Hyperspectral Data Processing Algorithm Design and Analysis
, by Chang, Chein-I
- ISBN: 9780471690566 | 0471690562
- Cover: Hardcover
- Copyright: 4/8/2013
This book is intended to be a sequel to the author?s previous work, Hyperspectral Imaging: Techniques for Spectral Detection and Classification. It is comprised of five major parts and covers a wealth of information, including: a discussion of OSP, FPGA designs for OSP and CEM, Kalman filter-based linear unmixing, least squares fully constrained linear mixture analysis, interference rejection for linear unmixing, signal-composed interference-annihilated theory, techniques for linear unmixing, spectral coding, and more.
CHEIN-I CHANG, PhD, is a Professor in the Department of Computer Science and Electrical Engineering at the University of Maryland, Baltimore County. He established the Remote Sensing Signal and Image Processing Laboratory and conducts research in designing and developing signal processing algorithms for hyperspectral imaging, medical imaging, and documentation analysis. A Fellow of IEEE and SPIE, Dr. Chang has published over 125 refereed journal articles, including more than forty papers in the IEEE Transaction on Geoscience and Remote Sensing. In addition to authoring Hyperspectral Imaging: Techniques for Spectral Detection and Classification, as well as editing two books, Hyperspectral Data Exploitation: Theory and Applications and Recent Advances in Hyperspectral Signal and Imaging Processing and co-editing one book, High Performance Computing in Remote Sensing, he holds five patents and has several pending.
PREFACE xxiii
1 OVERVIEWAND INTRODUCTION 1
1.1 Overview 2
1.2 Issues of Multispectral and Hyperspectral Imageries 3
1.3 Divergence of Hyperspectral Imagery from Multispectral Imagery 4
1.3.1 Misconception: Hyperspectral Imaging is a Natural Extension of Multispectral Imaging 4
1.3.2 Pigeon-Hole Principle: Natural Interpretation of Hyperspectral Imaging 5
1.4 Scope of This Book 7
1.5 Book’s Organization 10
1.5.1 Part I: Preliminaries 10
1.5.2 Part II: Endmember Extraction 12
1.5.3 Part III: Supervised Linear Hyperspectral Mixture Analysis 13
1.5.4 Part IV: Unsupervised Hyperspectral Analysis 13
1.5.5 Part V: Hyperspectral Information Compression 15
1.5.6 Part VI: Hyperspectral Signal Coding 16
1.5.7 Part VII: Hyperspectral Signal Feature Characterization 17
1.5.8 Applications 17
1.5.8.1 Chapter 30: Applications of Target Detection 17
1.5.8.2 Chapter 31: Nonlinear Dimensionality Expansion to Multispectral Imagery 18
1.5.8.3 Chapter 32: Multispectral Magnetic Resonance Imaging 19
1.6 Laboratory Data to be Used in This Book 19
1.6.1 Laboratory Data 19
1.6.2 Cuprite Data 19
1.6.3 NIST/EPA Gas-Phase Infrared Database 19
1.7 Real Hyperspectral Images to be Used in this Book 20
1.7.1 AVIRIS Data 20
1.7.1.1 Cuprite Data 21
1.7.1.2 Purdue’s Indiana Indian Pine Test Site 25
1.7.2 HYDICE Data 26
1.8 Notations and Terminologies to be Used in this Book 29
I: PRELIMINARIES 31
2 FUNDAMENTALS OF SUBSAMPLE AND MIXED SAMPLE ANALYSES 33
2.1 Introduction 33
2.2 Subsample Analysis 35
2.2.1 Pure-Sample Target Detection 35
2.2.2 Subsample Target Detection 38
2.2.2.1 Adaptive Matched Detector (AMD) 39
2.2.2.2 Adaptive Subspace Detector (ASD) 41
2.2.3 Subsample Target Detection: Constrained Energy Minimization (CEM) 43
2.3 Mixed Sample Analysis 45
2.3.1 Classification with Hard Decisions 45
2.3.1.1 Fisher’s Linear Discriminant Analysis (FLDA) 46
2.3.1.2 Support Vector Machines (SVM) 48
2.3.2 Classification with Soft Decisions 54
2.3.2.1 Orthogonal Subspace Projection (OSP) 54
2.3.2.2 Target-Constrained Interference-Minimized Filter (TCIMF) 56
2.4 Kernel-Based Classification 57
2.4.1 Kernel Trick Used in Kernel-Based Methods 57
2.4.2 Kernel-Based Fisher’s Linear Discriminant Analysis (KFLDA) 58
2.4.3 Kernel Support Vector Machine (K-SVM) 59
2.5 Conclusions 60
