Advances in Hyperspectral Image Processing Techniques (IEEE Press)

Cover
Title Page
Copyright Page
Contents
Editor Biography
List of Contributors
Preface
Part I General Theory
Chapter 1 Introduction: Two Fundamental Principles Behind Hyperspectral Imaging
1.1 Introduction
1.2 Why Is Hyperspectral Imaging?
1.3 Two Principles for Hyperspectral Imaging
1.3.1 Pigeon-Hole Principle
1.3.2 Orthogonality Principle
1.4 What Are the Issues of Hyperspectral Imaging?
1.5 Determination of p by Virtual Dimensionality via Pigeon-Hole Principle
1.6 Order Determination of Low Rank and Sparse Matrices by Virtual Dimensionality via Pigeon-Hole Principle
1.7 Band Selection by Pigeon-Hole Principle
1.8 Band Selection by a Hyperspectral Band Channel via Pigeon-Hole Principle
1.9 Band Sampling via Pigeon-Hole Principle
1.10 Spectral Unmixing via Orthogonality Principle
1.11 Target Detection by Orthogonality Principle
1.11.1 ATGP
1.11.1.1 Automatic Target Generation Process (ATGP)
1.11.2 Constrained Energy Minimization (CEM)
1.12 Anomaly Detection by Orthogonality Principle
1.13 Endmember Finding by Orthogonality Principle
1.13.1 Pixel Purity Index (PPI)
1.13.2 Vertex Component Analysis (VCA)
1.13.3 Simplex Growing Algorithm (SGA)
1.14 Low Rank and Sparse Representation by OSP via Orthogonality Principle
1.15 Hyperspectral Classification
1.15.1 Hyperspectral Mixed Pixel Classification (HMPC)
1.15.2 Number of Sampled Bands Lower Than Number of Classes
1.15.3 Potential and Promise of Band Sampling in HMPC
1.16 Conclusion
References
Chapter 2 Overview of Hyperspectral Imaging Remote Sensing from Satellites
2.1 Hyperspectral Imaging Remote Sensing from Airplanes to Satellites
2.1.1 History of Development of Airborne Hyperspectral Imagers
2.1.2 Early Development of Spaceborne Hyperspectral Imagers
2.2 Development of Spaceborne Hyperspectral Imagers in the Last Two Decades
2.2.1 Survey of Spaceborne Hyperspectral Imagers
Acronyms List
2.2.2 Brief Description of Spaceborne Hyperspectral Imagers
2.2.2.1 Visible Imagers and Spectrographic Imagers (UVISI) Onboard the MSX Satellite
2.2.2.2 HyperSpectral Imager (HSI) for the LEWIS Mission
2.2.2.3 MODIS Onboard Terra and Aqua Satellites
2.2.2.4 Hyperion Onboard NASA's EO-1 Satellite
2.2.2.5 CHRIS Onboard ESA's PROBA Satellite
2.2.2.6 MERIS Onboard ESA's ENVISAT Satellite
2.2.2.7 VIRTIS for ESA's Rosetta, Venus-Express, and NASA-Dawn Planetary Missions
2.2.2.8 CRISM Aboard Mars Reconnaissance Orbiter
2.2.2.9 Moon Mineralogy Mapper for Mapping Lunar Surface
2.2.2.10 Fourier Transform Hyperspectral Imager Onboard Chinese Environment Satellite
2.2.2.11 HySI Onboard Indian Mini Satellite-1
2.2.2.12 ARTEMIS Onboard TacSat-3
2.2.2.13 HICO Onboard the International Space Station
2.2.2.14 Visible and Near-infrared Imaging Spectrometer Aboard Chang'E 3 Spacecraft
2.2.2.15 Ocean and Land Color Imager (OLCI) on Sentinel-3A
2.2.2.16 Miniature High-Resolution Imaging Spectrometer on GHGSat-D
2.2.2.17 Aalto-1 Spectral Imager .(AaSI) on a 3U Nanosatellite
2.2.2.18 DLR Earth Sensing Imaging Spectrometer on the International Space Station
2.2.2.19 HyperScout Hyperspectral Camera on ESA's Nanosatellite GomX-4B
2.2.2.20 Advanced Hyperspectral Imager (AHSI) on Chinese Gaofen-5 Satellite
2.2.2.21 Italian Hyperspectral Satellite PRISMA
2.2.2.22 Hyperspectral Imager Suite Onboard the International Space Station
2.2.2.23 German Spaceborne Hyperspectral Imager EnMAP
2.2.2.24 ESA's Moons and Jupiter Imaging Spectrometer (MAJIS)
2.3 Conclusion
References
Chapter 3 Efficient Hardware Implementation for Hyperspectral Anomaly and Target Detection
