| |
 |
| Paper: |
Decomposition of stellar populations in simulated disk galaxies using the TensorFlow/Keras framework |
| Volume: |
535, Astronomical Data Analysis Software and Systems XXXI |
| Page: |
335 |
| Authors: |
Kunsagi-Mate, S.; Csabai, I. |
| Abstract: |
In our work we have developed a source separation method using the Tensor-flow/Keras framework. The method is designed to decompose simulated disk galaxies into old, red, and young, blue, stellar population using broadband images. In our study we estimate the broadband spectral energy distribution (SED) and morphology in two distinct steps. We developed a neural network (sedNN) in Keras that can predict the SEDs directly from the color distribution of galaxies, and we used these SEDs as input in the morphology determination of the two sources in a TensorFlow model. We trained and tested the network on simulated galaxies where we found that sedNN can predict the SEDs with 2-3% accuracy, which is about 2-4 times better than using the widely used Scarlet deblending algorithm. |
|
|
|
|
|
