[1]吕伏,张旭,张紫扬.多尺度特征细化的小样本图像分类[J].智能系统学报,2026,21(3):675-687.[doi:10.11992/tis.202505026]
LYU Fu,ZHANG Xu,ZHANG Ziyang.Multi-scale feature refinement for few-shot image classification[J].CAAI Transactions on Intelligent Systems,2026,21(3):675-687.[doi:10.11992/tis.202505026]
点击复制
LYU Fu,ZHANG Xu,ZHANG Ziyang.Multi-scale feature refinement for few-shot image classification[J].CAAI Transactions on Intelligent Systems,2026,21(3):675-687.[doi:10.11992/tis.202505026]
多尺度特征细化的小样本图像分类
《智能系统学报》[ISSN 1673-4785/CN 23-1538/TP] 卷:
21
期数:
2026年第3期
页码:
675-687
栏目:
学术论文—机器学习
出版日期:
2026-05-05
- Title:
- Multi-scale feature refinement for few-shot image classification
- Keywords:
- few-shot learning; metric learning; feature fusion; feature refinement; image classification; prototypical networks; multi-scale feature; deep learning
- 分类号:
- TP391.4
- 文献标志码:
- 2026-3-10
- 摘要:
- 在小样本图像分类任务中，特征融合虽有助于提升有限样本的利用效率，但易引入冗余信息，影响模型判别能力。为此，提出一种多尺度特征融合细化网络(multi-scale feature fusion refinement network, MFRNet)，通过系统性的冗余抑制机制优化特征表达。构建双流注意力融合模块，结合跨层级特征交互与自适应全局注意力，实现空间与通道维度的协同融合，有效抑制初始冗余。引入可分离通道注意力机制，利用深度可分离卷积解耦空间滤波与通道交互，进一步压缩冗余信息，提升特征判别性。设计自适应特征细化网络，通过双级压缩-释放机制与异构化特征分解，实现冗余的层级化过滤与判别性特征的保留。在5个不同粒度数据集上的实验表明，MFRNet在两种实验设置下均优于现有主流方法，展现出显著性能优势。
- Abstract:
- In few-shot image classification tasks, feature fusion improves the utilization of limited samples but can readily introduce redundant information and degrade the model’s discriminative ability. To address this, this paper proposes a multi-scale feature fusion refinement network (MFRNet) that optimizes feature representation through a systematic redundancy suppression mechanism. First, a dual-stream attention fusion module is constructed, combining cross-level feature interaction with adaptive global attention to achieve collaborative fusion across spatial and channel dimensions and effectively suppress initial redundancy. Second, a separable channel attention mechanism is introduced, using deep separable convolutions to decouple spatial filtering from channel interaction, further compressing redundant information and improving feature discriminability. Finally, an adaptive feature refinement network is designed to achieve hierarchical filtering of redundancy and retention of discriminative features through a two-stage compression–release mechanism and heterogeneous feature decomposition. Experiments on five datasets of varying granularity show that MFRNet outperforms existing mainstream methods in both experimental settings, demonstrating significant performance advantages.
- 参考文献/References:
-
[1] LU Jiang, GONG Pinghua, Ye Jieping, et al. A survey on machine learning from few samples[J]. Pattern recognition, 2023, 139: 109480
[2] SONG Yisheng, WANG Ting, CAI Puyu, et al. A comprehensive survey of few-shot learning: evolution, applications, challenges, and opportunities[J]. ACM computing surveys, 2023, 55(13): 1-40
[3] WANG Y, YAO Q, KWOK J T, et al. Generalizing from a few examples: a survey on few-shot learning[J]. ACM computing surveys (CSUR), 2020, 53(3): 1-34
[4] 周伯俊, 陈峙宇. 基于深度元学习的小样本图像分类研究综述[J]. 计算机工程与应用, 2024, 60(8): 1-15 ZHOU Bojun, CHEN Zhiyu. Survey of few-shot image classification based on deep meta-learning[J]. Computer engineering and applications, 2024, 60(8): 1-15
[5] RAVI S, LAROCHELLE H. Optimization as a model for few-shot learning[C]//International Conference on Learning Representations. Toulon: ICLR, 2017.
[6] MISHRA N, ROHANINEJAD M, CHEN X, et al. A simple neural attentive meta-learner[EB/OL]. (2017-07-01)[2025-01-01]. https://arxiv.org/abs/1707.03141.
