Remote sensing digital image analysis sixth edition john alan richards 2024 Scribd Download

Page 1


https://ebookmass.com/product/remote-sensing-digital-imageanalysis-sixth-edition-john-alan-richards/

Instant digital products (PDF, ePub, MOBI) ready for you

Download now and discover formats that fit your needs...

Remote Sensing Digital Image Analysis 6th ed. 2022 Edition

John A. Richards

https://ebookmass.com/product/remote-sensing-digital-imageanalysis-6th-ed-2022-edition-john-a-richards/

ebookmass.com

Remote Sensing and Image Interpretation, 7th Edition 7th Edition, (Ebook PDF)

https://ebookmass.com/product/remote-sensing-and-imageinterpretation-7th-edition-7th-edition-ebook-pdf/

ebookmass.com

Thermal and Optical Remote Sensing John O Odindi

https://ebookmass.com/product/thermal-and-optical-remote-sensing-johno-odindi/

ebookmass.com

The Case Writer’s Toolkit 1st ed. Edition June Gwee

https://ebookmass.com/product/the-case-writers-toolkit-1st-ed-editionjune-gwee/

ebookmass.com

Discovering Psychology: The Science of Mind 2nd Edition, (Ebook PDF)

https://ebookmass.com/product/discovering-psychology-the-science-ofmind-2nd-edition-ebook-pdf/

ebookmass.com

May We Rise: A Dark College Bully Romance (A Mayfair University Novel Book 1) K.G. Reuss

https://ebookmass.com/product/may-we-rise-a-dark-college-bullyromance-a-mayfair-university-novel-book-1-k-g-reuss/

ebookmass.com

Technology Ventures: From Idea to Enterprise 5th Edition, (Ebook PDF)

https://ebookmass.com/product/technology-ventures-from-idea-toenterprise-5th-edition-ebook-pdf/

ebookmass.com

Navy Lies: A Novel (White lies can be harmless, but the most dangerous are...) Monica Arya

https://ebookmass.com/product/navy-lies-a-novel-white-lies-can-beharmless-but-the-most-dangerous-are-monica-arya/

ebookmass.com

Spine

Secrets 3rd Edition Devlin Vincent J. (Ed.)

https://ebookmass.com/product/spine-secrets-3rd-edition-devlinvincent-j-ed/

ebookmass.com

The Deepest Black Randall Silvis

https://ebookmass.com/product/the-deepest-black-randall-silvis-2/

ebookmass.com

RemoteSensingDigitalImageAnalysis

JohnA.Richards RemoteSensingDigital ImageAnalysis

SixthEdition

TheAustralianNationalUniversity Canberra,ACT,Australia

ISBN978-3-030-82326-9ISBN978-3-030-82327-6(eBook) https://doi.org/10.1007/978-3-030-82327-6

1st –5th editions:©Springer-VerlagBerlinHeidelberg1986,1993,1999,2006,2013 6th edition:©TheEditor(s)(ifapplicable)andTheAuthor(s),underexclusivelicensetoSpringer NatureSwitzerlandAG2022

Thisworkissubjecttocopyright.AllrightsaresolelyandexclusivelylicensedbythePublisher,whether thewholeorpartofthematerialisconcerned,specificallytherightsoftranslation,reprinting,reuse ofillustrations,recitation,broadcasting,reproductiononmicrofilmsorinanyotherphysicalway,and transmissionorinformationstorageandretrieval,electronicadaptation,computersoftware,orbysimilar ordissimilarmethodologynowknownorhereafterdeveloped.

Theuseofgeneraldescriptivenames,registerednames,trademarks,servicemarks,etc.inthispublication doesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfromtherelevant protectivelawsandregulationsandthereforefreeforgeneraluse.

Thepublisher,theauthorsandtheeditorsaresafetoassumethattheadviceandinformationinthisbook arebelievedtobetrueandaccurateatthedateofpublication.Neitherthepublishernortheauthorsor theeditorsgiveawarranty,expressedorimplied,withrespecttothematerialcontainedhereinorforany errorsoromissionsthatmayhavebeenmade.Thepublisherremainsneutralwithregardtojurisdictional claimsinpublishedmapsandinstitutionalaffiliations.

ThisSpringerimprintispublishedbytheregisteredcompanySpringerNatureSwitzerlandAG Theregisteredcompanyaddressis:Gewerbestrasse11,6330Cham,Switzerland

Thisbookisdedicatedtothememoryofthe lateDavidLandgrebe,ProfessorEmeritusof PurdueUniversity.

Asateacher,mentor,friendandcolleagueto generationsofengineersandscientistsin remotesensing,Davetouchedandinfluenced thelivesandcareersofmanytheworldover.

Thedevelopmentofquantitativemethodsfor theanalysisofremotesensingimagedata owesmuchtoDave’sleadership.Helaidthe foundationsfortheapplicationof classificationtechniquestotheearthsciences thatweusetoday.

DavidLandgrebewasoneofthetrueand inspirationalpioneersofourfield.

Preface

Inafieldlikeremotesensingimageanalysis,whichchangessoquickly,revisinga long-standingtextbookisachallenge.Itisimportanttoincorporatecontemporary techniques,whilenotdiscardingproceduresfromthepastwhich,whileapparently supersededbynewermethods,neverthelessretainvalueandareoftensimplertouse. Also,someprocessingoperationsfromthepastcanbecomeimportantagainasdata typesandvolumeschange.Streamingmethodsforclusteringareanexamplewith thetrendnowtoverylargedatasets.

Aswiththepreviousedition,judgementshavehadtobemadeaboutwhattoleave out,whattoretainandwhattoadd.Thosejudgementshavebeenmadeagainstthe intendedpurposeofthebook.Fromthebeginning,ithasbeendesignedasateaching textfortheseniorundergraduateandpostgraduatestudent,andasafundamental treatmentforthoseengagedintheapplicationofdigitalimageanalysisinremote sensingprojectsorinremotesensingimageprocessingresearch.

