Deep-learning classification of eclipsing binaries

0585883 - ASÚ 2025 RIV SK eng J - Journal Article
Parimucha, Š. - Gajdoš, Pavol - Markus, Y. - Kudak, V.
Deep-learning classification of eclipsing binaries.
Contributions of the Astronomical Observatory Skalnaté Pleso. Roč. 54, č. 2 (2024), s. 167-170. ISSN 1335-1842. E-ISSN 1336-0337
Institutional support: RVO:67985815
Keywords : binaries stars * eclipsing * deep-learning
OECD category: Astronomy (including astrophysics,space science)
Impact factor: 0.5, year: 2022
Method of publishing: Open access

We present a deep-learning model for the classification of eclipsing binaries. Our classifier provides a tool for the categorization of light curves of eclipsing binaries into four classes: detached systems with and without spots, and over-contact systems with and without spots. The classifier was trained on 200 000 synthetic light curves created using ELISa code. We randomly selected 100 light curves from the GAIA catalogue, which were fitted for evaluation purposes, and their morphologies were determined. We tested several classifiers and found that the best-performing classifier combined a Long Short-Term Memory (LSTM) layer and two one-dimensional convolutional neural networks. The precision from the evaluation set was 97% compared with the predicted precision of 94% for the validation of synthetic data. Our classifier is more likely to successfully process data from subsequent large observational surveys.

Permanent Link: https://hdl.handle.net/11104/0353524