Speaker
Description
The advent of next-generation radio surveys, such as those conducted with LOFAR, ASKAP, and the upcoming SKA, is expected to revolutionise our understanding of the radio universe. However, the vast volume and complexity of data generated by these instruments present significant challenges for traditional manual cataloguing and data analysis techniques. To fully exploit these datasets, the integration of artificial intelligence and the use of high-performance computing infrastructures is essential.
While machine learning has already been applied to the detection of compact objects and radio galaxies, detecting diffuse radio sources, such as radio halos and relics in galaxy clusters, remains a critical challenge. Conventional approaches often involve computationally expensive pre-processing steps, such as the subtraction of compact sources prior to detecting diffuse emission, an increasingly impractical task as dataset sizes continue to grow. To address these issues, we employ Radio-UNet, a fully convolutional neural network based on the U-Net architecture, designed to enhance the automated identification of diffuse radio sources in galaxy clusters. Trained on synthetic radio observations derived from cosmological simulations, our model was validated on diffuse emission in galaxy clusters observed by the LOFAR Two-Metre Sky Survey (LoTSS). Furthermore, our network has proven to surpass human capability, facilitating the discovery of diffuse radio emission in a particular galaxy cluster where an unprecedentedly large-scale radio structure was detected and studied in detail. In this talk, I will present the achieved results (Stuardi et al., 2024, 2025) and showcase the ongoing effort to apply this network to the LoTSS dr3, creating the largest catalogue of clusters with diffuse radio emission to date.