[7] FINN C, ABBEEL P, LEVINE S. Model-agnostic meta-learning for fast adaptation of deep networks[C]//International Conference on Machine Learning. Sydney: PMLR, 2017: 1126-1135.
[8] NICHOL A, ACHIAM J, SCHULMAN J. On first-order meta-learning algorithms[EB/OL]. (2018-03-08)[2025-01-01]. https://arxiv.org/abs/1803.02999.
[9] VINYALS O, BLUNDELL C, LILLICRAP T, et al. Matching networks for one shot learning[C]//Advances in Neural Information Processing Systems. Barcelona: NIPS, 2016, 29.
[10] SNELL J, SWERSKY K, ZEMEL R. Prototypical networks for few-shot learning[C]//Advances in Neural Information Processing Systems. Long Beach: NIPS, 2017, 30.
[11] SUNG F, YANG Y, ZHANG L, et al. Learning to compare: Relation network for few-shot learning[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City: CVPR, 2018: 1199-1208.
[12] SATORRAS V G, ESTRACH J B. Few-shot learning with graph neural networks[C]//International Conference on Learning Representations. Vancouver: ICLR, 2018.
[13] BOIMAN O, SHECHTMAN E, IRANI M. In defense of nearest-neighbor based image classification[C]//2008 IEEE Conference on Computer Vision and Pattern Recognition. Anchorage: IEEE, 2008: 1-8.
[14] LI Wenbin, WANG Lei, XU Jinglin, 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. Long Beach: CVPR, 2019: 7260-7268.
[15] LIU Yang, ZHENG Tu, SONG Jie, et al. Dmn4: Few-shot learning via discriminative mutual nearest neighbor neural network[C]//Proceedings of the AAAI Conference on Artificial Intelligence. online: AAAI, 2022: 1828-1836.
[16] 文浪, 苟光磊, 白瑞峰, 等. 基于邻域融合和特征增强的小样本细粒度图像分类[J]. 计算机工程, 2026, 52(2): 158-166 WEN Lang, GOU Guanglei, BAI Ruifeng, et al. Few-shot fine-grained image classification based on neighborhood fusion and feature enhancement[J]. Computer engineering, 2026, 52(2): 158-166
[17] 李小菲, 苟光磊, 韩岩奇, 等. 多尺度知识引导局部增强的小样本细粒度图像分类方法[J]. 智能系统学报, 2024, 19(5): 1157-1167 LI Xiaofei, GOU Guanglei, HAN Yanqi, et al. Multiscale knowledge-guided local enhancement method for few-shot fine-grained image classification[J]. CAAI transactions on intelligent systems, 2024, 19(5): 1157-1167
[18] ZHANG Xuchao, GAO Yifeng, LIN J, et al. Tapnet: multivariate time series classification with attentional prototypical network[C]//Proceedings of the AAAI Conference on Artificial Intelligence. New York: AAAI, 2020: 6845-6852.
[19] QIAO Qian, XIE Yu, ZENG Ziyin, et al. Talds-net: task-aware adaptive local descriptors selection for few-shot image classification[C]//IEEE International Conference on Acoustics, Speech and Signal Processing. Seoul: IEEE, 2024: 3750-3754.
[20] YANG Yajie, FENG Yuxuan, ZHU Li, et al. Feature fusion network based on few-shot fine-grained classification[J]. Frontiers in neurorobotics, 2023, 17: 1301192
[21] KIM J, CHI M. SAFFNet: self-attention-based feature fusion network for remote sensing few-shot scene classification[J]. Remote sensing, 2021, 13(13): 2532
[22] HOU Ruibing, CHANG Hong, MA Bingpeng, et al. Cross attention network for few-shot classification[EB/OL]. (2019-10-17)[2025-01-01]. https://arxiv.org/abs/1910.07677.
