This document briefly describes the means by which the Transients Project source extraction & measurement code (henceforth may be used to obtain a list of sources found in a collection of images stored as FITS files. It does not attempt to act as a complete reference to the TKP codebase.


Pyse provides the following capabilities:

  • Identification of sources in astronomical images:
    • By a simple thresholding technique (ie, locating contiguous islands of pixels above some multiple of the noise in the image), or
    • By making use of a False Detection Rate (FDR) algorithm (Hopkins et al., AJ, 123, 1086, 2002).
  • Deblending merged sources.
  • Quick estimation of source properties based on the calculation of moments.
  • Fitting of identified sources with elliptical Gaussians for accurate measurement of source properties.
  • All measurements made are accompanied by a comprehensive error analysis.

For details of all algorithms implemented, the reader is referred to the PhD thesis by Spreeuw (University of Amsterdam, 2010).

It is worth emphasizing that there are a number of differences compared to projects such as, for example, BDSM. In particular, the code is made available in the form or Python modules, primarily designed for integration into a pipeline or other script, rather than for use as an interactive analysis environment. Further, it is reasonable to assume that astronomical transients are unresolved, so the code does not attempt to decompose complex, extended sources into a multiple component model.

Command Line Usage

A script is available to make it possible to test the basic functionality of the code. It does not make all the features listed above available.

Assuming exists on your $PATH, it is involed by simply providing a list of filenames:

$ file1.fits ... fileN.fits

For each file specified, a list of sources identified is printed to the screen.

By default, source extraction is carried out by thresholding: that is, identifying islands of pixels which exceed a particular multiple of the RMS noise.

A list of available command line option may be obtained with the -h/--help option:

Source extraction for radio-synthesis images

usage: [-h] [--detection DETECTION] [--analysis ANALYSIS] [--fdr]
               [--alpha ALPHA] [--deblend-thresholds DEBLEND_THRESHOLDS]
               [--grid GRID] [--margin MARGIN] [--radius RADIUS] [--bmaj BMAJ]
               [--bmin BMIN] [--bpa BPA] [--force-beam]
               [--detection-image DETECTION_IMAGE] [--fixed-posns FIXED_POSNS]
               [--fixed-posns-file FIXED_POSNS_FILE] [--ffbox FFBOX]
               [--skymodel] [--csv] [--regions] [--rmsmap] [--sigmap]
               [--residuals] [--islands]
               files [files ...]
Positional arguments:
files Image files for processing
--detection=10 Detection threshold
--analysis=3 Analysis threshold
--fdr=False Use False Detection Rate algorithm
--alpha=0.01 FDR Alpha
 Number of deblending subthresholds; 0 to disable
--grid=64 Background grid segment size
--margin=0 Margin applied to each edge of image (in pixels)
--radius=0 Radius of usable portion of image (in pixels)
--bmaj Set beam: Major axis of beam (deg)
--bmin Set beam: Minor axis of beam (deg)
--bpa Set beam: Beam position angle (deg)
 Force fit axis lengths to beam size
 Find islands on different image
--fixed-posns List of position coordinates to force-fit (decimal degrees, JSON, e.g [[123.4,56.7],[359.9,89.9]]) (Will not perform blind extraction in this mode)
 Path to file containing a list of positions to force-fit (Will not perform blind extraction in this mode)
--ffbox=3.0 Forced fitting positional box size as a multiple of beam width.
 Generate sky model
--csv=False Generate csv text file for use in programs such as TopCat
 Generate DS9 region file(s)
--rmsmap=False Generate RMS map
--sigmap=False Generate significance map
 Generate residual maps
 Generate island maps

The --detection argument specifies the multiple of the RMS noise which is required for detection; ie, setting --detection=5 is equivalent to requiring pixels used for detecting sources to be at 5 sigma.

The --analysis argument specifies the significance level used when performing fitting. It should be lower than --detection, such that once islands have been identified a larger number of pixels is included for the fitting process.

However, if the --fdr option is given, a False Detection Rate algorithm is used instead. In this case, an additional --alpha argument may be given to specify the \(\alpha\) parameter (as defined by Hopkins).

Note that if --fdr is specified, any values given for --detection and --analysis are not used. Conversely, if --fdr is not specified, any value given for --alpha is not used.

