Table 2 Quantitative comparisons of the Potsdam dataset with other models.
Method | Per-class F1 score | Avg. F1 | OA | mIOU | ||||
|---|---|---|---|---|---|---|---|---|
Impervious surface | Building | Low vegetation | Tree | Car | ||||
UNet | 85.88 | 88.14 | 73.19 | 78.9 | 85.13 | 82.25 | 82.01 | 66.44 |
FCN + SE | 87.84 | 89.54 | 76.2 | 78.94 | 85.47 | 83.6 | 84.09 | 69.47 |
DeepLabV3 + | 87.66 | 91.11 | 75.96 | 77.32 | 86.5 | 83.71 | 84.41 | 70.08 |
PSPNet | 87.34 | 91 | 74.16 | 79.3 | 83.25 | 83.01 | 84.08 | 69.27 |
DANet | 85.89 | 87.5 | 75.98 | 79.08 | 86.69 | 83.02 | 82.34 | 67.27 |
CBAM | 86.98 | 88.94 | 76.14 | 76.94 | 86.42 | 83.08 | 83.22 | 68.44 |
Proposed | 88.81 | 92.03 | 76.89 | 79.31 | 86.96 | 84.8 | 85.4 | 71.44 |
