Open Access
Table 6.
Description of the global contaminant neural network architecture, including map dimensions.
| Layer | Size |
|---|---|
| Input | 400 × 400 × 1 |
| Conv | 400 × 400 × 16 |
| Maxpool | 200 × 200 × 16 |
| Conv | 200 × 200 × 32 |
| Maxpool | 100 × 100 × 32 |
| Conv | 100 × 100 × 64 |
| Maxpool | 50 × 50 × 64 |
| Conv | 50 × 50 × 128 |
| Maxpool | 25 × 25 × 128 |
| Conv | 25 × 25 × 128 |
| Maxpool | 13 × 13 × 128 |
| Flatten | 21 632 |
| Fully connected | 64 |
| Fully connected | 64 |
| Fully connected | 2 |
Notes. All convolution kernels are 9 × 9 and max-pooling kernels are 2 × 2. All activation functions (not shown for brevity) are ReLU, except in the output layer where predictions are done using softmax.
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.