lsst.meas.algorithms gbdb7798760+73815205df
Loading...
Searching...
No Matches
detection.py
Go to the documentation of this file.
2# LSST Data Management System
3#
4# Copyright 2008-2017 AURA/LSST.
5#
6# This product includes software developed by the
7# LSST Project (http://www.lsst.org/).
8#
9# This program is free software: you can redistribute it and/or modify
10# it under the terms of the GNU General Public License as published by
11# the Free Software Foundation, either version 3 of the License, or
12# (at your option) any later version.
13#
14# This program is distributed in the hope that it will be useful,
15# but WITHOUT ANY WARRANTY; without even the implied warranty of
16# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17# GNU General Public License for more details.
18#
19# You should have received a copy of the LSST License Statement and
20# the GNU General Public License along with this program. If not,
21# see <https://www.lsstcorp.org/LegalNotices/>.
22#
23
24__all__ = ("SourceDetectionConfig", "SourceDetectionTask", "addExposures")
25
26from contextlib import contextmanager
27
28import numpy as np
29
30import lsst.geom
31import lsst.afw.display as afwDisplay
32import lsst.afw.detection as afwDet
33import lsst.afw.geom as afwGeom
34import lsst.afw.image as afwImage
35import lsst.afw.math as afwMath
36import lsst.afw.table as afwTable
37import lsst.pex.config as pexConfig
38import lsst.pipe.base as pipeBase
39from lsst.utils.timer import timeMethod
40from .subtractBackground import SubtractBackgroundTask
41
42
43class SourceDetectionConfig(pexConfig.Config):
44 """Configuration parameters for the SourceDetectionTask
45 """
46 minPixels = pexConfig.RangeField(
47 doc="detected sources with fewer than the specified number of pixels will be ignored",
48 dtype=int, optional=False, default=1, min=0,
49 )
50 isotropicGrow = pexConfig.Field(
51 doc="Pixels should be grown as isotropically as possible (slower)",
52 dtype=bool, optional=False, default=False,
53 )
54 combinedGrow = pexConfig.Field(
55 doc="Grow all footprints at the same time? This allows disconnected footprints to merge.",
56 dtype=bool, default=True,
57 )
58 nSigmaToGrow = pexConfig.Field(
59 doc="Grow detections by nSigmaToGrow * [PSF RMS width]; if 0 then do not grow",
60 dtype=float, default=2.4, # 2.4 pixels/sigma is roughly one pixel/FWHM
61 )
62 returnOriginalFootprints = pexConfig.Field(
63 doc="Grow detections to set the image mask bits, but return the original (not-grown) footprints",
64 dtype=bool, optional=False, default=False,
65 )
66 thresholdValue = pexConfig.RangeField(
67 doc="Threshold for footprints; exact meaning and units depend on thresholdType.",
68 dtype=float, optional=False, default=5.0, min=0.0,
69 )
70 includeThresholdMultiplier = pexConfig.RangeField(
71 doc="Include threshold relative to thresholdValue",
72 dtype=float, default=1.0, min=0.0,
73 )
74 thresholdType = pexConfig.ChoiceField(
75 doc="specifies the desired flavor of Threshold",
76 dtype=str, optional=False, default="stdev",
77 allowed={
78 "variance": "threshold applied to image variance",
79 "stdev": "threshold applied to image std deviation",
80 "value": "threshold applied to image value",
81 "pixel_stdev": "threshold applied to per-pixel std deviation",
82 },
83 )
84 thresholdPolarity = pexConfig.ChoiceField(
85 doc="specifies whether to detect positive, or negative sources, or both",
86 dtype=str, optional=False, default="positive",
87 allowed={
88 "positive": "detect only positive sources",
89 "negative": "detect only negative sources",
90 "both": "detect both positive and negative sources",
91 },
92 )
93 adjustBackground = pexConfig.Field(
94 dtype=float,
95 doc="Fiddle factor to add to the background; debugging only",
96 default=0.0,
97 )
98 reEstimateBackground = pexConfig.Field(
99 dtype=bool,
100 doc="Estimate the background again after final source detection?",
101 default=True, optional=False,
102 )
103 background = pexConfig.ConfigurableField(
104 doc="Background re-estimation; ignored if reEstimateBackground false",
105 target=SubtractBackgroundTask,
106 )
107 tempLocalBackground = pexConfig.ConfigurableField(
108 doc=("A local (small-scale), temporary background estimation step run between "
109 "detecting above-threshold regions and detecting the peaks within "
110 "them; used to avoid detecting spuerious peaks in the wings."),
111 target=SubtractBackgroundTask,
112 )
113 doTempLocalBackground = pexConfig.Field(
114 dtype=bool,
115 doc="Enable temporary local background subtraction? (see tempLocalBackground)",
116 default=True,
117 )
118 tempWideBackground = pexConfig.ConfigurableField(
119 doc=("A wide (large-scale) background estimation and removal before footprint and peak detection. "
120 "It is added back into the image after detection. The purpose is to suppress very large "
121 "footprints (e.g., from large artifacts) that the deblender may choke on."),
122 target=SubtractBackgroundTask,
123 )
124 doTempWideBackground = pexConfig.Field(
125 dtype=bool,
126 doc="Do temporary wide (large-scale) background subtraction before footprint detection?",
127 default=False,
128 )
129 nPeaksMaxSimple = pexConfig.Field(
130 dtype=int,
131 doc=("The maximum number of peaks in a Footprint before trying to "
132 "replace its peaks using the temporary local background"),
133 default=1,
134 )
135 nSigmaForKernel = pexConfig.Field(
136 dtype=float,
137 doc=("Multiple of PSF RMS size to use for convolution kernel bounding box size; "
138 "note that this is not a half-size. The size will be rounded up to the nearest odd integer"),
139 default=7.0,
140 )
141 statsMask = pexConfig.ListField(
142 dtype=str,
143 doc="Mask planes to ignore when calculating statistics of image (for thresholdType=stdev)",
144 default=['BAD', 'SAT', 'EDGE', 'NO_DATA'],
145 )
146
147 def setDefaults(self):
148 self.tempLocalBackground.binSize = 64
149 self.tempLocalBackground.algorithm = "AKIMA_SPLINE"
150 self.tempLocalBackground.useApprox = False
151 # Background subtraction to remove a large-scale background (e.g., scattered light); restored later.
