面向高光谱图像跨域小样本分类的解耦置信原型网络

王雪松; 江文超; 孔毅; 程玉虎

doi:10.11834/jrs.20253151

模型与方法 | 浏览量 : 0 下载量: 216 CSCD: 0

R-PDF
PDF
导出
分享
收藏
专辑

面向高光谱图像跨域小样本分类的解耦置信原型网络
Hyperspectral image cross-domain few-shot classification based on disentangled confidence prototype network
2026年30卷第1期页码：156-169
收稿：2024-05-27，

纸质出版：2026-01-07
DOI： 10.11834/jrs.20253151
稿件说明：

移动端阅览

王雪松，江文超，孔毅，程玉虎.2026.面向高光谱图像跨域小样本分类的解耦置信原型网络.遥感学报，30（1）： 156-169 DOI： 10.11834/jrs.20253151.

Wang X S，Jiang W C，Kong Y and Cheng Y H. 2026. Hyperspectral image cross-domain few-shot classification based on disentangled confidence prototype network. National Remote Sensing Bulletin， 30（1）：156-169 DOI： 10.11834/jrs.20253151.

摘要

针对标注的高光谱图像HSI（HyperSpectral Image）难以获取的问题，基于小样本学习的HSI分类方法备受关注。常用的小样本学习方法通常假设训练与测试样本分布一致，然而，由于受拍摄条件等因素的影响，不同HSI间往往存在分布差异，导致传统的小样本学习方法难以获取较高的分类性能。为此，本文提出了一种基于解耦置信原型网络的高光谱图像跨域小样本分类方法。首先，使用3D残差卷积网络提取样本的深度特征以充分挖掘HSI的空间—光谱信息；然后，借助解耦网络对深度特征进行功能分离，以实现对域不变与域特定特征进行更专注的表征；再次，通过置信原型网络筛选置信度高的查询集样本，并重新计算更可靠的类别原型；最后，通过综合利用高置信度类别原型与原始的类别原型，实现更准确的小样本分类。将本研究提出方法和其他已有研究方法在多个真实高光谱数据集进行实验验证对比，验证结果表明本文所提方法的有效性。

Abstract

Few-shot learning based methods have gained significant attention because of the difficulty in acquiring labeled hyperspectral images (HSIs). Common few-shot learning methods often assume that the distributions of training and testing sample are consistent. However

due to factors such as varied shooting conditions

distribution differences often exist between different HSIs

limiting the performance of conventional few-shot learning approaches in achieving high classification accuracy.

In this paper

we propose a novel cross-domain few-shot classification method for HSI that is based on a Disentangled Confidential Prototype Network (DCPN). Initially

a 3D residual convolutional network is used to extract deep embedded features from samples

thereby fully exploiting the spatial–spectral information of HSIs. Then

with the help of the disentangled network

these deep features undergo feature separation

enabling more focused representation of domain-invariant and domain-specific features. Additionally

a confidential prototype network is used to select high-confidence query set samples for recalculating more reliable class prototypes. More accurate few-shot classification is achieved by combining high-confidence class prototypes with original class prototypes. Experimental results on multiple real hyperspectral datasets validate the effectiveness of the proposed method．

Six real HSI datasets

namely

University of Pavia

Pavia Center

Salinas

Indian Pines

WHU Hi LongKou

and Chikusei

were selected for the experiment to validate the effectiveness of the method. Chikusei

which has the largest number of feature classes

was selected as the source domain

and the remaining five datasets were used as the target domains. DCPN achieves the best overall classification accuracy on all datasets and produces significantly fewer noise points and smoother classification maps than other methods do.

In this paper

we propose a cross-domain few-shot classification method for hyperspectral images based on the disentangled confidential prototype network. The method has the following advantages: (1) Domain-invariant features and domain-specific features are extracted by the DCPN

and only domain-invariant features are used for the few-shot classification task to reduce the negative impact from domain-specific feature information; (2) A confidential prototype network is proposed

which assigns corresponding weights to the unlabeled query set to recalculate the prototypes of each class and collaborates with the original support set class centers to jointly perform a high-quality few-shot classification task. Experimental results on six real HSI datasets show that the proposed method can achieve higher cross-domain classification accuracy and obtain smoother

more detailed classification results. Given that HSIs are often affected by atmospheric conditions

acquisition equipment

and other factors

how to obtain more robust domain-general and domain-specific features on the basis of our method remains a worthy topic for future work.

