应用Sinkhorn距离和图正则约束的高效解混算法

杨露露; 李春芝; 陈晓华; 王丽

doi:10.11834/jrs.20221126

浏览量 : 0 下载量: 17 CSCD: 0

R-PDF
PDF
导出
分享
收藏
专辑

应用Sinkhorn距离和图正则约束的高效解混算法
Efficient unmixing algorithm using Sinkhorn distance and graph regularization constraints
2022年页码：1-16
DOI： 10.11834/jrs.20221126

扫描看全文

引用
杨露露，李春芝，陈晓华，王丽.XXXX.应用Sinkhorn距离和图正则约束的高效解混算法.遥感学报，XX（XX）： 1-16

YANG Lulu，LI Chunzhi，CHEN Xiaohua，WANG Li. XXXX. Efficient unmixing algorithm using Sinkhorn distance and graph regularization constraints. National Remote Sensing Bulletin， XX（XX）：1-16
杨露露，李春芝，陈晓华，王丽.XXXX.应用Sinkhorn距离和图正则约束的高效解混算法.遥感学报，XX（XX）： 1-16 DOI： 10.11834/jrs.20221126.

YANG Lulu，LI Chunzhi，CHEN Xiaohua，WANG Li. XXXX. Efficient unmixing algorithm using Sinkhorn distance and graph regularization constraints. National Remote Sensing Bulletin， XX（XX）：1-16 DOI： 10.11834/jrs.20221126.

摘要

高光谱盲解混是解决混合像元问题的重要技术。其中，非负矩阵分解（Nonnegative matrix factorization，NMF）凭借其明确的物理意义，为无监督线性光谱解混的发展奠定了基础。由于传统NMF采用欧氏距离度量原始矩阵与重构矩阵之间的误差，不能有效挖掘各维度特征间关系，影响解混精度。为充分利用高光谱图像中丰富的相关特征，本文在Earth Mover’s Distance（EMD）的基础上引入熵正则约束，由Sinkhorn距离代替欧氏距离，建模不同维度特征之间的关系。进一步地，为刻画数据的流形结构，将图正则项作为丰度的约束条件，提出了一种基于Sinkhorn距离和图正则约束的非负矩阵解混算法（SDGNMF）。论文采用乘性迭代规则对提出的解混模型进行求解，在模拟数据集、Urban数据集以及Jasper数据集上进行实验，实验结果验证了所提出算法的有效性。

Abstract

Objective Hyperspectral remote sensing technology, as a new type of earth observation technology, provides rich spectral information of features and is capable of identifying and finely classifying feature targets. Limited by the spatial resolution, a single pixel in hyperspectral images contains multiple features making the mixed pixels widespread, which becomes the main problem that the accuracy of pixel-level applications is difficult to improve. Nonnegative matrix factorization (NMF), with its clear physical meaning, lays the foundation for the development of unsupervised linear spectral unmixing. For all that, traditional NMF often uses Euclidean distance as a similarity measure method. On the one hand, hyperspectral data are manifold distribution, simple linear measurement between two points cannot accurately represent the distance between data, which makes the sample internal features weakly correlated, resulting in the NMF algorithm having an inaccurate prediction of the high-dimensional spatial inaccurate prediction of the translational noise in high-dimensional space. On the other hand, the objective function constructed based on this method ignores the correlation characteristics in the image space, which inhibits the performance of the algorithm performance.Method Considering the correlation between data manifolds and features, this paper proposes a non-negative matrix factorization unmixing algorithm based on Sinkhorn distance and graph regularization constraint (SDGNMF). On the basis of fully exploiting the advantages of EMD, the algorithm imposes entropy regularization constraint on EMD, improves EMD to sinkhorn distance, and takes it as the standard of measuring error, which effectively reduces the computational complexity. In addition, EMD with entropy regularization constraint, that is, the representation of the model by sinkhorn distance, can better model the relationship between different dimensional features, and make full use of the correlation of features. In particular, based on the sinkhorn distance, this paper introduces the graph regularity constraint to further characterize the manifold structure of data. Compared with the unmixing model constructed by Euclidean distance, SDGNMF is relatively insensitive to the noise in hyperspectral data and can better extract the internal structural information of the data, thus improving the unmixing accuracy.Result The experiment was carried out on simulated datasets and real datasets. Experiments prove that the algorithm proposed in this paper has achieved excellent subspace learning results and has good robustness. Compared with several other algorithms, SDGNMF has the advantage of being able to retain the similar structure after iteration, and the correlation between the endmember features is fully considered, thus the similar substances distributed in adjacent regions can be separated. Therefore, SDGNMF can better display the details of local abundance and obtain a more realistic and perfect abundance map.Conclusion In general, the unmixing model proposed in this paper can overcome noise, and consider the correlation of features and data manifold structure at the same time. Experiments show that the algorithm proposed in this paper can effectively improve the unmixing accuracy of most hyperspectral remote sensing data, especially those with high feature correlation. However, the algorithm proposed in this paper still faces the problem of high computational complexity. In addition, the algorithm only considers the prior knowledge of abundance. Therefore, future work will focus on solving these problems.

