Research Article
Edge Prior Multilayer Segmentation Network Based on Bayesian Framework
Table 2
Confusion matrix of data set 2.
| Method | Category | Others | Shrub | Higher crop | Untilled glebe | Building | Forest | Average acc. |
| FCN8 | Others | 81.62 | 1.86 | 4.76 | 6.82 | 2.10 | 2.84 | 75.34 | Shrub | 33.45 | 65.23 | 0.87 | 0.12 | 0.19 | 0.14 | Higher crop | 47.69 | 1.17 | 43.98 | 4.85 | 1.42 | 0.90 | Untilled glebe | 11.88 | 0.09 | 2.72 | 82.89 | 1.45 | 0.97 | Building | 34.69 | 0.36 | 1.54 | 1.80 | 44.92 | 16.68 | Forest | 9.06 | 0.75 | 0.43 | 0.73 | 7.74 | 81.29 |
| DeepLab | Others | 73.29 | 0.76 | 25.85 | 0.02 | 0.06 | 0.01 | 78.83 | Shrub | 1.29 | 49.23 | 41.26 | 2.15 | 5.07 | 1.01 | Higher crop | 2.33 | 3.92 | 81.26 | 2.57 | 7.49 | 2.44 | Untilled glebe | 0.03 | 0.72 | 19.81 | 69.94 | 1.26 | 8.23 | Building | 0.15 | 1.16 | 10.73 | 2.34 | 84.60 | 1.02 | Forest | 0.07 | 0.22 | 4.75 | 5.33 | 0.62 | 89.00 |
| FCN8+DT+the fully connected CRF | Others | 82.77 | 1.49 | 2.76 | 7.28 | 3.23 | 2.47 | 79.48 | Shrub | 25.03 | 74.71 | 0.14 | 0.06 | 0.03 | 0.02 | Higher crop | 39.64 | 0.94 | 52.88 | 3.29 | 2.23 | 1.03 | Untilled glebe | 13.56 | 0.13 | 0.80 | 82.37 | 2.11 | 1.01 | Building | 26.19 | 0.31 | 0.14 | 0.48 | 69.91 | 2.97 | Forest | 10.04 | 0.08 | 0.04 | 0.27 | 3.56 | 86.01 |
| FCN8+HED+DT+the fully connected CRF | Others | 84.56 | 0.75 | 2.40 | 6.45 | 2.85 | 2.99 | 80.73 | Shrub | 24.68 | 74.72 | 0.22 | 0.26 | 0.09 | 0.03 | Higher crop | 39.16 | 0.55 | 53.20 | 3.88 | 2.21 | 1.00 | Untilled glebe | 14.01 | 0.14 | 0.75 | 81.93 | 2.19 | 0.98 | Building | 23.47 | 0.16 | 0.20 | 0.38 | 73.23 | 2.57 | Forest | 9.58 | 0.09 | 0.06 | 0.26 | 3.58 | 86.43 |
| Pro. approach | Others | 86.90 | 0.39 | 1.63 | 6.96 | 2.21 | 1.90 | 82.54 | Shrub | 30.82 | 69.00 | 0.17 | 0 | 0 | 0 | Higher crop | 40.44 | 0.11 | 53.11 | 3.37 | 2.25 | 0.71 | Untilled glebe | 10.08 | 0.07 | 0.65 | 85.92 | 2.48 | 0.80 | Building | 16.63 | 0.03 | 0.10 | 0.47 | 80.68 | 2.09 | Forest | 4.97 | 0.08 | 0.03 | 0.29 | 8.04 | 86.58 |
|
|