The Edge-on Galaxy candidates (data release 2) in Pan-STARRS1 dr2 found by ANN
Examples
Cone search:
- Leda style: J000055.99+202017.1
- Alternative form: 00:00:55.99 +20:20:17.1
- In degrees: 182.767746 +50.484854
- within 1.5 degreeis radius 12:11:04.33 +50:29:06.1 r=1.5
In the second release of the edge-on galaxies in the Pan-STARRS1 survey, the selection and classification of objects have been improved.
We used the same ideology as in the first version of the EGIPS catalog.
The artificial neural network was trained using a sample of the "best" edge-on galaxies from the EGIPS catalog.
The new version is expanded to the low surface brightness galaxies.
Team
- Alexandra Antipova (SAO RAS)
- Svyatoslav Borisov (University of Geneva, SAI MSU)
- Olga Kashibadze (SAO RAS)
- Danila Makarov (SAO RAS)
- Dmitry Makarov (SAO RAS)
- Lidia Makarova (SAO RAS)
- Alexander Marchuk (SPbU)
- Alexander Mosenkov (BYU, Pulkovo)
- Vladimir Reshetnikov (SPbU)
- Sergey Savchenko (SPbU)
- Ekaterina Shkodkina (SPbU)
- Iliya Tikhonenko (SPbU)
- Pavel Usachev (SPbU)
Quality of SExtractor-photometry
The quality of SExtractor-photometry was analyzed. Distribution of candidates by coordinate shift, sizes, positional angles, and ellipticities in all 5 filters has been analyzed. A detailed analysis of the ellipticity distribution revealed outliers in the region of extreme values in all filters. These values were associated with defects that were not excluded from the catalog on previous steps. It is shown that the sample of galaxies is weakly affected by the selection effects of the algorithm for galaxies larger than a > 5.Artificial tests
The completeness map shows a probability of detection of edge-on galaxies by our improved ANN algorithm depending on central surface brightness and exponential radial scale.
Gallery
Edge-on analogue of Malin 2Interesting interaction
Tadpoles
Statistics
| Selection | Objects | Unique |
|---|---|---|
| all | 284948 | 251971 |
| model1 | 279676 | 247141 |
| model2 | 150480 | 132528 |
| model2 & model1 | 145208 | 127649 |
| model2 & not model1 | 5272 | 5075 |
| trash | 4531 | 4348 |
ANN statistics
| ⟨ vote ⟩ | Objects |
|---|---|
| 0.95–1 | 41129 |
| 0.9–0.95 | 13373 |
| 0.8–0.9 | 20825 |
| 0.7–0.8 | 18045 |
| 0.6–0.7 | 17726 |
| 0.5–0.6 | 16568 |
| 0.4–0.5 | 4787 |
| 0.3–0.4 | 75 |
| 0.2–0.3 | 0 |
| 0.1–0.2 | 0 |
| 0.0–0.1 | 0 |
Visual inspection of candidates by the team members
We carried out the inspection of erroneous objects selected by the neural network. The sample of 25,775 galaxies included galaxies that were not examined by students and those that were marked as erroneous. The inspection statistics are available at the link.Visual inspection of candidates
A visual inspection of candidates was carried out in order to detect problematic cases of erroneous classification of objects by the neural network. As a rule, these are various kinds of defects in images, glare and rays from bright stars, configurations of stars and nearby galaxies. Statistics on visual inspections are available at this link. 26 students took part in this verification. It was assumed that all 150 thousand candidates would be examined twice. Unfortunately, this goal was achieved only for 37 thousand objects, and about 18 thousand were never inspected.New CNN model for selection of edge-on galaxies
Eleven models were trained on a "gold" and "silver" sample of the edge-on galaxies derived from galaxy visualization statistics. The galaxy selection algorithm has been improved:- the search is performed on the combined image obtained from three filters g, r and i;
- for further analysis, objects with a size >48 pix and 15 connected pixels exceeding the background level by 3.5 sigma are selected.
Very approximate statistics:
| Participant | # of fields |
|---|---|
| Savchenko | 44200 |
| Usachev | 35000 |
| Mosenkov | 20000 |
| Tikhonenko | 19800 |
| Reshetnikov | 19500 |
| Makarov Dmitry | 17500 |
| Makarova | 9700 |
| Marchuk | 9200 |
| Kashibadze | 3000 |
| Antipova | 1900 |
| Borisov | 1800 |
| Makarov Danila | 1000 |