关键词

高光谱解混非负矩阵分解Sinkhorn距离熵正则图正则

Keywords

hyperspectral unmixingnonnegative matrix factorization(NMF)Sinkhorn distanceentropy regularizationgraph regularization

references

Akhtar N and Mian A. 2017. RCMF: Robust constrained matrix factorization for hyperspectral unmixing. IEEE Transactions onGeoscience and Remote Sensing, 55(6): 3354-3366. [DOI: 10.1109/TGRS.2017.2669991http://dx.doi.org/10.1109/TGRS.2017.2669991]

Bajorski P. 2010. Second moment linear dimensionality as an alternative to virtual dimensionality. IEEE transactions on geoscience and remote sensing, 49(2)：672-678. [DOI:10.1109/ TGRS.2010.2057434http://dx.doi.org/10.1109/TGRS.2010.2057434]

Bruckstein A M, Elad M and Zibulevsky M. 2008. On the uniqueness of nonnegative sparse solutions to underdetermined systems of equations. IEEE Transactions on Information Theory, 54(11): 4813-4820. [DOI: 10.1109/TIT.2008.929920http://dx.doi.org/10.1109/TIT.2008.929920]

Cuturi M. 2013. Sinkhorn distances: Lightspeed computation of optimal transport. Advances in neural information processing systems, 26: 2292-2300.

Cuturi M and Avis D. 2014. Ground metric learning. The Journal of Machine Learning Research, 15(1): 533-564.

Feng X R, Li H C, Li J, Du Q, Plaza A and Emery W J. 2018. Hyperspectral unmixing using sparsity constrained deep non-negative matrix factorization with total variation. IEEE Transactions on Geoscience and Remote Sensing, 56(10): 6245-6257. [DOI: 10.1109/TGRS.2018.2834567http://dx.doi.org/10.1109/TGRS.2018.2834567]

Févotte C and Dobigeon N. 2015. Nonlinear hyperspectral unmixi-ng with robust nonnegative matrix factorization. IEEE Transa-ctions on Image Processing, 24(12): 4810-4819. [DOI: 10.1109/TIP.2015.2468177http://dx.doi.org/10.1109/TIP.2015.2468177]

Frogner C, Zhang C, Mobahi H, Araya M and Poggio T A. 2015. Learning with a Wasserstein loss. Advances in neural information processing systems, 28: 2053-2061.

Gan Y Q, Liu W H, Feng X P, Hu B L,Wen D S. 2019. A Sparse Constrained Graph Regularized Nonnegative Matrix Factorization Algorithm for Hyperspectral unmixing method. Spectroscopy and Spectral Analysis, 39(4): 1118-1127.

甘玉泉, 刘伟华, 冯向朋, 于涛, 胡炳樑, 汶德胜. 2019. 一种稀疏约束的图正则化非负矩阵光谱解混方法.光谱学与光谱分析, 39(4): 1118-1127 [DOI:10.3964/j.issn.1000-0593(2019)04-1118-10http://dx.doi.org/10.3964/j.issn.1000-0593(2019)04-1118-10]

Genevay A, Chizat L, Bach F, Cuturi M and Peyré G. 2019. Sample complexity of sinkhorn divergences. Proceedings of the 22nd International Conference on Artificial Intelligence and Statistics. Cambridge MA: JMLR: 1574-1583.

