Unsupervised hyperspectral unmixing based on robust nonnegative dictionary learning

doi:10.23919/JSEE.2022.000030

Journal of Systems Engineering and Electronics ›› 2022, Vol. 33 ›› Issue (2): 294-304.doi: 10.23919/JSEE.2022.000030

• ELECTRONICS TECHNOLOGY • Previous Articles Next Articles

Unsupervised hyperspectral unmixing based on robust nonnegative dictionary learning

Yang LI^1,²(), Bitao JIANG^1,^2,*(), Xiaobin LI²(), Jing TIAN²(), Xiaorui SONG²()

¹ Department of Space Information, Space Engineering University, Beijing 101400, China
² Beijing Institute of Remote Sensing Information, Beijing 100192, China

Received:2020-09-30 Accepted:2021-11-24 Online:2022-05-06 Published:2022-05-06
Contact: Bitao JIANG E-mail:yangli.cs@outlook.com;bitao_jiang@163.com;lixb14@tsinghua.org.cn;jingtian@nudt.edu.cn;songxr@buaa.edu.cn
About author:|LI Yang was born in 1993. She received her B.S. degree in computer science from Dalian University of Technology in 2015 and M.S. degree from National University of Defense Technology in 2017. She is currently a research assistant in Beijing Institute of Remote Sensing Information. Her research interests include high performance computing, image processing, and machine learning applied to remote sensing. E-mail: yangli.cs@outlook.com||JIANG Bitao was born in 1967. She received her Ph.D. degree in radar signal processing from Chinese Academy of Sciences in 2007. She is currently a researcher in Beijing Institute of Remote Sensing Information. Her research interests include radar image procesing, signal processing, and machine learning applied to remote sensing. E-mail: bitao_jiang@163.com||LI Xiaobin was born in 1978. He received his B.S. and M.S. degrees in 2002 and 2004, respectively, from National University of Defense Technology, and his Ph.D. degree in 2020 from Tsinghua University. He is currently an associate research fellow in Beijing Institute of Remote Sensing Information. His research interests include object detection, object recognition, and image processing applied to remote sensing. E-mail: lixb14@tsinghua.org.cn||TIAN Jing was born in 1980. She received her Ph.D. degree in Automatic Control Science from National University of Defense Technology in 2007. She is currently a researcher in Beijing Institute of Remote Sensing Information. Her research focuses on spaceborne image intelligent interpretation. E-mail: jingtian@nudt.edu.cn||SONG Xiaorui received her B.S. degree in instrumental science from Beihang University in 2013 and her Ph.D. degree from Space Engineering University in 2019. She is currently a research assistant in Beijing Institute of Remote Sensing Information. Her research interests include hyperspectural image processing and machine learning applied to remote sensing.E-mail: songxr@buaa.edu.cn
Supported by:
This work was supported by the National Natural Science Foundation of China (61801513).

Abstract

Abstract:

Considering the sparsity of hyperspectral images (HSIs), dictionary learning frameworks have been widely used in the field of unsupervised spectral unmixing. However, it is worth mentioning here that existing dictionary learning method-based unmixing methods are found to be short of robustness in noisy contexts. To improve the performance, this study specifically puts forward a new unsupervised spectral unmixing solution. For the reason that the solution only functions in a condition that both endmembers and the abundances meet non-negative constraints, a model is built to solve the unsupervised spectral unmixing problem on the account of the dictionary learning method. To raise the screening accuracy of final members, a new form of the target function is introduced into dictionary learning practice, which is conducive to the growing robustness of noisy HSI statistics. Then, by introducing the total variation (TV) terms into the proposed spectral unmixing based on robust nonnegative dictionary learning (RNDLSU), the context information under HSI space is to be cited as prior knowledge to compute the abundances when performing sparse unmixing operations. According to the final results of the experiment, this method makes favorable performance under varying noise conditions, which is especially true under low signal to noise conditions.

