Table 3 Quantitative comparison of METEOR with several state-of-the-art methods (rows) across different heterogeneous Earth observation datasets (columns).
From: Meta-learning to address diverse Earth observation problems across resolutions
5-Shot problem | Human influence | Crop type mapping | Land cover classification | Marine debris | Urban scenes | ||
|---|---|---|---|---|---|---|---|
Dataset | AnthPr.43 | DENETHOR42 | DFC2020-KR39 | EuroSAT40 | fl. obj.6 | NWPU-Urban41 | |
Spatial res. | 10 m | 3 m | 10 m | 10 m | 10 m | <1 m | |
Spectral res. | 10 bands | 4 bands | 13 bands | 13 bands | 12 bands | 3 bands | |
No. of classes | 2 | 3 | 5 | 10 | 2 | 5 | |
No. of training imgs | 10 | 15 | 25 | 50 | 10 | 25 | |
Model | Rank (↓) | Accuracy (↑) | |||||
METEOR | 3.6 | 83.7 | 75.6 | 87.7 | 60.9 | 90.8 | 57.4 |
SWAV36 | 4.2 | 96.7 | 69.8 | 54.2 | 67.7 | 65.4 | 70.4 |
MOSAIKS29 | 4.3 | 86.4 | 76.4 | 82.3 | 57.9 | 88.8 | 54.0 |
DINO37 | 5.0 | 91.2 | 66.2 | 56.6 | 61.3 | 65.1 | 70.6 |
SECO35 | 4.7 | 91.4 | 61.7 | 67.6 | 62.7 | 65.9 | 67.4 |
SSLTRANSRS16 | 5.3 | 90.7 | 65.5 | 76.3 | 59.7 | 78.9 | 52.1 |
SSL4EO34 | 5.5 | 96.2 | 58.0 | 80.2 | 59.1 | 82.4 | 49.9 |
BASELINE | 6.8* | 89.0 | 60.8 | 87.4 | 39.8 | 69.8 | 36.7 |
PROTO17 | 8.3** | 59.7 | 56.2 | 76.9 | 46.1 | 67.3 | 39.1 |
IMAGENET | 8.8* | 83.7 | 59.7 | 50.8 | 42.7 | 64.1 | 60.5 |
SCRATCH | 9.5** | 64.8 | 61.1 | 66.5 | 25.7 | 64.4 | 32.3 |
