Improved Hyperspectral Unmixed Initialization Method Based on Non-Negative Matrix Factorization

Pengfei Huang; Xiangbing Kong; Haitao Jing

doi:10.3788/LOP57.061020

Journals >Laser & Optoelectronics Progress >Volume 57 >Issue 6 >Page 061020 > Article

Laser & Optoelectronics Progress
Vol. 57, Issue 6, 061020 (2020)

Improved Hyperspectral Unmixed Initialization Method Based on Non-Negative Matrix Factorization

Pengfei Huang^1、2, Xiangbing Kong^2、*, and Haitao Jing¹

Author Affiliations

¹School of Surveying and Land Information Engineering, Henan Polytechnic University, Jiaozuo, Henan 454150, China

²Yellow River Institute of Hydraulic Research, Zhengzhou, Henan 450000, China

show less

DOI: 10.3788/LOP57.061020 Cite this Article Set citation alerts

Pengfei Huang, Xiangbing Kong, Haitao Jing. Improved Hyperspectral Unmixed Initialization Method Based on Non-Negative Matrix Factorization[J]. Laser & Optoelectronics Progress, 2020, 57(6): 061020 Copy Citation Text

show less

Fig. 1. Initialization flow chart of IISSF

Download full size

Fig. 2. Segmentation image and confidence index. (a) Segmentation image; (b) confidence index

Download full size

Fig. 3. Endmember spectra. (a) Acmite; (b) Alunite; (c) Ammonio; (d) Analcime; (e) Andalusite

Download full size

Fig. 4. Endmember abundance

Download full size

Fig. 5. Cuprite Image

Download full size

Fig. 6. Endmember abundance estimated by RESGSNMF. (a) Sphene; (b) Pyrope; (c) Alunite; (d) Muscovite; (e) Kaolinite_2; (f) Chalcedony; (g) Dumortierite; (h) Montmorillonite; (i) Kaolinite_1; (j) Andradite; (k) Nontronite; (l) Buddingtonite

Download full size

Fig. 7. RMSE of reconstructed image

Download full size

SNR /dB	20	25	30	35	40	45	50
VCAFCLS	0.1394	0.0594	0.0313	0.0262	0.0217	0.0217	0.022
ASCNMF	0.0832	0.0479	0.0277	0.0261	0.0215	0.0216	0.0217
SGSNMF	0.0357	0.0252	0.0224	0.0215	0.0136	0.0072	0.0045
RESGSNMF	0.0355	0.0249	0.0219	0.0216	0.0101	0.0057	0.0030

Table 1. RMSE at different SNR

SNR /dB	20	25	30	35	40	45	50
VCAFCLS	0.0902	0.0479	0.0269	0.0175	0.0089	0.0047	0.0024
ASCNMF	0.0468	0.0294	0.0178	0.0102	0.0066	0.0037	0.0021
SGSNMF	0.0201	0.0183	0.0131	0.0045	0.0026	0.0024	0.0022
RESGSNMF	0.0201	0.0174	0.0096	0.0043	0.0020	0.0017	0.0017

Table 2. SAD at different SNR

Endmember type	VCAFCLS	ASCNMF	SGSNMF	RESGSNMF
Alunite	0.1021	0.1127	0.1222	0.1143
Andradite	0.6908	0.1057	0.0699	0.0691
Buddingtonite	1.0904	0.0947	0.0966	0.117
Dumortierite	0.1001	0.125	0.0735	0.1003
Kaolinite_1	0.0838	0.6064	0.1249	0.179
Kaolinite_2	0.0758	0.0676	0.0799	0.0765
Muscovite	0.1746	0.6873	1.1197	1.1181
Montmorillonite	0.061	0.063	0.0686	0.0597
Nontronite	0.1072	0.077	0.1164	0.095
Pyrope	0.8972	0.3081	0.2339	0.0948
Sphene	0.0833	0.0531	0.0556	0.0604
Chalcedony	0.0896	0.0858	0.088	0.0907
Average of SAD	0.2963	0.1988	0.1874	0.1812

Table 3. SAD of 12 kinds of endmembers for different methods

Method	VCAFCLS	ASCNMF	SGSNMF	RESGSNMF
Running time /s	10.200	17.353	25.241	25.031

Table 4. Running time of four different methods

Pengfei Huang, Xiangbing Kong, Haitao Jing. Improved Hyperspectral Unmixed Initialization Method Based on Non-Negative Matrix Factorization[J]. Laser & Optoelectronics Progress, 2020, 57(6): 061020

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information