3.1 Introduction
3.2 Hyperspectral Anomaly and Target Detection
3.2.1 DPBS-CEM
3.2.2 DBN-RXD
3.2.3 Fast-ATGP
3.2.4 Fast-MGD
3.3 Model-Based Design
3.3.1 What is Model-Based Design?
3.3.2 FPGA Development Based on MBD
3.3.3 Examples of IP Design Based on HLS
3.3.3.1 Efficient Off-Chip Storage Access IP
3.3.3.2 Parallel Matrix Multiplication IP
3.3.3.3 Matrix Dot-Product-Plus IP
3.3.3.4 Erosion/Dilation IP
3.4 System Integration Framework Design
3.4.1 Efficient FPGA Implementation
3.4.1.1 FPGA Implementation of DPBS-CEM
3.4.1.2 FPGA Implementation of DBN-RXD
3.4.1.3 FPGA Implementation of Fast-ATGP
3.4.1.4 FPGA Implementation of Fast-MGD
3.5 Experiments and Discussions
3.5.1 Hyperspectral Image Data Set
3.5.1.1 TE1 Data Set
3.5.1.2 HyMap Data Set
3.5.1.3 Airport-Beach-Urban. .(ABU) Data Set
3.5.1.4 Cuprite Data Set
3.5.1.5 San Diego Data Set
3.5.1.6 HYDICE Data Set
3.5.2 Experiments of DPBS-CEM
3.5.2.1 Detection Accuracy
3.5.2.2 Acceleration Performance
3.5.3 Experiments of DBN-RXD
3.5.3.1 Detection Accuracy
3.5.3.2 Acceleration Performance
3.5.4 Experiments of Fast-ATGP
3.5.4.1 Detection Accuracy
3.5.4.2 Results for the AVIRIS Cuprite Scene
3.5.5 Experiments of Fast-MGD
3.5.5.1 Detection Accuracy
3.5.5.2 Performance Evaluation
3.6 Conclusion
References
Part II Band Selection for Hyperspectral Imaging
Chapter 4 Constrained Band Selection for Hyperspectral Imaging
4.1 Introduction
4.2 Constrained BS
4.2.1 Band Vector-Constrained BS
4.2.1.1 Band Correlation Minimization (BCM)
4.2.1.2 Band Dependence Minimization (BDM)
4.2.1.3 Band Correlation Constraint (BCC)
4.2.1.4 Band Dependence Constraint (BDC)
4.2.2 Band Image-Constrained BS
4.3 BCBS Experiments
4.3.1 HYDICE Data
4.3.1.1 Target Detection
4.3.1.2 Unsupervised Mixed Pixel Classification
4.3.2 AVIRIS Cuprite Data
4.4 Target-Constrained BS
4.4.1 Target-Constrained Band Prioritization
4.4.1.1 Single Band Minimum Variance Band Prioritization by TCBS
4.4.1.2 Leave-One-Out Maximum Variance Band Prioritization by TCBS
4.4.2 Constrained-Target Band Selection
4.4.2.1 Sequential Feed-Forward TCBS
4.4.2.2 Sequential Backward TCBS
4.5 TCBS Experiments
4.6 Conclusion
References
Chapter 5 Band Subset Selection for Hyperspectral Imaging
5.1 Introduction
5.2 Simultaneous Multiple Band Selection
5.3 Search Strategies for BSS
5.3.1 Sequential Band Subset Selection
5.3.2 Successive Band Subset Selection
5.4 Channel Capacity BSS
5.5 Multiple Band-Constrained Band Subset Selection
5.5.1 Constrained BSS (CBSS)
5.5.2 Search Algorithms for CBSS
5.5.2.1 Sequential CBSS (SQ CBSS)
5.5.2.2 Successive CBSS (SC CBSS)
5.6 Application-Specified BSS (AS-BSS)
5.6.1 Application to Hyperspectral Classification
5.6.2 LCMV Criterion for BSS
5.6.3 LCMV-BSS Algorithms
5.6.3.1 SQ LCMV-CBSS
5.6.3.2 SC LCMV-CBSS
5.7 Experiments
5.7.1 MBC-BSS
5.7.2 MTC-BSS
5.7.2.1 Purdue Indiana Indian Pines Scene
5.7.2.2 Salinas
5.7.2.3 ROSIS Data
5.8 Conclusion
References
Chapter 6 Progressive Band Selection Processing for Hyperspectral Image Classification
