Research Article
Composite Clustering Sampling Strategy for Multiscale Spectral-Spatial Classification of Hyperspectral Images
Table 5
Classification accuracy and test set independence rate with different sampling strategies of the Pavia University dataset.
| Sampling strategy | 3D-CNN | 1D-3D-2D-CNN | Test set-independent sample rate (%) | OA (%) | AA (%) | Kappa | Time (min.) | OA (%) | AA (%) | Kappa | Time (min.) |
| Area | 63.36 | 72.01 | 0.5581 | 17.94 | 66.12 | 79.12 | 0.5904 | 3.19 | 98.60 | Random | 97.54 | 96.90 | 0.9673 | 17.90 | 98.36 | 97.92 | 0.9782 | 3.44 | 3.31 | | 84.37 | 86.66 | 0.7979 | 17.80 | 84.91 | 88.71 | 0.8067 | 3.13 | 95.11 | | 87.31 | 85.70 | 0.8322 | 17.94 | 87.35 | 87.16 | 0.8337 | 3.65 | 94.10 | | 90.40 | 89.40 | 0.8710 | 18.05 | 92.12 | 91.21 | 0.8967 | 3.11 | 94.41 | | 88.78 | 89.73 | 0.8481 | 18.15 | 91.21 | 90.72 | 0.8843 | 3.28 | 94.08 | | 86.62 | 87.11 | 0.8236 | 17.99 | 88.83 | 88.56 | 0.8554 | 3.26 | 93.63 | | 88.35 | 87.32 | 0.8478 | 18.19 | 91.69 | 90.78 | 0.8912 | 3.34 | 92.94 | | 89.41 | 88.11 | 0.8594 | 17.71 | 91.27 | 89.94 | 0.8857 | 3.40 | 92.69 | | 90.52 | 88.15 | 0.8739 | 17.91 | 91.48 | 89.26 | 0.8876 | 3.52 | 92.27 | | 89.88 | 88.43 | 0.8646 | 17.97 | 92.84 | 91.35 | 0.9060 | 3.70 | 92.00 | | 89.52 | 88.59 | 0.8594 | 17.80 | 92.17 | 90.54 | 0.8968 | 3.70 | 91.16 | | 90.27 | 89.16 | 0.8701 | 17.60 | 92.45 | 91.46 | 0.9014 | 3.42 | 90.62 | | 92.06 | 90.95 | 0.8939 | 17.59 | 93.34 | 92.40 | 0.9124 | 3.92 | 90.52 | | 92.32 | 91.77 | 0.8973 | 17.76 | 93.60 | 93.29 | 0.9162 | 3.47 | 90.15 | | 90.54 | 89.55 | 0.8734 | 17.68 | 93.36 | 92.65 | 0.9127 | 3.87 | 89.87 | | 90.10 | 89.65 | 0.8689 | 17.87 | 90.41 | 90.88 | 0.8756 | 3.89 | 89.28 |
|
|