Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://www.atnf.csiro.au/research/workshops/2010/sourcefinding/Jurek.pdf Äàòà èçìåíåíèÿ: Mon May 10 06:33:12 2010 Äàòà èíäåêñèðîâàíèÿ: Sun Jun 27 07:00:40 2010 Êîäèðîâêà:

HI source finding algorithms
Comparing the general purpose Duchamp algorithm to a purpose built HI source finding algorithm


Talk Outline
· Common elements of source finding algorithms · The Duchamp algorithm
· · · · Algorithm Strengths Draw-backs Improvements

· An alternative HI source finder algorithm
· Key differences · Algorithm

· Preliminary work · Conclusion


Common elements of general source finding algorithms
· Define/calculate detection and growing criteria
· Thresholds or False Detection Rate

· · · · ·



Pre-condition data Scan through data and apply detection criterion Grow detections using growing criterion Merge detections Apply size criterion


The Duchamp Algorithm


Duchamp: Algorithm
· Pre-condition data (optional)
· Blank pixel removal · Baseline removal using wavelet reconstruction · Define channels to ignore · Wavelet reconstruction using a' trous wavelet procedure (priority)
OR

· Smooth in frequency space
OR

· Smooth spatially

· Set detection and growing criteria
· User specified (priority)
OR

· FDR or calculated from globally determined mean and rms values


Duchamp: Algorithm
· Raster scan data
· Travel along planes or channels and apply detection criterion · If a voxel satisfies the detection criterion
· Flag it · Check it's proximity to all previous detections and merge accordingly
· Can be turned off for efficiency, but default is ON.

· Merge detections
· Apply proximity test (again) to all detections

· Grow detections · Merge detections again · Apply size criterion
· Can be done prior to first round of merging


Duchamp: Strengths
· · · · · A truly general source finding algorithm Makes minimal assumptions Extremely flexible source detection IT EXISTS! and IT WORKS! Output is feature rich


Duchamp: Draw-backs
· Efficiency decreases with the number of detections
· Searching for faint sources is very inefficient

· Default is to run a merging routine every time a detection is made
· Compared to every! previous detection

· Merging is carried out multiple times · Size criterion is applied at the very end
· Inefficient but necessary

· Global detection and growing criteria are used
· Noise varies throughout the cube · Detect `crud' in some regions, miss detections in others

· Side-lobes + high dynamic range = problems · Multiple detections of single source


Duchamp: Improvements
· Sub-sample channels when raster scanning
· Sampling set to size criterion · Minimise detections that eventually would fail size criterion

· Define a data volume to check for previous detections
· To be used when initial merging not turned off

· Grow detections, merge (just the once), apply size and detection threshold criteria
· Apply growth criterion out to merging distance to fold in initial merging pass

· Use a local measure of noise · Flag probable side-lobes
· Sort detections according to intensity · Search through detections and for all objects within a moving, adaptive window (within intensity range of side-lobes), if within location of side-lobes flag it


A purpose built HI source finder algorithm


Key differences
· Treat datacube as a set of spectra rather than a collection of voxels · Use shape information rather than a detection threshold
· Can potentially detect faint objects that a detection threshold would miss · Recover `true' extent of source compared to using growth threshold

· Implicit is the assumption that every detection has a discernible shape · Assume that we have a well defined psf


Key differences
· Treat datacube as a set of spectra rather than a collection of voxels · Use shape information rather than a detection threshold
· Can potentially detect faint objects that a detection threshold would miss · Recover `true' extent of source compared to using growth threshold

· Implicit is the assumption that every detection has a discernible shape · Assume that we have a well defined psf


Key differences
· Treat datacube as a set of spectra rather than a collection of voxels · Use shape information rather than a detection threshold
· Can potentially detect faint objects that a detection threshold would miss · Recover `true' extent of source compared to using growth threshold

· Implicit is the assumption that every detection has a discernible shape · Assume that we have a well defined psf


Key differences
· Treat datacube as a set of spectra rather than a collection of voxels · Use shape information rather than a detection threshold
· Can potentially detect faint objects that a detection threshold would miss · Recover `true' extent of source compared to using growth threshold

· Implicit is the assumption that every detection has a discernible shape · Assume that we have a well defined psf


Specific HI source finding algorithm
· Divide data cube amongst GPUs · Sub-sample the data cube · For a given spectrum
· Pre-condition using iterative median smoothing · Use wavelet analysis to construct the noise spectrum + baselines and remove · Detect objects using shape information
· Cross-correlation? · Wavelet analysis? · Gamma test? (Even if just for measure of noise in spectrum)


Specific HI source finding algorithm
· For each detection, scan neighbouring positions in spiral pattern to determine the volume containing the detection
· Have a frequency range to process for neighbours · Well-known (and SOLVED) mouse navigating a maze problem
· The solution provides a `shrink-wrapped' volume

· Merge detections · Merge GPU results · Apply size criterion
· May have been incorporated earlier

· Flag probable side-lobes


Preliminary work
· · · · Prototyping iterative median smoothing as a pre-conditioner Using the WSRT simulated datacube Comparing to performance of Hanning filtering Results
· Quantitatively, residuals cf. input spectrum are reduced by ~20-40% · Comparable to Hanning filtering, but doesn't add/remove structure in the cases where Hanning filtering does


Conclusion
· · · · Duchamp is a great general purpose source finder The efficiency of Duchamp could be improved Side lobes are ignored by Duchamp Proposing to treat datacube as a set of spectra and use shape information to find HI sources · Development underway