Speaker
Description
(SKA-Low) will map the distribution of neutral hydrogen during reionisation and produce a tremendous amount of 3D tomographic data. These image cubes will be subject to instrumental limitations, such as systematic noise, foreground contamination, radio frequency interference (RFI) and limited resolution. The challenge of this astronomy image is the undesired astronomical and instrumental noise that outshines the 21-cm signal. Therefore, when studying the properties of the 21-cm signal for EoR, considering foregrounds and instrumental imprint on the data is of great importance, as they are orders of magnitude larger than the actual signal and further increase the data cleaning complexity.
Here we present SegU-Net, a stable and reliable method for identifying neutral and ionised regions in these images. SegU-Net is a U-Net architecture-based convolutional neural network (CNN) for image segmentation. It can segment our image data into meaningful features (ionised and neutral regions) with greater accuracy than previous methods. We can estimate the true ionisation history from our mock observation of SKA with an observation time of 1000 h with more than 87 per cent accuracy. Our network can be used to recover various topological summary statistics that characterise the non-Gaussian nature of the reionisation process.
Moreover, we also show an extended version of SegU-Net that can recover the 21-cm signal from the foreground contaminated tomographic dataset. The updated version of our network employs the segmented maps as position and shape prior to a guided recovery of the simulated 21-cm. We finally derive summary statistics from evaluating the applicability of the expected data from SKA.