Ghamisi P, Plaza J, Chen Y, Li J and Plaza A J. 2017. Advanced spectral classifiers for hyperspectral images: A review. IEEE Geoscience and Remote Sensing Magazine, 5(1): 8-32. [DOI: 10.1109/MGRS.2016.2616418http://dx.doi.org/10.1109/MGRS.2016.2616418]

Han H W, Wang G X, Wang M Z,Miao J Q,Guo S,Chen L,Zhang M Y and Guo K. 2020. Hyperspectral Unmixing Via Nonconvex Sparse and Low-Rank Constraint. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 13: 5704-5718. [DOI: 10.1109/JSTARS.2020.3021520http://dx.doi.org/10.1109/JSTARS.2020.3021520]

He W, Zhang H and Zhang L. 2016. Sparsity-regularized robust n-onnegative matrix factorization for hyperspectral unmixing. I-EEE Journal of Selected Topics in Applied Earth Observatio-ns and Remote Sensing, 9(9): 4267-4279. [DOI:10.1109/JSTARS.2016.2519498http://dx.doi.org/10.1109/JSTARS.2016.2519498]

Hendrix E M, Garcia I, Plaza J, Martin G and Plaza A. 2011. A new minimum volume enclosing algorithm for endmember identification and abundance estimation in hyperspectral data. IEEE Transactions on Geoscience and Remote Sensing, 50(7): 2744-2757. [DOI: 10.1109/TGRS.2011.2174443http://dx.doi.org/10.1109/TGRS.2011.2174443]

Heylen R, Parente M and Gader P. 2014. A review of nonlinear hyperspectral unmixing methods. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 7(6): 1844-1868.[DOI:10.1109/JSTARS.2014.2320576http://dx.doi.org/10.1109/JSTARS.2014.2320576]

Khan M J, Khan H S, Yousaf A, Khurshid K and Abbas A. 2018. Modern trends in hyperspectral image analysis: a review. IEEE Access, 6: 14118-14129.[DOI:10.1109/ACCESS.2018.2812999http://dx.doi.org/10.1109/ACCESS.2018.2812999]

Lan J H, Zou J L, Hao Y S, Zeng Y L, Zhang Y Z and Dong M W. 2018. Research progress on unmixing of hyperspectral remote sensing imagery. Journal of Remote Sensing, 22(1): 13-27.

蓝金辉, 邹金霖, 郝彦爽, 曾溢良, 张玉珍, 董铭巍. 2018. 高光谱遥感影像混合像元分解研究进展. 遥感学报, 22(1): 13-27 [DOI: 10. 11834/jrs.20186502http://dx.doi.org/10.11834/jrs.20186502]

Lee D D and Seung H S. 1999. Learning the parts of objects by non-negative matrix factorization. Nature, 401(6755): 788-791.

Lee D D and Seung H S. 2001. Algorithms for non-negative matrix factorization. Proceedings of the 13th International Conference on Neural Information Processing Systems, Cambridge: MIT Press: 556–562.

Li C Z and Chen X H. 2020. Error Approximation of Hyperspectral Unmixing via Correntropy-Induced Metric. IEEE Geoscien-ce and Remote Sensing Letters, 17(2): 282-286. [DOI: 10.1109/LGRS.2019.2920687http://dx.doi.org/10.1109/LGRS.2019.2920687]

Li C Z, Gu Y G, Chen X H, Zhang Y and Ruan L J. 2021. Hyperspectral Unmixing Via Latent Multi-Heterogeneous Subspace. IEEE Transactions on Geoscience and Remote Sensing, 59(1):563-577. [DOI: 10.1109/TGRS.2020.2996249http://dx.doi.org/10.1109/TGRS.2020.2996249]

Li J, Bioucas-Dias J M, Plaza A and Liu L. 2016. Robust collaborative nonnegative matrix factorization for hyperspectral unmixing. IEEE Transactions on Geoscience and Remote Sensing, 54(10): 6076-6090. [DOI: 10.1109/TGRS.2016.2580702http://dx.doi.org/10.1109/TGRS.2016.2580702]

Liu X L, Xia W, Wang B, and Zhang L M. 2010. An approach based on constrained nonnegative matrix factorization to unmix hyperspectral data. IEEE Transactions on Geoscience and Remote Sensing, 49(2): 757-772. [DOI: 10.1109/TGRS.2016.2580702http://dx.doi.org/10.1109/TGRS.2016.2580702]

Lu X Q, Wu H, Yuan Y, Yan P K and Li X L. 2013. Manifold regularized sparse NMF for hyperspectral unmixing. IEEE Transactions on Geoscience and Remote Sensing, 51(5): 2815-2826. [DOI: 10.1109/TGRS.2012.2213825http://dx.doi.org/10.1109/TGRS.2012.2213825]

