Speaker
Description
Solar radio bursts are crucial for understanding electron beam dynamics in the solar corona. Automated detection and classification of these events can aid both scientific understanding and operational space weather forecasting, where early identification of radio burst activity can provide insight into flares and coronal mass ejections. Utilising pre-trained models from PyTorch and Ultralytics, a segmentation approach for analysing solar radio bursts was created. These deep learning models were then fine-tuned using data from the Irish-LOFAR (I-LOFAR) radio telescope and the Incremental Development of LOFAR Space-weather (IDOLS) project. Unlike traditional classification or the coarse localisation offered by object detection, these methods can simultaneously detect and localise solar radio bursts within radio spectrograms. Furthermore, they specifically enable the extraction of physically meaningful parameters, which facilitate detailed characterisation of burst evolution and provide stronger constraints for statistical studies. In this talk, we walk through the design decisions behind our approach, the care required to construct a reliable training set, the method itself, and some preliminary results. Finally we will discuss current limitations and next steps, including how the work could be extended to real-time support and possibly other instruments.