Table 4 Impact of network architecture parameters—including convolutional layers, batch normalization, dropout rate, and hidden neurons—on the performance of the proposed CNN models.
From: Machine learning approach for wheat variety identification using single-seed imaging
CNN model | Convolutional layers | Batch norm | Dropout rate | Number of neurons | Accuracy | Validation curve |
|---|---|---|---|---|---|---|
GAP | 4 | Yes | 0.3 | 128 | – | Overfit |
GAP | 4 | Yes | 0.3 | 256 | – | Overfit |
GAP | 4 | Yes | 0.3 | 512 | – | Overfit |
GAP | 3 | Yes | 0.3 | 128 | – | Overfit |
FCL | 3 | Yes | 0.5 | 256 | – | Overfit |
FCL | 3 | Yes | 0.3 | 512 | – | Overfit |
FCL | 3 | No | 0.3 | 128 | – | Overfit |
FCL | 3 | No | 0.5 | 128 | – | Overfit |
FCL | 4 | No | 0.3 | 128 | – | Overfit |
GAP | 4 | No | 0.3 | 256 | 89.35 | Converged |
GAP | 3 | No | 0.3 | 64 | 88.94 | Converged |
GAP | 4 | No | 0 | 128 | 90.88 | Converged |
GAP | 4 | No | 0.3 | 128 | 91.15 | Converged |
GAP | 4 | No | 0.5 | 128 | 90.02 | Converged |
GAP | 3 | No | 0 | 128 | 88.11 | Converged |
GAP | 3 | No | 0.3 | 128 | 92.02 | Converged |
GAP | 3 | No | 0.5 | 128 | 92.19 | Converged |
GAP | 2 | No | 0.5 | 64 | 85.31 | Underfitting |
GAP | 2 | No | 0.5 | 128 | 86.09 | Underfitting |
