Acta Photonica Sinica, Volume. 50, Issue 7, 113(2021)
Hyperspectral Unmixing Based on Constrained Nonnegative Matrix Factorization
Figures & Tables(12)
Fig. 1. Spectral signatures used to generate the simulated data and abundance map of endmember 1-4
Fig. 2. Performance of AEC-NMF with respect to parameters
Fig. 3. Performance of AEC-NMF with respect to parameters
Fig. 5. Comparison of the library spectra （blue） with the endmember signatures extracted by AEC-NMF （red） on the jasper Ridge data set and the estimated fractional abundance map for each endmember
Fig. 6. Cuprite hyperspectral data and twelve kinds of endmember spectral signatures
Fig. 7. Comparison of the library spectra （blue） with the endmember signatures extracted by AEC-NMF （red） on the Cuprite data set
Table 1. SAD values of different methods with the simulated data
View tableView in ArticleTable 1. SAD values of different methods with the simulated data
SNR/dB AEC-NMF TV-RSNMF L1/2NMF VCA-FCLS 15 0.034 7 0.038 3 0.044 0 0.056 4 25 0.009 8 0.012 5 0.012 8 0.015 3 35 0.004 4 0.004 6 0.004 5 0.004 9
Table 2. RMSE values of different methods with the simulated data
View tableView in ArticleTable 2. RMSE values of different methods with the simulated data
SNR/dB AEC-NMF TV-RSNMF L1/2NMF VCA-FCLS 15 0.037 5 0.041 8 0.049 2 0.047 3 25 0.009 1 0.010 4 0.015 8 0.016 0 35 0.005 7 0.005 8 0.005 9 0.007 4
Table 3. SAD values of different methods with the real jasperRidge data set
View tableView in ArticleTable 3. SAD values of different methods with the real jasperRidge data set
Method AEC-NMF TV-RSNMF L1/2NMF VCA-FCLS Tree 0.041 3 0.044 0 0.153 0 0.155 9 Soil 0.163 3 0.186 9 0.092 6 0.245 3 Water 0.020 8 0.019 3 0.121 0 0.133 6 Road 0.082 8 0.083 9 0.060 4 0.106 9 Mean 0.077 0 0.083 5 0.106 8 0.160 4
Table 4. RMSE values of different methods with the real jasperRidge data set
View tableView in ArticleTable 4. RMSE values of different methods with the real jasperRidge data set
Method AEC-NMF TV-RSNMF L1/2NMF VCA-FCLS Tree 0.082 0 0.086 9 0.160 7 0.159 9 Soil 0.189 9 0.190 1 0.206 8 0.208 5 Water 0.094 7 0.098 0 0.128 1 0.130 0 Road 0.128 4 0.124 1 0.122 3 0.123 3 Mean 0.123 7 0.124 8 0.154 5 0.155 2
Table 5. SAD values of different methods with the real Cuprite data set
View tableView in ArticleTable 5. SAD values of different methods with the real Cuprite data set
Method AEC-NMF TV-RSNMF L1/2NMF VCA-FCLS Alunite 0.103 1 0.106 4 0.092 1 0.085 9 Andradite 0.087 6 0.087 8 0.065 2 0.058 2 Buddingtonite 0.091 5 0.096 4 0.064 8 0.072 4 Dumortierite 0.098 7 0.111 2 0.097 2 0.097 8 Kaolinite1 0.131 2 0.131 6 0.126 8 0.122 2 Kaolinite2 0.043 6 0.044 9 0.044 0 0.045 8 Muscovite 0.127 6 0.127 9 1.166 7 1.152 2 Montmorillonite 0.069 6 0.069 8 0.072 0 0.071 7 Nontronite 0.090 7 0.090 4 0.117 3 0.107 0 Pyrope 0.088 0 0.088 1 0.189 7 0.178 3 Sphene 1.013 0 1.013 0 0.082 6 0.087 6 Chalcedony 0.074 0 0.076 1 0.191 9 0.167 5 Mean 0.168 2 0.170 3 0.192 5 0.187 2
Tools
Get Citation
Copy Citation Text
Xiangxiang JIA, Baofeng GUO, Fanchang DING, Wenjie XU. Hyperspectral Unmixing Based on Constrained Nonnegative Matrix Factorization[J]. Acta Photonica Sinica, 2021, 50(7): 113
Paper Information
Category: Image Processing
Received: Dec. 21, 2020
Accepted: Apr. 16, 2021
Published Online: Sep. 1, 2021
The Author Email:
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20