You are on page 1of 12

Scientific Journal of Earth Science

December 2013, Volume 3, Issue 4, PP.107-118

An Improved Non-negative Matrix Factorization


Method of Blind Unmixing for Hyperspectral
Imagery
Jingjing Cao, Li Zhuo
Center of Integrated Geographic Information Analysis, Guangdong Key Laboratory for Urbanization and Geo-simulation, School
of Geography and Planning, Sun Yat-sen University, Guangzhou 510275, PR China
Email:

zhuoli@mail.sysu.edu.cn

Abstract
An improved non-negative matrix factorization method of blind unmixing for hyperspectral imagery (ATGP-NMF) was proposed
in this paper, concerning the fact that the blind unmixing of Non-negative Matrix Factorization(NMF) is easily reduced to the local
minimum, by which the spectra and abundance of the target endmember that were obtained by using Automatic Target Generation
Process (ATGP) algorithm based on unsupervised orthogonal subspace projection and Non-negative Least Squares (NNLS) were
then regarded as initial values by NMF to get the corresponding endmember. The validity and feasibility of the proposed method
were analyzed based on the data of both simulation and remote sensing imagery; and then the result was compared with that from
VCA-FCLS algorithm which extracted the endmember matrix by using the Vertex Component Analysis (VCA) algorithm and the
abundance matrix by using the Fully Constrained Least Squares (FCLS) algorithm. It was indicated that the optimization of the
target endmember initial value not only promotes the algorithm accuracy, but also strengthens its feasibility in the ATGP-NMF
algorithm.
Keywords: Hyperspectral Remote Sensing; Mixed Pixel; Target Endmember; Non-negative Matrix Factorization; Blind Unmixing

510275

(NMF)

(ATGP-NMF)(ATGP)(NNLS)
NMF
VCA-FCLS ATGP-NMF

S2012010010517
2011-2014-
B08008
- 107 http://www.j-es.org/

[1]

[2]Blind Source Separation, BSS


[3, 4]BSS

Independent Component Analysis, IC [5-10]


Non-negative matrix factorization, NMF[11-13]Complexity Analysis, CA[14]
Sparse Component Analysis, SCA[15, 16]NMF
NMF

NMF [17] NMF [11] NMF [18]


[19][13] NMF
NMF
NMF

Automatic Target Generation


Process-Non-negative matrix factorization, ATGP-NMF NMF

1
NMF 20 Lee Seung 1999 Nature
[20]
[21]
X nm r
Anr Sr m

X nm Anr Srm Enm


(1)
n m r (r min(n, m)) Anr
r X nm Enm
NMF

X nm Anr Srm

(2)

NMF NMF

(4) NMF
A S

Euc( X , AS )

1
X AS
2

1
2
( X ij ( AS )ij )

2 ij

min f ( A, S ) X AS
2

aij 0, sij 0

- 108 http://www.j-es.org/

(3)

2
F

(4)

2
F

Frobenius

S pb S pb

Alp Alp

( AT X ) pb
( AT AS ) pb

( XS T )lp
( ASS T )lp

(5)

(6)

NMF

2 ATGP-NMF

NMF NMF

ATGP [22] NMF NMF

2.1
- 109 http://www.j-es.org/


[23]

Harsanyi, Farrand Chang

-Neyman-Pearson

Hsrsanyi-Farrand-Chang, HFCHFC

[24]
PCA, ICA, Akaike Akaike Information Criterion, AIC
[25]

Minimum Description Length, MDL

HFC

2.2
ATGP[22,

26]

ATGP

(1) x HFC q
(2)

T
t0 t0 arg max[ x x]
x

(3) t0 U 0 t0 PU0 =Pt0

t1 arg max[( PU0 x) ( PU0 x)]

(4) i i 1 ( i 1,2,

, q 1 ) i ti ti arg max[( PUi1 x)T ( PUi1 x)]

(5) ti Ui [t0 , t1 ,

Ui

, ti ] P I U iU

#
i

(6) q
[t0 , t1 ,

, tq 1 ]

2.3 NMF
NMF

NMF
(1) A S ATGP NMF A NNLS
S
(2)

