| |
 |
| Paper: |
Deep Generative Models for Simulating Radio Datasets |
| Volume: |
538, ADASS XXXII |
| Page: |
194 |
| Authors: |
Renato Sortino; Eva Sciacca; Andrea De Marco; Giuseppe Fiameni; Simone Riggi; Concetto Spampinato |
| DOI: |
10.26624/JCBW1244 |
| Abstract: |
In the context of the Square Kilometre Array (SKA) preparatory scientific
phases and SKA precursor radio astronomy image processing, several works analyze
multiple solutions to make the best use of the data that will be produced by the up-coming SKA telescope. Evaluating the quality and performances of these solutions
requires a lot of annotated data, usually obtained by applying Gaussian models to simulate background noise and source fluxes. In this work, we propose the application of
deep learning-based generative models to generate images of radio sources of different morphologies starting from the relative segmentation map to generate both the pair
image-annotation, for realistic data simulation purposes. The approaches based on deep
learning rely on pattern recognition of real data and are capable of offering a higher degree of variability of the synthetically generated data, for Gaussian-based approaches.
Additionally, these models can be exploited to share radio data for prototyping novel
SKA detection and classification tools without compromising any privacy policy linked
to a specific observing program avoiding sharing the real data but the synthetic ones. |
|
|
|
|
|