Marrinan T and Gillis N. 2021. Hyperspectral Unmixing with Rar-e Endmembers via Minimax Nonnegative Matrix Factorization//Proceedings of the 28th European Signal Processing Confe-rence. Piscataway, NJ: IEEE :1015-1019. [DOI: 10.23919/Eu-sipco47968.2020.9287456http://dx.doi.org/10.23919/Eu-sipco47968.2020.9287456]

Mei S H, Bi Q Q, Ji J Y, Hou J H and Du Q. 2016. Spectral variation alleviation by low-rank matrix approximation for hyperspectral image analysis. IEEE Geoscience and Remote Sensing Letters, 13(6): 796-800. [DOI: 10.1109/LGRS.2016.2545341http://dx.doi.org/10.1109/LGRS.2016.2545341]

Miao L and Qi H. 2007. Endmember extraction from highly mixed data using minimum volume constrained nonnegative matrix factorization. IEEE Transactions on Geoscience and Remote Sensing, 45(3): 765-777. [DOI: 10.1109/TGRS.2006.888466http://dx.doi.org/10.1109/TGRS.2006.888466]

Nascimento J M P and Dias J M B. 2005. Does independent component analysis play a role in unmixing hyperspectral data? IEEE Transactions on Geoscience and Remote Sensing, 43(1): 175-187. [DOI: 10.1109/TGRS.2004.839806http://dx.doi.org/10.1109/TGRS.2004.839806]

Qian W, Hong B, Cai D, He X and Li X. 2016. Non-Negative Matrix Factorization with Sinkhorn Distance//Proceedings of the 25th International Joint Conferences on Artificial Intelligence. San Mateo, CA: Margan Kaufmann: 1960-1966.

Qian Y, Jia S, Zhou J and Robles-Kelly A. 2011. Hyperspectral unmixing via $L_{1 / 2}$ http://html.publish.founderss.cn/rc-pub/api/common/picture?pictureId=32295657&type=http://html.publish.founderss.cn/rc-pub/api/common/picture?pictureId=32295655&type=3.809999942.87866688sparsity-constrained nonnegative matrix factorization. IEEE Transactions on Geoscience and Remote Sensing, 49(11): 4282-4297. [DOI: 10.1109/TGRS.2011.2144605http://dx.doi.org/10.1109/TGRS.2011.2144605]

Qin J, Lee H, Chi J. T, Drumetz L, Chanussot J, Lou Y and Ber-tozzi A. L. 2020. Blind Hyperspectral Unmixing Based on Graph Total Variation Regularization. IEEE Transactions on Geoscience and Remote Sensing, 59(4): 3338-3351. [DOI: 10.1109/TGRS.2020.3020810http://dx.doi.org/10.1109/TGRS.2020.3020810]

Rathnayake B, Ekanayake E M M B, Weerakoon K, Godaliyadda G M R I, Ekanayake M P B and Herath H M V R. 2020. Graph-Based Blind Hyperspectral Unmixing via Nonnegative Matrix Factorization. IEEE Transactions on Geoscience and Remote Sensing, 58(9): 6391- 6409. [DOI: 10.1109/TGRS.2020.2976799http://dx.doi.org/10.1109/TGRS.2020.2976799]

Rubner Y, Tomasi C and Guibas L J. 2000. The earth mover's distance as a metric for image retrieval. International journal of computer vision, 40(2): 99-121.

Sandler R and Lindenbaum M. 2011. Nonnegative matrix factorization with earth mover's distance metric for image analysis. IEEE Transactions on Pattern Analysis and Machine Intelligence, 33(8): 1590-1602. [DOI: 10.1109/TPAMI.2011.18http://dx.doi.org/10.1109/TPAMI.2011.18]

Tong Q X, Zhang B, Zhang L F. 2016. Current progress of hyperspectral remote sensing in China. Journal of Remote Sensing, 20(05): 689-707.

童庆禧, 张兵, 张立福. 2016. 中国高光谱遥感的前沿进展. 遥感学报, 20(05): 689-707 [DOI:10.11834/jrs.20166264http://dx.doi.org/10.11834/jrs.20166264]

Tsinos C G, Rontogiannis A A and Berberidis K. 2017. Distribute-d blind hyperspectral unmixing via joint sparsity and low-ra-nk constrained non-negative matrix factorization. IEEE Transa-ctions on Computational Imaging, 3(2): 160-174. [DOI: 10.1109/TCI.2017.2693967http://dx.doi.org/10.1109/TCI.2017.2693967]