0 A S

S pb S pb

Alp Alp

( AT X ) pb
( AT AS ) pb

( XS T )lp
( ASS T )lp

(7)

(8)

(3) A S(7) S (8) A

- 110 http://www.j-es.org/

(4) 1
(9) S S 1

S pb

S pb
P

(9)

pb

(5) A S 0

Euc( k ) ( X , X ( k ) )

1
X X ( k )
2

(10)

X ( k ) k X

2.4

[27]

3
ATGP-NMF 2
2
VCA-FCLS NMF
VCA [28,

29]

FCLS [30]

VCA-FCLS
Nascimento Dias[31]Spectral Angle Distance, SAD
Spectral Information Divergence, SID
Root Mean Square Error, RMSE[27] SE[32] d[33]
(1) SAD
N

Ai Bi

AB
1
1
i 1

SAD( A, B) cos
(11)
cos N
N

A
B

Ai Ai Bi Bi
i 1
i 1

A B N

(2) SID

SID( A, B) D( A || B) D( B || A)

(12)

A B
(3) RMSE
N

RMSE

(S
i 1 j 1

ij

Sij )2

(13)

Sij Sij i j N
- 111 http://www.j-es.org/

(4) SE
N

SE

a
i 1 j 1

N p c1 , c2 ,
aij i c j

ij

aij

(14)

p N
, c p aij i c j

(5) d
N

d 1.0 N

MSE
1.0
PE

(O P )
t 1

( P O O O )
t 1

(15)

MSE PE P O N
[33]
d Willmott [01] d

R 2

3.1
ENVI 5.0 United States Geological Survey, USGS 5
BruciteChabazite(Olivine)Spessartine
Witherite420 2

2 USGS

- 112 http://www.j-es.org/

(a) VCA-FLCS

(b) ATGP-NMF

3
(a)VCA-FLCS (d)ATGP-NMF ( USGS )

Dirichlet 1
Signal Noise RatioSNR 30dB 1296

VCA-FCLS
ATGP-NMF

VCA-FCLS
ATGP-NMF

Brucite
0.0028
0.0056

Brucite
0.097010-4
0.069610-3

VCA-FCLS
ATGP-NMF

1 SAD
Chabazite
Olivine
Spessartine
0.0037
0.0053
0.0042
0.0124
0.0121
0.0080

Witherite
0.0021
0.0046

Mean
0.0036
0.0085

2 SID
Chabazite
Olivine
Spessartine
0.202410-4
0.631010-4
0.282110-4
0.290510-3
0.601910-3
0.094410-3

Witherite
0.045710-4
0.021910-3

Mean
0.251610-4
0.215710-3

3
RMSE
SE
0.0067
0.0047
0.0170
0.0121
- 113 http://www.j-es.org/

d
0.9999
0.9991


1 2 3 ATGP-NMF VCA-FCLS
SADSIDRMSE SE 1
VCA-FCLS

3.2
3.2.1

Hyperion 8
196196 356nm2577nm 10nm 30m30m 242
3.2.2

(1)
1m1m
3m3m ENVI 5.0 Feature Extraction K-
[34]

K-Nearest Neighbor, KNN

ArcGIS 30m30m

(2)
11

30
3.2.3

ENVI Hyperion HFC


VD 7
Plaza Chang

[35]

-3
PF =10 VD

7 VD
PF

10-1

10-2

10-3

10-4

10-5

10-6

VD

11

11

10

10

- 114 http://www.j-es.org/

HFC VD=10 ATGP-NMF


300 9

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

(i)

(j)

(k)

9 ATGP-NMF (a)(b)(c)(d)(e)(f)(g)
(h)(i) 1#(j) 2#(k)
8

3.2.4

(SAD)

(SID)