3 THREE-DIMENSIONAL RECEIVER OPERATING CHARACTERISTICS (3D ROC) ANALYSIS 63
3.1 Introduction 63
3.2 Neyman–Pearson Detection Problem Formulation 65
3.3 ROC Analysis 67
3.4 3D ROC Analysis 69
3.5 Real Data-Based ROC Analysis 72
3.5.1 How to Generate ROC Curves from Real Data 72
3.5.2 How to Generate Gaussian-Fitted ROC Curves 73
3.5.3 How to Generate 3D ROC Curves 75
3.5.4 How to Generate 3D ROC Curves for Multiple Signal Detection and Classification 77
3.6 Examples 78
3.6.1 Hyperspectral Imaging 79
3.6.1.1 Hyperspectral Target Detection 79
3.6.1.2 Linear Hyperspectral Mixture Analysis 80
3.6.2 Magnetic Resonance (MR) Breast Imaging 83
3.6.2.1 Breast Tumor Detection 84
3.6.2.2 Brain Tissue Classification 87
3.6.3 Chemical/Biological Agent Detection 91
3.6.4 Biometric Recognition 95
3.7 Conclusions 99
4 DESIGN OF SYNTHETIC IMAGE EXPERIMENTS 101
4.1 Introduction 102
4.2 Simulation of Targets of Interest 103
4.2.1 Simulation of Synthetic Subsample Targets 103
4.2.2 Simulation of Synthetic Mixed-Sample Targets 104
4.3 Six Scenarios of Synthetic Images 104
4.3.1 Panel Simulations 104
4.3.2 Three Scenarios for Target Implantation (TI) 106
4.3.2.1 Scenario TI1 (Clean Panels Implanted into Clean Background) 106
4.3.2.2 Scenario TI2 (Clean Panels Implanted into Noisy Background) 107
4.3.2.3 Scenario TI3 (Gaussian Noise Added to Clean Panels Implanted into Clean Background) 108
4.3.3 Three Scenarios for Target Embeddedness (TE) 108
4.3.3.1 Scenario TE1 (Clean Panels Embedded in Clean Background) 109
4.3.3.2 Scenario TE2 (Clean Panels Embedded in Noisy Background) 109
4.3.3.3 Scenario TE3 (Gaussian Noise Added to Clean Panels Embedded in Background) 110
4.4 Applications 112
4.4.1 Endmember Extraction 112
4.4.2 Linear Spectral Mixture Analysis (LSMA) 113
4.4.2.1 Mixed Pixel Classification 114
4.4.2.2 Mixed Pixel Quantification 114
4.4.3 Target Detection 114
4.4.3.1 Subpixel Target Detection 114
4.4.3.2 Anomaly Detection 122
4.5 Conclusions 123
5 VIRTUAL DIMENSIONALITY OF HYPERSPECTRAL DATA 124
5.1 Introduction 124
5.2 Reinterpretation of VD 126
5.3 VD Determined by Data Characterization-Driven Criteria 126
5.3.1 Eigenvalue Distribution-Based Criteria 127
5.3.1.1 Thresholding Energy Percentage 127
5.3.1.2 Thresholding Difference between Normalized Correlation Eigenvalues and Normalized Covariance Eigenvalues 128
5.3.1.3 Finding First Sudden Drop in the Normalized Eigenvalue Distribution 128
5.3.2 Eigen-Based Component Analysis Criteria 128
5.3.2.1 Singular Value Decomposition (SVD) 128
5.3.2.2 Principal Components Analysis (PCA) 129
5.3.3 Factor Analysis: Malinowski’s Error Theory 129
5.3.4 Information Theoretic Criteria (ITC) 130
5.3.4.1 AIC 131
5.3.4.2 MDL 131
5.3.5 Gershgorin Radius-Based Methods 131
5.3.5.1 Thresholding Gershgorin Radii 134
5.3.5.2 Thresholding Difference Gershgorin Radii between RLL and KLL 134
5.3.6 HFC Method 135
5.3.7 Discussions on Data Characterization-Driven Criteria 138
5.4 VD Determined by Data Representation-Driven Criteria 140
5.4.1 Orthogonal Subspace Projection (OSP) 140
5.4.2 Signal Subspace Estimation (SSE) 142
5.4.3 Discussions on OSP and SSE/HySime 143
5.5 Synthetic Image Experiments 144
5.5.1 Data Characterization-Driven Criteria 144
5.5.1.1 Target Implantation (TI) Scenarios 145
5.5.1.2 Target Embeddedness (TE) Scenarios 146
5.5.2 Data Representation-Driven Criteria 149
5.6 VD Estimated for Real Hyperspectral Images 155
5.7 Conclusions 163
6 DATA DIMENSIONALITY REDUCTION 168
6.1 Introduction 168
6.2 Dimensionality Reduction by Second-Order Statistics-Based Component Analysis Transforms 170
6.2.1 Eigen Component Analysis Transforms 170
6.2.1.1 Principal Components Analysis 170
6.2.1.2 Standardized Principal Components Analysis 172
6.2.1.3 Singular Value Decomposition 174
6.2.2 Signal-to-Noise Ratio-Based Components Analysis Transforms 176
6.2.2.1 Maximum Noise Fraction Transform 176
6.2.2.2 Noise-Adjusted Principal Component Transform 177
