Coverage for python/lsst/pipe/tasks/maskStreaks.py: 21%

283 statements  

« prev     ^ index     » next       coverage.py v7.4.0, created at 2024-01-04 13:31 +0000

1# This file is part of pipe_tasks. 

2# 

3# Developed for the LSST Data Management System. 

4# This product includes software developed by the LSST Project 

5# (https://www.lsst.org). 

6# See the COPYRIGHT file at the top-level directory of this distribution 

7# for details of code ownership. 

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 GNU General Public License 

20# along with this program. If not, see <https://www.gnu.org/licenses/>. 

21 

22__all__ = ["MaskStreaksConfig", "MaskStreaksTask", "setDetectionMask"] 

23 

24import lsst.pex.config as pexConfig 

25import lsst.pipe.base as pipeBase 

26import lsst.kht 

27from lsst.utils.timer import timeMethod 

28 

29import numpy as np 

30import scipy 

31import textwrap 

32import copy 

33from skimage.feature import canny 

34from sklearn.cluster import KMeans 

35import warnings 

36from dataclasses import dataclass 

37 

38 

39def setDetectionMask(maskedImage, forceSlowBin=False, binning=None, detectedPlane="DETECTED", 

40 badMaskPlanes=("NO_DATA", "INTRP", "BAD", "SAT", "EDGE"), detectionThreshold=5): 

41 """Make detection mask and set the mask plane. 

42 

43 Creat a binary image from a masked image by setting all data with signal-to- 

44 noise below some threshold to zero, and all data above the threshold to one. 

45 If the binning parameter has been set, this procedure will be preceded by a 

46 weighted binning of the data in order to smooth the result, after which the 

47 result is scaled back to the original dimensions. Set the detection mask 

48 plane with this binary image. 

49 

50 Parameters 

51 ---------- 

52 maskedImage : `lsst.afw.image.maskedImage` 

53 Image to be (optionally) binned and converted. 

54 forceSlowBin : `bool`, optional 

55 Force usage of slower binning method to check that the two methods 

56 give the same result. 

57 binning : `int`, optional 

58 Number of pixels by which to bin image. 

59 detectedPlane : `str`, optional 

60 Name of mask with pixels that were detected above threshold in image. 

61 badMaskPlanes : `set`, optional 

62 Names of masks with pixels that are rejected. 

63 detectionThreshold : `float`, optional 

64 Boundary in signal-to-noise between non-detections and detections for 

65 making a binary image from the original input image. 

66 """ 

67 data = maskedImage.image.array 

68 weights = 1 / maskedImage.variance.array 

69 mask = maskedImage.getMask() 

70 

71 detectionMask = ((mask.array & mask.getPlaneBitMask(detectedPlane))) 

72 badPixelMask = mask.getPlaneBitMask(badMaskPlanes) 

73 badMask = (mask.array & badPixelMask) > 0 

74 fitMask = detectionMask.astype(bool) & ~badMask 

75 

76 fitData = np.copy(data) 

77 fitData[~fitMask] = 0 

78 fitWeights = np.copy(weights) 

79 fitWeights[~fitMask] = 0 

80 

81 if binning: 

82 # Do weighted binning: 

83 ymax, xmax = fitData.shape 

84 if (ymax % binning == 0) and (xmax % binning == 0) and (not forceSlowBin): 

85 # Faster binning method 

86 binNumeratorReshape = (fitData * fitWeights).reshape(ymax // binning, binning, 

87 xmax // binning, binning) 

88 binDenominatorReshape = fitWeights.reshape(binNumeratorReshape.shape) 

89 binnedNumerator = binNumeratorReshape.sum(axis=3).sum(axis=1) 

90 binnedDenominator = binDenominatorReshape.sum(axis=3).sum(axis=1) 

91 else: 

92 # Slower binning method when (image shape mod binsize) != 0 

93 warnings.warn('Using slow binning method--consider choosing a binsize that evenly divides ' 

94 f'into the image size, so that {ymax} mod binning == 0 ' 

