Neural network for stellar spectrum normalization

We present a deep fully convolutional neural network trained
in the task of stellar spectrum normalization. We show that the proposed
model is able to fit spectral continuum, including non-smooth
instrumental pseudo-continuum, wide hydrogen and narrow
blended spectral lines. Proposed solution gives an opportunity to
automate this step of stellar spectrum pre-processing and achieve the
accuracy similar to that of careful manual normalization. This approach
may greatly simplify automated high-resolution spectra analysis.


Theme – Machine Learning, Statistics, and Algorithms, Data Processing Pipelines and Science-Ready Data