Inhalt

•  
   Acknowledgements
•  
   Zusammenfassung
•  
   Abstract
•  
   Contents
•  
   List of Figures
•  
  1 Introduction
•  
   1.1 Motivation
•  
   1.2 Main Goals of This Work
•  
   1.3 Contributions of the Dissertation
•  
   1.4 Outline of the Dissertation
•  
  2 Data Characteristics and Basic Mathematics
•  
  2.1 Data Characteristics
•  
   2.1.1 Introduction to Remote Sensing
•  
  2.1.2 Synthetic Aperture Radar
•  
   2.1.2.1 Radar Principle
•  
  2.1.3 SAR Statistical Properties
•  
   2.1.3.1 Speckle Effect
•  
   2.1.3.2 Statistical Properties of the Backscattered Signal
•  
  2.1.4 Speckle Reduction
•  
   2.1.4.1 Multi-Look Processing
•  
  2.2 Basic Mathematics
•  
   2.2.1 Probability Distributions
•  
   2.2.1.1 Copula-based Joint Probability Modeling
•  
   2.2.1.2 Gaussian Mixture Models (GMM)
•  
   2.2.1.3 Bayes' Rule
•  
  2.2.2 Parameter Estimation Methods
•  
   2.2.2.1 Method of Moments (MoM)
•  
   2.2.2.2 Method of Maximum Likelihood Estimation (MLE)
•  
   2.2.2.3 Method of Maximum A Posterior Estimation (MAP)
•  
   2.2.2.4 Method of Log-Cumulants (MoLC)
•  
   2.2.2.5 Expectation Maximization (EM)
•  
  2.2.3 Numerical Optimization Methods
•  
   2.2.3.1 Gradient Descent
•  
   2.2.3.2 Newton's Method
•  
   2.2.3.3 Stochastic Gradient Descent
•  
  2.2.4 Sampling Methods
•  
   2.2.4.1 Markov Chain Monte Carlo (MCMC)
•  
   3 State of the Art
•  
   3.1 Earth Observation Meets Computer Vision
•  
   3.2 Hierarchical Representation
•  
   3.2.1 Feature Hierarchy
•  
   3.2.2 Semantic Hierarchy
•  
   3.3 Description of Images
•  
   3.3.1 Feature Extraction
•  
   3.3.1.1 Multi-Spectral Information
•  
   3.3.1.2 Textural Information
•  
   3.3.1.3 Geometric Information
•  
   3.3.2 Feature Encoding
•  
   3.3.3 The Curse of Dimensionality
•  
   3.3.4 Distance Metrics
•  
   3.3.4.1 Fractional and Minkowski Distances
•  
   3.3.4.2 Distance Metric Learning
•  
   3.4 Machine Learning
•  
   3.4.1 Classic Machine Learning Methods
•  
   3.4.2 New Trends in Semi-supervised Learning
•  
   3.4.3 Object Extraction-based Semantic Exploration
•  
   3.5 Conclusions and Proposed Concepts
•  
   3.5.1 Applied Dataset
•  
   3.5.2 Semi-supervised Learning
•  
  4 Application and Evaluation of a Hierarchical Patch Clustering Method for Image Patches
•  
   4.1 Approach
•  
   4.2 Methodology
•  
   4.2.1 Feature Extraction
•  
   4.2.2 Hierarchical Clustering
•  
  4.2.3 Modified G-means Algorithm
•  
   4.2.3.1 Gaussian Hypothesis Testing
•  
   4.2.3.2 Anderson-Darling Test
•  
   4.2.3.3 Feature Vector Projection
•  
  4.2.4 Comparative Similarity Measures
•  
   4.2.4.1 Fractional Distance Metric
•  
   4.2.4.2 Minkowski Distance Metric
•  
  4.2.5 Evaluation
•  
   4.2.5.1 Visual Evaluation
•  
   4.2.5.2 Internal Evaluation
•  
   4.2.5.3 External Evaluation
•  
   4.2.6 Comparative Clustering Methods
•  
  4.3 Results
•  
   4.3.1 Datasets
•  
   4.3.2 Experimental Settings
•  
   4.3.3 Parameter Settings
•  
   4.3.4 Visual Evaluation
•  
   4.3.5 Internal Evaluation
•  
   4.3.6 External Evaluation
•  
   4.3.6.1 Analysis of Absolute Homogeneity
•  
   4.3.6.2 Analysis of Relative Homogeneity
•  
   4.3.6.3 Analysis of Cluster Numbers
•  
  4.3.7 Comparative Experiments
•  
   4.3.7.1 Different Clustering Methods
•  
   4.3.7.2 Different Features
•  
   4.4 Conclusions
•  
  5 Semi-supervised Semantic Image Patch Annotation
•  
   5.1 Methodology
•  
   5.1.1 Creation of a Reference Dataset
•  
  5.1.2 Semi-supervised Learning
•  
   5.1.2.1 Cluster-then-Label
•  
   5.1.2.2 Supervised Learning Within Clusters
•  
   5.1.3 K-Medoids Algorithm Implementation
•  
  5.1.4 Evaluation
•  
   5.1.4.1 Quantitative Evaluation
•  
   5.1.4.2 Visual Evaluation
•  
  5.2 Results
•  
  5.2.1 Image Data Selection and Subsampling
•  
   5.2.1.1 Data Selection
•  
   5.2.1.2 Data Pre-Processing
•  
   5.2.2 Parameter Settings
•  
  5.2.3 Quantitative Evaluations
•  
   5.2.3.1 Internal Evaluation
•  
   5.2.3.2 External Evaluation
•  
   5.2.3.3 Additional Evaluations
•  
  5.2.4 Visual Evaluations
•  
   5.2.4.1 Tree Structure
•  
   5.2.4.2 Feature Space Visualization
•  
   5.2.4.3 Cluster Centroid Patches
•  
   5.2.4.4 Cluster Homogeneity
•  
  5.3 Conclusions
•  
   5.3.1 Clustering
•  
   5.3.2 Classifiers
•  
   5.3.3 Semi-supervised Learning and Manually Annotated Reference Data
•  
   5.3.4 Semi-Annotation/Labeling
•  
  6 Pixel-Level Bayesian Classification and Active Learning Based Object Extraction
•  
   6.1 Pixel-Level Bayesian Classification
•  
   6.1.1 Modeling of Speckle Statistics Feature
•  
   6.1.2 Image Intensity Modeling
•  
   6.1.3 Combined Intensity - Speckle Statistics Feature Model
•  
   6.1.4 Bayesian Classification
•  
  6.1.5 Experiments
•  
   6.1.5.1 Brief Dataset Description
•  
   6.1.5.2 Evaluation
•  
  6.2 Active Learning Based Object Extraction
•  
   6.2.1 Definition of Non-Locality
•  
  6.2.2 SVM-based Active Learning
•  
   6.2.2.1 Version Space
•  
   6.2.2.2 Sample Selection Strategies
•  
   6.2.2.3 Prototype Implementation
•  
   6.2.2.4 Evaluation and Discussion
•  
   6.3 Summary
•  
  7 Conclusions
•  
   7.1 Summary
•  
   7.2 Future Works
•  
  Appendix A
•  
   A.1 Summary of Machine Learning Algorithms
•  
  A.2 Support Vector Machines (SVMs)
•  
   A.2.1 Linearly Separable Binary Classifications
•  
   A.2.2 Non-Linearly Separable Classifications
•  
   A.2.3 Application
•  
  Appendix B
•  
   B.1 Extraction of Common Objects
•  
   B.2 Extraction of Specific Objects
•  
   References