Thepresentationlevelisforthemathematicalnon-specialist.Becausemostoperationalusersofremotesensingcomefromtheearthsciencescommunities,thetext ispitchedatalevelcommensuratewiththeirbackground.Thatisimportantbecause therecognisedauthoritiesindigitalimageanalysisandmachinelearningtendto befromengineering,computerscienceandmathematics.Althoughfamiliaritywith acertainlevelofmathematicsandstatisticscannotbeavoided,thetreatmenthere progressesthroughanalysescarefully,withmanyhand-workedexamples,sothatany lackofdepthinmathematicalbackgroundshouldnottakeawayfromunderstanding theimportantaspectsofimageanalysisandinterpretation.

Althoughtheprincipalfocusofthetreatmentisondigitalimageinterpretationand theanalyticaltechniquesthatmakethatpossible,thematerialislocatedwithinthe domainofremotesensingapplications.Thatmeansprojectobjectivesareasimportantasfindingthebest-performingalgorithm.Algorithmsneedtobeincorporated intomethodologiesthatcangenerateoptimalresultsfromacarefulcombinationof procedures,andinwhichthestepsofchoosingreferencematerialtosupportthe processandforassessingaccuracy,maybejustasimportantasalgorithmperformance.Whilealgorithmperformanceisakeyobjectiveinthemachinelearning

remotesensingresearchcommunity,itisprojectoutcomesthatdrivetheremote sensingapplicationsspecialist.Thatisakeyemphasisofthisbook.

Althoughthechapterscanbeusedindividually,thematerialispresentedina sequentialmanner.Apersonwithlittleornobackgroundinremotesensingimage interpretationcanstartwiththeearlychaptersinordertoappreciatekeyconceptsin remotesensingandimageformation,howerrorsariseinrecordedimageryandhow theycanbecorrected.Theremainingchaptersthenworkprogressivelythroughthe majoranalyticalmethodsfundamentaltodigitalimageanalysis,finishingupwith meansbywhichmethodologiescanbedevisedtotackleremotesensingprojects.

Overtheyears,manypeoplehaveeitherdirectlyorindirectlycontributedtothis book.ThelateDavidLandgrebe,towhomthiseditionisdedicated,wasafriendand colleaguewhodidmuchtoshapemythinkingabouttheapplicationofquantitative methodsinremotesensing.Hepioneeredmanyoftheideasthatendedupinone wayoranotherinpartsofthisbook.

MycolleagueAssociateProfessorXiupingJiahasbeenagreatcollaboratorover theyears,commencingwhensheundertookherPh.D.Manyofthemethodspresented herehavebeentheresultofafruitfulresearchpartnershipforwhichIexpressmy sinceregratitudetoher.

Dr.TerryCocks,formerManagingDirectorofHyVistaCorporationPtyLtd, Australia,kindlymadeavailableHyMaphyperspectralimageryofPerth,Western Australia,toallowmanyoftheexamplescontainedinthisandthepreviousedition tobegenerated.

IamindebtedtoJasonBrownofCapellaSpacewithwhoseencouragementthis sixtheditionwasprepared;otherwise,itmaynothavehappened.Healsokindly providedtheradarimageryusedinChap. 1

Lastly,Iacknowledgethededication,supportandencouragementofmywife Glenda.Herperseveranceandunderstandinghavebeenenormouslyimportant, andhavemadethejobofwritingthisneweditionfulfillingandsatisfying, notwithstandingthedemandsitmadeonfamilytime.

Canberra,Australia June2021

1SourcesandCharacteristicsofRemoteSensingImageData .......1

1.1EnergySourcesandWavelengthRanges....................1

1.2PrimaryDataCharacteristics..............................5

1.3RemoteSensingPlatforms................................6

1.4WhatEarthSurfacePropertiesAreMeasured?...............11

1.4.1SensingintheVisibleandReflectedInfrared Ranges........................................11

1.4.2SensingintheThermalInfraredRange............14

1.4.3SensingintheMicrowaveRange..................14

1.5SpatialDataSourcesinGeneralandGeographic InformationSystems.....................................18

1.6ScaleinDigitalImageData...............................21

1.7DigitalEarth...........................................21

1.8HowThisBookIsArranged..............................23

1.9BibliographyonSourcesandCharacteristicsofRemote SensingImageData.....................................25

2.2SourcesofRadiometricDistortion.........................32 2.3InstrumentationErrors...................................32

2.4EffectoftheSolarRadiationCurveandtheAtmosphere onRadiometry..........................................35

2.5CompensatingfortheSolarRadiationCurve................37

2.6InfluenceoftheAtmosphere..............................38

2.7EffectoftheAtmosphereonRemoteSensingImagery........42

2.8CorrectingAtmosphericEffectsinBroadWaveband

2.9CorrectingAtmosphericEffectsinNarrowWaveband Systems...............................................45

2.10Empirical,DataDrivenMethodsforAtmospheric Correction.............................................49

2.10.1HazeRemovalbyDarkSubtraction...............50

2.10.2TheFlatFieldMethod...........................50

2.10.3TheEmpiricalLineMethod......................51

2.10.4LogResiduals..................................52

2.11SourcesofGeometricDistortion...........................53

2.12TheEffectofEarthRotation..............................54

2.13TheEffectofVariationsinPlatformAltitude,Attitude andVelocity............................................56

2.14TheEffectofSensorFieldofView:PanoramicDistortion.....56

2.15TheEffectofEarthCurvature.............................59

2.16GeometricDistortionCausedbyInstrumentation Characteristics..........................................60

