Table 6 Reevaluating RaVAEn. The final change predictions were derived by using \(\min ()\) on the change predictions from the memory. In RaVAEn-Floods, the geo-index used was NDWI; in RaVAEn-Wildfires, the NBR; and in RaVAEn-Landslides and RaVAEn-Hurricanes, NDVI. Highlighted are the best scores for STTORM-CD and RaVAEn models for each metric. All values of the evaluation metrics are reported as percentages.
From: STTORM-CD low-demand and high-impact disaster monitoring onboard satellites using change detection
Dataset | RaVAEn-Landslides | RaVAEn-Wildfires | RaVAEn-Hurricanes | RaVAEn-Floods | ||||
|---|---|---|---|---|---|---|---|---|
Tiles count | 626 | 27865 | 11773 | 11253 | ||||
Changed tiles [%] | 23.64 | 57.89 | 27.57 | 21.70 | ||||
Metric | AUPRC \(\uparrow\) | F1 \(\uparrow\) | AUPRC \(\uparrow\) | F1 \(\uparrow\) | AUPRC \(\uparrow\) | F1 \(\uparrow\) | AUPRC \(\uparrow\) | F1 \(\uparrow\) |
Geo-index | 71.69 | 67.92 | 95.16 | 88.60 | 79.12 | 72.27 | 52.53 | 50.76 |
Cosine baseline | 68.06 | 67.48 | 92.52 | 87.40 | 69.55 | 70.57 | 65.63 | 70.30 |
RaVAEn – small | 83.26 | 75.74 | 93.16 | 89.88 | 75.87 | 67.06 | 70.92 | 66.91 |
RaVAEn – medium | 84.72 | 75.91 | 93.18 | 89.95 | 73.52 | 65.24 | 69.73 | 65.33 |
RaVAEn – large | 83.12 | 74.68 | 93.49 | 90.07 | 71.85 | 63.52 | 68.11 | 64.86 |
STTORM-CD – small | 80.72 | 75.79 | 90.46 | 89.30 | 63.88 | 56.88 | 81.67* | 71.99* |
STTORM-CD – medium | 82.22 | 73.13 | 89.15 | 88.32 | 69.05 | 62.73 | 79.04* | 72.02* |
STTORM-CD – large | 82.60 | 74.29 | 93.42 | 89.33 | 70.75 | 64.36 | 79.79* | 71.84* |