6.3 Dimensionality Reduction by High-Order Statistics-Based Components Analysis Transforms 179
6.3.1 Sphering 179
6.3.2 Third-Order Statistics-Based Skewness 181
6.3.3 Fourth-Order Statistics-Based Kurtosis 182
6.3.4 High-Order Statistics 182
6.3.5 Algorithm for Finding Projection Vectors 183
6.4 Dimensionality Reduction by Infinite-Order Statistics-Based Components Analysis Transforms 184
6.4.1 Statistics-Prioritized ICA-DR (SPICA-DR) 187
6.4.2 Random ICA-DR 188
6.4.3 Initialization Driven ICA-DR 189
6.5 Dimensionality Reduction by Projection Pursuit-Based Components Analysis Transforms 190
6.5.1 Projection Index-Based Projection Pursuit 191
6.5.2 Random Projection Index-Based Projection Pursuit 192
6.5.3 Projection Index-Based Prioritized Projection Pursuit 193
6.5.4 Initialization Driven Projection Pursuit 194
6.6 Dimensionality Reduction by Feature Extraction-Based Transforms 195
6.6.1 Fisher’s Linear Discriminant Analysis 195
6.6.2 Orthogonal Subspace Projection 196
6.7 Dimensionality Reduction by Band Selection 196
6.8 Constrained Band Selection 197
6.9 Conclusions 198
II: ENDMEMBER EXTRACTION 201
7 SIMULTANEOUS ENDMEMBER EXTRACTION ALGORITHMS (SM-EEAs) 207
7.1 Introduction 208
7.2 Convex Geometry-Based Endmember Extraction 209
7.2.1 Convex Geometry-Based Criterion: Orthogonal Projection 209
7.2.2 Convex Geometry-Based Criterion: Minimal Simplex Volume 214
7.2.2.1 Minimal-Volume Transform (MVT) 214
7.2.2.2 Convex Cone Analysis (CCA) 214
7.2.3 Convex Geometry-Based Criterion: Maximal Simplex Volume 215
7.2.3.1 Simultaneous N-FINDR (SM N-FINDR) 216
7.2.3.2 Iterative N-FINDR (IN-FINDR) 216
7.2.3.3 Various Versions of Implementing IN-FINDR 218
7.2.3.4 Discussions on Various Implementation Versions of IN-FINDR 222
7.2.3.5 Comparative Study Among Various Versions of IN-FINDR 222
7.2.3.6 Alternative SM N-FINDR 223
7.2.4 Convex Geometry-Based Criterion: Linear Spectral Mixture Analysis 225
7.3 Second-Order Statistics-Based Endmember Extraction 228
7.4 Automated Morphological Endmember Extraction (AMEE) 230
7.5 Experiments 231
7.5.1 Synthetic Image Experiments 231
7.5.1.1 Scenario TI1 (Endmembers Implanted in a Clean Background) 232
7.5.1.2 Scenario TI2 (Endmembers Implanted in a Noisy Background) 233
7.5.1.3 Scenario TI3 (Noisy Endmembers Implanted in a Noisy Background) 234
7.5.1.4 Scenario TE1 (Endmembers Embedded into a Clean Background) 235
7.5.1.5 Scenario TE2 (Endmembers Embedded into a Noisy Background) 235
7.5.1.6 Scenario TE3 (Noisy Endmembers Embedded into a Noisy Background) 236
7.5.2 Cuprite Data 237
7.5.3 HYDICE Data 237
7.6 Conclusions 239
8 SEQUENTIAL ENDMEMBER EXTRACTION ALGORITHMS (SQ-EEAs) 241
8.1 Introduction 241
8.2 Successive N-FINDR (SC N-FINDR) 244
8.3 Simplex Growing Algorithm (SGA) 244
8.4 Vertex Component Analysis (VCA) 247
8.5 Linear Spectral Mixture Analysis-Based SQ-EEAs 248
8.5.1 Automatic Target Generation Process-EEA (ATGP-EEA) 248
8.5.2 Unsupervised Nonnegativity Constrained Least-Squares-EEA
(UNCLS-EEA) 249
8.5.3 Unsupervised Fully Constrained Least-Squares-EEA (UFCLS-EEA) 250
8.5.4 Iterative Error Analysis-EEA (IEA-EEA) 251
8.6 High-Order Statistics-Based SQ-EEAS 252
8.6.1 Third-Order Statistics-Based SQ-EEA 252
8.6.2 Fourth-Order Statistics-Based SQ-EEA 252
8.6.3 Criterion for kth Moment-Based SQ-EEA 253
8.6.4 Algorithm for Finding Projection Vectors 253
8.6.5 ICA-Based SQ-EEA 254
8.7 Experiments 254
8.7.1 Synthetic Image Experiments 255
8.7.2 Real Hyperspectral Image Experiments 258
8.7.2.1 Cuprite Data 258
8.7.2.2 HYDICE Data 260
8.8 Conclusions 262
9 INITIALIZATION-DRIVEN ENDMEMBER EXTRACTION ALGORITHMS (ID-EEAs) 265
9.1 Introduction 265
9.2 Initialization Issues 266
9.2.1 Initial Conditions to Terminate an EEA 267
9.2.2 Selection of an Initial Set of Endmembers for an EEA 267
9.2.3 Issues of Random Initial Conditions Demonstrated by Experiments 268
