Doppler Imaging and Deep Neural Networks

Tomasz Różański and Maja Jabłońska

The Doppler imaging (DI) is an effective inversion method for the reconstruction of stellar surface maps. The time series of high-resolution spectra used in this approach allow inferring the temperature and overabundance spots, the magnetic field, and the non-radial pulsations. Typically, the DI problem is solved with regularized chi-square minimization that produces solutions that fit observed variations in line profiles. Here we propose a different approach to this task based on the deep neural networks (DNN). We test the method on a synthetic dataset and show that DNN models can give results comparable to traditional techniques in a much shorter time, which is important due to the rapidly increasing number of available data.

Proceedings of the Polish Astronomical Society, vol. 12, 206-208 (2022)

Download full article as PDF file: