Research Article
Application Research for Fusion Model of Pseudolabel and Cross Network
Table 1
Experimental results of PL generation part.
| Hyperparameters | Model | VGG16 [11] | ResNet50 [12] (%) | Transfer network | VGG16 (%) | ResNet50 (%) |
| Resolution | 32 × 32 | 53.06 | 48.34 | 53.56 | 53.16 | 64 × 64 | 53.36 | 53.36 | 53.51 | 54.57 | 72 × 72 | 54.02 | 53.71 | 53.71 | 52.46 | 96 × 96 | 53.16 | 54.27 | 55.12 | 54.72 | 100 × 100 | 58.84 | 55.47 | 54.82 | 54.92 | 128 × 128 | 60.43 | 58.33 | 60.34 | 58.84 |
| Learning rate | 0.001 | 53.51 | 53.97 | 53.16 | 52.06 | 0.003 | 60.43 | 58.33 | 60.34 | 58.84 | 0.01 | 53.12 | 54.02 | 54.07 | 53.66 | 0.03 | 56.12 | 52.51 | 54.47 | 53.82 | 0.1 | 44.28 | 54.07 | 53.56 | 53.31 | 0.3 | 44.18 | 44.28 | 44.78 | 44.43 |
|
|
Better results in terms of recognition rates were highlighted in bold.
|