实验思路及记录
选择Loess、Karst、Mountain、Desert、Flat、Hill、Glacier七种地貌类型的数据,每种地貌选择5个区域,分别切成64与500像素大小的小块。
使用TfaSR预训练好的模型直接应用到七种地貌上,比较与双线性三次插值结果的坡度、高程精度指标的差异。
| 地貌 | TfaSR_RMSE_Slope | TfaSR_RMSE_Elevation | Cubic_RMSE_Slope | Cubic_RMSE_Elevation | Change_Elevation | Change_Slope |
|---|---|---|---|---|---|---|
| Loess | 0 | 0 | 0 | 0 | 0 | 0 |
| Karst | 0 | 0 | 0 | 0 | 0 | 0 |
| Hill | 0 | 0 | 0 | 0 | 0 | 0 |
| Mountain | 0 | 0 | 0 | 0 | 0 | 0 |
| Desert | 0 | 0 | 0 | 0 | 0 | 0 |
| Glacier | 0 | 0 | 0 | 0 | 0 | 0 |
| Plain | 0 | 0 | 0 | 0 | 0 | 0 |
Baseline: SRCNN,SRResNet,TfaSR 1、分别先在TfaSR数据集上进行预训练;
| Loess | Pre_SRCNN | Pre_SRResNet | Pre_TfaSR | MF_SRCNN | MF_SRResNet | MF_TfaSR |
|---|---|---|---|---|---|---|
| RMSE_Ele_Loess | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Karst | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Hill | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Mountain | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Desert | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Glacier | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Plain | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Loess | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Karst | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Hill | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Mountain | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Desert | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Glacier | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Plain | 0 | 0 | 0 | 0 | 0 | 0 |
2、使用预训练好的模型直接应用到七种地貌上; 3、分别测试七种地貌,每次选六种地貌进行元迁移学习,在第七种地貌上进行测试。
1.与迁移学习的对比:每个地貌中,选取3个区域的数据做迁移性训练,剩下2个区域的数据做测试,保证迁移学习的更新次数与元迁移测试的次数相同;
| T_SRCNN | T_SRResNet | T_TfaSR | MF_SRCNN | MF_SRResNet | MF_TfaSR | |
|---|---|---|---|---|---|---|
| RMSE_Ele_Loess | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Karst | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Hill | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Mountain | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Desert | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Glacier | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Plain | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Loess | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Karst | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Hill | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Mountain | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Desert | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Glacier | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Plain | 0 | 0 | 0 | 0 | 0 | 0 |
2.消融实验:地形知识对模型性能的影响
| No_SRCNN | No_SRResNet | No_TfaSR | MF_SRCNN | MF_SRResNet | MF_TfaSR | |
|---|---|---|---|---|---|---|
| RMSE_Ele_Loess | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Karst | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Hill | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Mountain | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Desert | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Glacier | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Ele_Plain | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Loess | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Karst | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Hill | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Mountain | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Desert | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Glacier | 0 | 0 | 0 | 0 | 0 | 0 |
| RMSE_Slope_Plain | 0 | 0 | 0 | 0 | 0 | 0 |
3.地形复杂度与迁移性可能的联系。