2.16.1SensorScanNonlinearities.......................61

2.16.2FiniteScanTimeDistortion......................61

2.16.3AspectRatioDistortion..........................61

2.17CorrectionofGeometricDistortion........................62

2.18UseofMappingFunctionsforImageCorrection.............62

2.18.1MappingPolynomialsandtheUseofGround ControlPoints..................................63

2.18.2BuildingaGeometricallyCorrectImage...........64

2.18.3ResamplingandtheNeedforInterpolation.........65

2.18.4TheChoiceofControlPoints.....................67

2.18.5ExampleofRegistrationtoaMapGrid............68

2.19MathematicalRepresentationandCorrection ofGeometricDistortion..................................70

2.19.1AspectRatioCorrection.........................70

2.19.2EarthRotationSkewCorrection..................71

2.19.3ImageOrientationtoNorth–South................72

2.19.4CorrectingPanoramicEffects....................72

2.19.5CombiningtheCorrections.......................72

2.20ImagetoImageRegistration..............................73

2.20.1RefiningtheLocalisationofControlPoints.........73

2.20.2ExampleofImagetoImageRegistration...........75

2.21OtherImageGeometryOperations.........................78

2.21.1ImageRotation.................................78

2.21.2ScaleChangingandZooming....................78

2.22BibliographyonCorrectingandRegisteringImages..........79

2.23Problems...............................................80

3InterpretingImages ...........................................87

3.1Introduction............................................87

3.2Photointerpretation......................................88

3.2.1FormsofImageryforPhotointerpretation..........89

3.2.2ComputerEnhancementofImagery forPhotointerpretation..........................90

3.3QuantitativeAnalysis:FromDatatoLabels.................91

3.4ComparingQuantitativeAnalysisandPhotointerpretation.....93

3.5TheFundamentalsofQuantitativeAnalysis.................94

3.5.1PixelVectorsandSpectralSpace..................94

3.5.2LinearClassifiers...............................98

3.5.3StatisticalClassifiers............................98

3.6Sub-classesandSpectralClasses..........................101

3.7UnsupervisedClassification..............................103

3.8BibliographyonInterpretingImages.......................103

3.9Problems...............................................104

4RadiometricEnhancementofImages ...........................107

4.1Introduction............................................107

4.1.1PointOperationsandLookUpTables.............107

4.1.2ScalarandVectorImages........................108

4.2TheImageHistogram....................................108

4.3ContrastModification....................................109

4.3.1HistogramModificationRule.....................109

4.3.2LinearContrastModification.....................110

4.3.3SaturatingLinearContrastEnhancement...........111

4.3.4AutomaticContrastEnhancement.................112

4.3.5LogarithmicandExponentialContrast Enhancement..................................113

4.3.6PiecewiseLinearContrastModification............113

4.4HistogramEqualisation..................................113

4.4.1UseoftheCumulativeHistogram.................113

4.4.2AnomaliesinHistogramEqualisation.............120

4.5HistogramMatching.....................................122

4.5.1Principle......................................122

4.5.2ImagetoImageContrastMatching................123

4.5.3MatchingtoaMathematicalReference............126

4.6DensitySlicing.........................................126

4.6.1BlackandWhiteDensitySlicing..................126

4.6.2ColourDensitySlicingandPseudocolouring.......127

4.7BibliographyonRadiometricEnhancementofImages........129

4.8Problems...............................................131

5GeometricProcessingandEnhancement:ImageDomain Techniques ...................................................135

5.1Introduction............................................135

5.2NeighbourhoodOperationsinImageFiltering...............136

5.3ImageSmoothing.......................................138

5.3.1MeanValueSmoothing..........................138

5.3.2MedianFiltering................................139

5.3.3ModalFiltering................................140

5.4SharpeningandEdgeDetection...........................140

5.4.1SpatialGradientMethods........................141

5.4.1.1TheRobertsOperator..................143

5.4.1.2TheSobelOperator....................143

5.4.1.3ThePrewittOperator...................144

5.4.1.4TheLaplacianOperator................145

5.4.2SubtractiveSmoothing(UnsharpMasking).........147

5.5EdgeDetection.........................................147

5.6LineandSpotDetection..................................150

5.7ThinningandLinking....................................150

5.8GeometricProcessingasaConvolutionOperation...........151

5.9ImageDomainTechniquesComparedwithUsing theFourierTransform....................................153

5.10GeometricPropertiesofImages...........................154

5.10.1MeasuringGeometricProperties..................155

5.10.2DescribingTexture..............................156

5.11MorphologicalAnalysis..................................159

5.11.1Erosion.......................................161

5.11.2Dilation.......................................162

5.11.3OpeningandClosing............................163

5.11.4BoundaryExtraction............................164

5.11.5OtherMorphologicalOperations andApplications...............................166

5.12ObjectandShapeRecognition............................166

5.13BibliographyonGeometricProcessingandEnhancement: ImageDomainTechniques...............................167

5.14Problems...............................................168

6SpectralDomainImageTransforms .............................171

6.1Introduction............................................171

6.2ImageArithmeticandVegetationIndices...................172

6.3ThePrincipalComponentsTransform......................174

6.3.1TheMeanVectorandtheCovarianceMatrix........174

6.3.2AZeroCorrelation,RotationalTransform..........179

6.3.3TheEffectofanOriginShift.....................184

6.3.4ExampleandSomePracticalConsiderations........185

6.3.5ApplicationofPrincipalComponentsinImage EnhancementandDisplay.......................187

6.3.6TheTaylorMethodofContrastEnhancement.......189

6.3.7UseofPrincipalComponentsforImage Compression...................................193

6.3.8ThePrincipalComponentsTransform inChangeDetectionApplications.................194

6.3.9UseofPrincipalComponentsforFeature Reduction.....................................198