9.2.3.1 HYDICE Experiments 268
9.2.3.2 AVIRIS Experiments 270
9.3 Initialization-Driven EEAs 271
9.3.1 Initial Endmember-Driven EEAs 272
9.3.1.1 Finding Maximum Length of Data Sample Vectors 272
9.3.1.2 Finding Sample Mean of Data Sample Vectors 273
9.3.2 Endmember Initialization Algorithm for SM-EEAs 274
9.3.2.1 SQ-EEAs 274
9.3.2.2 Maxmin-Distance Algorithm 275
9.3.2.3 ISODATA 275
9.3.3 EIA-Driven EEAs 275
9.4 Experiments 278
9.4.1 Synthetic Image Experiments 278
9.4.2 Real Image Experiments 281
9.5 Conclusions 283
10 RANDOM ENDMEMBER EXTRACTION ALGORITHMS (REEAs) 287
10.1 Introduction 287
10.2 Random PPI (RPPI) 288
10.3 Random VCA (RVCA) 290
10.4 Random N-FINDR (RN-FINDR) 290
10.5 Random SGA (RSGA) 292
10.6 Random ICA-Based EEA (RICA-EEA) 292
10.7 Synthetic Image Experiments 293
10.7.1 RPPI 293
10.7.2 Various Random Versions of IN-FINDR 296
10.7.2.1 Scenario TI2 297
10.7.2.2 Scenario TI3 299
10.7.2.3 TE2 301
10.7.2.4 TE3 Scenario 303
10.8 Real Image Experiments 305
10.8.1 HYDICE Image Experiments 305
10.8.1.1 RPPI 306
10.8.1.2 RN-FINDR 306
10.8.2 AVIRIS Image Experiments 309
10.8.2.1 RPPI 309
10.8.2.2 RN-FINDR 310
10.9 Conclusions 313
11 EXPLORATION ON RELATIONSHIPS AMONG ENDMEMBER EXTRACTION ALGORITHMS 316
11.1 Introduction 316
11.2 Orthogonal Projection-Based EEAs 318
11.2.1 Relationship among PPI, VCA, and ATGP 319
11.2.1.1 Relationship Between PPI and ATGP 319
11.2.1.2 Relationship Between PPI and VCA 320
11.2.1.3 Relationship Between ATGP and VCA 321
11.2.1.4 Discussions 322
11.2.2 Experiments-Based Comparative Study and Analysis 323
11.2.2.1 Synthetic Image Experiment: TI2 323
11.2.2.2 Real Image Experiments 325
11.3 Comparative Study and Analysis Between SGA and VCA 330
11.4 Does an Endmember Set Really Yield Maximum Simplex Volume? 339
11.5 Impact of Dimensionality Reduction on EEAs 344
11.6 Conclusions 348
III: SUPERVISED LINEAR HYPERSPECTRAL MIXTURE ANALYSIS 351
12 ORTHOGONAL SUBSPACE PROJECTION REVISITED 355
12.1 Introduction 355
12.2 Three Perspectives to Derive OSP 358
12.2.1 Signal Detection Perspective Derived from (d,U)-Model and OSP-Model 359
12.2.2 Fisher’s Linear Discriminant Analysis Perspective from OSP-Model 360
12.2.3 Parameter Estimation Perspective from OSP-Model 362
12.2.4 Relationship Between dLS ap ðrÞ and Least-Squares Linear Spectral Mixture Analysis 362
12.3 Gaussian Noise in OSP 364
12.3.1 Signal Detector in Gaussian Noise Using OSP-Model 365
12.3.2 Gaussian Maximum Likelihood Classifier Using OSP-Model 366
12.3.3 Gaussian Maximum Likelihood Estimator 367
12.3.4 Examples 367
12.4 OSP Implemented with Partial Knowledge 372
12.4.1 CEM 373
12.4.1.1 d Is Orthogonal to U (i.e., P? Ud ¼ d) and R ¼ I (i.e., Spectral Correlation is Whitened) 374
12.4.1.2 An Alternative Approach to Implementing CEM 374
12.4.1.3 CEM Implemented in Conjunction with P? U 375
12.4.1.4 CEM Implemented in Conjunction with P? U in White Noise 376
12.4.2 TCIMF 377
12.4.2.1 D ¼ mp ¼ d with nD ¼ 1 and U ¼ m1m2 mp1 with nU ¼ p 1 378
12.4.2.2 D ¼ mp ¼ d with nD ¼ 1 and U ¼ m1 m2 mp1 with nU ¼ p 1 and R ¼ I 378
12.4.2.3 D ¼ d and U ¼ Ø (i.e., Only the Desired Signature d is Available) 378
12.4.3 Examples 379
12.5 OSP Implemented Without Knowledge 383
12.6 Conclusions 390
13 FISHER’S LINEAR SPECTRAL MIXTURE ANALYSIS 391
13.1 Introduction 391
13.2 Feature Vector-Constrained FLSMA (FVC-FLSMA) 392
13.3 Relationship Between FVC-FLSMA and LCMV, TCIMF, and CEM 395
13.4 Relationship Between FVC-FLSMA and OSP 396
13.5 Relationship Between FVC-FLSMA and LCDA 396
13.6 Abundance-Constrained Least Squares FLDA (ACLS-FLDA) 397
13.7 Synthetic Image Experiments 398
13.8 Real Image Experiments 402
13.8.1 Image Background Characterized by Supervised Knowledge 402
13.8.2 Image Background Characterized by Unsupervised Knowledge 405
13.9 Conclusions 409
14 WEIGHTED ABUNDANCE-CONSTRAINED LINEAR SPECTRAL MIXTURE ANALYSIS 411
14.1 Introduction 411
