Spectral Classification of Unresolved Binary Stars with Artificial Neural Networks

Citations Over TimeTop 10% of 2000 papers

Abstract

An artificial neural network technique has been developed to perform two-dimensional spectral classification of the components of binary stars. The spectra are based on the 15 Å resolution near-infrared (NIR) spectral classification system described by Torres-Dodgen & Weaver. Using the spectrum with no manual intervention except wavelength registration, a single artificial neural network (ANN) can classify these spectra with Morgan-Keenan types with an average accuracy of about 2.5 types (subclasses) in temperature and about 0.45 classes in luminosity for up to 3 mag of difference in luminosity. The error in temperature classification does not increase substantially until the secondary contributes less than 10% of the light of the system. By following the coarse-classification ANN with a specialist ANN, the mean absolute errors are reduced to about 0.5 types in temperature and 0.33 classes in luminosity. The resulting ANN network was applied to seven binary stars.

Related Papers

• → Statistical relations between stellar spectral and luminosity classes and stellar effective temperature and surface gravity(2020)14 cited
• → Luminosity classification of stars of late spectral type from spectra of moderate dispersion(1956)1 cited
• → Prediction of Spectral and Luminosity Classes from Spectral Indices with Artificial Neural Networks(2000)1 cited
• → The precision of spectroscopic luminosity in A-F type stars and the classification of Am stars(1997)
• → Color-color diagrams in near infrared: (J-H)/(H-K). II.(2018)