6.4TheNoiseAdjustedPrincipalComponentsTransform........198

6.5TheKauth-ThomasTasseledCapTransform................201

6.6TheKernelPrincipalComponentsTransform................205

6.7HSIImageDisplay......................................208

6.8PanSharpening.........................................210

6.9BibliographyonSpectralDomainImageTransforms.........211 6.10Problems...............................................212

7.1Introduction............................................217

7.2SpecialFunctions.......................................218

7.2.1TheComplexExponentialFunction...............218

7.2.2TheImpulseorDeltaFunction...................220

7.2.3TheHeavisideStepFunction.....................221

7.3TheFourierSeries.......................................222

7.4TheFourierTransform...................................224

7.5TheDiscreteFourierTransform...........................227

7.5.1PropertiesoftheDiscreteFourierTransform........229

7.5.2ComputingtheDiscreteFourierTransform.........230

7.6Convolution............................................230

7.6.1TheConvolutionIntegral........................230

7.6.2ConvolutionwithanImpulse.....................231

7.6.3TheConvolutionTheorem.......................231

7.6.4DiscreteConvolution............................232

7.7SamplingTheory........................................233

7.8TheDiscreteFourierTransformofanImage................236

7.8.1TheTransformEquations........................236

7.8.2EvaluatingtheFourierTransformofanImage......237

7.8.3TheConceptofSpatialFrequency................238

7.8.4DisplayingtheDFTofanImage..................238

7.9ImageProcessingUsingtheFourierTransform..............239

7.10ConvolutioninTwoDimensions...........................241

7.11OtherFourierTransforms................................242

7.12LeakageandWindowFunctions...........................243

7.13TheWaveletTransform..................................244

7.13.1Background....................................244

7.13.2OrthogonalFunctionsandInnerProducts..........245

7.13.3WaveletsasBasisFunctions......................246

7.13.4DyadicWaveletswithCompactSupport...........247

7.13.5ChoosingtheWavelets..........................248

7.13.6FilterBanks...................................248

7.13.6.1SubBandFiltering, andDownsampling....................248

7.13.6.2ReconstructionfromtheWavelets, andUpsampling.......................252

7.13.6.3RelationshipBetweentheLow andHighPassFilters...................253

7.13.7ChoiceofWavelets.............................254

7.14TheWaveletTransformofanImage.......................256

7.15ApplicationsoftheWaveletTransforminRemote SensingImageAnalysis..................................257

7.16BibliographyonSpatialDomainImageTransforms..........259

7.17Problems...............................................260

8SupervisedClassificationTechniques

8.1Introduction............................................263

8.2TheEssentialStepsinSupervisedClassification.............264

8.3MaximumLikelihoodClassification.......................267

8.3.1Bayes’Classification............................267

8.3.2TheMaximumLikelihoodDecisionRule..........267

8.3.3MultivariateNormalClassModels................269

8.3.4DecisionSurfaces..............................270

8.3.5Thresholds....................................271

8.3.6NumberofTrainingPixelsRequired..............273

8.3.7TheHughesPhenomenonandtheCurse ofDimensionality..............................274

8.3.8AnExample...................................276

8.4GaussianMixtureModels................................278

8.5MinimumDistanceClassification..........................284

8.5.1TheCaseofLimitedTrainingData................284

8.5.2TheDiscriminantFunction.......................285

8.5.3DecisionSurfacesfortheMinimumDistance Classifier......................................286

8.5.4Thresholds....................................286

8.5.5DegenerationofMaximumLikelihood toMinimumDistanceClassification...............286

8.5.6ClassificationTimeComparison oftheMaximumLikelihoodandMinimum DistanceRules.................................287

8.6ParallelepipedClassification..............................288

8.7MahalanobisClassification...............................289

8.8Non-parametricClassification.............................290

8.9TableLookUpClassification.............................291

8.10 k NN(NearestNeighbour)Classification....................291

8.11TheSpectralAngleMapper...............................292

8.12Non-parametricClassificationfromaGeometricBasis........293

8.12.1LinearClassificationandtheConcept ofaWeightVector..............................293

8.12.2TestingClassMembership.......................294

8.13TrainingaLinearClassifier...............................295

8.14TheSupportVectorMachine:LinearlySeparableClasses.....295

8.15TheSupportVectorMachine:OverlappingClasses...........300

8.16TheSupportVectorMachine:NonlinearlySeparable DataandKernels........................................303

8.17Multi-categoryClassificationwithBinaryClassifiers.........306

8.18ApplyingtheSupportVectorClassifier.....................307

8.18.1InitialChoices.................................307

8.18.2GridSearchingforParameterDetermination........308 8.18.3DataCenteringandScaling......................309

8.18.4Examples......................................309

8.19CommitteesofClassifiers................................312

8.19.1Bagging.......................................313

8.19.2BoostingandAdaBoost.........................313

8.20NetworksofClassifiers:TheArtificialNeuralNetwork.......315

8.20.1TheProcessingElement.........................316

8.20.2TrainingtheNeural Network—Backpropagation......................317

8.20.3ChoosingtheNetworkParameters................323 8.20.4Example......................................323

8.21TheConvolutionalNeuralNetwork........................326

8.21.1TheBasicTopologyoftheConvolutional NeuralNetwork................................328

8.21.2DetectingSpatialStructure.......................332

8.21.3Stride.........................................332

8.21.4PoolingorDown-Sampling......................333

8.21.5TheReLUActivationFunction...................333

8.21.6HandlingtheOutputsofaCNN...................334

8.21.7MultipleFiltersintheConvolutionLayer..........335

8.21.8SimplifiedRepresentationoftheCNN.............336

8.21.9MultispectralandHyperspectralInputs toaCNN......................................336

8.21.10ASpectral-SpatialExampleoftheUse oftheCNN....................................339

8.21.11AvoidingOverfitting............................340

8.21.12Variations.....................................341

8.22RecurrentNeuralNetworks...............................343

8.22.1Multi-temporalRemoteSensing..................343 8.22.2ImportanceofMemory..........................343

