J/MNRAS/491/2280 Neural networks for spectral classification (Sharma+, 2020)
Application of convolutional neural networks for stellar spectral
classification.
Sharma K., Kembhavi A., Kembhavi A., Sivarani T., Abraham S., Vaghmare K.
<Mon. Not. R. Astron. Soc., 491, 2280-2300 (2020)>
=2020MNRAS.491.2280S 2020MNRAS.491.2280S (SIMBAD/NED BibCode)
ADC_Keywords: Models ; MK spectral classification ; Spectral types ;
Spectra, optical
Keywords: methods: data analysis - techniques: spectroscopic - catalogues -
stars: general
Abstract:
Due to the ever-expanding volume of observed spectroscopic data from
surveys such as SDSS and LAMOST, it has become important to apply
artificial intelligence (AI) techniques for analysing stellar spectra
to solve spectral classification and regression problems like the
determination of stellar atmospheric parameters Teff, logg, and
[Fe/H]. We propose an automated approach for the classification of
stellar spectra in the optical region using convolutional neural
networks (CNNs). Traditional machine learning (ML) methods with
'shallow' architecture (usually up to two hidden layers) have been
trained for these purposes in the past. However, deep learning methods
with a larger number of hidden layers allow the use of finer details
in the spectrum which results in improved accuracy and better
generalization. Studying finer spectral signatures also enables us to
determine accurate differential stellar parameters and find rare
objects. We examine various machine and deep learning algorithms like
artificial neural networks, Random Forest, and CNN to classify stellar
spectra using the Jacoby Atlas, ELODIE, and MILES spectral libraries
as training samples. We test the performance of the trained networks
on the Indo-U.S. Library of Coude Feed Stellar Spectra (CFLIB). We
show that using CNNs, we are able to lower the error up to 1.23
spectral subclasses as compared to that of two subclasses achieved in
the past studies with ML approach. We further apply the trained model
to classify stellar spectra retrieved from the SDSS data base with
SNR>20.
Description:
Developing any supervised classification model using machine/deep
learning techniques requires a standard data set containing a large
number of examples with known, reliable, and homogeneous
classification which is used for the training and testing purposes.
Since our primary goal is to classify large spectral data bases such
as SDSS/LAMOST, we select stellar libraries which have the maximum
overlap in the wavelength coverage with the target data bases. With
this primary constraint, we choose four spectral libraries,
Jacoby-Hunter-Christian (JHC) Atlas (Jacoby, Hunter & Christian
1984ApJS...56..257J 1984ApJS...56..257J, Cat. III/92), ELODIE (V3.1 Prugniel et al.
2007astro.ph..3658P 2007astro.ph..3658P), MILES (Sanchez-Blazquez et al.
2006MNRAS.371..703S 2006MNRAS.371..703S, Cat. J/MNRAS/371/703), and CFLIB (Valdes et al.
2004ApJS..152..251V 2004ApJS..152..251V, Cat. J/ApJS/152/251), for training and testing of
the ML/DL classifiers.
For CFLIB, sometimes also referred to as Indo-US library, 6917 spectra
of 1273 stars were obtained at resolution of ∼1Å full width at
half-maximum (FWHM) over the region 3460-9464Å. In some observing
runs, the unavailability of exact same grating settings caused gaps in
the wavelength coverage. For our analysis, we chose only those stars
for which the full wavelength coverage is available or the missing
region is small (<50Å) and/or lies beyond the optical region. This
gives us a sample of 850 stars (see Table 5).
File Summary:
--------------------------------------------------------------------------------
FileName Lrecl Records Explanations
--------------------------------------------------------------------------------
ReadMe 80 . This file
table5.dat 26 850 Expected and predicted class for CFLIB spectra
table8.dat 28 48084 Expected and predicted classes for the SDSS
spectra
--------------------------------------------------------------------------------
See also:
III/92 : A Library of Stellar Spectra (Jacoby+ 1984)
J/MNRAS/371/703 : MILES library of empirical spectra (Sanchez-Blazquez+, 2006)
J/ApJS/152/251 : Indo-US library of coude feed stellar spectra (Valdes+, 2004)
V/154 : Sloan Digital Sky Surveys (SDSS), Release 16 (DR16)
(Ahumada+, 2020)
Byte-by-byte Description of file: table5.dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 10 A10 --- Name Star name
12- 17 A6 --- ExpClass Expected class for CFLIB spectra
19- 24 A6 --- PreClass Predicted class for CFLIB spectra
26 A1 --- CFlag [0/1] Consistency between the two spectral
classifications (1)
--------------------------------------------------------------------------------
Note (1): Flag as follows:
0 = Good agreement between the expected and predicted classes
(within 3σ) (835/850)
1 = The classifications are inconsistent by more than five subclasses
(15/850)
--------------------------------------------------------------------------------
Byte-by-byte Description of file: table8.dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 4 I4 --- Plate SDSS Plate identifier
6- 10 I5 --- MJD SDSS MJD identifier
12- 15 I4 --- Fiber SDSS Fiber identifier
17- 22 A6 --- SDSSClass SDSS class for the spectra
24- 28 A5 --- PreClass Predicted class for the SDSS spectra
--------------------------------------------------------------------------------
History:
From electronic version of the journal
(End) Ana Fiallos [CDS] 15-Feb-2023