|
High resolution spectra of stars provide a wealth of information regarding the composition and properties of their atmospheres. However, the analysis of such spectra often requires normalization, which involves dividing the observed flux element-wise by a modeled pseudo-continuum. The modeling of this pseudo-continuum is a non-trivial task, as the shape is influenced by various factors such as the spectrum of the observed star, interstellar absorption, absorption in the Earth’s atmosphere, the response function of the spectrograph, and the reduction pipeline. These factors can introduce numerous low- and high-frequency distortions that are difficult to model automatically, resulting in time-consuming manual normalization. One limitation of automatic methods is the assumption that the local flux maximum is a good approximation of the pseudo-continuum, which is often not true in the case of spectra with emission features or wavelength ranges where spectral lines heavily blend. To address these challenges, we propose a method for stellar spectrum normalization based on a deep convolutional neural network called SUPPNet. One of the key advantages of this method is its ability to effectively handle spectra with both high and low signal-to-noise ratios, various rotational velocities, and spectrograph’s resolutions. The algorithm is open source and available online: https://github.com/RozanskiT/suppnet. |
|