Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz

AKPIK 3: Invited Talks

AKPIK 3.2: Hauptvortrag

Dienstag, 17. März 2026, 11:30–12:00, KS H C

Deep Learning-Based Imaging of MeerKAT Observations — •Kevin Schmitz — TU Dortmund University, Dortumd, Germany — Lamarr Institute for ML & AI, Dortmund, Germany

Modern radio interferometers such as MeerKAT produce vast amounts of data that enable high-resolution imaging of astrophysical sources but also pose significant challenges for image reconstruction. Classical deconvolution algorithms like CLEAN struggle with noise artifacts and limited scalability. The deep-learning framework radionets addresses these issues by combining simulated observations with neural-network-based image reconstruction.
Using the Radio Interferometer Measurement Equation (RIME), realistic simulations model signal propagation and systematic effects in the measured data. These synthetic datasets allow convolutional neural networks to be trained under controlled conditions, enabling quantitative studies of image quality, uncertainty estimates, and computational efficiency.
The framework reconstructs sparse data directly in Fourier space before transforming them into clean sky images, achieving higher positional accuracy and flux recovery than standard tools, especially for wide-field observations containing both compact sources and diffuse emission. I will present its application to real MeerKAT observations, compare the results with classical methods, and illustrate how physics-guided machine learning improves data quality and interpretation in radio astronomy.

Keywords: radio astronomy; deep learning; simulations; data analysis