Table 3.
Classification accuracies (%) of the three SVM classifiers trained with the reduced number of integrated spatial–spectral–temporal features compared to the deep spatial features for each ResNet for the OMNIAHCOV dataset.
| Deep Features | L-SVM | Q-SVM | C-SVM |
|---|---|---|---|
| ResNet-18 | |||
| Spatial | 92.7 (0.05) | 92.63 (0.02) | 92.30 (0.07) |
| Spatial-Spectral-Temporal | 98.19 (0.10) | 98.21 (0.02) | 97.98 (0.15) |
| ResNet-50 | |||
| Spatial | 94.17 (0.02) | 94.13 (0.02) | 94.13 (0) |
| Spatial-Spectral-Temporal | 99.23 (0.11) | 99.32 (0.06) | 99.38 (0) |
| ResNet-101 | |||
| Spatial | 95.36 (0) | 95.41 (0.06) | 95.35 (0.04) |
| Spatial-Spectral-Temporal | 99.05 (0) | 98.97 (0.02) | 98.97 (0.02) |