研究文章
超限分辨重建方法的软件基于卷积神经网络的遥感图像
表2
PNSR和SSIM的结果在不同的方法在不同的数据集。
|
| 图像 |
规模 |
双三次的 |
SRCNN |
ESPCN |
VDSR |
我们的 |
|
| 建筑 |
×2 |
27.25/0.9192 |
29.11/0.9376 |
29.88/0.9397 |
31.05/0.9567 |
32.57/0.9664 |
| ×3 |
23.32/0.8293 |
24.85/0.8326 |
26.12/0.8606 |
26.52/0.8821 |
27.81/0.9095 |
| ×4 |
22.25/0.7716 |
23.11/0.7522 |
23.55/0.7537 |
22.72/0.7983 |
25.48/0.8274 |
| Denseresidential |
×2 |
27.89/0.9176 |
30.22/0.9349 |
30.70/0.9380 |
30.77/0.9456 |
32.54/0.9566 |
| ×3 |
23.61/0.8246 |
25.54/0.8403 |
26.86/0.8609 |
26.85/0.8759 |
28.33/0.9012 |
| ×4 |
22.91/0.7761 |
23.42/0.7247 |
24.68/0.7659 |
24.57/0.7862 |
25.64/0.8421 |
| 飞机 |
×2 |
30.83/0.9547 |
32.17/0.9615 |
32.31/0.9617 |
31.93/0.9657 |
34.73/0.9740 |
| ×3 |
27.09/0.8732 |
27.98/0.8844 |
29.49/0.8978 |
30.16/0.9111 |
31.55/0.9248 |
| ×4 |
25.39/0.8323 |
26.05/0.8366 |
27.17/0.8456 |
27.64/0.8659 |
28.83/0.8852 |
|
|