Key words: hyperspectral image (HSI), nonnegative dictionary learning, norm loss function, unsupervised unmixing

Yang LI, Bitao JIANG, Xiaobin LI, Jing TIAN, Xiaorui SONG. Unsupervised hyperspectral unmixing based on robust nonnegative dictionary learning[J]. Journal of Systems Engineering and Electronics, 2022, 33(2): 294-304.

Figures/Tables 11

Table 1

Table 2

Fig 1

Fig 2

Table 3

Fig 3

Table 4

Fig 4

Fig 5

Fig 6

Fig 7

References 30

1	XU X, SHI Z W, PAN B L₀-based sparse hyperspectral unmixing using spectral information and a multi-objectives formulation . ISPRS Journal of Photogrammetry and Remote Sensing, 2018, 141, 46- 58. doi: 10.1016/j.isprsjprs.2018.04.008
2	LIU Y, GUO Y, LI F, et al Sparse dictionary learning for blind hyperspectral unmixing. IEEE Geoscience and Remote Sensing Letters, 2019, 16 (4): 578- 582. doi: 10.1109/LGRS.2018.2878036
3	ZHONG Y F, WANG X Y, XU Y, et al Mini-UAV-borne hyperspectral remote sensing: from observation and processing to applications. IEEE Geoscience and Remote Sensing Magazine, 2018, 6 (4): 46- 62. doi: 10.1109/MGRS.2018.2867592
4	PAN B, SHI Z W, XU X, et al Coinnet: copy initialization network for multispectral imagery semantic segmentation. IEEE Geoscience and Remote Sensing Letters, 2018, 16 (5): 816- 820.
5	TU B, LI N Y, FANG L Y, et al Hyperspectral image classification with a class-dependent spatial-spectral mixed metric. Pattern Recognition Letters, 2019, 123, 16- 22. doi: 10.1016/j.patrec.2019.02.025
6	MURPHY J M, JAMES M, MAGGION I, et al Unsupervised clustering and active learning of hyperspectral images with nonlinear diffusion. IEEE Trans. on Geoscience & Remote Sensing, 2019, 57 (3): 1829- 1845.
7	LI A L, QIN A Y, SHANG Z W, et al Spectral-spatial sparse subspace clustering based on three-dimensional edge-preserving filtering for hyperspectral image. International Journal of Pattern Recognition and Artificial Intelligence, 2019, 33 (3): 1955003.
8	ZOU Z X, SHI Z W Hierarchical suppression method for hyperspectral target detection. IEEE Trans. on Geoscience & Remote Sensing, 2015, 54 (1): 330- 342.
9	ZHU D H, DU B, ZHANG L P Target dictionary construction-based sparse representation hyperspectral target detection methods. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2019, 12 (4): 1254- 1264. doi: 10.1109/JSTARS.2019.2902430
10	XIE W Y, SHI Y Z, LI Y S, et al High-quality spectral-spatial reconstruction using saliency detection and deep feature enhancement. Pattern Recognition, 2019, (88): 139- 152.
11	JIAO C Z, CHEN C, MCGARVEY R G, et al Multiple instance hybrid estimator for hyperspectral target characterization and sub-pixel target detection. ISPRS Journal of Photogrammetry and Remote Sensing, 2018, 146, 235- 250. doi: 10.1016/j.isprsjprs.2018.08.012
12	WU C, DU B, ZHANG L P Hyperspectral anomalous change detection based on joint sparse representation. ISPRS Journal of Photogrammetry and Remote Sensing, 2018, 146, 137- 150. doi: 10.1016/j.isprsjprs.2018.09.005
13	LIU S C, BRUZZONE L, BOVOLO F, et al Unsupervised multitemporal spectral unmixing for detecting multiple changes in hyperspectral images. IEEE Trans. on Geoscience & Remote Sensing, 2016, 54 (5): 2733- 48.
