Dortmund 2021 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz

AKPIK 2: AKPIK II: Deep Learning

AKPIK 2.8: Talk

Wednesday, March 17, 2021, 17:45–18:00, AKPIKa

A Neural Network Architecture for Radio Imaging — •Stefan Fröse and Kevin Schmidt — TU Dortmund, Dortmund, Deutschland

In radio astronomy, an array of correlated antennas, called a radio interferometer, is used to produce high resolution images of the sky. The measurements take place in the complex Fourier space due to the pairwise correlation of antennas. Therefore, the amount of information to receive from such an array is restricted by the number of antennas. The missing information in the uv-plane will be reconstructed using a Neural Network. The architecture of this network is similar to the architecture used for the task of superresolution. Superresolution is an approach for upscaling images from a low resolution to a high resolution. This method is transferred to the task of measuring a source in the Fourier space and filling in the missing information. The core of this architecture is a residual approach to the network. This can be written as y = F(x) + x with the set of measurements x, the set of complete information y and the mapping function F(x). The task of the network is to learn the underlying mapping function. The neural network is able to learn the correct mapping for simulated radio images and also shows convergence for more complex images with large- and small-scale structures.