Classification Methods for Remotely Sensed Data, Second Edition » (2nd Edition)

Since the publishing of the first edition of Classification Methods for Remotely Sensed Data in 2001, the field of pattern recognition has expanded in many new directions that make use of new technologies to capture data and more powerful computers to mine and process it. What seemed visionary but a decade ago is now being put to use and refined in commercial applications as well as military ones.

Keeping abreast of these new developments, Classification Methods for Remotely Sensed Data, Second Edition provides a comprehensive and up-to-date review of the entire field of classification methods applied to remotely sensed data. This second edition provides sevenfully revised chapters and two new chapters covering support vector machines (SVM) and decision trees.

The book also provides updated discussions and descriptions of Earth observation missions along with updated bibliographic references. After an introduction to the basics, the text provides a detailed discussion of different approaches to image classification, including maximum likelihood, fuzzy sets, and artificial neural networks.

This cutting-edge resource:

• Presents a number of approaches to solving the problem of allocation of data to one of several classes (Section 4.4)
• Covers potential approaches to the use of decision trees (Section 6.0)
• Describes developments such as boosting and random forest generation (section 6.8)
• Reviews lopping branches that do not contribute to the effectiveness of the decision trees (section 6.7)

Complete with detailed comparisons, experimental results, and discussions for each classification method introduced, this book will bolster the work of researchers and developers by giving them access to new developments. It also provides students with a solid foundation in remote sensing data classification methods.