14.2 Abundance-Constrained LSMA (AC-LSMA) 413
14.3 Weighted Least-Squares Abundance-Constrained LSMA 413
14.3.1 Weighting Matrix Derived from a Parameter Estimation Perspective 414
14.3.1.1 MD-Weighted AC-LSMA 415
14.3.1.2 LCMV-Weighted AC-LSMA 415
14.3.2 Weighting Matrix Derived from Fisher’s Linear Discriminant Analysis Perspective 416
14.3.3 Weighting Matrix Derived from an Orthogonal Subspace Projection Perspective 417
14.3.3.1 OSP-Weighted AC-LSMA 417
14.3.3.2 SSP-Weighted AC-LSMA 418
14.4 Synthetic Image-Based Computer Simulations 419
14.5 Real Image Experiments 426
14.6 Conclusions 432
15 KERNEL-BASED LINEAR SPECTRAL MIXTURE ANALYSIS 434
15.1 Introduction 434
15.2 Kernel-Based LSMA (KLSMA) 436
15.2.1 Kernel Least Squares Orthogonal Subspace Projection (KLSOSP) 436
15.2.2 Kernel-Based Non-Negative Constraint Least Square (KNCLS) 438
15.2.3 Kernel-Based Fully Constraint Least Square (KFCLS) 439
15.2.4 A Note on Kernelization 440
15.3 Synthetic Image Experiments 441
15.4 AVIRIS Data Experiments 444
15.4.1 Radial Basis Function Kernels 449
15.4.2 Polynomial Kernels 452
15.4.3 Sigmoid Kernels 454
15.5 HYDICE Data Experiments 460
15.6 Conclusions 462
IV: UNSUPERVISED HYPERSPECTRAL IMAGE ANALYSIS 465
16 HYPERSPECTRAL MEASURES 469
16.1 Introduction 469
16.2 Signature Vector-Based Hyperspectral Measures for Target Discrimanition and Identification 470
16.2.1 Euclidean Distance 471
16.2.2 Spectral Angle Mapper 471
16.2.3 Orthogonal Projection Divergence 471
16.2.4 Spectral Information Divergence 471
16.3 Correlation-Weighted Hyperspectral Measures for Target Discrimanition and Identification 472
16.3.1 Hyperspectral Measures Weighted by A Priori Correlation 473
16.3.1.1 OSP-Based Hyperspectral Measures for Discrimination 473
16.3.1.2 OSP-Based Hyperspectral Measures for Identification 473
16.3.2 Hyperspectral Measures Weighted by A Posteriori Correlation 474
16.3.2.1 Covariance Matrix-Weighted Hyperspectral Measures 474
16.3.2.2 Correlation Matrix-Weighted Hyperspectral Measures 475
16.3.2.3 Covariance Matrix-Weighted Matched Filter Distance 475
16.3.2.4 Correlation Matrix-Weighted Matched Filter Distance 476
16.4 Experiments 477
16.4.1 HYDICE Image Experiments 477
16.4.2 AVIRIS Image Experiments 478
16.5 Conclusions 482
17 UNSUPERVISED LINEAR HYPERSPECTRAL MIXTURE ANALYSIS 483
17.1 Introduction 483
17.2 Least Squares-Based ULSMA 486
17.3 Component Analysis-Based ULSMA 488
17.4 Synthetic Image Experiments 490
17.4.1 LS-ULSMA 491
17.4.2 CA-ULSMA 499
17.5 Real-Image Experiments 503
17.5.1 LS-ULSMA 503
17.5.2 CA-ULSMA 505
17.5.3 Qualitative and Quantitative Analyses between ULSMA and SLSMA 511
17.6 ULSMAVersus Endmember Extraction 517
17.7 Conclusions 524
18 PIXEL EXTRACTION AND INFORMATION 526
18.1 Introduction 526
18.2 Four Types of Pixels 527
18.3 Algorithms Selected to Extract Pixel Information 528
18.4 Pixel Information Analysis via Synthetic Images 528
18.5 Real Image Experiments 534
18.5.1 AVIRIS Image Data 534
18.5.2 DAIS 7915 Image Data 537
18.6 Conclusions 539
V: HYPERSPECTRAL INFORMATION COMPRESSION 541
19 EXPLOITATION-BASED HYPERSPECTRAL DATA COMPRESSION 545
19.1 Introduction 545
19.2 Hyperspectral Information Compression Systems 547
19.3 Spectral/Spatial Compression 549
19.3.1 Dimensionality Reduction by Transform-Based Spectral Compression 550
19.3.1.1 Determination of Number of PCs/ICs to be Retained 551
19.3.1.2 PCA (ICA)/2D Compression 551
19.3.1.3 PCA (ICA)/3D Compression 552
19.3.1.4 Inverse PCA (Inverse ICA)/2D Compression 553
19.3.1.5 Inverse PCA (Inverse PCA)/3D Compression 553
19.3.1.6 Mixed Component Transforms for Hyperspectral Compression 554
19.3.2 Dimensionality Reduction by Band Selection-Based Spectral Compression 556
19.4 Progressive Spectral/Spatial Compression 557
19.5 3D Compression 557
19.5.1 3D-Multicomponent JPEG 557
19.5.2 3D-SPIHT Compression 558
19.6 Exploration-Based Applications 559
