Academia.eduAcademia.edu

Figure 9 – uploaded by Kevin Aguilar Arias

See full PDFdownloadDownload figure
Fig. 9. ERGAS in function of wavelength for the Hyperion dataset. (a) SC. (b) Spectral unmixing. (c) CNMF. (d) LSU (15 x 15). (e) Proposed (SUSC). Fig. 10. CC in function of wavelength for the Hyperion dataset. (a) SC. (b) Spectral unmixing. (c) CNMF. (d) LSU (15 x 15). (e) Proposed (SUSC).

Figure 9 ERGAS in function of wavelength for the Hyperion dataset. (a) SC. (b) Spectral unmixing. (c) CNMF. (d) LSU (15 x 15). (e) Proposed (SUSC). Fig. 10. CC in function of wavelength for the Hyperion dataset. (a) SC. (b) Spectral unmixing. (c) CNMF. (d) LSU (15 x 15). (e) Proposed (SUSC).

Related Figures (12)

Fig. 1. MSI used to construct the dictionary.
Fig. 1. MSI used to construct the dictionary.
“ig. 3. (a) Band 10 of the ROSIS ground truth image. (b) LRHSI. (c) Band 2 of the MSI. (d) PAN image. Spatial resolution enhancement results of band 10 of he HSI: (e) GS; (f) GSA; (g) SFIM; (h) MTF-GLP-HPM; and (i) Proposed method (SUSC).
“ig. 3. (a) Band 10 of the ROSIS ground truth image. (b) LRHSI. (c) Band 2 of the MSI. (d) PAN image. Spatial resolution enhancement results of band 10 of he HSI: (e) GS; (f) GSA; (g) SFIM; (h) MTF-GLP-HPM; and (i) Proposed method (SUSC).
Fig. 4. (a) Band 170 of the Hyperion ground truth image. (b) LRHSI. (c) Band 2 of the MSI. Spatial resolution enhancement results of band 170 of the HS (d) SC; (e) spectral unmixing; (f) LSU (15 x 15); (g) LSU (10 x 10); (h) LSU (5 x 5); (i) CNMF; and (j) proposed method (SUSC).
Fig. 4. (a) Band 170 of the Hyperion ground truth image. (b) LRHSI. (c) Band 2 of the MSI. Spatial resolution enhancement results of band 170 of the HS (d) SC; (e) spectral unmixing; (f) LSU (15 x 15); (g) LSU (10 x 10); (h) LSU (5 x 5); (i) CNMF; and (j) proposed method (SUSC).
SIMULATION RESULTS OF THE DIFFERENT METHODS FOR THE ROSIS DATASET TABLE I After constructing the proper dictionary, the HRHSI can be obtained from a linear combination of a small number of atoms from the dictionary. The matrix with the obtained coefficients is called the sparse code [31]. It is obtained by
SIMULATION RESULTS OF THE DIFFERENT METHODS FOR THE ROSIS DATASET TABLE I After constructing the proper dictionary, the HRHSI can be obtained from a linear combination of a small number of atoms from the dictionary. The matrix with the obtained coefficients is called the sparse code [31]. It is obtained by
Fig. 5. Performance of the SUSC algorithm versus A for the ROSIS dataset. (a) PSNR. (b) ERGAS. (c) SAM. (d) CC.
Fig. 5. Performance of the SUSC algorithm versus A for the ROSIS dataset. (a) PSNR. (b) ERGAS. (c) SAM. (d) CC.
Fig. 6. Performance of the SUSC algorithm versus \ for the Hyperion dataset. (a) PSNR. (b) ERGAS. (c) SAM. (d) CC With the proposed regularization term, the LIP can be rewrit- Aen aQ
Fig. 6. Performance of the SUSC algorithm versus \ for the Hyperion dataset. (a) PSNR. (b) ERGAS. (c) SAM. (d) CC With the proposed regularization term, the LIP can be rewrit- Aen aQ
Fig. 7. (a) PSNR; (b) ERGAS; and (c) CC in function of wavelength for the ROSIS dataset.
Fig. 7. (a) PSNR; (b) ERGAS; and (c) CC in function of wavelength for the ROSIS dataset.
downsampled by a factor of 4 (/). For the simulations, ground truth subimages are selected with size 120 x 120 x 93 leading to LRHSI with size 30 x 30 x 93 and MSI with size 120 x 120 x 4. For the dictionary, twenty PAN QuickBird images” with a spatial resolution 0.7 m, which are unrelated and do not overlap with the test images, have been used. These images are downsampled by a mean filter with a factor of 2 in order to have the same spatial resolution as the HRHSI. Fig. 3 shows for one of the subimages band 10 of the ground truth image and the LRHSI, band 2 of the MSI, and the PAN image. SUSC is applied to these images, and the reconstructed HRHSI is compared with the ground truth images.
downsampled by a factor of 4 (/). For the simulations, ground truth subimages are selected with size 120 x 120 x 93 leading to LRHSI with size 30 x 30 x 93 and MSI with size 120 x 120 x 4. For the dictionary, twenty PAN QuickBird images” with a spatial resolution 0.7 m, which are unrelated and do not overlap with the test images, have been used. These images are downsampled by a mean filter with a factor of 2 in order to have the same spatial resolution as the HRHSI. Fig. 3 shows for one of the subimages band 10 of the ground truth image and the LRHSI, band 2 of the MSI, and the PAN image. SUSC is applied to these images, and the reconstructed HRHSI is compared with the ground truth images.
Fig. 8. PSNR in function of wavelength for the Hyperion dataset. (a) SC. (b) Spectral unmixing. (c) CNMF. (d) LSU (15 x 15). (e) Proposed (SUSC).
Fig. 8. PSNR in function of wavelength for the Hyperion dataset. (a) SC. (b) Spectral unmixing. (c) CNMF. (d) LSU (15 x 15). (e) Proposed (SUSC).
Fig. 11. Obtained pixel spectra of different methods for Hyperion dataset.
Fig. 11. Obtained pixel spectra of different methods for Hyperion dataset.
Related topics:
Computer ScienceArtificial IntelligenceImage fusionGeomatic EngineeringHyperspectral ImagingElectrical and Electronic EngineeringMultispectral image
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved