 6541-6550.
[15] CHU W H, LI Y J, CHANG J C, et al. Spot and learn: a maximum-entropy patch sampler for few-shot image classification[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C.,USA:IEEE Press,2020: 6244-6253.
[16] SCHWARTZ E, KARLINSKY L, FERIS R, et al. Baby steps towards few-shot learning with multiple semantics[J]. Pattern Recognition Letters, 2022, 160: 142-147.
[17] SNELL J, SWERSKY K, ZEMEL R. Prototypical networks for few-shot learning[EB/OL].[2025-01-05]. https://arxiv.org/abs/1703.05175.
[18] VINYALS O, BLUNDELL C, LILLICRAP T P, et al. Matching networks for one shot learning[EB/OL].[2025-01-05].https://arxiv.org/abs/1606.04080.
[19] SUNG F, YANG Y X, ZHANG L, et al. Learning to compare: relation network for few-shot learning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington D.C.,USA:IEEE Press,2018: 1199-1208.
[20] LI W B, WANG L, XU J L, et al. Revisiting local descriptor based image-to-class measure for few-shot learning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C.,USA:IEEE Press,2020: 7253-7260.
[21] LIFCHITZ Y, AVRITHIS Y, PICARD S, et al. Dense classification and implanting for few-shot learning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C.,USA:IEEE Press,2020: 9250-9259.
[22] FINN C, ABBEEL P, LEVINE S. Model-agnostic meta-learning for fast adaptation of deep networks[EB/OL].[2025-01-05].https://arxiv.org/abs/1703.03400.
[23] ANTONIOU A, EDWARDS H, STORKEY A. How to train-your MAML[EB/OL].[2025-01-05].https://arxiv.org/abs/1810.09502.
[24] NGUYEN Q H, NGUYEN C Q, LE D D, et al. Enhancing few-shot image classification with cosine transformer[J]. IEEE Access, 2023, 11: 79659-79672.
[25] 许华杰, 梁书伟. 采用特征图增强原型的小样本图像分类方法[J]. 计算机科学与探索, 2024, 18(4): 990-1000. XU H J, LIANG S W. Few-shot image classification method with feature maps enhancement prototype[J]. Journal of Frontiers of Computer Science & Technology, 2024, 18(4): 990-1000. (in Chinese)
[26] LI X X, WU J J, SUN Z, et al. BSNet: bi-similarity network for few-shot fine-grained image classification[J]. IEEE Transactions on Image Processing, 2021, 30: 1318-1331.
[27] HUANG H X, ZHANG J J, ZHANG J, et al. Low-rank pairwise alignment bilinear network for few-shot fine-grained image classification[J]. IEEE Transactions on Multimedia, 2021, 23: 1666-1680.
[28] TIAN S, TANG H, DAI L Q. Coupled patch similarity network for one-shot fine-grained image recognition[C]//Proceedings of the IEEE International Conference on Image Processing (ICIP). Washington D.C.,USA:IEEE Press,2021: 2478-2482.
[29] ZHAO P, LI Y, TANG B W, et al. Feature relocation network for fine-grained image classification[J]. Neural Networks, 2023, 161: 306-317.
[30] MA Z X, CHEN Z D, ZHAO L J, et al. Cross-layer and cross-sample feature optimization network for few-shot fine-grained image classification[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2024, 38(5): 4136-4144.
[31] LI W B, XU J L, HUO J, et al. Distribution consistency based covariance metric networks for few-shot learning[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2019, 33(1): 8642-8649.
[32] WANG C F, SONG S J, YANG Q S, et al. Fine-grained few shot learning with foreground object transformation[J]. Neurocomputing, 2021, 466: 16-26.
[33] QI Y, SUN H, LIU N Z, et al. A task-aware dual similarity network for fine-grained few-shot learning[EB/OL].[2025-01-05].https://arxiv.org/abs/2210.12348.
[34] SONG Q J, ZHOU S Y, XU L W. Learning more discriminative local descriptors for few-shot learning[EB/OL].[2025-01-05].https://arxiv.org/abs/2305.08721.
[35] YANG Y J, FENG Y X, ZHU L, et al. Feature fusion network based on few-shot fine-grained classification[J]. Frontiers in Neurorobotics, 2023, 17: 1301192.
[36] SUN Z, ZHENG W, GUO P F. KLSANet: key local semantic alignment network for few-shot image classification[J]. Neural Networks, 2024, 178: 106456.
[37] WERTHEIMER D, TANG L M, HARIHARAN B. Few-shot classification with feature map reconstruction networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C.,USA:IEEE Press,2021: 8008-8017.
[38] LI X X, SONG Q, WU J J, et al. Locally-enriched cross-reconstruction for few-shot fine-grained image classification[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2023, 33(12): 7530-7540.
[39] LIU Y C, SHAO Z R, HOFFMANN N. Global attention mechanism: retain information to enhance channel-spatial interactions[EB/OL].[2025-01-05].https://arxiv.org/abs/2112.05561.

选择文件类型/文献管理软件名称

选择包含的内容