Table 4 Architecture comparison.
From: Land use classification using multi-year Sentinel-2 images with deep learning ensemble network
Model | Encoder | Decoder | Feature fusion | TTA | Novel element |
|---|---|---|---|---|---|
UNet | Basic CNN | Yes | No | No | Standard UNet |
ResUNet | CNN + Residual | Yes | No | No | Residual connections |
Attention-UNet | CNN + Attention Gate | Yes | No | No | Attention in skip paths |
IRUNet (ours) | Inception ResNetV2 | Yes | Yes | Yes | Multi-scale fusion + TTA |
