Research Article
Convolution Neural Network Based on Two-Dimensional Spectrum for Hyperspectral Image Classification
Table 8
Comparison of CNNs with other algorithms in Indian Pines, Salinas, KSC, and Botswana.
| Dataset | Performance | CNN (vector) | PPF (vector) | BASS (blocks) | 2D-spectrum CNN |
| Indian Pines | OA | 86.44 | 94.34 | 96.77 | 98.26 | Kappa | 0.856 | 0.922 | 0.955 | 0.978 |
| Salinas | OA | 89.28 | 94.80 | 95.36 | 97.28 | Kappa | 0.878 | 0.931 | 0.928 | 0.962 |
| KSC | OA | 88.38 | 93.18 | 94.53 | 96.22 | Kappa | 0.870 | 0.912 | 0.939 | 0.956 |
| Botswana | OA | 89.45 | 91.57 | 95.42 | 97.93 | Kappa | 0.885 | 0.908 | 0.939 | 0.963 |
|
|
Vector: one-dimensional spectral vectors. Blocks: small-area pixel division blocks.
|