152 # Want to keep it from exceeding the deblender size limit of 1 Mpix, so half that is reasonable.
153 self.tempWideBackground.binSize = 512
154 self.tempWideBackground.algorithm = "AKIMA_SPLINE"
155 self.tempWideBackground.useApprox = False
156 # Ensure we can remove even bright scattered light that is DETECTED
157 for maskPlane in ("DETECTED", "DETECTED_NEGATIVE"):
158 if maskPlane in self.tempWideBackground.ignoredPixelMask:
159 self.tempWideBackground.ignoredPixelMask.remove(maskPlane)
160
161
162class SourceDetectionTask(pipeBase.Task):
163 """Create the detection task. Most arguments are simply passed onto pipe.base.Task.
164
165 Parameters
166 ----------
167 schema : `lsst.afw.table.Schema`
168 Schema object used to create the output `lsst.afw.table.SourceCatalog`
169 **kwds
170 Keyword arguments passed to `lsst.pipe.base.task.Task.__init__`
171
172 If schema is not None and configured for 'both' detections,
173 a 'flags.negative' field will be added to label detections made with a
174 negative threshold.
175
176 Notes
177 -----
178 This task can add fields to the schema, so any code calling this task must ensure that
179 these columns are indeed present in the input match list.
180 """
181
182 ConfigClass = SourceDetectionConfig
183 _DefaultName = "sourceDetection"
184
185 def __init__(self, schema=None, **kwds):
186 pipeBase.Task.__init__(self, **kwds)
187 if schema is not None and self.config.thresholdPolarity == "both":
188 self.negativeFlagKey = schema.addField(
189 "flags_negative", type="Flag",
190 doc="set if source was detected as significantly negative"
191 )
192 else:
193 if self.config.thresholdPolarity == "both":
194 self.log.warning("Detection polarity set to 'both', but no flag will be "
195 "set to distinguish between positive and negative detections")
196 self.negativeFlagKey = None
197 if self.config.reEstimateBackground:
198 self.makeSubtask("background")
199 if self.config.doTempLocalBackground:
200 self.makeSubtask("tempLocalBackground")
201 if self.config.doTempWideBackground:
202 self.makeSubtask("tempWideBackground")
203
204 @timeMethod
205 def run(self, table, exposure, doSmooth=True, sigma=None, clearMask=True, expId=None):
206 r"""Run source detection and create a SourceCatalog of detections.
207
208 Parameters
209 ----------
211 Table object that will be used to create the SourceCatalog.
212 exposure : `lsst.afw.image.Exposure`
213 Exposure to process; DETECTED mask plane will be set in-place.
214 doSmooth : `bool`
215 If True, smooth the image before detection using a Gaussian of width
216 ``sigma``, or the measured PSF width. Set to False when running on
217 e.g. a pre-convolved image, or a mask plane.
218 sigma : `float`
219 Sigma of PSF (pixels); used for smoothing and to grow detections;
220 if None then measure the sigma of the PSF of the exposure
221 clearMask : `bool`
222 Clear DETECTED{,_NEGATIVE} planes before running detection.
223 expId : `int`
224 Exposure identifier; unused by this implementation, but used for
225 RNG seed by subclasses.
226
227 Returns
228 -------
229 result : `lsst.pipe.base.Struct`
230 The `~lsst.pipe.base.Struct` contains:
231
232 ``sources``
233 The detected sources (`lsst.afw.table.SourceCatalog`)
234 ``fpSets``
235 The result returned by `detectFootprints()`
236 (`lsst.pipe.base.Struct`).
