6–10 Jul 2026
University of the Western Cape
Africa/Johannesburg timezone
**Tours now open!** Registration is now closed - All registration payments are due before 23:39 SAST on 26 June.

Identification and cleaning of contaminants for neutral hydrogen intensity mapping using machine learning

8 Jul 2026, 11:00
20m
Lecture Hall C5 (University of the Western Cape)

Lecture Hall C5

University of the Western Cape

Oral Presentation Track D - Astrophysics & Space Science Astrophysics & Space Science

Speaker

Mosima Masipa (University of the Western Cape)

Description

Radio frequency Interference (RFI) can contaminate data collected by radio telescopes, making it difficult to distinguish between the target cosmological signal and artificial signals caused by RFI sources. Manually flagging RFI is time-consuming,so we turn to machine learning algorithms as a possible solution to detect/flag these RFI signals. We implemented a UNet, which is a Convolutional Neural Network(CNN) model to automate RFI flagging/detection in the MeerKLASS 2021 L-band survey data. A UNet model was chosen due to its suitability for semantic segmentation tasks. We leverage advanced machine learning techniques to automate RFI flagging in radio astronomy. Machine learning algorithms, more specifically, deep learning algorithms have improved the identification of Radio Frequency Interference. Recent and ongoing studies have shown that these algorithms, for example,U-Nets and ResNet can outperform non-machine learning methods regarding RFI flagging. With upcoming surveys such as the Square Kilometre Array (SKA), there is a need to develop new and automated methods that can detect and flag RFI. Since MeerKAT is a precursor instrument to the SKA-mid, it is a perfect instrument of choice to test these machine learning techniques as they can be easily adapted for SKA later. These algorithms can be applied to both simulated and real data, in both single-dish and interferometric data. This work involves using machine learning methods that are already established to do RFI flagging, and also exploring new unsupervised/self-supervised machine learning techniques for RFI flagging. Previous studies have mainly focused on simulated data and we use observational data that has been labeled using different RFI techniques to train machine learning models to detect and flag RFI.

Apply for student award at which level: PhD
Consent on use of personal information: Abstract Submission Yes, I ACCEPT

Author

Mosima Masipa (University of the Western Cape)

Presentation materials