8.22.3TheRecurrentNeuralNetwork(RNN) Architecture...................................344

8.22.4TrainingtheRNN..............................346

8.23ContextClassification....................................346

8.23.1TheConceptofSpatialContext...................346

8.23.2ContextClassificationbyImagePre-processing.....348

8.23.3PostClassificationFiltering......................349

8.23.5HandlingSpatialContextbyMarkovRandom

8.24BibliographyonSupervisedClassificationTechniques........359

9.2SimilarityMetricsandClusteringCriteria...................370

9.3 k MeansClustering......................................372

9.3.1The k MeansAlgorithm.........................373 9.4IsodataClustering.......................................374

9.4.1MergingandDeletingClusters...................375

9.4.2SplittingElongatedClusters......................375

9.5ChoosingtheInitialClusterCentres........................375

9.6Costof k MeansandIsodataClustering....................376

9.7UnsupervisedClassification..............................376

9.8AnExampleofClusteringwiththe k MeansAlgorithm.......377

9.9ASinglePassClusteringTechnique........................378

9.9.1TheSinglePassAlgorithm.......................379

9.9.2AdvantagesandLimitationsoftheSinglePass Algorithm.....................................380

9.9.3StripGenerationParameter......................381

9.9.4VariationsontheSinglePassAlgorithm...........381

9.9.5AnExampleofClusteringwiththeSinglePass Algorithm.....................................381

9.10HierarchicalClustering..................................383

9.10.1AgglomerativeHierarchicalClustering............383

9.11OtherClusteringMetrics.................................383

9.12SomeAlternativeClusteringTechniques....................385

9.12.1HistogramPeakSelection........................385

9.12.2MountainClustering............................385

9.12.3kMediansClustering...........................386

9.12.4kMedoidsClustering...........................386

9.13ClusteringLargeDataSets...............................388

9.13.1TheKTreesAlgorithm..........................389

9.13.2DBSCAN.....................................393

9.14ClusterSpaceClassification..............................395

9.15BibliographyonClusteringandUnsupervised Classification...........................................399

10.1TheNeedforFeatureReduction...........................403

10.2ApproachestoFeatureReduction..........................405

10.3FeatureReductionbySpectralTransforms..................406

10.3.1FeatureReductionUsingthePrincipal ComponentsTransform..........................406

10.3.2FeatureReductionUsingtheCanonical AnalysisTransform.............................409

10.3.2.1Within-ClassandAmong-Class Covariance...........................409

10.3.2.2ASeparabilityMeasure................411

10.3.2.3TheGeneralisedEigenvalue Equation.............................411

10.3.2.4AnExample..........................413

10.3.3DiscriminantAnalysisFeatureExtraction (DAFE).......................................415

10.3.4Non-parametricDiscriminantAnalysis(NDA)......417

10.3.5DecisionBoundaryFeatureExtraction(DBFE).....421

10.3.6Non-parametricWeightedFeatureExtraction (NWFE).......................................422

10.4FeatureReductionbyBlockDiagonalisingtheCovariance Matrix.................................................423

10.5FeatureSelection........................................429

10.5.1MeasuresofSeparability........................429

10.5.2Divergence....................................430

10.5.2.1Definition............................430

10.5.2.2DivergenceofaPairofNormal Distributions..........................432

10.5.2.3UsingDivergenceforFeature Selection.............................432

10.5.2.4AProblemwithDivergence.............433

10.5.3TheJeffries-Matusita(JM)Distance...............433

10.5.3.1Definition............................433

10.5.3.2ComparisonofDivergenceandJM Distance.............................435

10.5.4TransformedDivergence.........................436

10.5.4.1Definition............................436

10.5.4.2TransformedDivergence andtheProbabilityofCorrect Classification.........................436

10.5.4.3UseofTransformedDivergence inClustering..........................437

10.5.5SeparabilityMeasuresforMinimumDistance Classification..................................437

10.6DistributionFreeFeatureSelection—ReliefF................438

11.2AnOverviewofClassification............................448

11.2.1SupervisedClassification........................448

11.2.1.1SelectionofTrainingData..............448 11.2.1.2FeatureSelection......................449

11.2.1.3ClassifierOutputsandAccuracy Checking.............................450

11.2.2UnsupervisedClassification......................450

11.2.3Semi-supervisedClassificationandTransfer Learning......................................452

11.3EffectofResamplingonClassification.....................453

11.4AHybridSupervised/UnsupervisedMethodology............454

11.4.1OutlineoftheMethod...........................454

11.4.2ChoosingtheImageSegmentstoCluster...........455

11.4.3RationalisingtheNumberofSpectralClasses.......456

11.4.4AnExample...................................456

11.4.5HybridClassificationwithOtherSupervised Algorithms....................................459

11.5ClusterSpaceClassification..............................461

11.6AssessingClassificationAccuracy.........................462

11.6.1UseofaTestingSetofPixels.....................462

11.6.2TheErrorMatrix...............................463

11.6.3QuantifyingtheErrorMatrix.....................464

11.6.4TheKappaCoefficient..........................467

11.6.5NumberofTestingSamplesRequired forAssessingMapAccuracy.....................472

11.6.6NumberofTestingSamplesRequired forPopulatingtheErrorMatrix...................476

11.6.7PlacingConfidenceLimitsonAssessed Accuracy......................................478

11.6.8CrossValidationAccuracyAssessment andtheLeaveOneOutMethod...................479

11.7DecisionTreeClassifiers.................................479

11.7.1CART(ClassificationandRegressionTrees)........482

11.7.2RandomForests................................485

11.7.3ProgressiveTwo-ClassDecisionClassifier.........487

11.8ImageInterpretationThroughSpectroscopyandSpectral LibrarySearching.......................................488