237
238 Raises
239 ------
240 ValueError
241 Raised if flags.negative is needed, but isn't in table's schema.
242 lsst.pipe.base.TaskError
243 Raised if sigma=None, doSmooth=True and the exposure has no PSF.
244
245 Notes
246 -----
247 If you want to avoid dealing with Sources and Tables, you can use
248 `detectFootprints()` to just get the
250 """
251 if self.negativeFlagKey is not None and self.negativeFlagKey not in table.getSchema():
252 raise ValueError("Table has incorrect Schema")
253 results = self.detectFootprints(exposure=exposure, doSmooth=doSmooth, sigma=sigma,
254 clearMask=clearMask, expId=expId)
255 sources = afwTable.SourceCatalog(table)
256 sources.reserve(results.numPos + results.numNeg)
257 if results.negative:
258 results.negative.makeSources(sources)
259 if self.negativeFlagKey:
260 for record in sources:
261 record.set(self.negativeFlagKey, True)
262 if results.positive:
263 results.positive.makeSources(sources)
264 results.fpSets = results.copy() # Backward compatibility
265 results.sources = sources
266 return results
267
268 def display(self, exposure, results, convolvedImage=None):
269 """Display detections if so configured
270
271 Displays the ``exposure`` in frame 0, overlays the detection peaks.
272
273 Requires that ``lsstDebug`` has been set up correctly, so that
274 ``lsstDebug.Info("lsst.meas.algorithms.detection")`` evaluates `True`.
275
276 If the ``convolvedImage`` is non-`None` and
277 ``lsstDebug.Info("lsst.meas.algorithms.detection") > 1``, the
278 ``convolvedImage`` will be displayed in frame 1.
279
280 Parameters
281 ----------
282 exposure : `lsst.afw.image.Exposure`
283 Exposure to display, on which will be plotted the detections.
284 results : `lsst.pipe.base.Struct`
285 Results of the 'detectFootprints' method, containing positive and
286 negative footprints (which contain the peak positions that we will
287 plot). This is a `Struct` with ``positive`` and ``negative``
288 elements that are of type `lsst.afw.detection.FootprintSet`.
289 convolvedImage : `lsst.afw.image.Image`, optional
290 Convolved image used for thresholding.
291 """
292 try:
293 import lsstDebug
294 display = lsstDebug.Info(__name__).display
295 except ImportError:
296 try:
297 display
298 except NameError:
299 display = False
300 if not display:
301 return
302
303 afwDisplay.setDefaultMaskTransparency(75)
304
305 disp0 = afwDisplay.Display(frame=0)
306 disp0.mtv(exposure, title="detection")
307
308 def plotPeaks(fps, ctype):
309 if fps is None:
310 return
311 with disp0.Buffering():
312 for fp in fps.getFootprints():
313 for pp in fp.getPeaks():
314 disp0.dot("+", pp.getFx(), pp.getFy(), ctype=ctype)
315 plotPeaks(results.positive, "yellow")
316 plotPeaks(results.negative, "red")
317
318 if convolvedImage and display > 1:
319 disp1 = afwDisplay.Display(frame=1)
320 disp1.mtv(convolvedImage, title="PSF smoothed")
321
322 disp2 = afwDisplay.Display(frame=2)
323 disp2.mtv(afwImage.ImageF(np.sqrt(exposure.variance.array)), title="stddev")
324
325 def applyTempLocalBackground(self, exposure, middle, results):
326 """Apply a temporary local background subtraction
327
328 This temporary local background serves to suppress noise fluctuations
329 in the wings of bright objects.
330
331 Peaks in the footprints will be updated.
332
333 Parameters
334 ----------
335 exposure : `lsst.afw.image.Exposure`
336 Exposure for which to fit local background.
338 Convolved image on which detection will be performed
339 (typically smaller than ``exposure`` because the
340 half-kernel has been removed around the edges).
341 results : `lsst.pipe.base.Struct`
342 Results of the 'detectFootprints' method, containing positive and
343 negative footprints (which contain the peak positions that we will
344 plot). This is a `Struct` with ``positive`` and ``negative``
345 elements that are of type `lsst.afw.detection.FootprintSet`.
346 """
347 # Subtract the local background from the smoothed image. Since we
348 # never use the smoothed again we don't need to worry about adding
349 # it back in.
350 bg = self.tempLocalBackground.fitBackground(exposure.getMaskedImage())
351 bgImage = bg.getImageF(self.tempLocalBackground.config.algorithm,
352 self.tempLocalBackground.config.undersampleStyle)
353 middle -= bgImage.Factory(bgImage, middle.getBBox())
354 if self.config.thresholdPolarity != "negative":
355 results.positiveThreshold = self.makeThreshold(middle, "positive")
356 self.updatePeaks(results.positive, middle, results.positiveThreshold)
357 if self.config.thresholdPolarity != "positive":
358 results.negativeThreshold = self.makeThreshold(middle, "negative")
359 self.updatePeaks(results.negative, middle, results.negativeThreshold)
360
361 def clearMask(self, mask):
362 """Clear the DETECTED and DETECTED_NEGATIVE mask planes
363
364 Removes any previous detection mask in preparation for a new
365 detection pass.
366
367 Parameters
368 ----------
369 mask : `lsst.afw.image.Mask`
370 Mask to be cleared.
371 """
372 mask &= ~(mask.getPlaneBitMask("DETECTED") | mask.getPlaneBitMask("DETECTED_NEGATIVE"))
373
374 def calculateKernelSize(self, sigma):
375 """Calculate size of smoothing kernel
376
377 Uses the ``nSigmaForKernel`` configuration parameter. Note
378 that that is the full width of the kernel bounding box
379 (so a value of 7 means 3.5 sigma on either side of center).