95 f'and {xmax} mod binning == 0', stacklevel=2) 

96 xarray = np.arange(xmax) 

97 yarray = np.arange(ymax) 

98 xmesh, ymesh = np.meshgrid(xarray, yarray) 

99 xbins = np.arange(0, xmax + binning, binning) 

100 ybins = np.arange(0, ymax + binning, binning) 

101 numerator = fitWeights * fitData 

102 binnedNumerator, *_ = scipy.stats.binned_statistic_2d(ymesh.ravel(), xmesh.ravel(), 

103 numerator.ravel(), statistic='sum', 

104 bins=(ybins, xbins)) 

105 binnedDenominator, *_ = scipy.stats.binned_statistic_2d(ymesh.ravel(), xmesh.ravel(), 

106 fitWeights.ravel(), statistic='sum', 

107 bins=(ybins, xbins)) 

108 binnedData = np.zeros(binnedNumerator.shape) 

109 ind = binnedDenominator != 0 

110 np.divide(binnedNumerator, binnedDenominator, out=binnedData, where=ind) 

111 binnedWeight = binnedDenominator 

112 binMask = (binnedData * binnedWeight**0.5) > detectionThreshold 

113 tmpOutputMask = binMask.repeat(binning, axis=0)[:ymax] 

114 outputMask = tmpOutputMask.repeat(binning, axis=1)[:, :xmax] 

115 else: 

116 outputMask = (fitData * fitWeights**0.5) > detectionThreshold 

117 

118 # Clear existing Detected Plane: 

119 maskedImage.mask.array &= ~maskedImage.mask.getPlaneBitMask(detectedPlane) 

120 

121 # Set Detected Plane with the binary detection mask: 

122 maskedImage.mask.array[outputMask] |= maskedImage.mask.getPlaneBitMask(detectedPlane) 

123 

124 

125@dataclass 

126class Line: 

127 """A simple data class to describe a line profile. The parameter `rho` 

128 describes the distance from the center of the image, `theta` describes 

129 the angle, and `sigma` describes the width of the line. 

130 """ 

131 

132 rho: float 

133 theta: float 

134 sigma: float = 0 

135 

136 

137class LineCollection: 

138 """Collection of `Line` objects. 

139 

140 Parameters 

141 ---------- 

142 rhos : `np.ndarray` 

143 Array of `Line` rho parameters. 

144 thetas : `np.ndarray` 

145 Array of `Line` theta parameters. 

146 sigmas : `np.ndarray`, optional 

147 Array of `Line` sigma parameters. 

148 """ 

149 

150 def __init__(self, rhos, thetas, sigmas=None): 

151 if sigmas is None: 

152 sigmas = np.zeros(len(rhos)) 

153 

154 self._lines = [Line(rho, theta, sigma) for (rho, theta, sigma) in 

155 zip(rhos, thetas, sigmas)] 

156 

157 def __len__(self): 

158 return len(self._lines) 

159 

160 def __getitem__(self, index): 

161 return self._lines[index] 

162 

163 def __iter__(self): 

164 return iter(self._lines) 

165 

166 def __repr__(self): 

167 joinedString = ", ".join(str(line) for line in self._lines) 

168 return textwrap.shorten(joinedString, width=160, placeholder="...") 

169 

170 @property 

171 def rhos(self): 

172 return np.array([line.rho for line in self._lines]) 

173 

174 @property 

175 def thetas(self): 

176 return np.array([line.theta for line in self._lines]) 

177 

178 def append(self, newLine): 

179 """Add line to current collection of lines. 

180 

181 Parameters 

182 ---------- 

183 newLine : `Line` 

184 `Line` to add to current collection of lines 

185 """ 

186 self._lines.append(copy.copy(newLine)) 

187 

188 

189class LineProfile: 

190 """Construct and/or fit a model for a linear streak. 

191 

192 This assumes a simple model for a streak, in which the streak 

193 follows a straight line in pixels space, with a Moffat-shaped profile. The 

194 model is fit to data using a Newton-Raphson style minimization algorithm. 

