Speaker
Description
Interferometric imaging plays a critical role in the measurement and interpretation of galaxy clusters in the radio wave band, where diffuse emission traces large-scale physical processes such as feedback from active galactic nuclei. However, the fidelity of reconstructed astronomical images depends on the imaging and deconvolution algorithms used to recover the sky brightness from incomplete Fourier sampling. Traditional approaches, such as CLEAN and its variants, often struggle to accurately reconstruct extended structures. More recent methods based on compressed sensing or machine learning can significantly improve image quality, but they are often computationally inefficient for the high-throughput data flows expected from next-generation radio telescopes. Canada committed to becoming a full member of the Square Kilometre Array (SKA) Observatory in 2023. Without algorithms capable of handling SKA-scale data, astronomers may not be able to achieve the expected scientific return from such next-generation telescope in a timely manner. Therefore, we develop a novel Transformer-based imaging algorithm to improve the robustness and fidelity of interferometric reconstructions of extended sources on large scales. The method is trained and validated using both simulated and real datasets, allowing the model to learn structural features of extended celestial sources. Applied to observations of the Perseus galaxy cluster, we benchmark its performance against widely used deconvolution algorithms using field-specific quality assessors. The algorithm provides a promising pathway toward high-fidelity, large-scale imaging with upcoming radio telescopes, enabling more accurate studies of galaxy clusters and other extended astrophysical systems.