14	NASCIMENTO J M B, DIAS J M B Vertex component analysis: a fast algorithm to unmix hyperspectral data. IEEE Trans. on Geoscience & Remote Sensing, 2005, 43, 898- 910.
15	MIAO L D, QI H R Endmember extraction from highly mixed data using minimum volume constrained nonnegative matrix factorization. IEEE Trans. on Geoscience & Remote Sensing, 2007, 45 (3): 765- 77.
16	LI J, BIOUCAS-DIAS J M, PLAZA A, et al Robust collabo-rative nonnegative matrix factorization for hyperspectral unmixing (R-CoNMF). IEEE Trans. on Geoscience & Remote Sensing, 2016, 54 (10): 6076- 6090.
17	MIAO L D, QI H R, ZU H A maximum entropy approach to unsupervised mixed-pixel decomposition. IEEE Trans. on Image Processing, 2007, 16 (4): 1008- 1021. doi: 10.1109/TIP.2006.891350
18	QI L, LI J, GAO X B, et al A novel joint dictionary framework for sparse hyperspectral unmixing incorporating spectral library. Neurocomputing, 2019, 356, 97- 106. doi: 10.1016/j.neucom.2019.04.053
19	DRUMETZ L, MEYER T R, CHANUSSOT J, et al Hyperspectral image unmixing with endmember bundles and group sparsity inducing mixed norms. IEEE Trans. on Image Processing, 2019, 28 (7): 3435- 3450. doi: 10.1109/TIP.2019.2897254
20	YAO J, MENG D Y, ZHAO Q, et al Nonconvex-sparsity and nonlocal-smoothness based blind hyperspectral unmixing. IEEE Trans. on Image Processing, 2019, 28 (6): 2991- 3006. doi: 10.1109/TIP.2019.2893068
21	ZHANG X R, SUN Y J, ZHANG J Y, et al Hyperspectral unmixing via deep convolutional neural networks. IEEE Geoscience and Remote Sensing Letters, 2018, 15 (11): 1755- 1759. doi: 10.1109/LGRS.2018.2857804
22	SU Y C, LI J, PLAZA A, et al DAEN: deep autoencoder networks for hyperspectral unmixing. IEEE Trans. on Geoscience & Remote Sensing, 2019, 57 (7): 4309- 4321.
23	SONG X R, WU L D, HAO H X Blind hyperspectral sparse unmixing based on online dictionary learning. Proc. of the International Society for Optics and Photonics-Image and Signal Processing for Remote Sensing XXIV, 2018, 107890K.
24	SONG X R, WU L D. A novel hyperspectral endmember extraction algorithm based on online robust dictionary learning, Remote Sensing, 2019, 11: 1792.
25	QU Y, QI H R uDAS: an untied denoising autoencoder with sparsity for spectral unmixing. IEEE Trans. on Geoscience & Remote Sensing, 2019, 57, 1698- 1712.
26	PAN Q H, KONG D G, DING C, et al Robust non-negative dictionary learning. Proc. of the 28th AAAI Conference on Artificial Intelligence, 2014, 2027- 2032.
27	IORDACHE M D, BIOUCAS-DIAS J M, PLAZA A Total variation spatial regularization for sparse hyperspectral unmixing. IEEE Trans. on Geoscience & Remote Sensing, 2012, 50 (11): 4484- 4502.
28	BIOUCAS-DIAS J M, NASCIMENTO J M P Hyperspectral subspace identification. IEEE Trans. on Geoscience & Remote Sensing, 2008, 46 (8): 2435- 2445.
29	SU Y C, LI J, ANTONIO P, et al DAEN: deep autoencoder networks for hyperspectral unmixing. IEEE Trans. on Geoscience & Remote Sensing, 2019, (7): 1- 13.
30	WANG Y, PAN C H, XIANG S M, et al Robust hyperspectral unmixing with correntropy-based metric. IEEE Trans. on Image Processing, 2015, 24 (11): 4027- 4040. doi: 10.1109/TIP.2015.2456508