11.9EndMembersandUnmixing..............................490

11.10IsThereaBestClassifier?................................492

11.10.1SegmentingtheSpectralSpace...................492

11.10.2ComparingtheClassifiers........................494

11.11BibliographyonImageClassificationinPractice.............497

11.12Problems...............................................500

12MultisourceImageAnalysis

12.2StackedVectorAnalysis..................................504

12.3StatisticalMultisourceMethods...........................505

12.3.1JointStatisticalDecisionRules...................505

12.3.2CommitteeClassifiers...........................507

12.3.3OpinionPoolsandConsensusTheory.............508

12.3.4UseofPriorProbabilities........................509

12.3.5SupervisedLabelRelaxation.....................510

12.4TheTheoryofEvidence..................................510

12.4.1TheConceptofEvidentialMass..................511

12.4.2CombiningEvidencewiththeOrthogonalSum.....513

12.4.3DecisionRules.................................515

12.5Knowledge-BasedImageAnalysis.........................515

12.5.1EmulatingPhotointerpretationtoUnderstand KnowledgeProcessing..........................516

12.5.2TheStructureofaKnowledge-BasedImage AnalysisSystem................................517

12.5.3RepresentingKnowledgeinaKnowledge-Based ImageAnalysisSystem..........................518

12.5.4ProcessingKnowledge—TheInferenceEngine.....520

12.5.5RulesasJustifiersofaLabellingProposition.......521

12.5.6EndorsingaLabellingProposition................522

12.5.7AnExample...................................523

12.6OperationalMultisourceAnalysis.........................525

12.7BibliographyonMultisourceImageAnalysis................528 12.8Problems...............................................530

AppendixA:SatelliteAltitudesandPeriods ..........................535

AppendixB:BinaryRepresentationofDecimalNumbers

AppendixC:EssentialResultsfromVectorandMatrixAlgebra

AppendixD:SomeFundamentalMaterialfromProbability andStatistics ......................................................551

AppendixE:PenaltyFunctionDerivationoftheMaximum LikelihoodDecisionRule ...........................................555

Chapter1

SourcesandCharacteristicsofRemote

SensingImageData

Abstract Thewavelengthrangescommonlyusedforimagingtheearth’ssurface arediscussed,includingreflectedsunlight,thermalemissionfromtheearthitself andthemicrowaveradiationusedinimagingradars.Theideaofmeasuringenergy comingfromtheearth’ssurfaceinasetofwavebandssimultaneouslyiscovered, leadingtotheconceptofamultispectralimage,orahyperspectralimageifthe numberofwavebandsisverylarge.Remotesensingplatformsanddifferentsensor typesarecovered,alongwiththeearthsurfacecharacteristicsthatcanbedetected withremotesensinginstruments.Imagescaleisconsideredandthelocationof remotesensingwithinthe fieldsofgeographicinformationsystemsanddigitalearth modelsisintroduced.

1.1EnergySourcesandWavelengthRanges

Inremotesensingenergycomingupfromtheearth'ssurfaceismeasuredusinga sensormountedonaspacecraftorotherelevatedplatform.Thatmeasurementis usedtoconstructanimageofthelandscapebeneaththeplatform,asdepictedin Fig. 1.1

Inprinciple,anyenergycomingfromtheearth’ssurfacecanbeusedtoforman image.Mostoftenitisreflectedsunlightsothattheimagerecordedis,inmany ways,similartotheviewwewouldhaveoftheearth’ssurfacefromanaircraft, eventhoughthewavelengthsusedinremotesensingareoftenoutsidetherangeof humanvision.Theupwellingenergycouldalsobefromtheearthitselfactingasa radiatorbecauseofitsown finitetemperature.Alternatively,itcouldbeenergythat isscattereduptoasensorhavingbeenradiatedontothesurfacebyanartifi cial source,suchasalaseroraradar.

Providedanenergysourceisavailable,almostanywavelengthcouldbeusedto imagethecharacteristicsoftheearth’ssurface.Thereis,however,afundamental limitation,particularlywhenimagingfromspacecraftaltitudes.Theearth’satmospheredoesnotallowthepassageofradiationatallwavelengths.Energyatsome wavelengthsisabsorbedbythemolecularconstituentsoftheatmosphere.

© TheAuthor(s),underexclusivelicensetoSpringerNatureSwitzerlandAG2022 J.A.Richards, RemoteSensingDigitalImageAnalysis, https://doi.org/10.1007/978-3-030-82327-6_1

Fig.1.1 Signal flowina remotesensingsystem

upwelling radia on from the landscape sensor

signal transmission to the ground

ground recep on and processing data in image form ready for use

Wavelengthsforwhichthereislittleornoatmosphericabsorptionform atmosphericwindows.Figure 1.2 showsthetransmittanceoftheearth’satmosphereona pathbetweenspaceandtheearthoveraverybroadrangeoftheelectromagnetic spectrum.Thepresenceofasignifi cantnumberofatmosphericwindowsinthe visibleandinfraredregionsofthespectrumisevident,asisthealmostcomplete transparencyoftheatmosphereatradiowavelengths.Thewavelengthsusedfor imaginginremotesensingareclearlyconstrainedtotheseatmosphericwindows. Theyincludetheso-called optical wavelengthscoveringthevisibleandinfrared, thethermalwavelengthsandtheradiowavelengthsthatareusedinradarand passivemicrowaveimagingoftheearth’ssurface.

Whateverwavelengthrangeisusedtoimagetheearth’ssurface,theoverall systemisacomplexoneinvolvingthescatteringoremissionofenergyfromthe surface,followedbytransmissionthroughtheatmospheretoinstrumentsmounted ontheremotesensingplatform.Thedataisthentransmittedtotheearth’ssurface, afterwhichitisprocessedintoimageproductsreadyforapplicationbytheuser. ThatdatachainisshowninFig. 1.1.Itisfromthepointofimageacquisition onwardsthatthisbookisconcerned.Wewanttounderstandhowthedata,once availableinimageformat,canbeinterpreted.

