Fig. 1. Framework of GAN for hyperspectral image classification
Fig. 2. Sructure of residual unit
Fig. 3. Flowchart of hyperspectral image classification method using residual generative adversarial network
Fig. 4. Structure for residual block of generator
Fig. 5. Network structure of generator
Fig. 6. Structure for residual block of discriminator. (a) Block structure with convolution residuals; (b) Block structure without convolution residuals
Fig. 7. Network structure of discriminator
Fig. 8. Pavia University dataset. (a) Pseudo color image; (b) ground datum map
Fig. 9. Salinas dataset. (a) Pseudo color image; (b) ground datum map
Fig. 10. Indian Pines dataset. (a) Pseudo color image; (b) ground datum map
Fig. 11. Hyperspectral image classification result map of Indian Pines dataset. (a) CAE-SVM; (b) 2DCNN; (c) 3DCNN; (d) ResNet;(e) proposed method
Fig. 12. Hyperspectral image classification result map of Pavia University dataset. (a) CAE-SVM; (b) 2DCNN; (c) 3DCNN; (d) ResNet;(e) proposed method
Fig. 13. Hyperspectral image classification result map of Salinas dataset. (a) CAE-SVM; (b) 2DCNN; (c) 3DCNN; (d) ResNet;(e) proposed method
Class | Name | Color | Sample |
---|
1 | Asphalt | | 6631 | 2 | Meadows | | 18649 | 3 | Gravel | | 2099 | 4 | Trees | | 3064 | 5 | Painted metal sheets | | 1345 | 6 | Bare Soil | | 5029 | 7 | Bitumen | | 1330 | 8 | Self-Blocking Bricks | | 3682 | 9 | Shadows | | 947 |
|
Table 1. Categories and number of Pavia University hyperspectral dataset
Class | Name | Color | Sample |
---|
1 | Brocoli_green_weeds_1 | | 2009 | 2 | Brocoli_green_weeds_2 | | 3726 | 3 | Fallow | | 1976 | 4 | Fallow_rough_plow | | 1394 | 5 | Fallow_smooth | | 2678 | 6 | Stubble | | 3959 | 7 | Celery | | 3579 | 8 | Grapes_untrained | | 11271 | 9 | Soil_vinyard_develop | | 6203 | 10 | Corn_senesced_green_weeds | | 3278 | 11 | Lettuce_romaine_4wk | | 1068 | 12 | Lettuce_romaine_5wk | | 1927 | 13 | Lettuce_romaine_6wk | | 916 | 14 | Lettuce_romaine_7wk | | 1070 | 15 | Vinyard_untrained | | 7268 | 16 | Vinyard_vertical_trellis | | 1807 |
|
Table 2. Categories and number of Salinas hyperspectral dataset
Class | Name | Color | Sample |
---|
1 | Alfalfa | | 46 | 2 | Corn-notill | | 1428 | 3 | Corn-mintill | | 830 | 4 | Corn | | 237 | 5 | Grass-pasture | | 483 | 6 | Grass-trees | | 730 | 7 | Grass-pasture-mowed | | 28 | 8 | Hay-windrowed | | 478 | 9 | Oats | | 20 | 10 | Soybean-notill | | 972 | 11 | Soybean-mintill | | 2455 | 12 | Soybean-clean | | 593 | 13 | Wheat | | 205 | 14 | Woods | | 1265 | 15 | Buildings-Grass-Trees-Drives | | 386 | 16 | Stone-Steel-Towers | | 93 |
|
Table 3. Categories and number of Indian Pines dataset
Index | GAN | GAN with residual genenrator | GAN with residual discriminator | Proposed method |
---|
IP | PU | SA | IP | PU | SA | IP | PU | SA | IP | PU | SA |
---|
OA | 95.85 | 96.59 | 96.57 | 96.52 | 97.11 | 97.53 | 97.21 | 97.34 | 97.93 | 98.84 | 99.00 | 99.09 | AA | 94.10 | 94.42 | 98.02 | 98.50 | 95.77 | 98.69 | 96.38 | 96.25 | 98.77 | 98.48 | 98.74 | 99.45 | K | 95.27 | 95.47 | 96.18 | 96.03 | 96.16 | 97.25 | 96.82 | 96.46 | 97.69 | 98.69 | 98.67 | 98.98 |
|
Table 4. Comparison of classification results of ablation experiments
Index | CAE-SVM | 2DCNN | 3DCNN | ResNet | Proposed method |
---|
OA | 76.81 | 85.93 | 93.85 | 97.05 | 98.84 | AA | 75.52 | 87.23 | 94.13 | 96.78 | 98.48 | K | 75.28 | 84.30 | 92.77 | 96.63 | 98.69 |
|
Table 5. Comparison of classification accuracy of Indian Pines hyperspectral dataset
Index | CAE-SVM | 2DCNN | 3DCNN | ResNet | Proposed |
---|
OA | 84.28 | 90.81 | 95.69 | 97.24 | 99.00 | AA | 70.85 | 86.25 | 94.23 | 96.39 | 98.74 | K | 78.67 | 87.71 | 94.29 | 96.35 | 98.67 |
|
Table 6. Comparison of classification accuracy of Pavia University hyperspectral dataset
Index | CAE-SVM | 2DCNN | 3DCNN | ResNet | Proposed |
---|
OA | 84.34 | 91.22 | 95.41 | 97.20 | 99.09 | AA | 79.97 | 93.02 | 96.07 | 98.31 | 99.45 | K | 82.52 | 90.22 | 94.15 | 96.91 | 98.98 |
|
Table 7. Comparison of classification accuracy of Salinas hyperspectral dataset