SNR/dB	Noise type	Index	RCoNMF	GDME	EEORDL	uDAS	RNDLSU
30	Correlated	SAD/(°)	2.281	0.626 6	0.439 6	0.605 9	0.362 8
		AAD	7.218	8.677	4.152	5.299	3.678
		SRE/dB	14.97	14.71	22.62	19.49	22.92
	White	SAD/(°)	2.317	0.668 0	0.6314	0.5913	0.418 7
		AAD	7.138	8.696	4.329	5.283	3.905
		SRE/dB	15.01	14.73	20.97	20.11	22.61
25	Correlated	SAD/(°)	2.916	0.794 6	0.611 3	1.252	0.467 4
		AAD	10.87	9.349	7.267	10.42	5.545
		SRE/dB	12.53	14.28	17.96	13.66	19.58
	White	SAD/(°)	2.606	0.784 4	0.943 9	1.130	0.652 1
		AAD	10.80	9.244	8.032	9.917	6.282
		SRE/dB	12.70	14.42	17.04	14.42	18.58
20	Correlated	SAD/(°)	2.977	1.026	0.6810	2.362	0.603 8
		AAD	17.60	10.85	11.31	16.66	7.891
		SRE/dB	9.499	13.39	14.10	9.972	16.53
	White	SAD/(°)	2.991	1.043	1.721	2.272	0.775 7
		AAD	16.92	10.76	11.03	16.45	7.987
		SRE/dB	9.912	13.49	12.60	10.11	13.98

SNR/dB	Noise type	Index	RCoNMF	GDME	EEORDL	uDAS	RNDLSU
30	Correlated	SAD/(°)	5.198	3.255	2.035	2.309	1.887
		AAD	9.283	9.614	7.890	8.138	7.138
		SRE/dB	7.192	3.994	15.35	15.42	18.99
	White	SAD/(°)	5.331	3.015	2.235	2.490	2.176
		AAD	9.617	9.927	8.361	7.111	7.980
		SRE/dB	7.163	5.423	15.12	15.28	16.81
25	Correlated	SAD/(°)	6.023 2	2.982	3.035	3.119	1.898
		AAD	12.28	10.27	9.724	10.45	8.984
		SRE/dB	11.36	5.388	13.73	12.76	14.99
	White	SAD/(°)	6.083	3.007	2.578	3.132	1.956
		AAD	12.36	10.17	9.623	10.59	8.778
		SRE/dB	11.29	5.388	13.20	12.83	15.79
20	Correlated	SAD/(°)	7.284	3.620	3.098	8.881	2.315
		AAD	18.03	12.57	11.23	18.73	10.82
		SRE/dB	9.199	4.128	12.41	9.353	13.22
	White	SAD/(°)	7.027	2.938	2.908	5.336	2.332
		AAD	17.58	12.59	11.78	16.52	10.28
		SRE/dB	9.445	6.826	11.89	9.680	13.25

Performance	Value of $\lambda $
Performance	1	1×10^?1	1×10^?2	1×10^?3	1×10^?4	1×10^?5	1×10^?6	1×10^?7
SAD/(°)	0.6752	0.6377	0.6254	0.6119	0.6102	0.6079	0.6038	0.6038

Endmember	RCoNMF	GDME	EEORDL	uDAS	RNDLSU
Asphalt	0.245 0	0.123 6	0.105 5	0.219 2	0.080 4
Grass	0.620 6	0.561 0	0.125 8	0.336 4	0.045 2
Tree	0.358 7	0.160 8	0.054 1	0.172 0	0.052 4
Roof	0.164 2	0.042 2	0.107 5	0.282 5	0.049 1
Mean	0.347 1	0.221 9	0.098 2	0.252 5	0.056 8

Unsupervised hyperspectral unmixing based on robust nonnegative dictionary learning

RichHTML

PDF (PC)

Knowledge

Abstract

Cite this article

Share this article

Figures/Tables 11

References 30

Related Articles 0

Recommended Articles

Metrics

Comments