195 The initial guess for the line parameters is assumed to be fairly accurate, 

196 so only a narrow band of pixels around the initial line estimate is used in 

197 fitting the model, which provides a significant speed-up over using all the 

198 data. The class can also be used just to construct a model for the data with 

199 a line following the given coordinates. 

200 

201 Parameters 

202 ---------- 

203 data : `np.ndarray` 

204 2d array of data. 

205 weights : `np.ndarray` 

206 2d array of weights. 

207 line : `Line`, optional 

208 Guess for position of line. Data far from line guess is masked out. 

209 Defaults to None, in which case only data with `weights` = 0 is masked 

210 out. 

211 """ 

212 

213 def __init__(self, data, weights, line=None): 

214 self.data = data 

215 self.weights = weights 

216 self._ymax, self._xmax = data.shape 

217 self._dtype = data.dtype 

218 xrange = np.arange(self._xmax) - self._xmax / 2. 

219 yrange = np.arange(self._ymax) - self._ymax / 2. 

220 self._rhoMax = ((0.5 * self._ymax)**2 + (0.5 * self._xmax)**2)**0.5 

221 self._xmesh, self._ymesh = np.meshgrid(xrange, yrange) 

222 self.mask = (weights != 0) 

223 

224 self._initLine = line 

225 self.setLineMask(line) 

226 

227 def setLineMask(self, line): 

228 """Set mask around the image region near the line. 

229 

230 Parameters 

231 ---------- 

232 line : `Line` 

233 Parameters of line in the image. 

234 """ 

235 if line: 

236 # Only fit pixels within 5 sigma of the estimated line 

237 radtheta = np.deg2rad(line.theta) 

238 distance = (np.cos(radtheta) * self._xmesh + np.sin(radtheta) * self._ymesh - line.rho) 

239 m = (abs(distance) < 5 * line.sigma) 

240 self.lineMask = self.mask & m 

241 else: 

242 self.lineMask = np.copy(self.mask) 

243 

244 self.lineMaskSize = self.lineMask.sum() 

245 self._maskData = self.data[self.lineMask] 

246 self._maskWeights = self.weights[self.lineMask] 

247 self._mxmesh = self._xmesh[self.lineMask] 

248 self._mymesh = self._ymesh[self.lineMask] 

249 

250 def _makeMaskedProfile(self, line, fitFlux=True): 

251 """Construct the line model in the masked region and calculate its 

252 derivatives. 

253 

254 Parameters 

255 ---------- 

256 line : `Line` 

257 Parameters of line profile for which to make profile in the masked 

258 region. 

259 fitFlux : `bool` 

260 Fit the amplitude of the line profile to the data. 

261 

262 Returns 

263 ------- 

264 model : `np.ndarray` 

265 Model in the masked region. 

266 dModel : `np.ndarray` 

267 Derivative of the model in the masked region. 

268 """ 

269 invSigma = line.sigma**-1 

270 # Calculate distance between pixels and line 

271 radtheta = np.deg2rad(line.theta) 

272 costheta = np.cos(radtheta) 

273 sintheta = np.sin(radtheta) 

274 distance = (costheta * self._mxmesh + sintheta * self._mymesh - line.rho) 

275 distanceSquared = distance**2 

276 

277 # Calculate partial derivatives of distance 

278 drad = np.pi / 180 

279 dDistanceSqdRho = 2 * distance * (-np.ones_like(self._mxmesh)) 

280 dDistanceSqdTheta = (2 * distance * (-sintheta * self._mxmesh + costheta * self._mymesh) * drad) 

281 

282 # Use pixel-line distances to make Moffat profile 

283 profile = (1 + distanceSquared * invSigma**2)**-2.5 

284 dProfile = -2.5 * (1 + distanceSquared * invSigma**2)**-3.5 

285 

286 if fitFlux: 

287 # Calculate line flux from profile and data 

288 flux = ((self._maskWeights * self._maskData * profile).sum() 

289 / (self._maskWeights * profile**2).sum()) 

290 else: 