[23] 王雪松, 吕理想, 程玉虎, 等. 注意力集合表示的多尺度度量小样本图像分类[J]. 中国图象图形学报, 2024, 29(11): 3371-3382 WANG Xuesong, LYU Llixiang, CHENG Yuhu, et al. Attention set representation for multiscale measurement offew-shot image classification[J]. Journal of image and graphics, 2024, 29(11): 3371-3382
[24] 周彬, 鲜浩, 秦艺嘉, 等. 融合混合注意力机制的局部原型小样本分类模型[J]. 计算机科学与探索, 2026, 20(2): 584-595 ZHOU Bin, XIAN Hao, QIN Jiayi, et al. A local prototypes-based few-shot classification model embedded hybrid attention[J]. Journal of frontiers of computer science and technology, 2026, 20(2): 584-595
[25] 曹增辉, 陈浩, 曹雅慧. 基于类注意力的原型网络改进方法[J]. 自动化与信息工程, 2025, 46(1): 59-65 CAO Zenghui1, CHEN Hao1, CAO Yahui. Improvement method of prototype network based on class attention[J]. Automation & Information Engineering, 2025, 46(1): 59-65
[26] CHEN Zixuan, HE Zewei, LU Zheming. DEA-net: single image dehazing based on detail-enhanced convolution and content-guided attention[J]. IEEE transactions on image processing, 2024, 33: 1002-1015
[27] HU Jie, SHEN Li, SUN Gang. Squeeze-and-excitation networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City: CVPR, 2018: 7132-7141.
[28] LIU Guilin, REDA F A, SHIH K J, et al. Image inpainting for irregular holes using partial convolutions[C]//Proceedings of the European Conference on Computer Vision. Munich: ECCV, 2018: 85-100.
[29] WAH C, BRANSON S, WELINDER P, et al. The caltech-ucsd birds-200-2011 dataset: CNS-TR-2011-001[R]. [S. l. : s. n. ], 2011.
[30] DATASET E. Novel datasets for fine-grained image categorization[C]//First Workshop on Fine Grained Visual Categorization. Colorado Spring: CVPR, 2011: 2.
[31] KRAUSE J, STARK M, DENG J, et al. 3D object representations for fine-grained categorization[C]//Proceedings of the IEEE International Conference on Computer Vision Workshops. Sydney: ICCV, 2013: 554-561.
[32] ZHANG Chi, CAI Yujun, LIN Guosheng, et al. DeepEMD: few-shot image classification with differentiable earth mover’s distance and structured classifiers[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. online: CVPR, 2020: 12203-12213.
[33] WANg Xixi, Wang Xiao, JIANG Bo, et al. Few-shot learning meets transformer: Unified query-support transformers for few-shot classification[J]. IEEE transactions on circuits and systems for video technology, 2023, 33(12): 7789-7802
[34] LAI Jinxiang, YANG Siqian, LIU Wenlong, et al. tSF: transformer-based semantic filter for few-shot learning[C]//European Conference on Computer Vision. Cham: Springer Nature Switzerland, 2022: 1-19.
[35] HU Ling, WU Wei. Enhancing few-shot image classification with a multi-faceted self-supervised and contrastive learning approach[J]. IEEE access, 2024, 12: 164844-164861
[36] WU Jijie, CHANG Dongliang, SAIN A, et al. Bi-directional feature reconstruction network for fine-grained few-shot image classification[C]//Proceedings of the AAAI Conference on Artificial Intelligence. Washington, DC: AAAI, 2023: 2821-2829.
[37] MA Zhenxiang, CHEN Zhenduo, ZHAO Lijun, et al. Cross-layer and cross-sample feature optimization network for few-shot fine-grained image classification[C]//Proceedings of the AAAI Conference on Artificial Intelligence. Vancouver: AAAI, 2024: 4136-4144.
[38] WANG Maofa, YAN Bingchen. LDCA: local descriptors with contextual augmentation for few-shot learning[EB/OL]. (2024-01-24)[2025-01-01]. https://arxiv.org/abs/2401.13499.
[39] SONG Qijun, ZHOU Siyun, XU Liwei. Learning more discriminative local descriptors for few-shot lLearning[EB/OL]. (2023-05-15)[2025-01-01]. https://arxiv.org/abs/2305.08721.
[40] ZHA Zican, TANG Hao, SUN Yunlian, et al. Boosting few-shot fine-grained recognition with background suppression and foreground alignment[J]. IEEE transactions on circuits and systems for video technology, 2023, 33(8): 3947-3961
[41] WANG Guangpeng, WANG Yongxiong, PAN Zhiqun, et al. Vitfsl-baseline: a simple baseline of vision transformer network for few-shot image classification[J]. IEEE access, 2024, 12: 11836-11849
[42] ZHANG Bo, YUAN Jiakang, LI Baopu, et al. Learning cross-image object semantic relation in transformer for few-shot fine-grained image classification[C]//Proceedings of the 30th ACM International Conference on Multimedia. Lisboa: ACM, 2022: 2135-2144.