380 The value will be rounded up to the nearest odd integer.
381
382 Parameters
383 ----------
384 sigma : `float`
385 Gaussian sigma of smoothing kernel.
386
387 Returns
388 -------
389 size : `int`
390 Size of the smoothing kernel.
391 """
392 return (int(sigma * self.config.nSigmaForKernel + 0.5)//2)*2 + 1 # make sure it is odd
393
394 def getPsf(self, exposure, sigma=None):
395 """Retrieve the PSF for an exposure
396
397 If ``sigma`` is provided, we make a ``GaussianPsf`` with that,
398 otherwise use the one from the ``exposure``.
399
400 Parameters
401 ----------
402 exposure : `lsst.afw.image.Exposure`
403 Exposure from which to retrieve the PSF.
404 sigma : `float`, optional
405 Gaussian sigma to use if provided.
406
407 Returns
408 -------
410 PSF to use for detection.
411 """
412 if sigma is None:
413 psf = exposure.getPsf()
414 if psf is None:
415 raise RuntimeError("Unable to determine PSF to use for detection: no sigma provided")
416 sigma = psf.computeShape(psf.getAveragePosition()).getDeterminantRadius()
417 size = self.calculateKernelSize(sigma)
418 psf = afwDet.GaussianPsf(size, size, sigma)
419 return psf
420
421 def convolveImage(self, maskedImage, psf, doSmooth=True):
422 """Convolve the image with the PSF
423
424 We convolve the image with a Gaussian approximation to the PSF,
425 because this is separable and therefore fast. It's technically a
426 correlation rather than a convolution, but since we use a symmetric
427 Gaussian there's no difference.
428
429 The convolution can be disabled with ``doSmooth=False``. If we do
430 convolve, we mask the edges as ``EDGE`` and return the convolved image
431 with the edges removed. This is because we can't convolve the edges
432 because the kernel would extend off the image.
433
434 Parameters
435 ----------
436 maskedImage : `lsst.afw.image.MaskedImage`
437 Image to convolve.
439 PSF to convolve with (actually with a Gaussian approximation
440 to it).
441 doSmooth : `bool`
442 Actually do the convolution? Set to False when running on
443 e.g. a pre-convolved image, or a mask plane.
444
445 Returns
446 -------
447 results : `lsst.pipe.base.Struct`
448 The `~lsst.pipe.base.Struct` contains:
449
450 ``middle``
451 Convolved image, without the edges. (`lsst.afw.image.MaskedImage`)
452 ``sigma``
453 Gaussian sigma used for the convolution. (`float`)
454 """
455 self.metadata["doSmooth"] = doSmooth
456 sigma = psf.computeShape(psf.getAveragePosition()).getDeterminantRadius()
457 self.metadata["sigma"] = sigma
458
459 if not doSmooth:
460 middle = maskedImage.Factory(maskedImage, deep=True)
461 return pipeBase.Struct(middle=middle, sigma=sigma)
462
463 # Smooth using a Gaussian (which is separable, hence fast) of width sigma
464 # Make a SingleGaussian (separable) kernel with the 'sigma'
465 kWidth = self.calculateKernelSize(sigma)
466 self.metadata["smoothingKernelWidth"] = kWidth
467 gaussFunc = afwMath.GaussianFunction1D(sigma)
468 gaussKernel = afwMath.SeparableKernel(kWidth, kWidth, gaussFunc, gaussFunc)
469
470 convolvedImage = maskedImage.Factory(maskedImage.getBBox())
471
472 afwMath.convolve(convolvedImage, maskedImage, gaussKernel, afwMath.ConvolutionControl())
473 #
474 # Only search psf-smoothed part of frame
475 #
476 goodBBox = gaussKernel.shrinkBBox(convolvedImage.getBBox())
477 middle = convolvedImage.Factory(convolvedImage, goodBBox, afwImage.PARENT, False)
478 #
479 # Mark the parts of the image outside goodBBox as EDGE
480 #
481 self.setEdgeBits(maskedImage, goodBBox, maskedImage.getMask().getPlaneBitMask("EDGE"))
482
483 return pipeBase.Struct(middle=middle, sigma=sigma)
484
485 def applyThreshold(self, middle, bbox, factor=1.0, factorNeg=None):
486 r"""Apply thresholds to the convolved image
487
488 Identifies ``Footprint``\ s, both positive and negative.
489
490 The threshold can be modified by the provided multiplication
491 ``factor``.
492
493 Parameters
494 ----------
496 Convolved image to threshold.
497 bbox : `lsst.geom.Box2I`
498 Bounding box of unconvolved image.
499 factor : `float`
500 Multiplier for the configured threshold.
501 factorNeg : `float` or `None`
502 Multiplier for the configured threshold for negative detection polarity.
503 If `None`, will be set equal to ``factor`` (i.e. equal to the factor used
504 for positive detection polarity).
505
506 Returns
507 -------
508 results : `lsst.pipe.base.Struct`
509 The `~lsst.pipe.base.Struct` contains:
510
511 ``positive``
512 Positive detection footprints, if configured.
514 ``negative``
515 Negative detection footprints, if configured.
517 ``factor``
518 Multiplier for the configured threshold.
519 (`float`)
520 ``factorNeg``
521 Multiplier for the configured threshold for negative detection polarity.
522 (`float`)
523 """
524 if factorNeg is None:
525 factorNeg = factor
526 self.log.info("Setting factor for negative detections equal to that for positive "
527 "detections: %f", factor)
528 results = pipeBase.Struct(positive=None, negative=None, factor=factor, factorNeg=factorNeg,
529 positiveThreshold=None, negativeThreshold=None)
530 # Detect the Footprints (peaks may be replaced if doTempLocalBackground)
531 if self.config.reEstimateBackground or self.config.thresholdPolarity != "negative":
532 results.positiveThreshold = self.makeThreshold(middle, "positive", factor=factor)
533 results.positive = afwDet.FootprintSet(
534 middle,
535 results.positiveThreshold,
536 "DETECTED",
537 self.config.minPixels
538 )
539 results.positive.setRegion(bbox)
540 if self.config.reEstimateBackground or self.config.thresholdPolarity != "positive":
541 results.negativeThreshold = self.makeThreshold(middle, "negative", factor=factorNeg)
542 results.negative = afwDet.FootprintSet(
543 middle,
544 results.negativeThreshold,
545 "DETECTED_NEGATIVE",
546 self.config.minPixels
547 )
548 results.negative.setRegion(bbox)
549
550 return results
551
552 def finalizeFootprints(self, mask, results, sigma, factor=1.0, factorNeg=None):
553 """Finalize the detected footprints.
554
555 Grows the footprints, sets the ``DETECTED`` and ``DETECTED_NEGATIVE``
556 mask planes, and logs the results.
557
558 ``numPos`` (number of positive footprints), ``numPosPeaks`` (number
559 of positive peaks), ``numNeg`` (number of negative footprints),
560 ``numNegPeaks`` (number of negative peaks) entries are added to the
561 detection results.
562
563 Parameters
564 ----------
565 mask : `lsst.afw.image.Mask`
566 Mask image on which to flag detected pixels.
567 results : `lsst.pipe.base.Struct`
568 Struct of detection results, including ``positive`` and
569 ``negative`` entries; modified.
570 sigma : `float`
571 Gaussian sigma of PSF.
572 factor : `float`
573 Multiplier for the configured threshold. Note that this is only
574 used here for logging purposes.
575 factorNeg : `float` or `None`
576 Multiplier used for the negative detection polarity threshold.
577 If `None`, a factor equal to ``factor`` (i.e. equal to the one used
578 for positive detection polarity) is assumed. Note that this is only
579 used here for logging purposes.
580 """
581 factorNeg = factor if factorNeg is None else factorNeg
582 for polarity, maskName in (("positive", "DETECTED"), ("negative", "DETECTED_NEGATIVE")):
583 fpSet = getattr(results, polarity)
584 if fpSet is None:
585 continue
586 if self.config.nSigmaToGrow > 0:
587 nGrow = int((self.config.nSigmaToGrow * sigma) + 0.5)
588 self.metadata["nGrow"] = nGrow
589 if self.config.combinedGrow:
590 fpSet = afwDet.FootprintSet(fpSet, nGrow, self.config.isotropicGrow)
591 else:
592 stencil = (afwGeom.Stencil.CIRCLE if self.config.isotropicGrow else
593 afwGeom.Stencil.MANHATTAN)
594 for fp in fpSet:
595 fp.dilate(nGrow, stencil)
596 fpSet.setMask(mask, maskName)
597 if not self.config.returnOriginalFootprints:
598 setattr(results, polarity, fpSet)
599
600 results.numPos = 0
601 results.numPosPeaks = 0
602 results.numNeg = 0
603 results.numNegPeaks = 0
604 positive = ""
605 negative = ""
606
607 if results.positive is not None:
608 results.numPos = len(results.positive.getFootprints())
609 results.numPosPeaks = sum(len(fp.getPeaks()) for fp in results.positive.getFootprints())
610 positive = " %d positive peaks in %d footprints" % (results.numPosPeaks, results.numPos)
611 if results.negative is not None:
612 results.numNeg = len(results.negative.getFootprints())
613 results.numNegPeaks = sum(len(fp.getPeaks()) for fp in results.negative.getFootprints())
614 negative = " %d negative peaks in %d footprints" % (results.numNegPeaks, results.numNeg)
615
616 self.log.info("Detected%s%s%s to %g +ve and %g -ve %s",
617 positive, " and" if positive and negative else "", negative,
618 self.config.thresholdValue*self.config.includeThresholdMultiplier*factor,
619 self.config.thresholdValue*self.config.includeThresholdMultiplier*factorNeg,
620 "DN" if self.config.thresholdType == "value" else "sigma")
621
622 def reEstimateBackground(self, maskedImage, backgrounds):