19.6.1 Linear Spectral Mixture Analysis 559
19.6.2 Subpixel Target Detection 559
19.6.3 Anomaly Detection 560
19.6.4 Endmember Extraction 561
19.7 Experiments 561
19.7.1 Synthetic Image Experiments 562
19.7.2 Real Image Experiments 567
19.8 Conclusions 580
20 PROGRESSIVE SPECTRAL DIMENSIONALITY PROCESS 581
20.1 Introduction 582
20.2 Dimensionality Prioritization 584
20.3 Representation of Transformed Components for DP 585
20.3.1 Projection Index-Based PP 585
20.3.2 Mixed Projection Index-Based Prioritized PP (M-PIPP) 587
20.3.3 Projection Index-Based Prioritized PP (PI-PRPP) 587
20.3.4 Initialization-Driven PIPP (ID-PIPP) 588
20.4 Progressive Spectral Dimensionality Process 589
20.4.1 Progressive Principal Components Analysis 591
20.4.1.1 Simultaneous PCA 591
20.4.1.2 Progressive PCA 592
20.4.1.3 Sequential PCA 593
20.4.1.4 Initialization-Driven PCA 595
20.4.2 Progressive High-Order Statistics Component Analysis 596
20.4.3 Progressive Independent Component Analysis 596
20.5 Hyperspectral Compression by PSDP 597
20.5.1 Progressive Spectral Dimensionality Reduction 597
20.5.2 Progressive Spectral Dimensionality Expansion 597
20.6 Experiments for PSDP 598
20.6.1 Endmember Extraction 598
20.6.2 Land Cover/Use Classification 599
20.6.3 Linear Spectral Mixture Analysis 603
20.7 Conclusions 608
21 PROGRESSIVE BAND DIMENSIONALITY PROCESS 613
21.1 Introduction 614
21.2 Band Prioritization 615
21.3 Criteria for Band Prioritization 617
21.3.1 Second-Order Statistics-Based BPC 617
21.3.1.1 Variance-Based BPC 617
21.3.1.2 Signal-to-Noise-Ratio-Based BPC 618
21.3.2 High-Order Statistics-Based BPC 618
21.3.2.1 Skewness 618
21.3.2.2 Kurtosis 618
21.3.3 Infinite-Order Statistics-Based BPC 618
21.3.3.1 Entropy 619
21.3.3.2 Information Divergence 619
21.3.4 Classification-Based BPC 619
21.3.4.1 Fisher’s Linear Discriminant Analysis (FLDA)-Based BPC 619
21.3.4.2 OSP-Based BPC 620
21.3.5 Constrained Band Correlation/Dependence Minimization 620
21.3.5.1 Band Correlation/Dependence Minimization 621
21.3.5.2 Band Correlation Constraint 622
21.4 Experiments for BP 624
21.4.1 Applications Using Highest-Prioritized Bands 625
21.4.1.1 Unsupervised Linear Spectral Mixture Analysis 626
21.4.1.2 Endmember Extraction 632
21.4.2 Applications Using Least-Prioritized Bands 635
21.4.2.1 Unsupervised Linear Spectral Mixture Analysis 636
21.4.2.2 Endmember Extraction 637
21.4.3 Applications Using Mixing Highest-Prioritized and Least-Prioritized Bands 646
21.4.3.1 Unsupervised Linear Spectral Mixture Analysis 646
21.4.3.2 Endmember Extraction 646
21.5 Progressive Band Dimensionality Process 651
21.6 Hyperspectral Compresssion by PBDP 653
21.6.1 Progressive Band Dimensionality Reduction Via BP 654
21.6.2 Progressive Band Dimensionality Expansion Via BP 655
21.7 Experiments for PBDP 656
21.7.1 Endmember Extraction 656
21.7.2 Land Cover/Use Classification 658
21.7.3 Linear Spectral Mixture Analysis 660
21.8 Conclusions 662
22 DYNAMIC DIMENSIONALITYALLOCATION 664
22.1 Introduction 664
22.2 Dynamic Dimensionality Allocaction 665
22.3 Signature Discriminatory Probabilties 667
22.4 Coding Techniques for Determining DDA 667
22.4.1 Shannon Coding-Based DDA 667
22.4.2 Huffman Coding-Based DDA 668
22.4.3 Hamming Coding-Based DDA 669
22.4.4 Notes on DDA 669
22.5 Experiments for Dynamic Dimensionality Allocation 669
22.5.1 Reflectance Cuprite Data 670
22.5.2 Purdue’s Data 672
22.5.3 HYDICE Data 674
22.6 Conclusions 682
23 PROGRESSIVE BAND SELECTION 683
23.1 Introduction 683
23.2 Band De-Corrleation 684
23.2.1 Spectral Measure-Based BD 684
23.2.2 Orthogonalization-Based BD 685
23.3 Progressive Band Selection 686
23.3.1 PBS: BP Followed by BD 687
23.3.2 PBS: BD Followed by BP 687
23.4 Experiments for Progressive Band Selection 688
23.5 Endmember Extraction 688
23.6 Land Cover/Use Classification 690
23.7 Linear Spectral Mixture Analysis 694
23.8 Conclusions 715
VI: HYPERSPECTRAL SIGNAL CODING 717
