Figure 1.
Process flow of the proposed algorithm. The function and process of each block are detailed in
Section 2.
Figure 1.
Process flow of the proposed algorithm. The function and process of each block are detailed in
Section 2.
Figure 2.
Visualization results of the Indian Pines dataset at different bands. The first row represents the raw grayscale. The second row represents the noise map with added noise. The third row represents the quality improvement result. Each column from left to right represents (a) band 1, (b) band 10, (c) band 20, (d) band 30, (e) band 50, (f) band 120, and (g) band 170.
Figure 2.
Visualization results of the Indian Pines dataset at different bands. The first row represents the raw grayscale. The second row represents the noise map with added noise. The third row represents the quality improvement result. Each column from left to right represents (a) band 1, (b) band 10, (c) band 20, (d) band 30, (e) band 50, (f) band 120, and (g) band 170.
Figure 3.
Spectral curve at (40,40) in the Indian Pines dataset. Blue represents the original spectral curve, red represents the spectral curve of degraded data, and yellow represents the spectral curve after quality improvement.
Figure 3.
Spectral curve at (40,40) in the Indian Pines dataset. Blue represents the original spectral curve, red represents the spectral curve of degraded data, and yellow represents the spectral curve after quality improvement.
Figure 4.
Visualization results of the Pavia University dataset at different bands. The first row represents the original data. The second row represents the degraded map with added noise (σ = 5). The third row represents the quality improvement result. Each column from left to right represents (a) band 1, (b) band 10, (c) band 20, (d) band 30, (e) band 50, (f) band 80, and (g) band 100.
Figure 4.
Visualization results of the Pavia University dataset at different bands. The first row represents the original data. The second row represents the degraded map with added noise (σ = 5). The third row represents the quality improvement result. Each column from left to right represents (a) band 1, (b) band 10, (c) band 20, (d) band 30, (e) band 50, (f) band 80, and (g) band 100.
Figure 5.
The spectral curve at (311,311) in the PaviaU dataset. Blue represents the original spectral curve, red represents the spectral curve of degraded data, and yellow represents the spectral curve after quality improvement.
Figure 5.
The spectral curve at (311,311) in the PaviaU dataset. Blue represents the original spectral curve, red represents the spectral curve of degraded data, and yellow represents the spectral curve after quality improvement.
Figure 6.
Visualization results of the Salinas dataset at different bands. The first row represents the original data. The second row represents the noise map with added noise (σ = 10). The third row represents the quality improvement result. Each column from left to right represents (a) band 1, (b) band 10, (c) band 40, (d) band 80, (e) band 120, (f) band 160, and (g) band 210.
Figure 6.
Visualization results of the Salinas dataset at different bands. The first row represents the original data. The second row represents the noise map with added noise (σ = 10). The third row represents the quality improvement result. Each column from left to right represents (a) band 1, (b) band 10, (c) band 40, (d) band 80, (e) band 120, (f) band 160, and (g) band 210.
Figure 7.
Spectral curve at (151,151) in the Salinas dataset. Blue represents the original spectral curve, red represents the degraded spectral curve, and yellow represents the spectral curve after quality improvement.
Figure 7.
Spectral curve at (151,151) in the Salinas dataset. Blue represents the original spectral curve, red represents the degraded spectral curve, and yellow represents the spectral curve after quality improvement.
Figure 8.
Visualization comparison results of three public datasets using different methods. (a) Experimental results of band 50 in the Indian Pines dataset. (b) Experimental results of band 1 in the PaviaU dataset. (c) Experimental results of Band 11 in the Salinas dataset. Each line from left to right represents ground truth, degraded data, BM4D, BM4D + LS, BM4D + HE, and the proposed method.
Figure 8.
Visualization comparison results of three public datasets using different methods. (a) Experimental results of band 50 in the Indian Pines dataset. (b) Experimental results of band 1 in the PaviaU dataset. (c) Experimental results of Band 11 in the Salinas dataset. Each line from left to right represents ground truth, degraded data, BM4D, BM4D + LS, BM4D + HE, and the proposed method.
Figure 9.
The spectral curve results of different comparison algorithms. Blue represents the original spectral curve, red represents the spectral curve of degraded data, yellow represents the curve results using the BM4D algorithm, purple represents the curve results using the BM4D + LS algorithm, green represents the curve results using the BM4D + HE algorithm, and blue represents the spectral curve after quality improvement. (a) The spectral curve at (111,111) in the Indian Pines dataset. (b) The spectral curve at (171,171) in the PaviaU dataset. (c) The spectral curve at (101,101) in the Salinas dataset.
Figure 9.