- 相似文献/References:
-
[1]宋婉茹,赵晴晴,陈昌红,等.行人重识别研究综述[J].智能系统学报,2017,12(6):770.[doi:10.11992/tis.201706084]
SONG Wanru,ZHAO Qingqing,CHEN Changhong,et al.Survey on pedestrian re-identification research[J].CAAI Transactions on Intelligent Systems,2017,12():770.[doi:10.11992/tis.201706084]
[2]马岭,鲁越,蒋慧琴,等.基于小样本学习的LCD产品缺陷自动检测方法[J].智能系统学报,2020,15(3):560.[doi:10.11992/tis.201904020]
MA Ling,LU Yue,JIANG Huiqin,et al.An automatic small sample learning-based detection method for LCD product defects[J].CAAI Transactions on Intelligent Systems,2020,15():560.[doi:10.11992/tis.201904020]
[3]王鑫,郭鑫垚,魏巍,等.对抗样本三元组约束的度量学习算法[J].智能系统学报,2021,16(1):30.[doi:10.11992/tis.202009050]
WANG Xin,GUO Xinyao,WEI Wei,et al.Metric learning algorithm with adversarial sample triples constraints[J].CAAI Transactions on Intelligent Systems,2021,16():30.[doi:10.11992/tis.202009050]
[4]郭茂祖,王偲佳,王鹏跃,等.基于卫星图的小样本街区品质评估[J].智能系统学报,2022,17(6):1254.[doi:10.11992/tis.202111049]
GUO Maozu,WANG Sijia,WANG Pengyue,et al.Small sample block quality evaluation based on satellite images[J].CAAI Transactions on Intelligent Systems,2022,17():1254.[doi:10.11992/tis.202111049]
[5]季友昌,袁伟伟,毛善斌,等.不平衡小样本基于局部域对抗适应网络的发动机振动预测模型[J].智能系统学报,2023,18(5):1005.[doi:10.11992/tis.202210030]
JI Youchang,YUAN Weiwei,MAO Shanbin,et al.Partial domain adversarial adaptation networks for imbalanced small samples in aeroengine vibration prediction[J].CAAI Transactions on Intelligent Systems,2023,18():1005.[doi:10.11992/tis.202210030]
[6]李小菲,苟光磊,韩岩奇,等.多尺度知识引导局部增强的小样本细粒度图像分类方法[J].智能系统学报,2024,19(5):1157.[doi:10.11992/tis.202309003]
LI Xiaofei,GOU Guanglei,HAN Yanqi,et al.Multiscale knowledge-guided local enhancement method for few-shot fine-grained image classification[J].CAAI Transactions on Intelligent Systems,2024,19():1157.[doi:10.11992/tis.202309003]
[7]缪宛谕,苟光磊,钟声,等.多级决策优化关系网络的小样本学习方法[J].智能系统学报,2025,20(4):882.[doi:10.11992/tis.202406016]
MIAO Wanyu,GOU Guanglei,ZHONG Sheng,et al.Multi-level decision optimization in relational networks for few-shot learning method[J].CAAI Transactions on Intelligent Systems,2025,20():882.[doi:10.11992/tis.202406016]
[8]周凯锐,刘鑫,景丽萍,等.概念驱动的小样本判别特征学习方法[J].智能系统学报,2023,18(1):162.[doi:10.11992/tis.202203061]
ZHOU Kairui,LIU Xin,JING Liping,et al.Concept-driven discriminative feature learning for few-shot learning[J].CAAI Transactions on Intelligent Systems,2023,18():162.[doi:10.11992/tis.202203061]
备注/Memo
收稿日期:2025-5-26。
基金项目:国家自然科学基金面上项目(51874166, 52274206)；国家自然科学基金青年基金项目(51904144).
作者简介:吕伏，副教授，博士，主要研究方向为智能数据处理、图像与视觉信息计算、大数据与云计算。发表学术论文33篇。E-mail：38458786@qq.com。;张旭，硕士研究生，主要研究方向为小样本学习、图像分类。E-mail：2696357169@qq.com。;张紫扬，硕士研究生，主要研究方向为目标检测。E-mail：734759802@qq.com。
通讯作者:吕伏. E-mail：38458786@qq.com
更新日期/Last Update:
1900-01-01