Unveiling Stellar Nature Through Oscillation Pattern Recognition

Asteroseismology is a pivotal observational method of stellar physics, allowing for the characterization of stars based on their oscillation properties. In recent years, supervised machine learning algorithms have aided in stellar characterization as a powerful tool to categorize evolution based on interior oscillations. In this study, we use an unsupervised machine learning algorithm, coupled with data clustering techniques, to classify red giant stars using 2D asteroseismic data represented as images. With our classification of ~6000 (6381) Kepler red giants, we automatically identified stars belonging to the red giant branch, the red clump (core Helium-burning), and for the first time with automatic methods, the secondary red clump (non-degenerate core Helium-burning) as well as stars with low-amplitude power in their oscillation spectra. This iteration of the method has high agreement with previous classifications, where the red giant branch and core Helium-burning stars show 98.879% and 85.038% agreement, respectively. Among the core Helium-burning stars, 18.584% are secondary red clump stars, and among the red giant branch, 7.506% of them are low amplitude stars. As we are in an age dominated by large amounts of data to be analyzed in stellar physics, our automatic classification method shows promise in being capable and reliable to analyse millions of stars observed by Kepler, TESS, and PLATO.