623 """Estimate the background after detection
624
625 Parameters
626 ----------
627 maskedImage : `lsst.afw.image.MaskedImage`
628 Image on which to estimate the background.
629 backgrounds : `lsst.afw.math.BackgroundList`
630 List of backgrounds; modified.
631
632 Returns
633 -------
634 bg : `lsst.afw.math.backgroundMI`
635 Empirical background model.
636 """
637 bg = self.background.fitBackground(maskedImage)
638 if self.config.adjustBackground:
639 self.log.warning("Fiddling the background by %g", self.config.adjustBackground)
640 bg += self.config.adjustBackground
641 self.log.info("Resubtracting the background after object detection")
642 maskedImage -= bg.getImageF(self.background.config.algorithm,
643 self.background.config.undersampleStyle)
644
645 actrl = bg.getBackgroundControl().getApproximateControl()
646 backgrounds.append((bg, getattr(afwMath.Interpolate, self.background.config.algorithm),
647 bg.getAsUsedUndersampleStyle(), actrl.getStyle(), actrl.getOrderX(),
648 actrl.getOrderY(), actrl.getWeighting()))
649 return bg
650
651 def clearUnwantedResults(self, mask, results):
652 """Clear unwanted results from the Struct of results
653
654 If we specifically want only positive or only negative detections,
655 drop the ones we don't want, and its associated mask plane.
656
657 Parameters
658 ----------
659 mask : `lsst.afw.image.Mask`
660 Mask image.
661 results : `lsst.pipe.base.Struct`
662 Detection results, with ``positive`` and ``negative`` elements;
663 modified.
664 """
665 if self.config.thresholdPolarity == "positive":
666 if self.config.reEstimateBackground:
667 mask &= ~mask.getPlaneBitMask("DETECTED_NEGATIVE")
668 results.negative = None
669 elif self.config.thresholdPolarity == "negative":
670 if self.config.reEstimateBackground:
671 mask &= ~mask.getPlaneBitMask("DETECTED")
672 results.positive = None
673
674 @timeMethod
675 def detectFootprints(self, exposure, doSmooth=True, sigma=None, clearMask=True, expId=None):
676 """Detect footprints on an exposure.
677
678 Parameters
679 ----------
680 exposure : `lsst.afw.image.Exposure`
681 Exposure to process; DETECTED{,_NEGATIVE} mask plane will be
682 set in-place.
683 doSmooth : `bool`, optional
684 If True, smooth the image before detection using a Gaussian
685 of width ``sigma``, or the measured PSF width of ``exposure``.
686 Set to False when running on e.g. a pre-convolved image, or a mask
687 plane.
688 sigma : `float`, optional
689 Gaussian Sigma of PSF (pixels); used for smoothing and to grow
690 detections; if `None` then measure the sigma of the PSF of the
691 ``exposure``.
692 clearMask : `bool`, optional
693 Clear both DETECTED and DETECTED_NEGATIVE planes before running
694 detection.
695 expId : `dict`, optional
696 Exposure identifier; unused by this implementation, but used for
697 RNG seed by subclasses.
698
699 Returns
700 -------
701 results : `lsst.pipe.base.Struct`
702 A `~lsst.pipe.base.Struct` containing:
703
704 ``positive``
705 Positive polarity footprints.
707 ``negative``
708 Negative polarity footprints.
710 ``numPos``
711 Number of footprints in positive or 0 if detection polarity was
712 negative. (`int`)
713 ``numNeg``
714 Number of footprints in negative or 0 if detection polarity was
715 positive. (`int`)
716 ``background``
717 Re-estimated background. `None` if
718 ``reEstimateBackground==False``.
719 (`lsst.afw.math.BackgroundList`)
720 ``factor``
721 Multiplication factor applied to the configured detection
722 threshold. (`float`)
723 """
724 maskedImage = exposure.maskedImage
725
726 if clearMask:
727 self.clearMask(maskedImage.getMask())
728
729 psf = self.getPsf(exposure, sigma=sigma)
730 with self.tempWideBackgroundContext(exposure):
731 convolveResults = self.convolveImage(maskedImage, psf, doSmooth=doSmooth)
732 middle = convolveResults.middle
733 sigma = convolveResults.sigma
734
735 results = self.applyThreshold(middle, maskedImage.getBBox())
736 results.background = afwMath.BackgroundList()
737 if self.config.doTempLocalBackground:
738 self.applyTempLocalBackground(exposure, middle, results)
739 self.finalizeFootprints(maskedImage.mask, results, sigma)
740
741 # Compute the significance of peaks after the peaks have been
742 # finalized and after local background correction/updatePeaks, so
743 # that the significance represents the "final" detection S/N.
744 results.positive = self.setPeakSignificance(middle, results.positive, results.positiveThreshold)
745 results.negative = self.setPeakSignificance(middle, results.negative, results.negativeThreshold,
746 negative=True)
747
748 if self.config.reEstimateBackground:
749 self.reEstimateBackground(maskedImage, results.background)
750
751 self.clearUnwantedResults(maskedImage.getMask(), results)
752
753 self.display(exposure, results, middle)
754
755 return results
756
757 def setPeakSignificance(self, exposure, footprints, threshold, negative=False):
758 """Set the significance of each detected peak to the pixel value divided
759 by the appropriate standard-deviation for ``config.thresholdType``.
760
761 Only sets significance for "stdev" and "pixel_stdev" thresholdTypes;
762 we leave it undefined for "value" and "variance" as it does not have a
763 well-defined meaning in those cases.
764
765 Parameters
766 ----------
767 exposure : `lsst.afw.image.Exposure`
768 Exposure that footprints were detected on, likely the convolved,
769 local background-subtracted image.
