This project report offers and examines a novel anomaly detection algorithm for ground-to-ground, or air-to-ground, software applications demanding automatic target detection using hyperspectral (HS) data. Targets are artificial items in natural background clutter under unknown lighting and environmental situations. The application of statistical models here is solely for inspiration of specific formulas for determining anomaly output surfaces. Specifically, formulas from semiparametrics are employed to get novel forms for output surfaces, and alternate scoring algorithms are suggested to determine output surfaces which are similar to the ones from semiparametrics. Examination makes use of both simulated data and real HS data from a joint data collection effort involving the Army Research Laboratory and the Army Armament Research Development & Engineering Center. A data transformation technique is introduced to be used by the two-sample data structure univariate semiparametric and nonparametric scoring……
Contents
Chapter 1 Introduction.
1.1 Background
1.2 Application of Statistical Models
1.3 Statistical Models for Hyperspectral Data
1.4 Relevant Work
1.5 Overview of This Work
1.6 Significance of This Work
1.7 Organization of Dissertation
Chapter 2 HS Sensing, Data Characterization and Models
2.1 Background
2.2 Hyperspectral Sensors
2.3 Hyperspectral Sensing Model
2.4 Data Characterization
2.4.1 Event Probabilities Using Small Windows in X
2.4.2 Data Models for Small Windows
2.4.3 Data Models for Null and Alternative Hypotheses
2.4.4 Parameter Specifications
2.5 Summary
Chapter 3 HS Data Transformation
3.1 Introduction
3.2 A Data Transformation Method
3.3 I.I.D. Test Experiment Using Transformed Data
3.4 Summary and Conclusions
Chapter 4 Statistical Anomaly Detection
4.1. Introduction
4.2. Multivariate Techniques
4.2.1. Statistical Hypothesis Testing
4.2.1.1. Multivariate Normal Target Detection
4.2.1.2. Multivariate Normal Anomaly Detection
4.2.1.3. Reed-Xi (RX) Anomaly Detector
4.2.1.4. Kernel RX (KRX) Algorithm
4.2.2. Alternative Multivariate Anomaly Detection
4.2.2.1. Fisher’s Linear Discriminant (FLD)
4.2.2.2. Dominant Principal Component (DPC)
4.2.2.3. Eigen-Separation Transform (EST)
4.3. Univariate Techniques
4.3.1. Semiparametric (SemiP) Anomaly Detection
4.3.2. Alternative Univariate Methods for Anomaly Detection
4.3.2.1. Functional Approximation of SemiP (AsemiP)
4.3.2.2. Asymmetric Variance Test (AVT)
4.3.2.3. Analysis of Variance (ANOVA)
4.4. Summary
Chapter 5 Power Using Idealized Spectral Samples
5.1. Introduction
5.2. Heterogeneous Models to Study Detection Power
5.3. Simulation Plan
5.4. Summary of Results
5.4.1. Impact of Spectral Bias/Shape on Detection Performances
5.4.2. Impact of Spectral Mixtures on Detection Performances
5.5. Summary and Conclusions
Chapter 6 Power Using Idealized Top View Cubes
6.1. Introduction
6.2. Notations and Definitions
6.3. Simulation Plan and Construction of Cubes
6.3.1. Simulation Plan
6.3.2. Background Cube Construction
6.3.3. Background Target Cube Construction
6.4. Type I and Type II Error Estimations
6.4.1. Obtaining Cutoff Thresholds
6.4.2. Estimating Type I and Type II Errors
6.5. Summary of Results
6.6. Summary and Conclusions
Chapter 7 GV Anomaly Detection Using Real HS Data
7.1. Introduction
7.2. Description of the SOC-700 Hyperspectral Data
7.3. Autonomous Sampling of the Cluttered Environment
7.3.1. A Binomial Based Parallel Random Sampling Model
7.3.2. GV Anomaly Detection Using No Prior Information
7.3.3. Summary of Results
7.3.3.1. Initial Results Using No Prior Information
7.3.3.2. Adaptive Threshold Under Various Environment Conditions……
Source: University of Maryland