Yi-Lung Chiu (Institute of Astronomy, NTHU); Chi-Ting Ho (Physics Department, NTHU); Daw-Wei Wang (Physics Department, NTHU);Shih-Ping Lai (Institute of Astronomy, NTHU)
Accurate measurements of statistical properties, such as the star formation rate and the lifetime of young stellar objects (YSOs) in different stages, is essential for constraining the star formation theories. However, it is a difficult task to separate galaxies and YSOs based on spectral energy distributions(SEDs) alone, because they contain both thermal emission from stellar and dust around and no reliable theories can be applied to distinguish them. Here we develop a machine learning algorithm, named Spectrum Classifier of Astromonical Objects (SCAO), by using Convolutional Neural Network (CNN) to classify regular stars, galaxies, and YSOs, solely based on their SEDs. Superior to previous classifiers, our SCAO is solely trained by labeled data without a priori theoretical knowledge, and provides excellent results with high precision (>90%) and recall (>98%) for YSOs when data of only eight bands are included. Such high accurate performance is still maintained even using normalized data so that the distance effects are removed. Finally, we show that a good classification results can be made even from SED of three bands only in the long wavelengths regime, indicating the possible direction for future YSO research.