Abstract Blue horizontal-branch (BHB) stars are ideal tracers for studying the kinematics and structural properties of the Milky Way. With massive spectral data provided by the Large Sky Area Multi-Object Fiber Spectroscopic Telescope (LAMOST), we aim to identify more potential BHB stars using machine learning methods. In this study, we propose BHBNet, an advanced two-stage deep learning model integrating multiple techniques. By implementing Bayesian inference, it not only provides classification results but also quantifies uncertainty. In stage 1, a six-class classification model was constructed to initially identify BHB candidates, achieving a precision of 95.43% on the test set. In stage 2, a binary classification model constructed through the transfer learning method was employed to further refine the candidates, achieving a precision of 98.36% on the test set. Subsequently, by performing a two-stage search in LAMOST low-resolution survey DR10, we identified 6792 candidates. Nevertheless, since the completeness of this search result has not been assessed, these samples may not be adequate for statistical studies of the BHB population. We analyzed candidate properties including color, absolute magnitude, and spatial distribution, while estimating their atmospheric parameters. Eventually, by fitting Balmer line profiles, we identified 1605 new BHB stars compared to the previous studies by X.-X. Xue et al. and J. J. Vickers et al. Our study emphasizes the potential and effectiveness of using machine learning methods in identifying and analyzing BHB stars.
Zhang et al. (Fri,) studied this question.