Wang J and Chang C I. 2006. Applications of independent compo-nent analysis in endmember extraction and abundance quanti-fication for hyperspectral imagery. IEEE Transactions on Ge-oscience and Remote Sensing, 44(9): 2601-2616. [DOI: 10.1109/TGRS.2006.874135http://dx.doi.org/10.1109/TGRS.2006.874135]

Wang N, Du B and Zhang L P. 2013. An endmember dissimilarity constrained non-negative matrix factorization method for hyperspectral unmixing. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 6(2): 554-569. [DOI: 10.1109/JSTARS.2013.2242255http://dx.doi.org/10.1109/JSTARS.2013.2242255]

Wang W H, Qian Y T and Tang Y Y. 2016. Hypergraph-regularized sparse NMF for hyperspectral unmixing. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 9(2): 681-694. [DOI: 10.1109/JSTARS.2015.2508448http://dx.doi.org/10.1109/JSTARS.2015.2508448]

Wang X Y, Zhong Y F, Zhang L P and Xu Y Y. 2017. Spatial g-roup sparsity regularized nonnegative matrix factorization for hyperspectral unmixing. IEEE Transactions on Geoscience a-nd Remote Sensing, 55(11): 6287-6304. [DOI: 10.1109/TGRS.2017.2724944http://dx.doi.org/10.1109/TGRS.2017.2724944]

Yao J, Meng D, Zhao Q, Cao W and Xu Z. 2019. Nonconvex-sparsity and nonlocal-smoothness-based blind hyperspectral unmixing. IEEE Transactions on Image Processing, 28(6): 2991-3006. [DOI: 10.1109/TIP.2019.2893068http://dx.doi.org/10.1109/TIP.2019.2893068]

Yu Y and Sun W. 2007. Minimum distance constrained non-negative matrix factorization for the endmember extraction of hyperspectral images//Proceedings of the 5th International Symposium on Multispectral Image Processing and Pattern Recognition. Bellingham: SPIE Press: 6790: 679015.

Yuan Y, Fu M and Lu X Q. 2015. Substance dependence constrained sparse NMF for hyperspectral unmixing. IEEE Transactions on Geoscience and Remote Sensing, 53(6): 2975-2986. [DOI: 10.1109/TGRS.2014.2365953http://dx.doi.org/10.1109/TGRS.2014.2365953]

Zhang B, Chen Z C, Peng D L, Benediktsson J A, Liu B, Zou L, Li J and Plaza A. 2019. Remotely sensed big data: evolution in model development for information extraction [point of view]. Proceedings of the IEEE, 107(12): 2294-2301. [DOI: 10.1109/JPROC.2019.2948454http://dx.doi.org/10.1109/JPROC.2019.2948454]

Zhang B. 2016. Advancement of hyperspectral image processing a-nd information extraction. Journal of Remote Sensing, 20(05):1062-1090.

张兵. 2016. 高光谱图像处理与信息提取前沿. 遥感学报, 20(05): 1062-1090 [DOI:10.11834/jrs.20166179http://dx.doi.org/10.11834/jrs.20166179]

Zhang X R, Zhang J Y, Li C, Cheng C, Jiao L C and Zhou H. 2018. Hybrid unmixing based on adaptive region segmentatio-n for hyperspectral imagery. IEEE Transactions on Geoscien-ce and Remote Sensing, 56(7): 3861-3875. [DOI: 10.1109/TGRS. 2018.2815044http://dx.doi.org/10.1109/TGRS.2018.2815044]

Zhi T, Yang B, Chen Z and Wang B. 2017. Nonnegative matrix f-actorization with constraints on endmember and abundance f-or hyperspectral unmixing//Proceedings of the 2017 IEEE Int-ernational Geoscience and Remote Sensing Symposium. Pisc-ataway, NJ: IEEE: 1149-1152. [DOI: 10.1109/IGARSS.2017.8127161http://dx.doi.org/10.1109/IGARSS.2017.8127161]

Zhang Y M, Shu Z Q, Zhang J, You C Z, Weng Z H, Fan H Hand Ye F Y. 2020. Dual Graph regularized NMF with Sinkhorn Distance// Proceedings of the 19th International Symposium on Distributed Computing and Applications for Business Engineering and Science. Piscataway, NJ: IEEE: 146-149.

文章被引用时，请邮件提醒。

提交

生成对抗网络的无监督高光谱解混

重加权稀疏和全变差约束下的深度非负矩阵分解高光谱解混

空间与谱间相关性分析的NMF高光谱解混

考虑光谱信息和超像素分割的高光谱解混网络