Searching for young stellar objects through SEDs by machine learning

Accurate measurements of statistical properties, such as the star formation rate and lifetime young stellar objects (YSOs) in different stages, are essential for constraining theories. However, it is a difficult task to separate galaxies YSOs based on spectral energy distributions (SEDs) alone, because they contain both thermal emission from stars dust around them no reliable theories can be applied distinguish them. Here we compare machine learning algorithms develop Spectrum Classifier Astronomical Objects (SCAO), Fully Connected Neural Network (FCN), classify regular stars, galaxies, YSOs. Superior previous classifiers, SCAO solely trained by high quality data labeled Molecular Cores Planet-forming Disks (c2d) catalog without priori theoretical knowledge, provides excellent results with precision (>96%) recall (>98%) when only eight bands included. We systematically investigate effects observation errors distance effects, show that accuracy performance still maintained even using fluxes three (IRAC 3, a=IRAC 4, MIPS 1) long wavelengths regime, silicate absorption feature automatically detected SCAO. Finally, Spitzer Enhanced Imaging Products (SEIP), most complete observations, found 129219 YSO candidates. The website available at http://scao.astr.nthu.edu.tw.