The spectral curve results of different comparison algorithms. Blue represents the original spectral curve, red represents the spectral curve of degraded data, yellow represents the curve results using the BM4D algorithm, purple represents the curve results using the BM4D + LS algorithm, green represents the curve results using the BM4D + HE algorithm, and blue represents the spectral curve after quality improvement. (a) The spectral curve at (111,111) in the Indian Pines dataset. (b) The spectral curve at (171,171) in the PaviaU dataset. (c) The spectral curve at (101,101) in the Salinas dataset.
Figure 10.
Index curves for each band of the Indian Pines dataset. The blue, red, yellow, purple, and dotted green represent each band’s index with degraded data, BM4D, BM4D + LS, BM4D + HE and proposed results, respectively. (a) PSNR evaluation. (b) SSIM evaluation. (c) Brightness evaluation. (d) Contrast evaluation.
Figure 10.
Index curves for each band of the Indian Pines dataset. The blue, red, yellow, purple, and dotted green represent each band’s index with degraded data, BM4D, BM4D + LS, BM4D + HE and proposed results, respectively. (a) PSNR evaluation. (b) SSIM evaluation. (c) Brightness evaluation. (d) Contrast evaluation.
Figure 11.
Classification results on the Indian Pines dataset. (a–c) represent category labels of ground truth, classification result on degraded data, and classification result after quality enhancement, respectively.
Figure 11.
Classification results on the Indian Pines dataset. (a–c) represent category labels of ground truth, classification result on degraded data, and classification result after quality enhancement, respectively.
Figure 12.
Classification results on the PaviaU dataset. (a–c) represent category labels of ground truth, classification result on degraded data, and classification result after quality enhancement, respectively.
Figure 12.
Classification results on the PaviaU dataset. (a–c) represent category labels of ground truth, classification result on degraded data, and classification result after quality enhancement, respectively.
Figure 13.
Classification results on the Salinas dataset. (a–c) represent category labels of ground truth, classification result on degraded data, and classification result after quality enhancement, respectively.
Figure 13.
Classification results on the Salinas dataset. (a–c) represent category labels of ground truth, classification result on degraded data, and classification result after quality enhancement, respectively.
Table 1.
The results of five indicators on three pubic datasets.
Table 1.
The results of five indicators on three pubic datasets.
Class | Indian Pines | Salinas | PaviaU |
---|
C1 | Alfalfa | Brocoli_green_weeds_1 | Meadows |
C2 | Corn-notill | Brocoli_green_weeds_2 | Gravel |
C3 | Corn-mintill | Fallow | Trees |
C4 | Corn | Fallow_rough_plow | Painted metal sheets |
C5 | Grass-pasture | Fallow_smooth | Bare Soil |
C6 | Grass-trees | Stubble | Bitumen |
C7 | Grass-pasture-mowed | Celery | Self-Blocking Bricks |
C8 | Hay-windrowed | Grapes_untrained | Shadows |
C9 | Oats | Soil_vinyard_develop | Meadows |
C10 | Soybean-notill | Corn_senesced_green_weeds | - |
C11 | Soybean-mintill | Lettuce_romaine_4wk | - |
C12 | Soybean-clean | Lettuce_romaine_5wk | - |
C13 | Wheat | Lettuce_romaine_6wk | - |
C14 | Woods | Lettuce_romaine_7wk | - |
C15 | Buildings-Grass-Trees-Drives | Vinyard_untrained | - |
C16 | Stone-Steel-Towers | Vinyard_vertical_trellis | - |
Table 2.
The quantitative results of three pubic datasets. (Bold texts indicate the best performance).
Table 2.
The quantitative results of three pubic datasets. (Bold texts indicate the best performance).
| | SA | PSNR | SSIM | Brightness | Contrast |
---|
Indian Pines | GT | - | - | - | 0.1963 | 0.0339 |
degenerate | 9.7401 | 26.7679 | 0.4725 | 0.2004 | 0.0345 |
BM4D | 3.3897 | 35.2064 | 0.8775 | 0.2002 | 0.0321 |
BM4D + LS | 39.9986 | 9.6317 | 0.3205 | 0.4282 | 0.0451 |
BM4D + HE | 43.4669 | 7.2456 | 0.2092 | 0.4998 | 0.0834 |
Proposed algorithm | 2.8914 | 38.3781 | 0.9408 | 0.2670 | 0.0918 |
PaviaU | GT | - | - | - | 0.1736 | 0.0125 |
degenerate | 15.561 | 26.2975 | 0.4868 | 0.1748 | 0.0146 |
BM4D | 4.0191 | 30.2303 | 0.9295 | 0.1746 | 0.0121 |
BM4D + LS | 4.4843 | 36.0436 | 0.9364 | 0.1651 | 0.0127 |
BM4D + HE | 34.4843 | 8.00190 | 0.3947 | 0.4999 | 0.0860 |
Proposed algorithm | 3.6901 | 38.6992 | 0.9432 | 0.3506 | 0.0603 |
Salinas | GT | - | - | - | 0.1310 | 0.0142 |
degenerate | 15.4885 | 26.7626 | 0.3911 | 0.1350 | 0.0154 |
BM4D | 4.0345 | 33.4485 | 0.9214 | 0.1347 | 0.0134 |
BM4D + LS | 37.1998 | 12.7385 | 0.4668 | 0.3017 | 0.0288 |
BM4D + HE | 40.5397 | 6.73277 | 0.2363 | 0.5000 | 0.0845 |
Proposed algorithm | 3.1011 | 40.0777 | 0.9391 | 0.3151 | 0.0753 |
Table 3.