Preface to the Second Edition xi

Preface to the First Edition xiii

Author Biographies xix

Chapter 1 Remote Sensing in the Optical and Microwave Regions 1

1.1 Introduction to Remote Sensing 4

1.1.1 Atmospheric Interactions 5

1.1.2 Surface Material Reflectance 5

1.1.3 Spatial and Radiometric Resolution 8

1.2 Optical Remote Sensing Systems 10

1.3 Atmospheric Correction 11

1.3.1 Dark Object Subtraction 12

1.3.2 Modeling Techniques 13

1.3.2.1 Modeling the Atmospheric Effect 13

1.3.2.2 Steps in Atmospheric Correction 17

1.4 Correction for Topographic Effects 19

1.5 Remote Sensing in the Microwave Region 22

1.6 Radar Fundamentals 23

1.6.1 SLAR Image Resolution 24

1.6.2 Geometric Effects on Radar Images 26

1.6.3 Factors Affecting Radar Backscatter 29

1.6.3.1 Surface Roughness 29

1.6.3.2 Surface Conductivity 30

1.6.3.3 Parameters of the Radar Equation 30

1.7 Imaging Radar Polarimetry 31

1.7.1 Radar Polarization State 32

1.7.2 Polarization Synthesis 34

1.7.3 Polarization Signatures 35

1.8 Radar Speckle Suppression 37

1.8.1 Multilook Processing 37

1.8.2 Filters for Speckle Suppression 38

Chapter 2 Pattern Recognition Principles 41

2.1 Feature Space Manipulation 42

2.1.1 Tasseled Cap Transform 45

2.1.2 Principal Components Analysis 46

2.1.3 Minimum/Maximum Autocorrelation Factors (MAF) 50

2.1.4 Maximum Noise Fraction Transformation 51

2.2 Feature Selection 52

2.3 Fundamental Pattern Recognition Techniques 54

2.3.1 Unsupervised Methods 54

2.3.1.1 The k-means Algorithm 54

2.3.1.2 Fuzzy Clustering 56

2.3.2 Supervised Methods 57

2.3.2.1 Parallelepiped Methods 57

2.3.2.2 Minimum Distance Classifier 57

2.3.2.3 MaximumLikelihood Classifier 58

2.4 Combining Classifiers 61

2.5 Incorporation of Ancillary Information 62

2.5.1 Use of Texture and Context 63

2.5.2 Using Ancillary Multisource Data 63

2.6 Sampling Scheme and Sample Size 65

2.6.1 Sampling Scheme 66

2.6.2 Sample Size, Scale, and Spatial Variability 67

2.6.3 Adequacy of Training Data 69

2.7 Estimation of Classification Accuracy 69

2.8 Epilogue 74

Chapter 3 Artificial Neural Networks 77

3.1 Multilayer Perceptron 77

3.1.1 Back-Propagation 78

3.1.2 Parameter Choice, Network Architecture, and Input/Output Coding 82

3.1.3 Decision Boundaries in Feature Space 84

3.1.4 Overtraining and Network Pruning 88

3.2 Kohonen's Self-Organizing Feature Map 90

3.2.1 SOM Network Construction and Training 90

3.2.1.1 Unsupervised Training 91

3.2.1.2 Supervised Training 93

3.2.2 Examples of Self-Organization 94

3.3 Counter-Propagation Networks 98

3.3.1 Counter-Propagation Network Training 99

3.3.2 Training Issues 101

3.4 Hopfield Networks 101

3.4.1 Hopfield Network Structure 102

3.4.2 Hopfield Network Dynamics 102

3.4.3 Network Convergence 103

3.4.4 Issues Relating to Hopfield Networks 105

3.4.5 Energy and Weight Coding: An Example 106

3.5 Adaptive Resonance Theory (ART) 108

3.5.1 Fundamentals of the ART Model 109

3.5.2 Choice of Parameters 112

3.5.3 Fuzzy ARTMAP 113

3.6 Neural Networks in Remote Sensing Image Classification 116

3.6.1 An Overview 116

3.6.2 A Comparative Study 119

Chapter 4 Support Vector Machines 125

4.1 Linear Classification 126

4.1.1 The Separable Case 126

4.1.2 The Nonseparable Case 129

4.2 Nonlinear Classification and Kernel Functions 130

4.2.1 Nonlinear SVMs 130

4.2.2 Kernel Functions 132

4.3 Parameter Determination 135

4.3.1 t-Fold Cross-Validations 137

4.3.2 Bound on Leave-One-Out Error 138

4.3.3 Grid Search 140

4.3.4 Gradient Descent Method 142

4.4 Multiclass Classification 144

4.4.1 One-against-One, One-against-Others, and DAG 144

4.4.2 Multiclass SVMs 146

4.4.2.1 Vapnik's Approach 146

4.4.2.2 Methodology of Crammer and Singer 147

4.5 Feature Selection 149

4.6 SVM Classification of Remotely Sensed Data 150

4.7 Concluding Remarks 153

Chapter 5 Methods Based on Fuzzy Set Theory 155

5.1 Introduction to Fuzzy Set Theory 155

5.1.1 Fuzzy Sets: Definition 156

5.1.2 Fuzzy Set Operations 157