771 Footprints detected on the image.
772 threshold : `lsst.afw.detection.Threshold`
773 Threshold used to find footprints.
774 negative : `bool`, optional
775 Are we calculating for negative sources?
776 """
777 if footprints is None or footprints.getFootprints() == []:
778 return footprints
779 polarity = -1 if negative else 1
780
781 # All incoming footprints have the same schema.
782 mapper = afwTable.SchemaMapper(footprints.getFootprints()[0].peaks.schema)
783 mapper.addMinimalSchema(footprints.getFootprints()[0].peaks.schema)
784 mapper.addOutputField("significance", type=float,
785 doc="Ratio of peak value to configured standard deviation.")
786
787 # Copy the old peaks to the new ones with a significance field.
788 # Do this independent of the threshold type, so we always have a
789 # significance field.
790 newFootprints = afwDet.FootprintSet(footprints)
791 for old, new in zip(footprints.getFootprints(), newFootprints.getFootprints()):
792 newPeaks = afwDet.PeakCatalog(mapper.getOutputSchema())
793 newPeaks.extend(old.peaks, mapper=mapper)
794 new.getPeaks().clear()
795 new.setPeakCatalog(newPeaks)
796
797 # Compute the significance values.
798 if self.config.thresholdType == "pixel_stdev":
799 for footprint in newFootprints.getFootprints():
800 footprint.updatePeakSignificance(exposure.variance, polarity)
801 elif self.config.thresholdType == "stdev":
802 sigma = threshold.getValue() / self.config.thresholdValue
803 for footprint in newFootprints.getFootprints():
804 footprint.updatePeakSignificance(polarity*sigma)
805 else:
806 for footprint in newFootprints.getFootprints():
807 for peak in footprint.peaks:
808 peak["significance"] = 0
809
810 return newFootprints
811
812 def makeThreshold(self, image, thresholdParity, factor=1.0):
813 """Make an afw.detection.Threshold object corresponding to the task's
814 configuration and the statistics of the given image.
815
816 Parameters
817 ----------
818 image : `afw.image.MaskedImage`
819 Image to measure noise statistics from if needed.
820 thresholdParity: `str`
821 One of "positive" or "negative", to set the kind of fluctuations
822 the Threshold will detect.
823 factor : `float`
824 Factor by which to multiply the configured detection threshold.
825 This is useful for tweaking the detection threshold slightly.
826
827 Returns
828 -------
829 threshold : `lsst.afw.detection.Threshold`
830 Detection threshold.
831 """
832 parity = False if thresholdParity == "negative" else True
833 thresholdValue = self.config.thresholdValue
834 thresholdType = self.config.thresholdType
835 if self.config.thresholdType == 'stdev':
836 bad = image.getMask().getPlaneBitMask(self.config.statsMask)
837 sctrl = afwMath.StatisticsControl()
838 sctrl.setAndMask(bad)
839 stats = afwMath.makeStatistics(image, afwMath.STDEVCLIP, sctrl)
840 thresholdValue *= stats.getValue(afwMath.STDEVCLIP)
841 thresholdType = 'value'
842
843 threshold = afwDet.createThreshold(thresholdValue*factor, thresholdType, parity)
844 threshold.setIncludeMultiplier(self.config.includeThresholdMultiplier)
845 self.log.debug("Detection threshold: %s", threshold)
846 return threshold
847
848 def updatePeaks(self, fpSet, image, threshold):
849 """Update the Peaks in a FootprintSet by detecting new Footprints and
850 Peaks in an image and using the new Peaks instead of the old ones.
851
852 Parameters
853 ----------
855 Set of Footprints whose Peaks should be updated.
856 image : `afw.image.MaskedImage`
857 Image to detect new Footprints and Peak in.
858 threshold : `afw.detection.Threshold`
859 Threshold object for detection.
860
861 Input Footprints with fewer Peaks than self.config.nPeaksMaxSimple
862 are not modified, and if no new Peaks are detected in an input
863 Footprint, the brightest original Peak in that Footprint is kept.
864 """
865 for footprint in fpSet.getFootprints():
866 oldPeaks = footprint.getPeaks()
867 if len(oldPeaks) <= self.config.nPeaksMaxSimple:
868 continue
869 # We detect a new FootprintSet within each non-simple Footprint's
870 # bbox to avoid a big O(N^2) comparison between the two sets of
871 # Footprints.
872 sub = image.Factory(image, footprint.getBBox())
873 fpSetForPeaks = afwDet.FootprintSet(
874 sub,
875 threshold,
876 "", # don't set a mask plane
877 self.config.minPixels
878 )
879 newPeaks = afwDet.PeakCatalog(oldPeaks.getTable())
880 for fpForPeaks in fpSetForPeaks.getFootprints():
881 for peak in fpForPeaks.getPeaks():
882 if footprint.contains(peak.getI()):
883 newPeaks.append(peak)
884 if len(newPeaks) > 0:
885 del oldPeaks[:]
886 oldPeaks.extend(newPeaks)
887 else:
888 del oldPeaks[1:]
889
890 @staticmethod
891 def setEdgeBits(maskedImage, goodBBox, edgeBitmask):
892 """Set the edgeBitmask bits for all of maskedImage outside goodBBox
893
894 Parameters
895 ----------
896 maskedImage : `lsst.afw.image.MaskedImage`
897 Image on which to set edge bits in the mask.
898 goodBBox : `lsst.geom.Box2I`
899 Bounding box of good pixels, in ``LOCAL`` coordinates.
900 edgeBitmask : `lsst.afw.image.MaskPixel`
901 Bit mask to OR with the existing mask bits in the region
902 outside ``goodBBox``.