VCA-FCLS

ATGP-NMF

VCA-FCLS

ATGP-NMF

0.1707

0.1008

0.1206

0.0310

0.3848

0.1584

0.3342

0.3849

0.1158

0.0980

0.0196

0.0113

0.2332

0.0444

0.0582

0.0022

0.1544

0.0618

0.0303

0.0039

0.0449

0.0918

0.0019

0.0104

0.5870

0.1249

0.7253

0.0167

1#

0.1107

0.0759

0.0123

0.0073

2#

0.8300

0.0796

1.1788

0.0065

0.4485

0.0705

0.2403

0.0050

0.3080

0.0906

0.2722

0.0479

- 115 http://www.j-es.org/

8 VCA-FCLS ATGP-NMF 10 SAD


SID

VCA-FCLS
ATGP-NMF

9
RMSE
SE
0.2368
0.1356
0.1814
0.1155

d
0.5538
0.7382

8 VCA-FCLS ATGP-NMF 10 SAD


SIDATGP-NMF

10 ATGP-NMF
RMSESE d
9 VCA-FCLS ATGP-NMF
ATGP-NMF

ATGP-NMF

4
NMF ATGP-NMF
NMF NMF

VCA-FCLS
ATGP-NMF
VCA-FCLS

NMF
(1)
(2)(3)
(4)

REFERENCES
[1]

Li X W, Wang J D,Hu B X et al., "Application of Priori Knowledge in Remote Sensing Inversion." Science China (Series D),
1998, 28(1): 67-72

[2]

S. Haykin, Z. Chen, "The Cocktail Party Problem." Neural computation, 2005, 17: 1875-1902
- 116 http://www.j-es.org/

[3] Yang Wei, Chen Jin, Gong Peng et al. "A New Spectral Mixture Analysis Method Based on Spectral Correlation Matching."
Journal of Remote Sensing. 2008, 12: 454-461
[4] J. Chen, X. Jia, W. Yang et al., "Generalization of Subpixel Analysis for Hyperspectral Data with Flexibility in Spectral Similarity
Measures." Geoscience and Remote Sensing, IEEE Transactions on, 2009, 47: 2165-2171
[5] J. D. Bayliss, J. A. Gualtieri and R. F. Cromp, "Analyzing Hyperspectral Data with Independent Component Analysis." in 26th
AIPR Workshop: Exploiting New Image Sources and Sensors, 1998, 133-143
[6] Tao Xin, Fan Wenjie and Xu Xiru. "Blind Separation of Component Information from Hyperspectral Data." Acta Scientiarum
Naturalium Universitatis Pekinensis, 2008, 6: 19
[7] W. Xia, B. Wang and L. Zhang, "Blind Unmixing Based on Independent Component Analysis for Hyperspectral Imagery." Journal
of Infrared and Millimeter Waves, 2011, 30: 131-136
[8] T. Tanaka and A. Cichocki, "Subband Decomposition Independent Component Analysis and New Performance Criteria." in
Acoustics, Speech, and Signal Processing, 2004. Proceedings. (ICASSP'04). IEEE International Conference on, 2004, 5: V-541-4
[9] I. Kopriva and D. Seri, "Wavelet Packets Approach to Blind Separation of Statistically Dependent Sources." Neurocomputing,
2008, 71:1642-1655
[10] C. Wu, B. Du and L. Zhang, "Hyperspectral Change Detection Based on Independent Component Analysis." J. Remote Sens, 2012,
3: 545-561
[11] A. Pascual-Montano, J. M. Carazo, K. Kochi et al., "Non-smooth Non-negative Matrix Factorization (nsNMF)." Pattern Analysis
and Machine Intelligence, IEEE Transactions on, 2006, 28: 403-415
[12] Tao Xuetao. "Research of Spectral Unmixing Based on Linear Model for Multi-Channel Remote Sensing Images." 2008
[13] D. Cai, X. He, J. Han et al., "Graph Regularized Non-negative Matrix Factorization for Data Representation." Pattern Analysis and
Machine Intelligence, IEEE Transactions on, 2011, 33: 1548-1560
[14] J. V. Stone, "Blind Source Separation using Temporal Predictability." Neural computation, 2001, 13: 1559-1574
[15] P. Georgiev, F. Theis and A. Cichocki, "Sparse Component Analysis and Blind Source Separation of Underdetermined Mixtures."
Neural Networks, IEEE Transactions on, 2005, 16: 992-996
[16] Yu Xianchuan, Cao Tingting, Hu Dan et al., "Blind Image Separation Based on Wavelet Transformation and Sparse Component
Analysis." Journal of Beijing University of Posts and Telecommunications, 2010, 33(002): 58-63
[17] V. P. Pauca, J. Piper and R. J. Plemmons, "Nonnegative Matrix Factorization for Spectral Data Analysis." Linear algebra and its
applications, 2006, 416: 29-47
[18] A. Cichocki and R. Zdunek, "Regularized Alternating Least Squares Algorithms for Non-negative Matrix/Tensor Factorization." in
Advances in Neural Networks-ISNN 2007: Springer, 2007, 793-802
[19] Z. Yang, G. Zhou, S. Xie et al., "Blind Spectral Unmixing Based on Sparse Non-negative Matrix Factorization." Image Processing,
IEEE Transactions on, 2011, 20: 1112-1125
[20] D. D. Lee and H. S. Seung, "Learning the Parts of Objects by Non-negative Matrix Factorization." Nature, 1999, 401: 788-791
[21] Gu YanTing. "Application Research on Quantitative Prediction of Mineral Resources of Non-Negative Matrices Factorizaton
algorithm Based on GIS." 2010.
[22] H. Ren and C. Chang, "Automatic Spectral Target Recognition in Hyperspectral Imagery." Aerospace and Electronic Systems,
IEEE Transactions on, 2003, 39:1232-1249
[23] J. C. Harsanyi, W. H. Farrand and C. Chang, "Determining the Number and Identity of Spectral Endmembers: an Integrated
Approach using Neyman-Pearson Eigen-Thresholding and Iterative Constrained RMS Error Minimization." in Proceedings of the
Thematic Conference on Geologic Remote Sensing, 1993, 395-395
[24] H. Akaike, "A New Look at the Statistical Model Identification. "Automatic Control, IEEE Transactions on, 1974, 19: 716-723
[25] G. Schwarz, "Estimating the Dimension of a Model." The annals of statistics, 1978, 6: 461-464
[26] Wu Bo, Zhang Liangpei, and Li Xiangping. "Unsupervised Orthogonal Subspace Projection Approach to Unmix Hyperspectral
Imagery Automatically." Journal of Image and Graphics, 2004,9: 1392-1396
- 117 http://www.j-es.org/