291 # Approximately normalize the line 

292 flux = invSigma**-1 

293 if np.isnan(flux): 

294 flux = 0 

295 

296 model = flux * profile 

297 

298 # Calculate model derivatives 

299 fluxdProfile = flux * dProfile 

300 fluxdProfileInvSigma = fluxdProfile * invSigma**2 

301 dModeldRho = fluxdProfileInvSigma * dDistanceSqdRho 

302 dModeldTheta = fluxdProfileInvSigma * dDistanceSqdTheta 

303 dModeldInvSigma = fluxdProfile * distanceSquared * 2 * invSigma 

304 

305 dModel = np.array([dModeldRho, dModeldTheta, dModeldInvSigma]) 

306 return model, dModel 

307 

308 def makeProfile(self, line, fitFlux=True): 

309 """Construct the line profile model. 

310 

311 Parameters 

312 ---------- 

313 line : `Line` 

314 Parameters of the line profile to model. 

315 fitFlux : `bool`, optional 

316 Fit the amplitude of the line profile to the data. 

317 

318 Returns 

319 ------- 

320 finalModel : `np.ndarray` 

321 Model for line profile. 

322 """ 

323 model, _ = self._makeMaskedProfile(line, fitFlux=fitFlux) 

324 finalModel = np.zeros((self._ymax, self._xmax), dtype=self._dtype) 

325 finalModel[self.lineMask] = model 

326 return finalModel 

327 

328 def _lineChi2(self, line, grad=True): 

329 """Construct the chi2 between the data and the model. 

330 

331 Parameters 

332 ---------- 

333 line : `Line` 

334 `Line` parameters for which to build model and calculate chi2. 

335 grad : `bool`, optional 

336 Whether or not to return the gradient and hessian. 

337 

338 Returns 

339 ------- 

340 reducedChi : `float` 

341 Reduced chi2 of the model. 

342 reducedDChi : `np.ndarray` 

343 Derivative of the chi2 with respect to rho, theta, invSigma. 

344 reducedHessianChi : `np.ndarray` 

345 Hessian of the chi2 with respect to rho, theta, invSigma. 

346 """ 

347 # Calculate chi2 

348 model, dModel = self._makeMaskedProfile(line) 

349 chi2 = (self._maskWeights * (self._maskData - model)**2).sum() 

350 if not grad: 

351 return chi2.sum() / self.lineMaskSize 

352 

353 # Calculate derivative and Hessian of chi2 

354 derivChi2 = ((-2 * self._maskWeights * (self._maskData - model))[None, :] * dModel).sum(axis=1) 

355 hessianChi2 = (2 * self._maskWeights * dModel[:, None, :] * dModel[None, :, :]).sum(axis=2) 

356 

357 reducedChi = chi2 / self.lineMaskSize 

358 reducedDChi = derivChi2 / self.lineMaskSize 

359 reducedHessianChi = hessianChi2 / self.lineMaskSize 

360 return reducedChi, reducedDChi, reducedHessianChi 

361 

362 def fit(self, dChi2Tol=0.1, maxIter=100, log=None): 

363 """Perform Newton-Raphson minimization to find line parameters. 

364 

365 This method takes advantage of having known derivative and Hessian of 

366 the multivariate function to quickly and efficiently find the minimum. 

367 This is more efficient than the scipy implementation of the Newton- 

368 Raphson method, which doesn't take advantage of the Hessian matrix. The 

369 method here also performs a line search in the direction of the steepest 

370 derivative at each iteration, which reduces the number of iterations 

371 needed. 

372 

373 Parameters 

374 ---------- 

375 dChi2Tol : `float`, optional 

376 Change in Chi2 tolerated for fit convergence. 

377 maxIter : `int`, optional 

378 Maximum number of fit iterations allowed. The fit should converge in 

379 ~10 iterations, depending on the value of dChi2Tol, but this 

380 maximum provides a backup. 