Wetalkabouttherecordedimageryas imagedata,sinceitistheprimarydata sourcefromwhichweextractusableinformation.Oneoftheimportantcharacteristicsoftheimagedataacquiredbysensorsonaircraftorspacecraftplatformsis instrumenta on

Fig.1.2 Theelectromagneticspectrumandthetransmittanceoftheearth’satmosphere,showing thepositionsoftheatmosphericwindowsusedinopticalremotesensing

thatitisreadilyavailableindigitalformat.Spatiallyitiscomposedofdiscrete pictureelements,or pixels.Radiometrically thatisinbrightness itisquantised intodiscretelevels.

Possiblythemostsigni ficantcharacteristicoftheimagedataprovidedbya remotesensingsystemisthewavelength,orrangeofwavelengths,usedinthe imageacquisitionprocess.Ifreflectedsolarradiationismeasured,imagescan,in principle,beacquiredintheultraviolet,visibleandnear-to-middleinfraredranges ofwavelengths.Becauseofsigni ficantatmosphericabsorption,asseeninFig. 1.2, ultravioletmeasurementsarenotmadefromspacecraftaltitudes.Mostcommon opticalremotesensingsystemsrecorddatafromthevisiblethroughtothenearand mid-infraredrange:typically,thatcoversapproximately0.4–2.5 lm.

Theenergyemittedbytheearthitself,inthethermalinfraredrangeofwavelengths,canalsoberesolvedintodifferentwavelengthsthathelpunderstand propertiesofthesurfacebeingimaged.Figure 1.3 showswhytheserangesare important.Thesunasaprimarysourceofenergyhasasurfacetemperatureofabout 5950K.Theenergyitemitsasafunctionofwavelengthisdescribedtheoretically byPlanck’sblackbodyradiationlaw.AsseeninFig. 1.3 ithasitsmaximaloutput atwavelengthsjustshorterthan1 lmandisamoderatelystrongemitteroverthe range0.4–2.5 lmidenti fiedearlier.

Theearthcanalsobeconsideredasablackbodyradiator,withatemperatureof 300K.Itsemissioncurvehasamaximuminthevicinityof10 µmasseeninFig. 1.3 Asaresult,remotesensinginstrumentsdesignedtomeasuresurfacetemperature

Fig.1.3 Relativelevelsofenergyfromblackbodieswhenmeasuredatthesurfaceoftheearth: themagnitudeofthesolarcurvehasbeenreducedasaresultofthedistancetravelledbysolar radiationfromthesuntotheearth;alsoshownaretheboundariesbetweenthedifferentwavelength rangesusedinopticalremotesensing

typicallyoperatesomewhereintherangeof8–12 lm.AlsoshowninFig. 1.3 isthe blackbodyradiationcurvecorrespondingtoa firewithatemperatureof1000K.As observed,itsmaximumoutputisinthewavelengthrange3–5 lm.Accordingly,sensors designedtomapburning firesontheearth’ssurfacetypicallyoperateinthatrange.

Thevisible,reflectiveinfraredandthermalinfraredrangesofwavelengthrepresentonlypartofthestoryinremotesensing.Wecanalsoimagetheearthinthe microwaveorradiorange,typicalofthewavelengthsusedinmobilephones, satellitenavigationsystems,television,WiFi,Bluetoothandradar.Whiletheearth doesemititsownlevelofmicrowaveradiation,itisoftentoosmalltobemeasured formostremotesensingpurposes.Instead,energyisradiatedfromaplatformonto theearth’ssurface.Itisbymeasuringtheenergyscatteredbacktotheplatformthat imagedataisrecordedatmicrowavewavelengths.1 Suchasystemisreferredtoas active sincetheenergysourceisprovidedbytheplatformitself,orbyacompanion platform.Bycomparison,remotesensingmeasurementsthatdependonanenergy sourcesuchasthesunortheearthitselfarecalled passive

1 ForatreatmentofremotesensingatmicrowavewavelengthsseeJ.A.Richards, RemoteSensing withImagingRadar,Springer,Berlin,2009.

1.2PrimaryDataCharacteristics

Thepropertiesofdigitalimagedataofimportanceinimageprocessingandanalysis arethenumberandlocationofthespectralmeasurements(bandsorchannels),the spatialresolutiondescribedbythepixelsize,andthe radiometricresolution.These areshowninFig. 1.4.Radiometricresolutiondescribestherangeanddiscernible numberofdiscretebrightnessvalues.Itissometimesreferredtoas dynamicrange andisrelatedtothesignal-to-noiseratioofthedetectorsused.Frequently,radiometricresolutionisexpressedintermsofthenumberofbinarydigits,orbits, necessarytorepresenttherangeofavailablebrightnessvalues.Datawithan8bit radiometricresolutionhas256levelsofbrightness,whiledatawith12bitradiometricresolutionhas4096brightnesslevels.2

Thesizeoftherecordedimageframeisalsoanimportantproperty.Itis describedbythenumberofpixelsacrosstheframeor swath,orintermsofthe numbersofkilometrescoveredbytherecordedscene.Together,theframesizeof theimage,thenumberofspectralbands,theradiometricresolutionandthespatial resolutiondeterminethedatavolumegeneratedbyaparticularsensor.Thatsetsthe amountofdatatobeprocessed,atleastinprinciple.

