http://arxiv.org/abs/2304.02661
The upcoming Square Kilometre Array Observatory (SKAO) will produce images of neutral hydrogen distribution during the epoch of reionization by observing the corresponding 21-cm signal. However, the 21-cm signal will be subject to instrumental limitations such as noise, foreground contamination, and limited resolution, which pose a challenge for accurate detection. In this study, we present the \texttt{SegU-Net v2} framework, which is an enhanced version of our U-Net architecture-based convolutional neural network built for segmenting image data into meaningful features. This framework is designed to identify neutral and ionized regions in the 21-cm signal contaminated with foreground emission that is $\sim$3 order of magnitude larger. We demonstrate the effectiveness of our method by estimating the true ionization history from mock observations of SKA with an observation time of 1000 h, achieving an average classification accuracy of 71 per cent. As the photon sources driving reionization are expected to be located inside the ionised regions identified by \texttt{SegU-Net v2}, this tool can be used to identify locations for follow-up studies with infrared/optical telescopes to detect these sources. Additionally, we derive summary statistics, such as the size distribution of neutral islands, from evaluating the reliability of our method on the tomographic data expected from the SKA-Low. Our study suggests that \texttt{SegU-Net v2} can be a stable and reliable tool for analyzing the 3D tomographic data produced by the SKA and recovering important information about the non-Gaussian nature of the reionization process.
M. Bianco, S. Giri, D. Prelogović, et. al.
Fri, 7 Apr 23
49/50
Comments: 16 pages, 9 figures, 2 tables. Comments welcome
You must be logged in to post a comment.