Deep learning determination of stellar atmospheric fundamental parameters
Résumé
In order to estimate fundamental parameters (effective temperature, surface gravity and metallicity) of the large amount of stars in the PolarBase data base, we need a fast and reliable algorithm. With this aim, we developed a convolutional neural network able to derive this parameter triplet. Our neural network was trained on observed spectra from the PolarBase and Elodie data bases (M to F stars). We used the spectral region between 6095 and 6185 Angströms which has proved its efficiency in a number of previous studies. We analyzed the outcome of our approach for a sample of spectra from the same data bases. We discuss the accuracy and reliability of the neural network depending on the parameter domain, size and quality of the training data set.