Processing math: 100%

2504.02523

Total: 1

#1 Denoising medium resolution stellar spectra with U-Net convolutional neural networks [PDF] [Copy] [Kimi] [REL]

Authors: Balázs Pál, László Dobos

We investigated the use of a U-Net convolutional neural network for denoising simulated medium-resolution spectroscopic observations of stars. Simulated spectra were generated under realistic observational conditions resembling the Subaru Prime Focus Spectrograph (PFS). We found that our U-Net model effectively captured spectral features, achieving an average relative error of around 1% across a broad range of stellar parameters, despite a limited training set of only 1000 observations and a relatively short training period. Although U-Net did not reach the performance previously demonstrated by fully-connected denoising autoencoders (DAEs) consisting of dense layers and trained extensively on larger datasets, it outperformed dense networks trained under similarly constrained conditions. These results indicate that the U-Net architecture offers rapid, robust feature learning and may be particularly advantageous in scenarios involving initial denoising, subsequently refined by more accurate, but otherwise slower deep-learning models.

Subjects: Instrumentation and Methods for Astrophysics , Solar and Stellar Astrophysics

Publish: 2025-04-03 12:27:14 UTC