关键词

Keywords

references

Baur J , Dobler G , Bianco F , Sharma M and Karpf A . 2018 . Persistent hyperspectral observations of the urban lightscape // IEEE Global Conference on Signal and Information Processing . Anaheim : IEEE: 983 - 987 [ DOI: 10.1109/GlobalSIP.2018.8646419 http://dx.doi.org/10.1109/GlobalSIP.2018.8646419 ]

Cao J M , Katzir O , Jiang P , Lischinski D , Cohen-Or D , Tu C H and Li Y Y . 2021 . DiDA: iterative boosting of disentangled synthesis and domain adaptation // Proceedings of the 11th International Conference on Information Technology in Medicine and Education [ DOI: 10.1109/ITME53901.2021.00049 http://dx.doi.org/10.1109/ITME53901.2021.00049 ]

Chen H J , Lyu D N , Zhou X and Liu J . 2024 . Cross-domain transfer learning algorithm for few-shot ship recognition in remote-sensing images . National Remote Sensing Bulletin , 28 ( 3 ): 793 - 804

陈华杰 , 吕丹妮 , 周枭 , 刘俊 . 2024 . 遥感图像小样本舰船识别跨域迁移学习算法 . 遥感学报 , 28 ( 3 ): 793 - 804 [ DOI: 10.11834/jrs.20211368 http://dx.doi.org/10.11834/jrs.20211368 ]

Chen J H and Wang X L . 2022 . Multi-graph convolutional network for a remote sensing image few shot classification . National Remote Sensing Bulletin , 26 ( 10 ): 2029 - 2042

陈杰虎 , 汪西莉 . 2022 . 多图卷积网络的遥感图像小样本分类 . 遥感学报 , 26 ( 10 ): 2029 - 2042 [ DOI: 10.11834/jrs.20210522 http://dx.doi.org/10.11834/jrs.20210522 ]

Deng W Y , Lendasse A , Ong Y S , Tsang I W H , Chen L and Zheng Q H . 2019 . Domain adaption via feature selection on explicit feature map . IEEE Transactions on Neural Networks and Learning Systems , 30 ( 4 ): 1180 - 1190 [ DOI: 10.1109/TNNLS.2018.2863240 http://dx.doi.org/10.1109/TNNLS.2018.2863240 ]

Feng J , Chen J T , Liu L G , Cao X H , Zhang X R , Jiao L C and Yu T . 2019 . CNN-based multilayer spatial-spectral feature fusion and sample augmentation with local and nonlocal constraints for hyperspectral image classification . IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , 12 ( 4 ): 1299 - 1313 [ DOI: 10.1109/JSTARS.2019.2900705 http://dx.doi.org/10.1109/JSTARS.2019.2900705 ]

Fu Y Q , Fu Y W and Jiang Y G . 2021 . Meta-FDMixup: cross-domain few-shot learning guided by labeled target data //Proceedings of the 29th ACM International Conference on Multimedia. [s.l. ] : Association for Computing Machinery: 5326 - 5334 [ DOI: 10.1145/3474085.3475655 http://dx.doi.org/10.1145/3474085.3475655 ]

Ganin Y and Lempitsky V . 2014 . Unsupervised domain adaptation by backpropagation . // Proceedings of the 32nd International Conference on International Conference on Machine Learning . Lille : JMLR.org: 1180 - 1189

Guo Y H , Codella N , Karlinsky L , Codella J V , Smith J R , Saenko K , Rosing T and Feris R . 2020 . A broader study of cross-domain few-shot learning // 16th European Conference on Computer Vision . Glasgow : Springer: 124 - 141 [ DOI: 10.1007/978-3-030-58583-9_8 http://dx.doi.org/10.1007/978-3-030-58583-9_8 ]

Hang R L , Li Z , Liu Q S , Ghamisi P and Bhattacharyya S S . 2021 . Hyperspectral image classification with attention-aided CNNs . IEEE Transactions on Geoscience and Remote Sensing , 59 ( 3 ): 2281 - 2293 [ DOI: 10.1109/TGRS.2020.3007921 http://dx.doi.org/10.1109/TGRS.2020.3007921 ]

He K M , Zhang X Y , Ren S Q and Sun J . 2016 . Deep residual learning for image recognition // 2016 IEEE Conference on Computer Vision and Pattern Recognition . Las Vegas : IEEE: 770 - 778 [ DOI: 10.1109/CVPR.2016.90 http://dx.doi.org/10.1109/CVPR.2016.90 ]

Kong Y , Cheng Y H , Chen Y and Wang X S . 2024 . Joint classification of hyperspectral image and LiDAR data based on spectral prompt tuning . IEEE Transactions on Geoscience and Remote Sensing , 62 : 5521312 [ DOI: 10.1109/TGRS.2024.3417475 http://dx.doi.org/10.1109/TGRS.2024.3417475 ]

Li Z K , Liu M , Chen Y S , Xu Y M , Li W and Du Q . 2022 . Deep cross-domain few-shot learning for hyperspectral image classification . IEEE Transactions on Geoscience and Remote Sensing , 60 : 5501618 [ DOI: 10.1109/TGRS.2021.3057066 http://dx.doi.org/10.1109/TGRS.2021.3057066 ]

Liu B , Yu X C , Yu A Z , Zhang P Q , Wan G and Wang R R . 2019 . Deep few-shot learning for hyperspectral image classification . IEEE Transactions on Geoscience and Remote Sensing , 57 ( 4 ): 2290 - 2304 [ DOI: 10.1109/TGRS.2018.2872830 http://dx.doi.org/10.1109/TGRS.2018.2872830 ]