381 log : `lsst.utils.logging.LsstLogAdapter`, optional 

382 Logger to use for reporting more details for fitting failures. 

383 

384 Returns 

385 ------- 

386 outline : `np.ndarray` 

387 Coordinates and inverse width of fit line. 

388 chi2 : `float` 

389 Reduced Chi2 of model fit to data. 

390 fitFailure : `bool` 

391 Boolean where `False` corresponds to a successful fit. 

392 """ 

393 # Do minimization on inverse of sigma to simplify derivatives: 

394 x = np.array([self._initLine.rho, self._initLine.theta, self._initLine.sigma**-1]) 

395 

396 dChi2 = 1 

397 iter = 0 

398 oldChi2 = 0 

399 fitFailure = False 

400 

401 def line_search(c, dx): 

402 testx = x - c * dx 

403 testLine = Line(testx[0], testx[1], testx[2]**-1) 

404 return self._lineChi2(testLine, grad=False) 

405 

406 while abs(dChi2) > dChi2Tol: 

407 line = Line(x[0], x[1], x[2]**-1) 

408 chi2, b, A = self._lineChi2(line) 

409 if chi2 == 0: 

410 break 

411 if not np.isfinite(A).all(): 

412 fitFailure = True 

413 if log is not None: 

414 log.warning("Hessian matrix has non-finite elements.") 

415 break 

416 dChi2 = oldChi2 - chi2 

417 try: 

418 cholesky = scipy.linalg.cho_factor(A) 

419 except np.linalg.LinAlgError: 

420 fitFailure = True 

421 if log is not None: 

422 log.warning("Hessian matrix is not invertible.") 

423 break 

424 dx = scipy.linalg.cho_solve(cholesky, b) 

425 

426 factor, fmin, _, _ = scipy.optimize.brent(line_search, args=(dx,), full_output=True, tol=0.05) 

427 x -= factor * dx 

428 if (abs(x[0]) > 1.5 * self._rhoMax) or (iter > maxIter): 

429 fitFailure = True 

430 break 

431 oldChi2 = chi2 

432 iter += 1 

433 

434 outline = Line(x[0], x[1], abs(x[2])**-1) 

435 

436 return outline, chi2, fitFailure 

437 

438 

439class MaskStreaksConfig(pexConfig.Config): 

440 """Configuration parameters for `MaskStreaksTask`. 

441 """ 

442 

443 minimumKernelHeight = pexConfig.Field( 

444 doc="Minimum height of the streak-finding kernel relative to the tallest kernel", 

445 dtype=float, 

446 default=0.0, 

447 ) 

448 absMinimumKernelHeight = pexConfig.Field( 

449 doc="Minimum absolute height of the streak-finding kernel", 

450 dtype=float, 

451 default=5, 

452 ) 

453 clusterMinimumSize = pexConfig.Field( 

454 doc="Minimum size in pixels of detected clusters", 

455 dtype=int, 

456 default=50, 

457 ) 

458 clusterMinimumDeviation = pexConfig.Field( 

459 doc="Allowed deviation (in pixels) from a straight line for a detected " 

460 "line", 

461 dtype=int, 

462 default=2, 

463 ) 

464 delta = pexConfig.Field( 

465 doc="Stepsize in angle-radius parameter space", 

466 dtype=float, 

467 default=0.2, 

468 ) 

469 nSigma = pexConfig.Field( 

470 doc="Number of sigmas from center of kernel to include in voting " 

471 "procedure", 

472 dtype=float, 

473 default=2, 

474 ) 

475 rhoBinSize = pexConfig.Field( 

476 doc="Binsize in pixels for position parameter rho when finding " 

477 "clusters of detected lines", 

478 dtype=float, 

479 default=30, 

480 ) 

481 thetaBinSize = pexConfig.Field( 

482 doc="Binsize in degrees for angle parameter theta when finding " 

483 "clusters of detected lines", 

484 dtype=float, 

485 default=2, 

486 ) 

487 invSigma = pexConfig.Field( 

488 doc="Inverse of the Moffat sigma parameter (in units of pixels)" 

489 "describing the profile of the streak", 

490 dtype=float, 

491 default=10.**-1, 

492 ) 

