Table 2 Deep feature extraction and fusion summary.
Backbone model | Output layer | Feature dimension | Post-processing | Normalization | Final use |
|---|---|---|---|---|---|
DenseNet-121 | GAP | 1024 | Dense connectivity, pooled vector | StandardScaler + L2 | Fusion input |
ResNet-34 | GAP | 512 | Residual mapping, pooled vector | StandardScaler + L2 | Fusion input |
Fused Vector | Concatenation | 1536 | Optional PCA → ~480 comps | L2 norm | Classifier input |