24 BINARY CODING FOR SPECTRAL SIGNATURES 719
24.1 Introduction 719
24.2 Binary Coding 720
24.2.1 SPAM Binary Coding 720
24.2.2 Median Partition Binary Coding 721
24.2.3 Halfway Partition Binary Coding 722
24.2.4 Equal Probability Partition Binary Coding 722
24.3 Spectral Feature-Based Coding 723
24.4 Experiments 725
24.4.1 Computer Simulations 725
24.4.2 Real Hyperspectral Image Data 730
24.5 Conclusions 740
25 VECTOR CODING FOR HYPERSPECTRAL SIGNATURES 741
25.1 Introduction 741
25.2 Spectral Derivative Feature Coding 743
25.2.1 Re-interpretation of SPAM and SFBC 743
25.2.2 Spectral Derivative Feature Coding 744
25.2.3 AVIRIS Data Experiments 746
25.2.3.1 Signature Discrimination 747
25.2.3.2 Mixed Signature Classification 748
25.2.4 NIST Gas Data Experiments 749
25.2.4.1 Signature Discrimination 750
25.2.4.2 Mixed Signature Classification 751
25.3 Spectral Feature Probabilistic Coding 755
25.3.1 Arithmetic Coding 755
25.3.2 Spectral Feature Probabilistic Coding 756
25.3.3 AVIRIS Data Experiments 758
25.3.4 NIST Gas Data Experiments 760
25.4 Real Image Experiments 764
25.4.1 SDFC 764
25.4.2 SFPC 766
25.5 Conclusions 771
26 PROGRESSIVE CODING FOR SPECTRAL SIGNATURES 772
26.1 Introduction 772
26.2 Multistage Pulse Code Modulation 774
26.3 MPCM-Based Progressive Spectral Signature Coding 783
26.3.1 Spectral Discrimination 784
26.3.2 Spectral Identification 785
26.4 NIST-GAS Data Experiments 786
26.5 Real Image Hyperspectral Experiments 790
26.6 Conclusions 796
VII: HYPERSPECTRAL SIGNAL CHARACTERIZATION 797
27 VARIABLE-NUMBERVARIABLE-BAND SELECTION FOR HYPERSPECTRAL SIGNALS 799
27.1 Introduction 799
27.2 Orthogonal Subspace Projection-Based Band Prioritization Criterion 801
27.3 Variable-Number Variable-Band Selection 803
27.4 Experiments 806
27.4.1 Hyperspectral Data 806
27.4.1.1 Signature Discrimination 806
27.4.1.2 Signature Classification/Identification 809
27.4.1.3 Noise Effect on VNVBS 811
27.4.2 NIST-Gas Data 813
27.4.2.1 Signature Discrimination 813
27.4.2.2 Signature Classification/Identification 814
27.4.2.3 Signature Discrimination between Two Signatures with Different Numbers of Bands 816
27.5 Selection of Reference Signatures 819
27.6 Conclusions 819
28 KALMAN FILTER-BASED ESTIMATION FOR HYPERSPECTRAL SIGNALS 820
28.1 Introduction 820
28.2 Kalman Filter-Based Linear Unmixing 822
28.3 Kalman Filter-Based Spectral Characterization Signal-Processing Techniques 824
28.3.1 Kalman Filter-based Spectral Signature Estimator 825
28.3.2 Kalman Filter-Based Spectral Signature Identifier 826
28.3.3 Kalman Filter-Based Spectral Signature Quantifier 828
28.4 Computer Simulations Using AVIRIS Data 831
28.4.1 KFSSE 831
28.4.2 KFSSI 832
28.4.2.1 Subpixel Target Identification by KFSSI 832
28.4.2.2 Mixed Target Identification by KFSSI 838
28.4.3 KFSSQ 839
28.4.3.1 Subpixel Target Quantification by KFSSQ 839
28.4.3.2 Mixed Target Quantification by KFSSQ 840
28.5 Computer Simulations Using NIST-Gas Data 843
28.5.1 KFSSE 843
28.5.2 KFSSI 843
28.5.2.1 Subpixel Target Identification by KFSSI 843
28.5.2.2 Mixed Target Identification by KFSSI 848
28.5.3 KFSSQ 849
28.5.3.1 Subpixel Target Identification by KFSSQ 849
28.5.3.2 Mixed Target Quantification by KFSSQ 849
28.6 Real Data Experiments 852
28.6.1 KFSSE 852
28.6.2 KFSSI 852
28.6.3 KFSSQ 856
28.7 Conclusions 857
29 WAVELET REPRESENTATION FOR HYPERSPECTRAL SIGNALS 859
29.1 Introduction 859
29.2 Wavelet Analysis 860
29.2.1 Multiscale Approximation 860
29.2.2 Scaling Function 861
29.2.3 Wavelet Function 862
29.3 Wavelet-Based Signature Characterization Algorithm 863
29.3.1 Wavelet-Based Signature Characterization Algorithm for Signature Self-Tuning 863
29.3.2 Wavelet-Based Signature Characterization Algorithm for Signature Self-Correction 866
29.3.3 Signature Self-Discrimination, Classification, and Identification 867
29.4 Synthetic Image-Based Computer Simulations 868