6.1 Introduction
6.2 Measures of Class Classification Priority
6.3 p-Ary Huffman Coding Tree Construction
6.4 Iterative LCMV
6.4.1 Linearly Constrained Minimum Variance (LCMV)
6.4.2 Iterative Linearly Constrained Minimum Variance (ILCMV)
6.5 Class Signature Constrained Band Prioritization-Based Band Selection
6.6 Progressive Band Selection
6.7 Classification Measures
6.8 Real Images to be Used for Experiments
6.8.1 Purdue Indiana Indian Pines
6.8.2 Salinas
6.8.3 ROSIS Data
6.9 Experiments
6.9.1 Purdue Indiana Indian Pines
6.9.2 Salinas
6.9.3 University of Pavia
6.10 Conclusion
References
Part III Compressive Sensing for Hyperspectral Imaging
Chapter 7 Restricted Entropy and Spectrum Properties for Hyperspectral Imaging
7.1 Introduction
7.2 Compressive Sensing Review
7.3 Restricted Entropy Property
7.4 Restricted Spectrum Property
7.5 REP and RSP Hyperspectral Measures
7.6 Experiments
7.7 Conclusion
References
Chapter 8 Endmember Finding in Compressively Sensed Band Domain
8.1 Introduction
8.2 Compressive Hyperspectral Band Sensing
8.2.1 Compressive Sensing Framework
8.2.2 Compressive Sensing of Hyperspectral Bands
8.2.3 Universality Model
8.3 Simplex Volume Calculation
8.3.1 Simplex Volume via Singular Value Decomposition
8.3.2 Simplex Volume via Matrix Determinant
8.4 Restricted Simplex Volume Property
8.5 Two Sequential Algorithms for p-FINDR
8.5.1 SeQuential p-FINDR (SQ p-FINDR)
8.5.2 SuCcessive p-FINDR (SC p-FINDR)
8.5.3 SQ p-FINDR and SC p-FINDR in CSBD
8.6 Experiments
8.6.1 Experimental Setup
8.6.2 Algorithm Analysis on Experimental Data
8.7 Experimental Results and Discussions
8.7.1 SQ p-FINDR Experimental Result Analysis
8.7.2 SC p-FINDR Experimental Result Analysis
8.8 Conclusion
References
Chapter 9 Hyperspectral Image Classification in Compressively Sensed Band Domain
9.1 Introduction
9.2 Compressive Sensing Review
9.2.1 Compressive Sensing Framework
9.2.2 Compressive Sensing of Hyperspectral Bands
9.2.3 Universality (Universal Sensed Model)
9.3 Hyperspectral Image Classification
9.3.1 Linear Support Vector Machines
9.3.2 Kernel Support Vector Machines
9.3.3 Edge-Preserving Filters
9.4 Classification Measures
9.5 Experiments
9.5.1 Experimental Setup
9.5.2 Classification Accuracy Analysis
9.5.2.1 Purdue Indian Pines Scene
9.5.2.2 Salinas Scene
9.5.2.3 Pavia University
9.5.2.4 Pavia Centre
9.5.3 Classification Precision Analysis
9.5.3.1 Purdue Indian Pines Scene
9.5.3.2 Salinas and Pavia Scenes
9.5.4 Discussions on Individual Class Accuracies
9.5.5 Scene Complexity Analysis
9.5.6 Selecting an Appropriate Number of CSBs
9.6 Conclusion
References
Part IV Fusion for Hyperspectral Imaging
Chapter 10 Hyperspectral and LiDAR Data Fusion
10.1 Introduction
10.2 Deep Learning-Based HSI and LiDAR Data Classification
10.2.1 Two-Branch CNN for Joint Classification
10.2.2 Hierarchical Random Walk Network (HRWN)
10.2.3 Residual Network-Based Probability Reconstruction Fusion (RNPRF)
10.3 Experiments
10.3.1 Experimental Data
10.3.2 Classification Performance
10.4 Conclusions
