Table 3 Ablation study comparing UNET, RESUNET, and IRU-NET.
From: Land use classification using multi-year Sentinel-2 images with deep learning ensemble network
Method | Accuracy (%) | DSC (%) | IoU (%) | Recall (%) | Precision (%) | Kappa (%) |
|---|---|---|---|---|---|---|
Net | 93.92 | 78.21 | 78.92 | 73.49 | 87.66 | 0.823 |
ResUNet | 94.62 | 78.75 | 79.36 | 77.39 | 87.04 | 0.834 |
IRU-Net | 94.9 | 81.09 | 79.48 | 77.68 | 89.66 | 0.847 |
UNet + TTA | 94.81 | 79.12 | 78.87 | 72.96 | 89.75 | 0.831 |
ResUNet + TTA | 95.01 | 83.24 | 80.85 | 77.19 | 91.92 | 0.856 |
IRUNet + TTA | 98.21 | 88.96 | 88.96 | 89.19 | 94.71 | 0.872 |