29.4.1 Signature Self-Tuning and Self-Denoising 869
29.4.2 Signature Self-Discrimination, Self-Classification, and Self-Identification 870
29.5 Real Image Experiments 871
29.6 Conclusions 875
VIII: APPLICATIONS 877
30 APPLICATIONS OF TARGET DETECTION 879
30.1 Introduction 879
30.2 Size Estimation of Subpixel Targets 880
30.3 Experiments 881
30.3.1 Synthetic Image Experiments 881
30.3.2 HYDICE Image Experiments 886
30.4 Concealed Target Detection 891
30.5 Computer-Aided Detection and Classification Algorithm for Concealed Targets 892
30.6 Experiments for Concealed Target Detection 893
30.7 Conclusions 895
31 NONLINEAR DIMENSIONALITY EXPANSION TO MULTISPECTRAL IMAGERY 897
31.1 Introduction 897
31.2 Band Dimensionality Expansion 899
31.2.1 Rationale for Developing BDE 899
31.2.2 Band Expansion Process 901
31.3 Hyperspectral Imaging Techniques Expanded by BDE 902
31.3.1 BEP-Based Orthogonal Subspace Projection 903
31.3.2 BEP-Based Constrained Energy Minimization 903
31.3.3 BEP-Based RX-Detector 903
31.4 Feature Dimensionality Expansion by Nonlinear Kernels 904
31.4.1 FDE by Transformation 905
31.4.2 FDE by Classification 907
31.4.2.1 FDE by Classification using Sample Spectral Correlation 907
31.4.2.2 FDE by Classification using Intrapixel Spectral Correlation 908
31.5 BDE in Conjunction with FDE 909
31.6 Multispectral Image Experiments 909
31.7 Conclusion 918
32 MULTISPECTRAL MAGNETIC RESONANCE IMAGING 920
32.1 Introduction 920
32.2 Linear Spectral Mixture Analysis for MRI 923
32.2.1 Orthogonal Subspace Projection to MRI 925
32.2.2 Band Expansion Process-Based OSP 927
32.2.3 Unsupervised Orthogonal Subspace Projection 928
32.3 Linear Spectral Random Mixture Analysis for MRI 928
32.3.1 Source Separation-Based OC-ICA for MR Image Analysis 930
32.3.2 Band Expansion Process Over complete ICA for MR Image Analysis 931
32.3.2.1 Eigenvector-Prioritized ICA 931
32.3.2.2 High-Order Statistics-Based PICA 932
32.3.2.3 ATGP-Prioritized PCA 932
32.4 Kernel-Based Linear Spectral Mixture Analysis 933
32.5 Synthetic MR Brain Image Experiments 933
32.6 Real MR Brain Image Experiments 951
32.7 Conclusions 955
33 CONCLUSIONS 956
33.1 Design Principles for Nonliteral Hyperspectral Imaging Techniques 956
33.1.1 Pigeon-Hole Principle 956
33.1.1.1 Multispectral Imagery Versus Hyperspectral Imagery 957
33.1.1.2 Virtual Dimensionality 957
33.1.2 Principle of Orthogonality 963
33.2 Endmember Extraction 964
33.3 Linear Spectral Mixture Analysis 970
33.3.1 Supervised LSMA 970
33.3.2 Unsupervised LSMA 973
33.4 Anomaly Detection 974
33.5 Support Vector Machines and Kernel-Based Approaches 977
33.6 Hyperspectral Compression 981
33.7 Hyperspectral Signal Processing 984
33.7.1 Signal Coding 986
33.7.2 Signal Estimation 986
33.8 Applications 987
33.9 Further Topics 987
33.9.1 Causal Processing 987
33.9.2 Real-Time Processing 988
33.9.3 FPGA Designs for Hardware Implementation 989
33.9.4 Parallel Processing 990
33.9.5 Progressive Hyperspectral Processing 990
GLOSSARY 993
APPENDIX: ALGORITHM COMPENDIUM 997
REFERENCES 1052
INDEX 1071
What is included with this book?
The New copy of this book will include any supplemental materials advertised. Please check the title of the book to determine if it should include any access cards, study guides, lab manuals, CDs, etc.
The Used, Rental and eBook copies of this book are not guaranteed to include any supplemental materials. Typically, only the book itself is included. This is true even if the title states it includes any access cards, study guides, lab manuals, CDs, etc.
Please wait while the item is added to your bag...
Review
Flag Review
×
Digital License
You are licensing a digital product for a set duration. Durations are set forth in the product description, with "Lifetime" typically meaning five (5) years of online access and permanent download to a supported device. All licenses are non-transferable.
More details can be found here.