Imagepropertieslikepixelsizeandframesizearerelateddirectlytothetechnicalcharacteristicsofthesensorthatwasusedtorecordthedata.The instantaneous fieldofview (IFOV)ofthesensorisits fi nestangularresolution,asshownin Fig. 1.5.Whenprojectedontothesurfaceoftheearthattheoperatingaltitudeofthe platform,itdefinesthesmallestresolvableelementintermsofequivalentground metres,whichiswhatwerefertoaspixelsize.Similarly,the fieldofview (FOV)of thesensoristheangularextentoftheviewithasacrosstheearth’ssurface,againas

Fig.1.4

Fig.1.5 Definitionofimage spatialproperties,with commonunitsindicated

seeninFig. 1.5.Whenthatangleisprojectedontothesurfaceitdefinestheswath widthinequivalentgroundkilometres.Mostimageryisrecordedinacontinuous stripastheremotesensingplatformtravelsforward.Generally,particularlyfor spacecraftprograms,thestripiscutupintosegments,equalinlengthtotheswath width,sothatasquareimageframeisproduced.Foraircraftsystems,thedatais oftenleftinstripformatforthecomplete flightline flowninagivenmission.

1.3RemoteSensingPlatforms

Remotesensingcanbecarriedoutusingsatellites,aircraftordronesasplatformsto carrytheimaginginstruments.Inmanywaysthoseinstrumentshavesimilar characteristicsbutdifferencesinthealtitudeandstabilityoftheplatformcanleadto differingimageproperties.

Therearetwobroadclassesofsatelliteprogram:thosesatellitesthatorbitat geostationaryaltitudesabovetheearth’ssurface,generallyassociatedwithweather andclimatestudies,andthosewhichorbitmuchclosertotheearthandthatare generallyusedforearthsurfaceandoceanographicobservations.Thelowearth orbitingsatellitesareusuallyinasun-synchronousorbit.Thatmeansthattheorbital planeisdesignedsothatitprecessesabouttheearthatthesameratethatthesun appearstomoveacrosstheearth’ssurface.Inthismannerthesatelliteacquiresdata ataboutthesamelocaltimeoneachorbit.

Lowearthorbitingsatellitescanalsobeusedformeteorologicalstudies. Notwithstandingthedifferencesinaltitude,thewavebandsusedforgeostationary andearthorbitingsatellites,forweatherandearthobservation,areverycomparable. Themajordistinctionintheimagedatatheyprovidegenerallyliesinthespatial resolutionavailable.Whereasdataacquiredforearthresourcespurposeshaspixel

sizesoftheorderof1–10morso,thatusedformeteorologicalpurposes(bothat geostationaryandloweraltitudes)hasamuchlargerpixelsize,oftenoftheorderof 1km.

Theimagingtechnologiesusedinsatelliteremotesensingprogramshaveranged fromtraditionalcamerastoscannersthatrecordimagesoftheearth’ssurfaceby movingtheinstantaneous fieldofviewoftheinstrumentacrossthesurfaceto recordtheupwellingenergy.Typicalofthelattertechniqueisthatusedinthe Landsatprograminwhichamechanicalscannerrecordsdataatrightanglestothe directionofsatellitemotiontoproducerasterscansofdata.Theforwardmotionof thevehicleallowsanimagestriptobebuiltupfromtherasterscans.Thatprocessis showninFig. 1.6.Adispersiondevice,suchasaprismordiffractiongrating, integratedwiththesensor,separatestherecordedsignalintoanumberofwavebandsbydispersingtheradiationontosetsofdetectors;thereareasmanyseparate imagesrecordedoftheregionoftheearth’ssurfaceastherearedetectorsandthus wavebands.

Someweathersatellitesscantheearth’ssurfaceusingthespinofthesatellite itselfwhilethesensor ’spointingdirectionisvariedalongtheaxisofthesatellite. Theimagedataisthenrecordedinarasterscanfashion.

Withtheavailabilityofreliabledetectorarraysbasedonchargecoupleddevice (CCD)technology,analternativeimageacquisitionmechanismutiliseswhatis commonlycalleda “push-broom” technique.InthisapproachalinearCCDimaging arrayiscarriedonthesatellitenormaltotheplatformmotionasshowninFig. 1.7 Asthesatellitemovesforwardthearrayrecordsastripofimagedata,equivalentin widthtothe fieldofviewseenbythearray.Eachindividualdetectorrecordsastripin widthequivalenttothesizeofapixel.Becausethetimeoverwhichenergyemanatingfromtheearth’ssurfaceperpixelcanbelargerwithpushbroomtechnology thanwithmechanicalscanners,betterspatialresolutionisusuallyachieved.

TwodimensionalCCDarraysarealsoavailableand findapplicationinsatellite imagingsensors.However,ratherthanrecordatwo-dimensionalsnapshotimageof theearth’ssurface,thearrayisemployedinapushbroommanner;thesecond dimensionisusedtorecordsimultaneouslyanumberofdifferentwavebandsfor eachpixelviatheuseofamechanismthatdispersestheincomingradiationby wavelength.SuchanarrangementisshowninFig. 1.8.Oftenabout200channels arerecordedinthismannersothatthereflectioncharacteristicsoftheearth’s surfacearewellrepresentedinthedata.Suchdevicesareoftenreferredtoas imagingspectrometers andthedataisdescribedas hyperspectral,asagainst multispectral whenoftheorderof10wavebandsisrecorded.

Aircraftscannersoperateessentiallyonthesameprinciplesasthosefoundwith satellitesensors.BothmechanicalscannersandCCDarraysareemployed.

ThelogarithmicscaleusedinFig. 1.3 hidesthefactthateachofthecurves shownextendstoinfinity.Ifweignoreemissionsassociatedwithaburning fire,itis clearthattheemissionfromtheearthatlongerwavelengthsfarexceedsreflected solarenergy.Figure 1.9 re-plotstheearthcurvefromFig. 1.3 showingthatthereis continuousemissionofenergyrightouttothewavelengthswenormallyassociate withradiotransmissions.Inthemicrowaveenergyrange,wherethewavelengths

Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.