References
Chapter 11 Hyperspectral Data Fusion Using Multidimensional Information
11.1 Introduction of Remote Sensing Data Fusion
11.1.1 Basic Idea of Data Fusion
11.1.2 Why Do We Need Data Fusion?
11.2 Common Methods for Data Fusion
11.2.1 Fusion Methods Emphasizing on Spatial Resolution Enhancement
11.2.1.1 Component Constitution Method
11.2.1.2 Multiresolution Analysis Method
11.2.2 Fusion Methods Emphasizing on Spectral Resolution Enhancement
11.2.2.1 Linear Optimization Method
11.2.2.2 Nonlinear Optimization Method
11.2.3 Fusion Methods Emphasizing on Temporal Resolution Enhancement
11.2.3.1 Weight-Based Method
11.2.3.2 Linear Optimization Method
11.2.3.3 Nonlinear Optimization Method
11.2.4 Quality Assessment of Data Fusion
11.2.4.1 Quality Assessment with Reference Image
11.2.4.2 Quality Assessment Without Reference Image
11.3 Enhancing Spectral Resolution of Multispectral Data Using Deep Learning Method
11.3.1 Data Fusion Methods Using Deep Learning
11.3.2 Spectral Resolution Enhancement Method via Convolutional Neural Networks (SRECNN)
11.3.3 Application 1: Extending the Swath of Hyperspectral Data
11.3.4 Application 2: Cloud Removal of GF-5 Hyperspectral Data
11.4 Multidimensional Datasets (MDD)
11.4.1 Introduction of Multidimensional Datasets (MDD)
11.4.1.1 Temporal Sequential in Band (TSB)
11.4.1.2 Temporal Sequential in Pixel (TSP)
11.4.1.3 Temporal Interleaved by Band (TIB)
11.4.1.4 Temporal Interleaved by Pixel (TIP)
11.4.1.5 Temporal Interleaved by Spectrum (TIS)
11.4.2 Spatial-Temporal Data Fusion via Multidimensional Datasets
11.4.2.1 Basic Architecture
11.4.2.2 Dataset Experiment
11.5 A Fusion-Related Case Study: Comparison of Fusion Methods on GF-5 Hyperspectral Data
11.5.1 Backgrounds
11.5.2 Experiment
11.5.3 Visual Analysis
11.5.4 Index Evaluation
11.5.5 Classification Application
11.5.6 Discussion
11.6 Conclusion
References
Chapter 12 Fusion of Band Selection Methods for Hyperspectral Imaging
12.1 Introduction
12.2 Band Selection Fusion
12.2.1 Simultaneous Band Selection Fusion
12.2.2 Progressive Band Selection Fusion
12.3 Experiments
12.3.1 Linear Spectral Unmixing
12.3.2 Hyperspectral Image Classification
12.3.2.1 AVIRIS Data
12.3.2.2 ROSIS Data
12.4 Conclusion
References
Part V Hyperspectral Data Unmixing
Chapter 13 Model-Inspired Deep Neural Networks for Hyperspectral Unmixing
13.1 Model-Based and Learning-Based Spectral Unmixing
13.1.1 Mode-Based Spectral Unmixing
13.1.2 Learning-Based Spectral Unmixing
13.2 Model-Inspired Learning for Spectral Unmixing
13.2.1 Why Should Model-Driven and Data-Driven Techniques Be Combined?
13.2.2 How Are Model-Inspired Unmixing Network Architectures Designed?
13.3 Unfolded Iterative Shrinkage-Thresholding Model for Supervised Abundance Estimation
13.3.1 Linear Mixture Model
13.3.2 Model Optimization
13.3.3 Unfolded ISTA for Abundance Estimation
13.3.4 Experimental Results on Synthetic Data
13.3.4.1 Data Generation
13.3.4.2 Evaluation Index
13.3.4.3 Experiment Setting
13.3.4.4 Impact of Sampling Strategy