If the --regions option is specified, a DS9-compatible region file is output showing the shapes & positions of the sources. It is named according to the input filename with the extension changed to .reg.

If the --residuals option is specified, a FITS file is produced showing the residuals left after the fitted sources have been subtraced from the input image. It is named according to the input filename with .residuals inserted before the extension.

If the --islands option is specified, a FITS file is produced showing the Gaussians which have been fitted in the data. It is named according to the input filename with .islands inserted before the extension. The sum of this file with that produced by --residuals above should total the input image.

If the --skymodel option is given, a skymodel file suitable for use with BBS will be generated. It is named according to the input filename with the extension changed to .skymodel.

If the --csv option is given, a comma-separated list of sources will be written to file. It is named according to the input filename with the extension changed to .csv.

If the --rmsmap option is given, a FITS file is produced showing the noise map which has been generated during the source-finding process. It is named according to the input filename with .rms inserted before the extension.

If the --sigmap option is given, a FITS file is produced showing the significance of each pixel: that is, the background-subtracted image pixel value divided by the RMS noise at that pixel. It is named according to the input filename with .sigmap inserted before the extension.

If the --deblend option is specified, will attempt to separate composite sources into multiple components and fit each one independently. The number of subthresholds used in this process can be specified using the --deblend-thresholds argument. Refer to Spreeuw’s thesis for a detailed description of the algorithm used.

--bmaj, --bmin and --bpa specify the shape of the restoring beam. They are equivalent to the BMAJ, BMIN and BPA FITS headers. Normally, the code will read the beam shape from the image metadata; however, if it is not available, it must be manually specified using these arguments or the process will abort.

When generating background and RMS maps of the image prior to source detection, it is segmented into a grid. The size of the grid can be specified using the --grid option. The optimal value is a compromise: it should be significantly larger than the most extended sources in the image, but small enough to account for small-scale variation across the image.

Sometimes, it is useful to exclude the edge regions of an image from processing. The --margin takes an argument given in pixels and masks off all portions of the image within the given distance of the edge before processing. The --radius argument is similar, but rather masks off all parts of the image more than the given distance from the centre. This options are cumulative.

If the --force-beam option is given, PySE will insist that all sources have axis lengths and position angles equal to the restoring beam parameters. This is (might be...) a good assumption if you are observing only point sources.

If the --detection-image option is specified, PySE will identify sources and the positions of pixels which comprise them on the deteciton image, but then use the corresponding pixels on the target images to perform measurements. Of course, the detection image and the target image(s) must have the same pixel dimensions. Note that only a single detection image may be specified, and the same pixels are then used on all target images. Note further that this --detection-image option is incompatible with --fdr.

It is possible to configure PySE to perform a fit to user-specified positions in the image _rather_ than “blindly” locating sources and attempting to fit them. (Note that it is not possible to do both at once: that requires invoking PySE twice.) This mode may be invoked either by using either of the --fixed-posns or --fixed-posns-file options. The former directly reads a list of positions from the command line; the latter accepts a filename, and reads the positions to fit from that. In both cases, the positions themselves are provided in JSON format, and should consist of a _list_ of RA, declination _pairs_ given in decimal degrees.

When fitting to a fixed position, a square “box” of pixels is chosen around the requested position, and the optimization procedure allows the source position to vary within that box. The size of the box may be changed with the --ffbox option. Note that this parameter is given in units of the major axis of the beam.

All of these arguments are optional (with the caveat that the beam shape must be provided if not included with the image).

Output Definition

The Gaussian fitted to sources is defined as:

\[peak * \exp(\ln(2.0) * ((x \cos(\theta) + y \sin(\theta)) / semiminor)^2 + ((y \cos(\theta) - x \sin(\theta)) / semimajor)^2)\]

In other words:

  • \(x\) and \(y\) are the Cartesian coordinates of the centre of the Gaussian;
  • \(peak\) is the value at the centre of the Gaussian;
  • \(theta\) is the position angle of the major axis measured counterclockwise from the y axis;
  • \(semimajor\) and \(semiminor\) are the half-widths at half-maximum of the Gaussian along its major and minor axes, respectively.