493 footprintThreshold = pexConfig.Field( 

494 doc="Threshold at which to determine edge of line, in units of " 

495 "nanoJanskys", 

496 dtype=float, 

497 default=0.01 

498 ) 

499 dChi2Tolerance = pexConfig.Field( 

500 doc="Absolute difference in Chi2 between iterations of line profile" 

501 "fitting that is acceptable for convergence", 

502 dtype=float, 

503 default=0.1 

504 ) 

505 detectedMaskPlane = pexConfig.Field( 

506 doc="Name of mask with pixels above detection threshold, used for first" 

507 "estimate of streak locations", 

508 dtype=str, 

509 default="DETECTED" 

510 ) 

511 streaksMaskPlane = pexConfig.Field( 

512 doc="Name of mask plane holding detected streaks", 

513 dtype=str, 

514 default="STREAK" 

515 ) 

516 

517 

518class MaskStreaksTask(pipeBase.Task): 

519 """Find streaks or other straight lines in image data. 

520 

521 Nearby objects passing through the field of view of the telescope leave a 

522 bright trail in images. This class uses the Kernel Hough Transform (KHT) 

523 (Fernandes and Oliveira, 2007), implemented in `lsst.houghtransform`. The 

524 procedure works by taking a binary image, either provided as put or produced 

525 from the input data image, using a Canny filter to make an image of the 

526 edges in the original image, then running the KHT on the edge image. The KHT 

527 identifies clusters of non-zero points, breaks those clusters of points into 

528 straight lines, keeps clusters with a size greater than the user-set 

529 threshold, then performs a voting procedure to find the best-fit coordinates 

530 of any straight lines. Given the results of the KHT algorithm, clusters of 

531 lines are identified and grouped (generally these correspond to the two 

532 edges of a strea) and a profile is fit to the streak in the original 

533 (non-binary) image. 

534 """ 

535 

536 ConfigClass = MaskStreaksConfig 

537 _DefaultName = "maskStreaks" 

538 

539 @timeMethod 

540 def find(self, maskedImage): 

541 """Find streaks in a masked image. 

542 

543 Parameters 

544 ---------- 

545 maskedImage : `lsst.afw.image.maskedImage` 

546 The image in which to search for streaks. 

547 

548 Returns 

549 ------- 

550 result : `lsst.pipe.base.Struct` 

551 Results as a struct with attributes: 

552 

553 ``originalLines`` 

554 Lines identified by kernel hough transform. 

555 ``lineClusters`` 

556 Lines grouped into clusters in rho-theta space. 

557 ``lines`` 

558 Final result for lines after line-profile fit. 

559 ``mask`` 

560 2-d boolean mask where detected lines are True. 

561 """ 

562 mask = maskedImage.getMask() 

563 detectionMask = (mask.array & mask.getPlaneBitMask(self.config.detectedMaskPlane)) 

564 

565 self.edges = self._cannyFilter(detectionMask) 

566 self.lines = self._runKHT(self.edges) 

567 

568 if len(self.lines) == 0: 

569 lineMask = np.zeros(detectionMask.shape, dtype=bool) 

570 fitLines = LineCollection([], []) 

571 clusters = LineCollection([], []) 

572 else: 

573 clusters = self._findClusters(self.lines) 

574 fitLines, lineMask = self._fitProfile(clusters, maskedImage) 

575 

576 # The output mask is the intersection of the fit streaks and the image detections 

577 outputMask = lineMask & detectionMask.astype(bool) 

578 

579 return pipeBase.Struct( 

580 lines=fitLines, 

581 lineClusters=clusters, 

582 originalLines=self.lines, 

583 mask=outputMask, 

584 ) 

585 

586 @timeMethod 

587 def run(self, maskedImage): 

588 """Find and mask streaks in a masked image. 

589 

590 Finds streaks in the image and modifies maskedImage in place by adding a 

591 mask plane with any identified streaks. 

592 

593 Parameters 

594 ---------- 

595 maskedImage : `lsst.afw.image.maskedImage` 

596 The image in which to search for streaks. The mask detection plane 

597 corresponding to `config.detectedMaskPlane` must be set with the 

598 detected pixels. 

