Using deep learning to characterize single-exposure double-line spectroscopic binaries

¹ Porter School of the Environment and Earth Sciences, Raymond and Beverly Sackler Faculty of Exact Sciences, Tel Aviv University, Tel Aviv 6997801, Israel
² School of Electrical and Computer Engineering, Iby and Aladar Fleischman Faculty of Engineering, Tel Aviv University, Tel Aviv 6997801, Israel
³ School of Computer Science and AI, Tel Aviv University, Tel Aviv 6997801, Israel
⁴ School of Physics and Astronomy, Raymond and Beverly Sackler Faculty of Exact Sciences, Tel Aviv University, Tel Aviv 6997801, Israel

^★ Corresponding authors: avrahambinn@gmail.com; shayz@tauex.tau.ac.il

Received: 10 February 2025
Accepted: 12 June 2025

Abstract

Distinguishing the component spectra of double-line spectroscopic binaries (SB2s) and extracting their stellar parameters is a complex and computationally intensive task that usually requires observations spanning several epochs that represent various orbital phases. This poses an especially significant challenge for large surveys such as Gaia or LAMOST, where the number of available spectra per target is often not enough for a proper spectral disentangling. We present a new approach for characterizing SB2 components from single-exposure spectroscopic observations. The proposed tool uses deep neural networks to extract the stellar parameters of the individual component spectra that comprise the single exposure, without explicitly disentangling them or extracting their radial velocities. The neural networks were trained, tested, and validated using simulated data resembling Gaia RVS spectra, which will be made available to the community in the coming Gaia data releases. We expect our tool to be useful in their analysis.