Table 8 Test metric results for each semantic segmentation method for the Botswana and Houston datasets while employing patching for different patch sizes. For TCNN-E-ILS T = 20. An explanation of the metrics can be found in "Experiment configurations" section
From: CEU-Net: ensemble semantic segmentation of hyperspectral images using clustering
Patch | Methods | OA | AA | Kappa | OA | AA | Kappa |
|---|---|---|---|---|---|---|---|
| Â | Â | Botswana (k=3) | Houston (k=2) | ||||
|  | HybridSN | 99.88 ± 0.15 | 99.89 ± 0.15 | 99.87 ± 0.15 | 98.28 ± 0.1 | 98.18 ± 0.1 | 98.19 ± 0.1 |
5 x 5 | U-Net | 98.65 ± 0.15 | 98.97 ± 0.15 | 98.65 ± 0.15 | 98.30 ± 0.1 | 98.34 ± 0.1 | 98.21 ± 0.1 |
|  | CEU-Net | 97.54 ± 0.15 | 97.81 ± 0.15 | 97.34 ± 0.15 | 94.47 ± 0.1 | 95.02 ± 0.1 | 94.02 ± 0.1 |
|  | HybridSN | 98.89 ± 0.15 | 98.97 ± 0.15 | 98.79 ± 0.15 | 97.5 ± 0.1 | 97.30 ± 0.1 | 96.30 ± 0.1 |
10 x 10 | U-Net | 97.92 ± 0.15 | 98.02 ± 0.15 | 97.97 ± 0.15 | 97.78 ± 0.1 | 97.69 ± 0.1 | 96.52 ± 0.1 |
|  | CEU-Net | 96.57 ± 0.15 | 96.45 ± 0.15 | 96.44 ± 0.15 | 97.92 ± 0.1 | 96.54 ± 0.1 | 96.34 ± 0.1 |
|  | HybridSN | 97.41 ± 0.15 | 97.55 ± 0.15 | 97.19 ± 0.15 | 98.05 ± 0.1 | 98.11 ± 0.1 | 97.89 ± 0.1 |
15 x 15 | U-Net | 90.88 ± 0.15 | 91.34 ± 0.15 | 90.1 ± 0.15 | 94.95 ± 0.1 | 95.51 ± 0.1 | 94.54 ± 0.1 |
|  | CEU-Net | 91.38 ± 0.15 | 91.55 ± 0.2 | 90.66 ± 0.2 | 93.94 ± 0.1 | 94.01 ± 0.1 | 92.97 ± 0.1 |
33 x 33 | TCNN-E-ILS | N/A | N/A | N/A | 88.33 ± 0.68 | 88.10 ± 0.86 | 87.39 ± 0.74 |