Open Access
Table 3
Step-by-step model architecture in the parameter-prediction model (using the spectrum and temperatures as input).
| Step | Layer | Input shape | Output shape | Details |
|---|---|---|---|---|
| 1 | Input (spectrum) | (1, 903) | (1, 903) | Raw spectrum |
| 2 | Input (temperatures) | (2) | (2) | Two predicted temperatures (lower first) |
| 3 | Conv1D | (1, 903) | (32, 450) | Kernel 7, Stride 2, Padding 3 |
| 4 | MaxPool | (32, 450) | (32, 225) | Kernel 3, Stride 2, Padding 1 |
| 5 | Dense block 1 | (32, 225) | (128, 225) | 2 Conv layers, Kernel 3, BatchNorm, ReLU |
| 6 | Transition block 1 | (128, 225) | (64, 112) | Conv 1 × 1, Pool 2 × 2 |
| 7 | Dense block 2 | (64, 112) | (256, 112) | 4 Conv layers, Kernel 3, BatchNorm, ReLU |
| 8 | Transition Block 2 | (256, 112) | (128, 56) | Conv 1 × 1, Pool 2 × 2 |
| 9 | Dense block 3 | (128, 56) | (512, 56) | 6 Conv layers, Kernel 3, BatchNorm, ReLU |
| 10 | Transition block 3 | (512, 56) | (256, 28) | Conv 1 × 1, Pool 2 × 2 |
| 11 | Dense block 4 | (256, 28) | (512, 28) | 4 Conv layers, Kernel 3, BatchNorm, ReLU |
| 12 | Adaptive pooling | (512, 28) | (512, 1) | Adaptive avg. pooling |
| 13 | Flatten | (512, 1) | (512) | Reshaped for fully connected layers |
| 14 | Fully Connected (temperatures) | (2) | (64) | Linear layer, ReLU |
| 15 | Fully Connected (merged) | (512 + 64) | (512) | Combined spectral and temperature features |
| 16 | Fully connected 1 | (512) | (512) | Linear layer, ReLU |
| 17 | Fully connected 2 | (512) | (2) | Parameters matching sorted temperatures |
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.