Hyperspectral image analysis

Contents......Page 6
1 Introduction......Page 8
2 Machine Learning Methods for Spatial and Temporal Parameter Estimation......Page 12
  2.1.2 Data and Model Challenges......Page 13
  2.1.3 Goals and Outline......Page 16
 2.2 Gap Filling and Multi-sensor Fusion......Page 17
  2.2.2 LMC-GP......Page 18
  2.2.4 Results......Page 21
 2.3 Distribution Regression for Multiscale Estimation......Page 24
  2.3.1 Kernel Distribution Regression......Page 26
  2.3.2 Data and Setup......Page 29
  2.3.3 Results......Page 31
 2.4 Global Parameter Estimation in the Cloud......Page 32
  2.4.1 Data and Setup......Page 33
  2.4.2 Results......Page 35
 2.5 Conclusions......Page 37
 References......Page 38
 3.1 Introduction......Page 43
  3.1.1 History of Deep Learning in Computer Vision......Page 45
  3.1.2 History of Deep Learning for HSI Tasks......Page 46
  3.1.3 Challenges......Page 47
  3.2.1 Perceptron......Page 48
  3.2.2 Multi-layer Neural Networks......Page 50
  3.2.3 Learning and Gradient Computation......Page 53
  3.3.1 Autoencoders......Page 57
  3.3.3 Recurrent Neural Networks......Page 59
  3.3.4 Long Short-Term Memory......Page 61
 3.4 Convolutional Neural Networks......Page 63
  3.4.1 Building Blocks of CNNs......Page 64
  3.4.2 CNN Flavors for HSI......Page 66
 3.5 Software Tools for Deep Learning......Page 69
 References......Page 70
4 Deep Learning for Hyperspectral Image Analysis, Part II: Applications to Remote Sensing and Biomedicine......Page 75
 4.1 Introduction......Page 76
  4.2.1 Remote Sensing Case Study: Urban Land Cover Classification......Page 78
  4.2.2 Biomedical Application: Tissue Histology......Page 79
  4.3.1 Practical Considerations......Page 82
  4.3.2 Related Developments in the Community......Page 83
  4.4.1 CNNs......Page 86
  4.4.2 RNNs......Page 88
  4.4.3 CRNNs......Page 89
  4.5.1 Remote Sensing Results......Page 90
  4.5.2 Biomedical Results......Page 101
 4.6 Design Choices and Hyperparameters......Page 108
  4.6.2 Pooling Layer Hyperparameters......Page 109
  4.6.3 Training Hyperparameters......Page 110
  4.6.4 General Model Hyperparameters......Page 112
  4.6.5 Regularization Hyperparameters......Page 113
 4.7 Concluding Remarks......Page 115
 References......Page 116
 5.1 Deep Learning—Challenges presented  by Hyperspectral Imagery......Page 122
 5.2 Robust Learning with Limited Labeled Data......Page 124
  5.2.1 Unsupervised Feature Learning......Page 125
  5.2.2 Semi-supervised learning......Page 127
  5.2.3 Active learning......Page 129
  5.3.1 Transfer Learning and Domain Adaptation......Page 132
  5.3.2 Transferring Knowledge—Beyond Classification......Page 138
 5.4 Data Augmentation......Page 139
 References......Page 141
 6.1 Motivating Examples for Multiple Instance Learning  in Hyperspectral Analysis......Page 146
 6.2 Introduction to Multiple Instance Classification......Page 150
  6.2.2 Axis-Parallel Rectangles, Diverse Density, and Other General MIL Approaches......Page 151
 6.3 Multiple Instance Learning Approaches for Hyperspectral Target Characterization and Sub-pixel Target Detection......Page 154
  6.3.1 Extended Function of Multiple Instances......Page 155
  6.3.2 Multiple Instance Spectral Matched Filter and Multiple Instance Adaptive Coherence/Cosine Detector......Page 157
  6.3.3 Multiple Instance Hybrid Estimator......Page 159
  6.3.4 Multiple Instance Learning for Multiple Diverse Hyperspectral Target Characterizations......Page 161