The classification results on different dividing ratios of three public datasets.
Table 3.
The classification results on different dividing ratios of three public datasets.
Datasets | 50% (%) | 25% (%) | 15% (%) |
---|
Accuracy | Kappa | Accuracy | Kappa | Accuracy | Kappa |
---|
Indian Pines | Degenerate | 93.776 | 92.9 | 81.865 | 79.3 | 72.922 | 68.9 |
Proposed algorithm | 96.702 | 96.3 | 94.224 | 93.4 | 86.111 | 84.1 |
Salinas | Degenerate | 90.120 | 89.0 | 89.342 | 88.5 | 76.720 | 74.5 |
Proposed algorithm | 96.586 | 96.2 | 92.815 | 91.9 | 89.600 | 88.4 |
PaviaU | Degenerate | 93.651 | 91.7 | 92.470 | 90.8 | 86.176 | 81.5 |
Proposed algorithm | 95.666 | 94.3 | 94.620 | 93.0 | 92.954 | 91.7 |
Average | Degenerate | 92.516 | 91.2 | 87.892 | 86.2 | 78.606 | 75.0 |
Proposed algorithm | 96.318 | 95.6 | 93.886 | 92.7 | 89.556 | 88.1 |
Table 4.
The classification accuracy of the 16 categories on the Indian Pines dataset.
Table 4.
The classification accuracy of the 16 categories on the Indian Pines dataset.
Accuracy (%) | C1 | C2 | C3 | C4 | C5 | C6 | C7 | C8 |
---|
Degenerate | 97.8 | 95.6 | 89.3 | 91.8 | 93.2 | 99.2 | 100 | 99.8 |
Proposed algorithm | 100 | 98.7 | 94.3 | 97.5 | 96.6 | 99.9 | 96.3 | 99.8 |
Accuracy (%) | C9 | C10 | C11 | C12 | C13 | C14 | C15 | C16 |
Degenerate | 94.7 | 92.8 | 94.9 | 93 | 99.5 | 98.5 | 88.3 | 98.9 |
Proposed algorithm | 100 | 98.4 | 97.9 | 98.8 | 99.5 | 99.9 | 89.1 | 96.7 |
Table 5.
The classification accuracy of the 9 categories on the PaviaU dataset.
Table 5.
The classification accuracy of the 9 categories on the PaviaU dataset.
Accuracy (%) | C1 | C2 | C3 | C4 | C5 | C6 | C7 | C8 | C9 |
---|
Degenerate | 96.3 | 95.2 | 88.9 | 98.1 | 100 | 92.9 | 92.3 | 93.5 | 100 |
Proposed algorithm | 97.6 | 96.7 | 92.0 | 98.7 | 100 | 97.5 | 97.2 | 95.9 | 99.7 |
Table 6.
The classification accuracy of the 16 categories on the Salinas dataset.
Table 6.
The classification accuracy of the 16 categories on the Salinas dataset.
Accuracy (%) | C1 | C2 | C3 | C4 | C5 | C6 | C7 | C8 |
---|
Degenerate | 97.7 | 99.5 | 95.7 | 99.8 | 97.3 | 99.5 | 99.6 | 83.6 |
Proposed algorithm | 98.4 | 100 | 99.9 | 99.9 | 99.9 | 99.6 | 99.7 | 95.2 |
Accuracy (%) | C9 | C10 | C11 | C12 | C13 | C14 | C15 | C16 |
Degenerate | 98.7 | 95.9 | 92.3 | 98.6 | 98.0 | 98.3 | 60.6 | 94.9 |
Proposed algorithm | 99.8 | 98.4 | 96.5 | 99.0 | 99.2 | 99.0 | 91.1 | 94.1 |