[27] Li Xi, Chen Xuehong, Chen Xiaoling et al., "Blind Unmixing of Hyperspectral Mixed Pixels Assisted by Wavelet Packet
Decomposition." Acta Photonica Sinica, 2011, 40(6): 835-842
[28] J. M. Nascimento and J. B. Dias, "Vertex Component Analysis: A fast Algorithm to Unmix Hyperspectral Data." Geoscience and
Remote Sensing, IEEE Transactions on, 2005, 43: 898-910
[29] J. M. P. Nascimento, "Unsupervised Hyperspectral Unmixing." Universidade Tcnica de Lisboa, 2006.
[30] D. C. Heinz, "Fully Constrained Least Squares Linear Spectral Mixture Analysis Method for Material Quantification in
Hyperspectral Imagery." Geoscience and Remote Sensing, IEEE Transactions on, 2001, 39: 529-545
[31] J. M. Nascimento and J. B. Dias, "Vertex Component Analysis: A Fast Algorithm to Unmix Hyperspectral Data." Geoscience and
Remote Sensing, IEEE Transactions on, 2005, 43: 898-910
[32] S. Jia and Y. Qian, "Constrained Non-negative Matrix Factorization for Hyperspectral Unmixing." Geoscience and Remote
Sensing, IEEE Transactions on, 2009, 47: 161-173
[33] D. R. Legates and G. J. McCabe, "Evaluating the Use of "Goodness-of-Fit" Measures in Hydrologic and Hydroclimatic Model
Validation." Water Resources Research, 1999, 35: 233-241
[34] T. Cover and P. Hart, "Nearest Neighbor Pattern Classification." Information Theory, IEEE Transactions on, 1967, 13: 21-27
[35] A. Plaza and C. Chang, "Impact of Initialization on Design of Endmember Extraction Algorithms." Geoscience and Remote
Sensing, IEEE Transactions on, 2006, 44: 3397-3407

1989-

1973-

Email: caojingjing8994@126.com

(CPGIS)

1
1 2
1 30
Email: zhuoli@mail.sysu.edu.cn

- 118 http://www.j-es.org/

You might also like