A Two-Stream Deep Fusion Framework for High-Resolution Aerial Scene Classification

Table 8

Comparison with the state-of-the-art methods on the NWPU-RESISC45 dataset.

Methods	Training ratios
	10%	20%
GIST [46]	15.90 ± 0.23	17.88 ± 0.22
LBP [46]	19.20 ± 0.41	21.74 ± 0.18
Color histograms [46]	24.84 ± 0.22	27.52 ± 0.14
BoVW + SPM [46]	27.83 ± 0.61	32.96 ± 0.47
LLC [46]	38.81 ± 0.23	40.03 ± 0.34
BoVW [46]	41.72 ± 0.21	44.97 ± 0.28
GoogLeNet [46]	76.19 ± 0.38	78.48 ± 0.26
VGGNet-16 [46]	76.47 ± 0.18	79.79 ± 0.15
AlexNet [46]	76.69 ± 0.21	79.85 ± 0.13
Ours	80.22 ± 0.22	83.16 ± 0.18