Estimation of Stellar Atmospheric Parameters from LAMOST DR8 Low-resolution Spectra with 20 ≤ S/N < 30

The accuracy of the estimated stellar atmospheric parameter decreases evidently with decreasing spectral signal-to-noise ratio (SNR) and there are a huge amount this kind observations, especially in case SNR$<$30. Therefore, it is helpful to improve estimation performance for these spectra work studied ($T_\texttt{eff}, \log~g$, [Fe/H]) problem LAMOST DR8 low-resolution 20$\leq$SNR$<$30. We proposed data-driven method based on machine learning techniques. Firstly, scheme detected parameter-sensitive features from by Least Absolute Shrinkage Selection Operator (LASSO), rejected ineffective data components irrelevant data. Secondly, Multi-layer Perceptron (MLP) was used estimate parameters LASSO features. Finally, LASSO-MLP evaluated computing analyzing consistency between its reference APOGEE (Apache Point Observatory Galactic Evolution Experiment) high-resolution spectra. Experiments show that Mean Errors (MAE) $T_\texttt{eff}, [Fe/H] reduced LASP (137.6 K, 0.195 dex, 0.091 dex) (84.32 0.137 0.063 dex), which indicate evident improvements estimation. In addition, 1,162,760 20$\leq$SNR$<$30 using LASSO-MLP, released catalog, learned model, experimental code, trained training test scientific exploration algorithm study.