Identification and classification of red giant stars with unsupervised deep learning method
Observation of mixed modes of oscillations in evolved stars has proven to very efficient to classify. Indeed, differences in internal structure that characterizes the evolutionary stage or stellar nature will affect the frequencies and amplitudes of the mixed modes. Moreover, precision of the observations of recent satellites such as CoRoT, Kepler and TESS, or even future ones, such as the ESA project PLATO, allow us to probe the interior of hundreds of thousands of stars, and millions more are expected in the coming years. Faced with so much data it is difficult to start a classification.
Unveilling stellar nature through oscillation pattern recognition
Le Saux A., Bugnet L., Smith K. et al. (in prep)
In this project we present an innovative deep-learning-based methodology used to automatically classify evolved solar-like stars according to their asteroseismic signal. It is based on an unsupervised intelligent method as we do not wish reduce the classification to previously known categories such as the evolutionary stage or stellar nature, but rather explore the diversity of solar-like oscillations. Building the echelle diagrams of stars allow to study solar-like oscillations in two dimensional images. As a result, we can use 2D deep learning classification methods based on image patterns recognition such as Convolutional Neural Networks on this representation of oscillations. Our classification allows to identify stars with low amplitude dipolar modes but also stars from the Red Giant Branch, the Red Clump and the Second Clump.