Table 1.
Quantitative results for all models according to their band combinations. Training time is measured for 120 epochs.
| Model | Encoder | Band combination | OA (%) | mIoU (%) | Training time (hours) | Inference time (seconds) | Epoch best model |
|---|---|---|---|---|---|---|---|
| U-Net | SK-ResNeXt50 | RGB | 79.010 | 53.161 | 4.129 | 0.010 | 113 |
| RG-NIR | 79.533 | 53.255 | 4.066 | 0.010 | 115 | ||
| RGB-NIR | 80.561 | 54.394 | 4.283 | 0.011 | 109 | ||
| U-Net | - | RGB | 75.025 | 48.814 | 3.602 | 0.003 | 118 |
| RG-NIR | 74.380 | 49.800 | 3.890 | 0.003 | 117 | ||
| RGB-NIR | 76.106 | 50.461 | 4.465 | 0.003 | 117 | ||
| ResNet50 | RGB | 78.859 | 52.740 | 3.395 | 0.006 | 116 | |
| RG-NIR | 78.279 | 52.547 | 3.212 | 0.006 | 116 | ||
| RGB-NIR | 77.522 | 52.208 | 3.539 | 0.006 | 115 | ||
| VGG16 | RGB | 78.417 | 52.814 | 3.326 | 0.009 | 119 | |
| RG-NIR | 78.157 | 52.699 | 3.223 | 0.009 | 117 | ||
| RGB-NIR | 78.482 | 52.921 | 3.552 | 0.006 | 120 | ||
| MobileNetV2 | RGB | 78.654 | 52.369 | 3.375 | 0.010 | 120 | |
| RG-NIR | 78.022 | 51.930 | 3.512 | 0.009 | 120 | ||
| RGB-NIR | 76.798 | 50.121 | 3.802 | 0.010 | 119 | ||
| DeepLabV3+ | ResNet50 | RGB | 79.669 | 53.484 | 3.443 | 0.005 | 118 |
| RG-NIR | 79.866 | 53.730 | 3.154 | 0.005 | 120 | ||
| RGB-NIR | 79.970 | 54.008 | 3.473 | 0.006 | 120 | ||
| MobileNetV2 | RGB | 79.144 | 51.906 | 2.840 | 0.005 | 120 | |
| RG-NIR | 78.145 | 51.136 | 2.854 | 0.006 | 120 | ||
| RGB-NIR | 79.337 | 51.949 | 3.083 | 0.006 | 118 | ||
| MobileNetV3 | RGB | 78.925 | 52.271 | 3.465 | 0.006 | 120 | |
| RG-NIR | 79.566 | 52.475 | 3.516 | 0.006 | 120 | ||
| RGB-NIR | 79.350 | 52.170 | 3.626 | 0.006 | 118 | ||
| PSPNet | ResNet50 | RGB | 76.436 | 48.875 | 2.920 | 0.003 | 119 |
| RG-NIR | 75.238 | 47.372 | 2.999 | 0.003 | 120 | ||
| RGB-NIR | 77.418 | 50.213 | 3.126 | 0.003 | 120 | ||
| VGG16 | RGB | 77.483 | 50.236 | 3.292 | 0.004 | 118 | |
| RG-NIR | 78.067 | 50.629 | 3.147 | 0.003 | 119 | ||
| RGB-NIR | 77.719 | 50.075 | 3.183 | 0.003 | 120 | ||
| MobileNetV2 | RGB | 75.739 | 46.193 | 2.580 | 0.003 | 119 | |
| RG-NIR | 76.159 | 46.371 | 2.610 | 0.003 | 115 | ||
| RGB-NIR | 76.493 | 46.695 | 2.959 | 0.003 | 120 | ||
| Ma-Net | ResNet50 | RGB | 75.623 | 48.006 | 3.789 | 0.003 | 115 |
| RG-NIR | 77.403 | 50.281 | 3.648 | 0.003 | 117 | ||
| RGB-NIR | 78.556 | 52.110 | 3.985 | 0.003 | 119 | ||
| VGG16 | RGB | 78.732 | 51.241 | 3.689 | 0.003 | 118 | |
| RG-NIR | 77.871 | 51.132 | 3.477 | 0.003 | 115 | ||
| RGB-NIR | 77.938 | 51.019 | 3.776 | 0.003 | 112 | ||
| MobileNetV2 | RGB | 79.313 | 50.883 | 3.552 | 0.003 | 120 | |
| RG-NIR | 78.329 | 50.722 | 3.544 | 0.003 | 120 | ||
| RGB-NIR | 78.858 | 50.783 | 3.888 | 0.003 | 120 | ||
| DeepLabV3 | ResNet50 | RGB | 79.940 | 54.106 | 4.308 | 0.006 | 120 |
| RG-NIR | 79.154 | 53.764 | 4.376 | 0.006 | 120 | ||
| RGB-NIR | 79.880 | 53.890 | 4.613 | 0.006 | 120 | ||
| MobileNetV2 | RGB | 78.682 | 51.843 | 3.724 | 0.008 | 118 | |
| RG-NIR | 78.825 | 51.795 | 3.895 | 0.006 | 118 | ||
| RGB-NIR | 79.111 | 51.957 | 3.965 | 0.006 | 120 | ||
| MobileNetV3 | RGB | 79.100 | 51.962 | 3.570 | 0.006 | 120 | |
| RG-NIR | 79.071 | 51.954 | 3.620 | 0.006 | 118 | ||
| RGB-NIR | 79.163 | 52.345 | 3.779 | 0.006 | 119 | ||
| SegFormer | - | RGB | 71.334 | 42.381 | 2.682 | 0.012 | 108 |
| RG-NIR | 71.716 | 43.100 | 2.684 | 0.012 | 119 | ||
| RGB-NIR | 72.794 | 44.264 | 2.976 | 0.013 | 111 |