599 

600 Returns 

601 ------- 

602 result : `lsst.pipe.base.Struct` 

603 Results as a struct with attributes: 

604 

605 ``originalLines`` 

606 Lines identified by kernel hough transform. 

607 ``lineClusters`` 

608 Lines grouped into clusters in rho-theta space. 

609 ``lines`` 

610 Final result for lines after line-profile fit. 

611 """ 

612 streaks = self.find(maskedImage) 

613 

614 maskedImage.mask.addMaskPlane(self.config.streaksMaskPlane) 

615 maskedImage.mask.array[streaks.mask] |= maskedImage.mask.getPlaneBitMask(self.config.streaksMaskPlane) 

616 

617 return pipeBase.Struct( 

618 lines=streaks.lines, 

619 lineClusters=streaks.lineClusters, 

620 originalLines=streaks.originalLines, 

621 ) 

622 

623 def _cannyFilter(self, image): 

624 """Apply a canny filter to the data in order to detect edges. 

625 

626 Parameters 

627 ---------- 

628 image : `np.ndarray` 

629 2-d image data on which to run filter. 

630 

631 Returns 

632 ------- 

633 cannyData : `np.ndarray` 

634 2-d image of edges found in input image. 

635 """ 

636 # Ensure that the pixels are zero or one. Change the datatype to 

637 # np.float64 to be compatible with the Canny filter routine. 

638 filterData = (image > 0).astype(np.float64) 

639 return canny(filterData, use_quantiles=True, sigma=0.1) 

640 

641 def _runKHT(self, image): 

642 """Run Kernel Hough Transform on image. 

643 

644 Parameters 

645 ---------- 

646 image : `np.ndarray` 

647 2-d image data on which to detect lines. 

648 

649 Returns 

650 ------- 

651 result : `LineCollection` 

652 Collection of detected lines, with their detected rho and theta 

653 coordinates. 

654 """ 

655 lines = lsst.kht.find_lines(image, self.config.clusterMinimumSize, 

656 self.config.clusterMinimumDeviation, self.config.delta, 

657 self.config.minimumKernelHeight, self.config.nSigma, 

658 self.config.absMinimumKernelHeight) 

659 self.log.info("The Kernel Hough Transform detected %s line(s)", len(lines)) 

660 

661 return LineCollection(lines.rho, lines.theta) 

662 

663 def _findClusters(self, lines): 

664 """Group lines that are close in parameter space and likely describe 

665 the same streak. 

666 

667 Parameters 

668 ---------- 

669 lines : `LineCollection` 

670 Collection of lines to group into clusters. 

671 

672 Returns 

673 ------- 

674 result : `LineCollection` 

675 Average `Line` for each cluster of `Line`s in the input 

676 `LineCollection`. 

677 """ 

678 # Scale variables by threshold bin-size variable so that rho and theta 

679 # are on the same scale. Since the clustering algorithm below stops when 

680 # the standard deviation <= 1, after rescaling each cluster will have a 

681 # standard deviation at or below the bin-size. 

682 x = lines.rhos / self.config.rhoBinSize 

683 y = lines.thetas / self.config.thetaBinSize 

684 X = np.array([x, y]).T 

685 nClusters = 1 

686 

687 # Put line parameters in clusters by starting with all in one, then 

688 # subdividing until the parameters of each cluster have std dev=1. 

689 # If nClusters == len(lines), each line will have its own 'cluster', so 

690 # the standard deviations of each cluster must be zero and the loop 

691 # is guaranteed to stop. 

692 while True: 

693 kmeans = KMeans(n_clusters=nClusters, n_init='auto').fit(X) 

694 clusterStandardDeviations = np.zeros((nClusters, 2)) 

695 for c in range(nClusters): 

696 inCluster = X[kmeans.labels_ == c] 

697 clusterStandardDeviations[c] = np.std(inCluster, axis=0) 

698 # Are the rhos and thetas in each cluster all below the threshold? 

