Table 4.
The classification accuracy (OA (%) and SD) of the examined method and the reference methods, with 80% and 50% of UC-Merced dataset as a training set.
| Method | 80% of UCM Dataset as a Training Set | 50% of UCM Dataset as a Training Set |
|---|---|---|
| CaffeNet [44] | 95.02 ± 0.81 | 93.98 ± 0.67 |
| GoogLeNet [44] | 94.31 ± 0.89 | 92.70 ± 0.60 |
| VGG-16 [44] | 95.21 ± 1.20 | 94.14 ± 0.69 |
| SRSCNN [58] | 95.57 | / |
| CNN-ELM [59] | 95.62 | / |
| salM3LBP-CLM [60] | 95.75 ± 0.80 | 94.21 ± 0.75 |
| TEX-Net-LF [61] | 96.62 ± 0.49 | 95.89 ± 0.37 |
| LGFBOVW [62] | 96.88 ± 1.32 | / |
| Fine-tuned GoogLeNet [63] | 97.10 | / |
| Fusion by addition [64] | 97.42 ± 1.79 | / |
| CCP-net [65] | 97.52 ± 0.97 | / |
| Two-stream Fusion [50] | 98.02 ± 1.03 | 96.97 ± 0.75 |
| DSFATN [66] | 98.25 | / |
| Deep CNN Transfer [37] | 98.49 | / |
| InceptionV3 mixed_8 (PCA) + Xception avg pooling (Ours) | 98.57 | 97.62 |
| GCFs+LOFs [57] | 99 ± 0.35 | 97.37 ± 0.44 |
| Inception-v3-CapsNet [54] | 99.05 ± 0.24 | 97.59 ± 0.16 |