Table 3.
Classification performance for multi-domain adversarial neural networks (MDANNs) with EfficientNetV2 backbone.
| Domain | Metric | 1-shot MDANN | 3-shot MDANN | 5-shot MDANN |
|---|---|---|---|---|
| Source | Accuracy (%) | 87.8 | 92.2 | 90.0 |
| Precision (%) | 88.0 | 92.4 | 90.5 | |
| Recall (%) | 87.8 | 92.2 | 90.0 | |
| F1-score (%) | 87.8 | 92.1 | 89.8 | |
| Target (BF) | Accuracy (%) | 60.0 | 68.3 | 76.7 |
| Precision (%) | 57.4 | 67.7 | 77.3 | |
| Recall (%) | 60 | 68.3 | 76.7 | |
| F1-score (%) | 57.8 | 65.1 | 76.2 | |
| Target (20×) | Accuracy (%) | 54.4 | 82.2 | 75.6 |
| Precision (%) | 60.6 | 84.0 | 75.3 | |
| Recall (%) | 54.4 | 82.2 | 75.6 | |
| F1-score (%) | 48.4 | 82.1 | 74.8 |
The source domain was PC and the target domains were BF and 20×. Results are shown for domain-adversarial training with 1-shot, 3-shot, and 5-shot labeled samples. Bold values indicate the highest accuracy for each domain and setting.