Automated detection of extended sources in radio maps: progress from the SCORPIO survey

TL;DR

The paper introduces CAESAR, an advanced algorithm for detecting and analyzing extended radio sources in interferometric maps, demonstrating improved performance over existing methods on real survey data.

Contribution

It presents CAESAR, a novel modular algorithm combining denoising, superpixel clustering, and parameterization for enhanced detection of extended radio sources.

Findings

CAESAR outperforms existing algorithms in detecting extended sources.

The method effectively handles diffuse emission and imaging artefacts.

Applied to ATCA data, it successfully identified known sources and diffuse regions.

Abstract

Automated source extraction and parameterization represents a crucial challenge for the next-generation radio interferometer surveys, such as those performed with the Square Kilometre Array (SKA) and its precursors. In this paper we present a new algorithm, dubbed CAESAR (Compact And Extended Source Automated Recognition), to detect and parametrize extended sources in radio interferometric maps. It is based on a pre-filtering stage, allowing image denoising, compact source suppression and enhancement of diffuse emission, followed by an adaptive superpixel clustering stage for final source segmentation. A parameterization stage provides source flux information and a wide range of morphology estimators for post-processing analysis. We developed CAESAR in a modular software library, including also different methods for local background estimation and image filtering, along with alternative algorithms for both compact and diffuse source extraction. The method was applied to real radio continuum data collected at the Australian Telescope Compact Array (ATCA) within the SCORPIO project, a pathfinder of the ASKAP-EMU survey. The source reconstruction capabilities were studied over different test fields in the presence of compact sources, imaging artefacts and diffuse emission from the Galactic plane and compared with existing algorithms. When compared to a human-driven analysis, the designed algorithm was found capable of detecting known target sources and regions of diffuse emission, outperforming alternative approaches over the considered fields.