Liu Y C , Yeh Y Y , Fu T C , Wang S D , Chiu W C and Wang Y C F . 2018 . Detach and adapt: learning cross-domain disentangled deep representation // IEEE/CVF Conference on Computer Vision and Pattern Recognition . Salt Lake City : IEEE: 8867 - 8876 [ DOI: 10.1109/CVPR.2018.00924 http://dx.doi.org/10.1109/CVPR.2018.00924 ]

Long M S , Cao Z J , Wang J M and Jordan M I . 2018 . Conditional adversarial domain adaptation // Proceedings of the 32nd International Conference on Neural Information Processing Systems . Montréal : Curran Associates Inc.: 1647 - 1657

Melgani F and Bruzzone L . 2004 . Classification of hyperspectral remote sensing images with support vector machines . IEEE Transactions on Geoscience and Remote Sensing , 42 ( 8 ): 1778 - 1790 [ DOI: 10.1109/TGRS.2004.831865 http://dx.doi.org/10.1109/TGRS.2004.831865 ]

Siebels K , Goïta K and Germain M . 2020 . Estimation of mineral abundance from hyperspectral data using a new supervised neighbor-band ratio unmixing approach . IEEE Transactions on Geoscience and Remote Sensing , 58 ( 10 ): 6754 - 6766 [ DOI: 10.1109/TGRS.2020.2969577 http://dx.doi.org/10.1109/TGRS.2020.2969577 ]

Sun Q R , Liu Y Y , Chua T S and Schiele B . 2019 . Meta-transfer learning for few-shot learning // IEEE/CVF Conference on Computer Vision and Pattern Recognition . Long Beach : IEEE: 403 - 412 [ DOI: 10.1109/CVPR.2019.00049 http://dx.doi.org/10.1109/CVPR.2019.00049 ]

Wang T , Zhang H S and Lin H . 2016 . High-level impervious surfaces classification in urban environments from hyperspectral imagery // Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing . Los Angeles : IEEE: 1 - 5 [ DOI: 10.1109/WHISPERS.2016.8071729 http://dx.doi.org/10.1109/WHISPERS.2016.8071729 ]

Wang Y Q , Yao Q M , Kwok J T and Ni L M . 2021 . Generalizing from a few examples: a survey on few-shot learning . ACM Computing Surveys , 53 ( 3 ): 63 [ DOI: 10.1145/3386252 http://dx.doi.org/10.1145/3386252 ]

Yu C Y , Liu C Y , Yu H Y , Song M P and Chang C I . 2021 . Unsupervised domain adaptation with dense-based compaction for hyperspectral imagery . IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , 14 : 12287 - 12299 [ DOI: 10.1109/JSTARS.2021.3128932 http://dx.doi.org/10.1109/JSTARS.2021.3128932 ]

Zhang M M , Li W and Du Q . 2018 . Diverse region-based CNN for hyperspectral image classification . IEEE Transactions on Image Processing , 27 ( 6 ): 2623 - 2634 [ DOI: 10.1109/TIP.2018.2809606 http://dx.doi.org/10.1109/TIP.2018.2809606 ]

Zhao X D , Zhang M M , Tao R , Li W , Liao W Z and Philips W . 2022 . Cross-domain classification of multisource remote sensing data using fractional fusion and spatial-spectral domain adaptation . IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , 15 : 5721 - 5733 [ DOI: 10.1109/JSTARS.2022.3190316 http://dx.doi.org/10.1109/JSTARS.2022.3190316 ]

Zhong C X , Zhang J P , Wu S F and Zhang Y . 2020 . Cross-scene deep transfer learning with spectral feature adaptation for hyperspectral image classification . IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , 13 : 2861 - 2873 . [ DOI: 10.1109/JSTARS.2020.2999386 http://dx.doi.org/10.1109/JSTARS.2020.2999386 ]

Zhou L , He C , Wang D W and Guo Z Q . 2024 . Few-shot object detection in optical remote sensing images . National Remote Sensing Bulletin , 28 ( 7 ): 1693 - 1701

周莲 , 何楚 , 汪鼎文 , 郭子琪 . 2024 . 光学遥感图像的小样本目标检测 . 遥感学报 , 28 ( 7 ): 1693 - 1701 [ DOI: 10.11834/jrs.20243209 http://dx.doi.org/10.11834/jrs.20243209 ]

Zhu C H , Chen X H , Chen J , Yuan Y H and Tang K . 2023 . A siamese nested-UNet for change detection in posterior probability space (SNU-PS) . National Remote Sensing Bulletin , 27 ( 9 ): 2006 - 2023

朱传海 , 陈学泓 , 陈晋 , 袁宇恒 , 唐凯 . 2023 . 基于分类后验概率空间的孪生Nested-UNet (SNU-PS) 变化检测网络 . 遥感学报 , 27 ( 9 ): 2006 - 2023 [ DOI: 10.11834/jrs.20233070 http://dx.doi.org/10.11834/jrs.20233070 ]