Assessing the Performance of Sub-Millimetre Compact Object Detection Algorithms

Abstract

Sub-millimetre astronomy is about to be transformed by the deployment of new detectors that can map larger images than previously possible. A particular issue in sub-millimetre astronomy is the automated detection of compact, irregular regions of emission known as clumps. There are numerous clump detection software packages freely available yet there is little consensus as to which is the most appropriate to use, as each package has its own systematic bias when performing clumpfinding. The purpose of this investigation was to investigate a number of these clumpfinding packages and determine where some of these biases may lie. The CUPID package is designed for the creation and detection of clumps within images. There are four algorithms for the detection of clumps; ClumpFind, FellWalker, GaussClumps, and Reinhold. Each algorithm was individually investigated using data from SCAMPS (the SCUBA Massive Precluster Survey), (Thompson et al., 2005) to determine the effect of changing their parameters; the algorithms were then compared against each other to examine how the results differed between them. Using Monte Carlo simulations, Gaussian artificial clumps (with known peak, size, location, and integrated flux) were inserted into an image and the algorithms were tested to determine which algorithm extracted the information with the greatest accuracy, and where the completeness limits lie with each algorithm. ClumpFind, FellWalker, and Reinhold detected a lower integrated flux level than was inserted; this effect was more evident in large, flat clumps. Due to the profile of the clump it was expected that GaussClumps would detect the integrated flux more correctly, as was the proven case.