Table 2.
Producer’s accuracy (PA), user’s accuracy (UA), F1 score (F1), test overall accuracy (OA) and kappa coefficient (kappa) of Support Vector Machine (SVM), CNN, U-Net and ASPP-Unet models based on WorldView-2 (WV2) test image patches, and training OA of each model based on the 50 WV2 training image patches.
| WV2 | SVM | CNN | U-Net | ASPP-Unet | Res_ASPP_Unet | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| PA (%) |
UA (%) |
F 1 | PA (%) |
UA (%) |
F 1 | PA (%) |
UA (%) |
F 1 | PA (%) |
UA (%) |
F 1 | PA (%) |
UA (%) |
F 1 | |
| Vegetation | 91.9 | 91.1 | 0.915 | 92.4 | 93.2 | 0.928 | 91.8 | 93.3 | 0.926 | 91.7 | 94.1 | 0.929 | 91.8 | 94.3 | 0.930 |
| Water | 98.9 | 55.9 | 0.714 | 97.8 | 90.8 | 0.942 | 97.5 | 79.4 | 0.875 | 98.5 | 87.7 | 0.928 | 90.2 | 91.0 | 0.906 |
| Road | 45.6 | 52.4 | 0.488 | 67.7 | 74.4 | 0.709 | 71.8 | 74.5 | 0.731 | 78.0 | 70.1 | 0.738 | 78.5 | 79.9 | 0.792 |
| Building | 58.9 | 49.1 | 0.536 | 83.5 | 71.9 | 0.773 | 83.1 | 81.3 | 0.822 | 81.3 | 84.5 | 0.830 | 85.1 | 85.9 | 0.856 |
| Shadow | 67.7 | 86.7 | 0.796 | 81.1 | 87.8 | 0.843 | 81.8 | 81.7 | 0.817 | 79.0 | 80.3 | 0.796 | 89.0 | 74.9 | 0.814 |
| Others | 41.1 | 72.7 | 0.525 | 13.1 | 79.2 | 0.224 | 0 | 0 | 0 | 0 | 0 | 0 | 39.2 | 84.4 | 0.536 |
| Test OA (%) | 71.4 | 83.6 | 84.7 | 85.2 | 87.1 | ||||||||||
| kappa | 0.606 | 0.774 | 0.788 | 0.793 | 0.820 | ||||||||||
| Training OA (%) | 79.3 | 89.5 | 85.5 | 86.4 | 87.6 | ||||||||||