Fine-tuned ResNet50 and vision transformers on polar-hexbin images classify eclipsing binaries as detached or overcontact with high accuracy on real data, but cannot reliably detect starspots.
Statistics of 700 individually studied W UMa stars
1 Pith paper cite this work. Polarity classification is still indexing.
abstract
We present a statistical study of the largest bibliographic compilation of stellar and orbital parameters of W UMa stars derived by light curve synthesis with Roche models. The compilation includes nearly 700 individually investigated objects from over 450 distinct publications. Almost 70% of this sample is comprised of stars observed in the last decade that have not been considered in previous statistical studies. We estimate the ages of the cataloged stars, model the distributions of their periods, mass ratios, temperatures and other quantities, and compare them with the data from CRTS, LAMOST and Gaia archives. As only a small fraction of the sample has radial velocity curves, we examine the reliability of the photometric mass ratios in totally and partially eclipsing systems and find that totally eclipsing W UMa stars with photometric mass ratios have the same parameter distributions as those with spectroscopic mass ratios. Most of the stars with reliable parameters have mass ratios below 0.5 and orbital periods shorter than 0.5 days. Stars with longer periods and temperatures above 7000 K stand out as outliers and shouldn't be labeled as W UMa binaries. The collected data is available as an online database at https://wumacat.aob.rs.
fields
cs.CV 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Morphological classification of eclipsing binary stars using computer vision methods
Fine-tuned ResNet50 and vision transformers on polar-hexbin images classify eclipsing binaries as detached or overcontact with high accuracy on real data, but cannot reliably detect starspots.