  6.3.5 Experimental Results for MIL in Hyperspectral Target Detection......Page 163
  6.4.1 Multiple Instance Choquet Integral Classifier Fusion......Page 171
  6.4.2 Multiple Instance Regression......Page 174
  6.4.3 Multiple Instance Multi-resolution and Multi-modal Fusion......Page 178
 6.5 Summary......Page 186
 References......Page 187
7 Supervised, Semi-supervised,  and Unsupervised Learning  for Hyperspectral Regression......Page 191
 7.1 Introduction to Hyperspectral Regression......Page 192
 7.2 Fundamentals of Hyperspectral Regression......Page 195
 7.3 Hyperspectral Regression at the Data Level......Page 197
  7.3.2 Dataset Shift......Page 198
  7.3.3 Dataset Splitting......Page 200
 7.4 Hyperspectral Regression at the Feature Level......Page 203
  7.4.1 Dimensionality Reduction......Page 204
  7.4.2 Clustering......Page 205
  7.4.3 Feature Engineering and Feature Selection......Page 208
  7.5.1 Supervised Learning Models......Page 209
  7.5.2 Semi-supervised Learning for Regression......Page 221
  7.5.3 Model Selection, Optimization, and Evaluation......Page 224
 7.6 Summary and Trends in Hyperspectral Regression......Page 227
  7.6.2 Trends at the Feature Level: Domain Knowledge......Page 228
 References......Page 229
 8.1 Introduction......Page 237
 8.2 Sparse Representation-Based HSI Classification......Page 238
 8.3 Advanced Models of Sparse Representation for Hyperspectral Image Classification......Page 240
  8.4.1 Model and Algorithm......Page 247
  8.4.2 Experimental Results and Discussion......Page 250
 References......Page 259
 9.1 Introduction......Page 262
 9.2 Learning from Multiple Kernels......Page 264
  9.2.1 General MKL......Page 265
  9.2.2 Strategies for MKL......Page 267
  9.3.1 Subspace MKL......Page 268
  9.3.2 Nonlinear MKL......Page 269
  9.3.3 Sparsity-Constrained MKL......Page 271
  9.3.4 Ensemble MKL......Page 274
  9.3.5 Heterogeneous Feature Fusion with MKL......Page 275
  9.3.6 MKL with Superpixel......Page 277
  9.4.1 Hyperspectral Data Sets......Page 278
  9.4.2 Experimental Settings and Evaluation......Page 279
  9.4.3 Spectral Classification......Page 284
  9.4.4 Spatial–Spectral Classification......Page 286
  9.4.5 Classification with Heterogeneous Features......Page 288
  9.4.6 Superpixel-Based Classification......Page 290
 9.5 Conclusion......Page 292
 References......Page 293
 10.1 Introduction......Page 297
 10.2 Low Dimensional Manifold Model......Page 298
  10.2.2 Model Formulation and Calculating the Manifold Dimension......Page 299
 10.3 Two Numerical Approaches of Solving the LDMM Model......Page 302
  10.3.1 The First Approach......Page 303
  10.3.2 The Second Approach......Page 305
  10.3.3 A Comparison of the Two Approaches......Page 311
  10.4.1 Experimental Setup......Page 313
  10.4.2 Reconstruction from Noise-Free Subsample......Page 314
  10.4.3 Reconstruction from Noisy Subsample......Page 316
 10.5 Conclusion......Page 317
 References......Page 318
 11.1 Introduction......Page 320
  11.2.1 Hyperspectral Imaging Techniques......Page 322
  11.2.2 Spectral Face Recognition......Page 324
  11.2.3 Spectral Band Selection......Page 325
 11.3 Sparsity......Page 326
  11.4.1 Network Pruning......Page 328
  11.4.3 Decomposition......Page 329
 11.7 Convolutional Neural Network......Page 330
  11.7.1 Convolutional Layer......Page 331
  11.7.3 Fully Connected Layer......Page 332
  11.7.5 Activation Function: ReLU......Page 333