13.3.4.5 Impact of Number of Layers
13.3.4.6 Impact of Number of Training Samples
13.3.4.7 Robustness to Noise
13.3.4.8 Running Time Comparison
13.3.5 Experimental Results on Real-World Data
13.4 Model-Inspired Network Architectures for Blind Unmixing
13.4.1 Unsupervised Model-Inspired NN for Blind Unmixing
13.4.2 Experimental Results on Synthetic Data
13.4.2.1 Evaluation Index
13.4.2.2 Impact of Number of Layers
13.4.2.3 Impact of Number of Training Samples
13.4.2.4 Robustness to Noise
13.4.2.5 Running Time Comparison
13.4.3 Experimental Results on Real-World Data
13.5 NMF-Inspired Sparse Autoencoder for Hyperspectral Unmixing
13.5.1 Model Optimization
13.5.2 Network Architecture
13.5.3 Experimental Results on Synthetic Data
13.5.3.1 Influence of the Number of Training Samples
13.5.3.2 Comparison with the State-of-the-Arts
13.5.4 Experimental Results on Real-World Data
13.6 Learning a Deep Alternating Neural Network for Hyperspectral Unmixing
13.6.1 Model Formulation
13.6.2 Model Optimization
13.6.3 SNMF-NET
13.6.3.1 Connection Between Proximal Gradient Method and DNN
13.6.3.2 Lp-NMF Inspired Deep Alternating Neural Network
13.6.4 Experimental Results on Synthetic Data
13.6.4.1 The Influence of Layers
13.6.4.2 Influence of the Number of Training Samples
13.6.4.3 Influence of Initialization
13.6.4.4 Robustness to Noise
13.6.4.5 Experiments on Real-world Data
13.7 Conclusion
References
Chapter 14 Analytical Fully Constrained Least Squares Linear Spectral Mixture Analysis
14.1 Introduction
14.2 Linear Spectral Mixture Analysis
14.3 Fully Constrained Least Squares Method
14.3.1 Abundance Sum-to-one-Constrained LSMA
14.3.2 Abundance Non-negativity-Constrained LSMA
14.3.3 Abundance Fully Constrained LSMA
14.4 Modified Fully Constrained Linear Squares Method
14.5 Analytical Non-negativity-Constrained Linear Squares Method
14.6 Analytical Fully Constrained Least Squares Method
14.7 Experiments
14.8 Conclusion
References
Chapter 15 Swarm Intelligence Optimization-Based Spectral Unmixing
15.1 Introduction
15.2 Pixel Mixing Models
15.3 Swarm Intelligence Optimization-Based Approaches
15.3.1 Approach Based on LMM
15.3.2 Approach Based on NCM
15.3.2.1 "Winner-Take-All" Version of the EM
15.3.2.2 PSO for Abundance Optimization
15.3.2.3 Unmixing Based on PSO-EM Algorithm
15.3.3 Approach Based on NLMM
15.4 Experiments
15.4.1 Experimental Data
15.4.1.1 Moffett Field Data
15.4.1.2 Cuprite Data
15.4.2 Results of Approaches Based on LMM
15.4.2.1 Pure Pixel-Based Algorithm
15.4.2.2 Minimum Volume-Based Algorithm
15.4.3 Results of Approaches Based on NCM
15.4.3.1 Experiment Using Moffett Field Data
15.4.3.2 Experiment Using Cuprite Data
15.4.4 Results of Approaches Based on NLMM
15.4.4.1 Experiment Using Moffett Field Data
15.4.4.2 Experiment Using Cuprite Data
15.5 Conclusion
Acknowledgments
References
Chapter 16 Spectral-Spatial Robust Nonnegative Matrix Factorization for Hyperspectral Unmixing
16.1 Introduction
16.2 Robust NMF
16.2.1 RNMF Using ℓ2,1 Norm
16.2.2 RNMF Using ℓ1, 2 Norm