699 if (clusterStandardDeviations <= 1).all(): 

700 break 

701 nClusters += 1 

702 

703 # The cluster centers are final line estimates 

704 finalClusters = kmeans.cluster_centers_.T 

705 

706 # Rescale variables: 

707 finalRhos = finalClusters[0] * self.config.rhoBinSize 

708 finalThetas = finalClusters[1] * self.config.thetaBinSize 

709 result = LineCollection(finalRhos, finalThetas) 

710 self.log.info("Lines were grouped into %s potential streak(s)", len(finalRhos)) 

711 

712 return result 

713 

714 def _fitProfile(self, lines, maskedImage): 

715 """Fit the profile of the streak. 

716 

717 Given the initial parameters of detected lines, fit a model for the 

718 streak to the original (non-binary image). The assumed model is a 

719 straight line with a Moffat profile. 

720 

721 Parameters 

722 ---------- 

723 lines : `LineCollection` 

724 Collection of guesses for `Line`s detected in the image. 

725 maskedImage : `lsst.afw.image.maskedImage` 

726 Original image to be used to fit profile of streak. 

727 

728 Returns 

729 ------- 

730 lineFits : `LineCollection` 

731 Collection of `Line` profiles fit to the data. 

732 finalMask : `np.ndarray` 

733 2d mask array with detected streaks=1. 

734 """ 

735 data = maskedImage.image.array 

736 weights = maskedImage.variance.array**-1 

737 # Mask out any pixels with non-finite weights 

738 weights[~np.isfinite(weights) | ~np.isfinite(data)] = 0 

739 

740 lineFits = LineCollection([], []) 

741 finalLineMasks = [np.zeros(data.shape, dtype=bool)] 

742 nFinalLines = 0 

743 for line in lines: 

744 line.sigma = self.config.invSigma**-1 

745 lineModel = LineProfile(data, weights, line=line) 

746 # Skip any lines that do not cover any data (sometimes happens because of chip gaps) 

747 if lineModel.lineMaskSize == 0: 

748 continue 

749 

750 fit, chi2, fitFailure = lineModel.fit(dChi2Tol=self.config.dChi2Tolerance, log=self.log) 

751 if fitFailure: 

752 self.log.warning("Streak fit failed.") 

753 

754 # Initial estimate should be quite close: fit is deemed unsuccessful if rho or theta 

755 # change more than the allowed bin in rho or theta: 

756 if ((abs(fit.rho - line.rho) > 2 * self.config.rhoBinSize) 

757 or (abs(fit.theta - line.theta) > 2 * self.config.thetaBinSize)): 

758 fitFailure = True 

759 self.log.warning("Streak fit moved too far from initial estimate. Line will be dropped.") 

760 

761 if fitFailure: 

762 continue 

763 

764 self.log.debug("Best fit streak parameters are rho=%.2f, theta=%.2f, and sigma=%.2f", fit.rho, 

765 fit.theta, fit.sigma) 

766 

767 # Make mask 

768 lineModel.setLineMask(fit) 

769 finalModel = lineModel.makeProfile(fit) 

770 # Take absolute value, as streaks are allowed to be negative 

771 finalModelMax = abs(finalModel).max() 

772 finalLineMask = abs(finalModel) > self.config.footprintThreshold 

773 # Drop this line if the model profile is below the footprint threshold 

774 if not finalLineMask.any(): 

775 continue 

776 fit.chi2 = chi2 

777 fit.finalModelMax = finalModelMax 

778 lineFits.append(fit) 

779 finalLineMasks.append(finalLineMask) 

780 nFinalLines += 1 

781 

782 finalMask = np.array(finalLineMasks).any(axis=0) 

783 nMaskedPixels = finalMask.sum() 

784 percentMasked = (nMaskedPixels / finalMask.size) * 100 

785 self.log.info("%d streak(s) fit, with %d pixels masked (%0.2f%% of image)", nFinalLines, 

786 nMaskedPixels, percentMasked) 

787 

788 return lineFits, finalMask