  11.7.6 VGG-19 Architecture......Page 334
  11.8.1 Proposed Structured Sparsity Learning for Generic Structures......Page 335
  11.8.3 Face Recognition Loss Function......Page 336
  11.8.4 Band Selection via Group Lasso......Page 337
  11.9.2 Initializing Parameters of the Network......Page 338
  11.9.4 Hyperspectral Face Datasets......Page 339
  11.9.5 Parameter Sensitivity......Page 341
  11.9.7 Band Selection......Page 342
  11.9.8 Effectiveness of SSL......Page 343
  11.9.9 Comparison......Page 345
 11.10 Conclusion......Page 346
 References......Page 347
12 Detection of Large-Scale and Anomalous Changes......Page 352
 12.1 Introduction......Page 353
 12.2 Change Detection: The Fundamentals......Page 354
  12.3.1 Change Vector Analysis and Related Methods......Page 356
  12.3.3 Transformation-Based Methods......Page 359
 12.4 Approaches for Anomalous Change Detection......Page 362
  12.4.1 Difference-Based Methods......Page 363
  12.4.3 Joint-Distribution Methods......Page 365
  12.4.4 Other Methods......Page 366
 12.5 Operational Considerations......Page 367
  12.6.1 Details on the Data......Page 368
  12.6.2 Illustrative Examples......Page 370
 References......Page 372
13 Recent Advances in Hyperspectral Unmixing Using Sparse Techniques  and Deep Learning......Page 377
  13.1.1 Spectral Unmixing......Page 378
  13.1.2 Sparse Unmixing......Page 379
  13.1.3 Deep Learning for Spectral Unmixing......Page 380
  13.1.4 Contributions of This Chapter......Page 381
  13.2.1 Sparse Versus Spectral Unmixing......Page 382
  13.2.3 Total Variation Regularization......Page 383
  13.2.5 Double Reweighted Regularization......Page 384
  13.2.6 Spectral–Spatial Weighted Regularization......Page 385
 13.3 Deep Learning for Hyperspectral Unmixing......Page 386
  13.3.2 Deep Auto-Encoder Network......Page 387
  13.3.3 Stacked Auto-Encoders for Initialization......Page 388
  13.3.4 Variational Auto-Encoders for Unmixing......Page 389
 13.4 Experiments and Analysis: Sparse Unmixing......Page 392
  13.4.2 Real Hyperspectral Data......Page 393
 13.5 Experiments and Analysis: Deep Learning......Page 396
  13.5.1 Mangrove Dataset......Page 398
  13.5.2 Samson Dataset......Page 399
 13.6 Conclusions and Future Work......Page 400
 References......Page 403
14 Hyperspectral–Multispectral Image Fusion Enhancement Based on Deep Learning......Page 406
 14.1 Introduction......Page 407
  14.2.1 HSI Pan-sharpening......Page 408
  14.2.2 HSI Super-Resolution......Page 409
  14.2.4 Deep Learning Based Image Enhancement......Page 410
  14.3.1 Spectral–Spatial Deep Feature Learning......Page 413
  14.3.2 HSI-MSI Fusion......Page 414
  14.3.3 Experimental Results......Page 416
 14.4 Conclusions and Discussions......Page 427
 References......Page 428
15 Automatic Target Detection for Sparse Hyperspectral Images......Page 433
  15.1.2 Hyperspectral Target Detection: Concept and Challenges......Page 434
  15.1.3 Goals and Outline......Page 437
  15.1.4 Summary of Main Notations......Page 438
 15.2 Related Works......Page 439
 15.3 Main Contribution......Page 440
  15.3.1 Recovering a Low-Rank Background Matrix and a Sparse Target Matrix by Convex Optimization......Page 441
 15.4 Experiments and Analysis......Page 445
  15.4.1 Construction of the Target Dictionary At......Page 448
  15.4.2 Target Detection Evaluation......Page 450
 15.5 Conclusion and Future Work......Page 455
 References......Page 456
Index......Page 461