16.3 Spectral-Spatial Robust Nonnegative Matrix Factorization (SSRNMF)
16.3.1 SSRNMF Model
16.3.2 Update Rules for SSRNMF
16.3.3 Implementation Issues
16.3.4 Computation Complexity Analysis
16.4 Experiments on Synthetic Data
16.4.1 Robustness Analysis to Noise Composition
16.4.2 Robustness Analysis to Noise Intensity
16.4.3 Performance Comparison When the Number of Endmembers Varies
16.5 Experiments on First Real Data
16.5.1 Results Without the Low-SNR Bands
16.5.2 Results with the Low-SNR Bands
16.6 Experiments on Second Real Data
16.7 Conclusion
References
Part VI Hyperspectral Image Classification
Chapter 17 Sparse Representation-Based Hyperspectral Image Classification
17.1 Introduction
17.2 Classic Representation-Based Models
17.2.1 Sparse Representation-Based Framework
17.2.2 Joint Representation-Based Framework
17.3 Sparse Representation-Based Hyperspectral Image Classification
17.3.1 Approach in the Spectral Domain
17.3.2 Approach in the Spectral-Spatial Domain
17.4 Experimental Results and Analysis
17.4.1 Experimental Data
17.4.1.1 Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) Indian Pines Scene
17.4.1.2 Reflective Optics Spectrographic Imaging System (ROSIS) University of Pavia Scene
17.4.2 Parameter Tuning
17.4.3 Results and Analysis
17.5 Conclusion
Acknowledgments
References
Chapter 18 Collaborative Classification Based on Hyperspectral Images
18.1 Introduction
18.1.1 Hyperspectral and Panchromatic Images
18.1.2 Hyperspectral and Multispectral Images
18.1.3 LWIR Hyperspectral and Visible Images
18.2 Problems and Challenges in Multisource Image Collaborative Processing
18.3 Collaborative Classification of Hyperspectral and High-Resolution Panchromatic Images
18.3.1 Methodology
18.3.2 Experiments
18.4 Collaborative Classification of Infrared Hyperspectral and Visible Images
18.4.1 Methodology
18.4.2 Experiments
18.5 Conclusions
References
Chapter 19 Class Feature-Weighted Hyperspectral Image Classification
19.1 Introduction
19.2 Class Feature Descriptors
19.2.1 Intra-Class Feature Descriptors for Class Variability
19.2.2 Inter-Class Feature Descriptors for Class Separability
19.2.3 Total Class Features
19.2.4 Calculation of CF Probabilities
19.3 Allocation of Class Training Sample Size
19.4 CFW-HSIC
19.5 Experiments
19.5.1 Purdue University's Indiana Indian Pines
19.5.2 University of Pavia
19.6 Novelties
19.7 Conclusion
References
Chapter 20 Target Detection Approaches to Hyperspectral Image Classification
20.1 Introduction
20.2 Signal Detection Theory
20.3 Binary Classification Theory
20.4 Multiple Hypotheses Testing Theory
20.4.1 Multi-Target Detection Problems
20.4.1.1 One-Against-All (Winner-Take-All) Approach
20.4.1.2 One-Against-One Approach
20.4.1.3 Multi-Class Classification Problems
20.5 Iterative Constrained Energy Minimization
20.6 Iterative Linearly Constrained Minimum Variance
20.7 How to Convert Detection to Classification
20.8 Real Image Experiments
20.8.1 Purdue University´s Indiana Indian Pines
20.8.2 University of Pavia
20.9 Conclusion
References
Index
EULA