5.2 Fuzzy C-Means Clustering Algorithm 159

5.3 Fuzzy Maximum Likelihood Classification 162

5.4 Fuzzy Rule Base 164

5.4.1 Fuzzification 165

5.4.2 Inference 169

5.4.3 Defuzzification 171

5.5 Image Classification Using Fuzzy Rules 173

5.5.1 Introductory Methodology 173

5.5.2 Experimental Results 178

Chapter 6 Decision Trees 183

6.1 Feature Selection Measures for Tree Induction 184

6.1.1 Information Gain 185

6.1.2 Gini Impurity Index 188

6.2 ID3, C4.5, and SEE5.0 Decision Trees 189

6.2.1 ID3 189

6.2.2 C4.5 193

6.2.3 SEE5.0 196

6.3 CHAID 197

6.4 CART 198

6.5 QUEST 201

6.5.1 Split Point Selection 201

6.5.2 Attribute Selection 203

6.6 Tree Induction from Artificial Neural Networks 204

6.7 Pruning Decision Trees 205

6.7.1 Reduced Error Pruning (REP) 207

6.7.2 Pessimistic Error Pruning (PEP) 207

6.7.3 Error-Based Pruning (EBP) 208

6.7.4 Cost Complexity Pruning (CCP) 209

6.7.5 Minimal Error Pruning (MEP) 212

6.8 Boosting and Random Forest 214

6.8.1 Boosting 214

6.8.2 Random Forest 215

6.9 Decision Trees in Remotely Sensed Data Classification 217

6.10 Concluding Remarks 220

Chapter 7 Texture Quantization 221

7.1 Fractal Dimensions 222

7.1.1 Introduction to Fractals 223

7.1.2 Estimation of the Fractal Dimension 224

7.1.2.1 Fractal Brownian Motion (FBM) 225

7.1.2.2 Box-Counting Methods and Multifractal Dimension 226

7.2 Frequency Domain Filtering 231

7.2.1 Fourier Power Spectrum 231

7.2.2 Wavelet Transform 235

7.3 Gray-Level Co-Occurrence Matrix (GLCM) 239

7.3.1 Introduction to the GLCM 239

7.3.2 Texture Features Derived from the GLCM 241

7.4 Multiplicative Autoregressive Random Fields 243

7.4.1 MAR Model: Definition 243

7.4.2 Estimation of the Parameters of the MAR Model 245

7.5 The Semivariogram and Window Size Determination 246

7.6 Experimental Analysis 249

7.6.1 Test Image Generation 249

7.6.2 Choice of Texture Features 250

7.6.2.1 Multifractal Dimension 250

7.6.2.2 Fourier Power Spectrum 250

7.6.2.3 Wavelet Transform 250

7.6.2.4 Gray-Level Co-Occurrence Matrix 250

7.6.2.5 Multiplicative Autoregressive Random Field 251

7.6.3 Segmentation Results 251

7.6.4 Texture Measure of Remote Sensing Patterns 252

Chapter 8 Modeling Context Using Markov Random Fields 255

8.1 Markov Random Fields and Gibbs Random Fields 256

8.1.1 Markov Random Fields 256

8.1.2 Gibbs Random Fields 257

8.1.3 MRF-GRF Equivalence 259

8.1.4 Simplified Form of MRF 261

8.1.5 Generation of Texture Patterns Using MRF 263

8.2 Posterior Energy for Image Classification 264

8.3 Parameter Estimation 267

8.3.1 Least Squares Fit Method 268

8.3.2 Results of Parameter Estimations 271

8.4 MAP-MRF Classification Algorithms 273

8.4.1 Iterated Conditional Modes 274

8.4.2 Simulated Annealing 275

8.4.3 Maximizer of Posterior Marginals 277

8.5 Experimental Results 278

Chapter 9 Multisource Classification 283

9.1 Image Fusion 284

9.1.1 Image Fusion Methods 284

9.1.2 Assessment of Fused Image Quality in the Spectral Domain 287

9.1.3 Performance Overview of Fusion Methods 288

9.2 Multisource Classification Using the Stacked-Vector Method 288

9.3 The Extension of Bayesian Classification Theory 290

9.3.1 An Overview 290

9.3.1.1 Feature Extraction 291

9.3.1.2 Probability or Evidence Generation 292

9.3.1.3 Multisource Consensus 292

9.3.2 Bayesian Multisource Classification Mechanism 292

9.3.3 A Refined Multisource Bayesian Model 294

9.3.4 Multisource Classification Using the Markov Random Field 295

9.3.5 Assumption of Intersource Independence 296

9.4 Evidential Reasoning 297

9.4.1 Concept Development 297

9.4.2 Belief Function and Belief Interval 299

9.4.3 Evidence Combination 302

9.4.4 Decision Rules for Evidential Reasoning 304

9.5 Dealing with Source Reliability 304

9.5.1 Using Classification Accuracy 305

9.5.2 Use of Class Separability 305

9.5.3 Data Information Class Correspondence Matrix 306

9.5.4 The Genetic Algorithm 307

9.6 Experimental Results 309

Bibliography 317

Index 349