Table B.1.
Classification (TP: true positive, TN: true negative, FP: false positive, FN: false negative) of the test set of 50 images for several models.
| Classification |
||||||
|---|---|---|---|---|---|---|
| Network Type | Model | TP | TN | FP | FN | Loss |
| Capsule | 2 | 23 | 20 | 5 | 2 | 0.041 |
| 3 | 25 | 0 | 25 | 0 | 0.202 | |
| 4 | 23 | 21 | 4 | 2 | 0.046 | |
| 5 | 23 | 18 | 7 | 2 | 0.056 | |
| ALED-m | 23 | 23 | 2 | 2 | 0.039 | |
| 6 | 23 | 23 | 2 | 2 | 0.041 | |
| 7 | 23 | 20 | 5 | 2 | 0.047 | |
| 8 | 24 | 21 | 4 | 1 | 0.037 | |
| 9 | 23 | 22 | 3 | 2 | 0.213 | |
| 10 | 23 | 19 | 6 | 2 | 0.045 | |
| Convolutional | 12 | 21 | 22 | 3 | 4 | 0.371 |
| 13 | 21 | 22 | 3 | 4 | 0.459 | |
| 14 | 17 | 16 | 9 | 8 | 0.644 | |
| 15 | 12 | 18 | 7 | 13 | 0.721 | |
Notes. For the capsule networks, an image was classified as containing a light echo if the length of the final capsule was greater than 0.5. For the CNNs, an image was classified as containing a light echo if the neuron corresponding to the light echo class was more active than the non-light-echo class neuron. The model loss is given by the margin loss for capsule networks, and cross-entropy loss for CNNs.
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.