903 """
904 msk = maskedImage.getMask()
905
906 mx0, my0 = maskedImage.getXY0()
907 for x0, y0, w, h in ([0, 0,
908 msk.getWidth(), goodBBox.getBeginY() - my0],
909 [0, goodBBox.getEndY() - my0, msk.getWidth(),
910 maskedImage.getHeight() - (goodBBox.getEndY() - my0)],
911 [0, 0,
912 goodBBox.getBeginX() - mx0, msk.getHeight()],
913 [goodBBox.getEndX() - mx0, 0,
914 maskedImage.getWidth() - (goodBBox.getEndX() - mx0), msk.getHeight()],
915 ):
916 edgeMask = msk.Factory(msk, lsst.geom.BoxI(lsst.geom.PointI(x0, y0),
917 lsst.geom.ExtentI(w, h)), afwImage.LOCAL)
918 edgeMask |= edgeBitmask
919
920 @contextmanager
921 def tempWideBackgroundContext(self, exposure):
922 """Context manager for removing wide (large-scale) background
923
924 Removing a wide (large-scale) background helps to suppress the
925 detection of large footprints that may overwhelm the deblender.
926 It does, however, set a limit on the maximum scale of objects.
927
928 The background that we remove will be restored upon exit from
929 the context manager.
930
931 Parameters
932 ----------
933 exposure : `lsst.afw.image.Exposure`
934 Exposure on which to remove large-scale background.
935
936 Returns
937 -------
938 context : context manager
939 Context manager that will ensure the temporary wide background
940 is restored.
941 """
942 doTempWideBackground = self.config.doTempWideBackground
943 if doTempWideBackground:
944 self.log.info("Applying temporary wide background subtraction")
945 original = exposure.maskedImage.image.array[:].copy()
946 self.tempWideBackground.run(exposure).background
947 # Remove NO_DATA regions (e.g., edge of the field-of-view); these can cause detections after
948 # subtraction because of extrapolation of the background model into areas with no constraints.
949 image = exposure.maskedImage.image
950 mask = exposure.maskedImage.mask
951 noData = mask.array & mask.getPlaneBitMask("NO_DATA") > 0
952 isGood = mask.array & mask.getPlaneBitMask(self.config.statsMask) == 0
953 image.array[noData] = np.median(image.array[~noData & isGood])
954 try:
955 yield
956 finally:
957 if doTempWideBackground:
958 exposure.maskedImage.image.array[:] = original
959
960
961def addExposures(exposureList):
962 """Add a set of exposures together.
963
964 Parameters
965 ----------
966 exposureList : `list` of `lsst.afw.image.Exposure`
967 Sequence of exposures to add.
968
969 Returns
970 -------
971 addedExposure : `lsst.afw.image.Exposure`
972 An exposure of the same size as each exposure in ``exposureList``,
973 with the metadata from ``exposureList[0]`` and a masked image equal
974 to the sum of all the exposure's masked images.
975 """
976 exposure0 = exposureList[0]
977 image0 = exposure0.getMaskedImage()
978
979 addedImage = image0.Factory(image0, True)
980 addedImage.setXY0(image0.getXY0())
981
982 for exposure in exposureList[1:]:
983 image = exposure.getMaskedImage()
984 addedImage += image
985
986 addedExposure = exposure0.Factory(addedImage, exposure0.getWcs())
987 return addedExposure
def getPsf(self, exposure, sigma=None)
Definition: detection.py:394
def makeThreshold(self, image, thresholdParity, factor=1.0)
Definition: detection.py:812
def setPeakSignificance(self, exposure, footprints, threshold, negative=False)
Definition: detection.py:757
def run(self, table, exposure, doSmooth=True, sigma=None, clearMask=True, expId=None)
Definition: detection.py:205
def convolveImage(self, maskedImage, psf, doSmooth=True)
Definition: detection.py:421
def applyTempLocalBackground(self, exposure, middle, results)
Definition: detection.py:325
def detectFootprints(self, exposure, doSmooth=True, sigma=None, clearMask=True, expId=None)
Definition: detection.py:675
def reEstimateBackground(self, maskedImage, backgrounds)
Definition: detection.py:622
def updatePeaks(self, fpSet, image, threshold)
Definition: detection.py:848
def applyThreshold(self, middle, bbox, factor=1.0, factorNeg=None)
Definition: detection.py:485
def setEdgeBits(maskedImage, goodBBox, edgeBitmask)
Definition: detection.py:891
def display(self, exposure, results, convolvedImage=None)
Definition: detection.py:268
def finalizeFootprints(self, mask, results, sigma, factor=1.0, factorNeg=None)
Definition: detection.py:552
def addExposures(exposureList)
Definition: detection.py:961