The ASKAP/EMU Source Finding Data Challenge

Abstract

The Evolutionary Map of the Universe (EMU) is a proposed radio continuum survey of the Southern Hemisphere up to declination +30• , with the Australian Square Kilometre Array Pathfinder (ASKAP). EMU will use an automated source identification and measurement approach that is demonstrably optimal, to maximise the reliability and robustness of the resulting radio source catalogues. As a step toward this goal we conducted a "Data Challenge" to test a variety of source finders on simulated images. The aim is to quantify the accuracy and limitations of existing automated source finding and measurement approaches. The Challenge initiators also tested the current ASKAPsoft source-finding tool to establish how it could benefit from incorporating successful features of the other tools. As expected, most finders show completeness around 100% at ≈10 σ dropping to about 10% by ≈5 σ . Reliability is typically close to 100% at ≈10 σ , with performance to lower sensitivities varying between finders. All finders show the expected trade-off, where a high completeness at low signal-to-noise gives a corresponding reduction in reliability, and vice versa. We conclude with a series of recommendations for improving the performance of the ASKAPsoft source-finding tool.

The FIRST Classifier: compact and extended radio galaxy classification using deep Convolutional Neural Networks

“…Generally, the recall is the best measure to check for over-fitting, while the precision provides a quantitative measure of the correctly classified sources. Recall and precision are also called reliability (or sensitivity) and completeness respectively (Hopkins et al 2015). The higher value of recall and precision the better performance of the model (Ivezic et al 2014).…”

Radio source extraction with ProFound

“…Both PyBDSF and AEGEAN have been used in past radio continuum observations (see e.g. Hurley-Walker et al 2017;Shimwell et al 2017) and have been compared to each other in previous tests for large survey data challenges (a comparison for use on simulated images in preparation for EMU was performed in Hopkins et al 2015). This gives default parameters for PyBDSF of thresh_isl=3.0 and thresh_pix=5.0.…”

“…There is also a peak in the histogram around a flux density per beam of 0 mJy/beam. This excess is again due to the Here PyBDSF and AEGEAN use the default parameters as described in Section 3, and used in Hopkins et al (2015). Shown are the entirety of the images downloaded from Leahy et al (1996).…”

“…Shown are the entirety of the images downloaded from Leahy et al (1996). Here PyBDSF and AEGEAN use the default parameters as described in Section 3, and used in Hopkins et al (2015). Shown are the entirety of the images downloaded from Leahy et al (1996). smooth sky model that ProFound uses and so small noise fluctuations may be included as part of the source.…”

PySE: Software for extracting sources from radio images

“…7. Comparing PySE to other source finding software Hopkins et al (2015) conducted the ASKAP/EMU Source Finder Data Challenge comparing the accuracy of many source finders when run on three simulated radio maps containing a mixture of blended, unblended, point and extended sources. Some of the source finding tools that have been tested within the Data Challenge were created to work on radio images, whereas others were designed to work on images at other wavelengths.…”