Explore anomaly detection techniques in hyperspectral imaging through this informative conference talk. Discover how hyperspectral imaging excels at detecting minor signals in heterogeneous mixtures, even when the signal of interest is small volumetrically but dominant in individual pixels. Learn about three methods for detecting minor target signals: generalized least squares (GLS) target detection, iterative GLS and extended least squares (ELS), and whitened principal components analysis (WPCA) targeted anomaly detection. Understand the mathematical similarities between these approaches and how they model clutter locally for flexible 'adaptive' models. Gain insights into targeted anomaly detection's expansion of the adaptive concept by characterizing target signals locally. Examine practical examples demonstrating the application of these techniques, including detection of contaminants in wheat gluten and hidden watermarks. Delve into topics such as PCA anomaly detection, T² weighting, iterative de-weighting, and the comparison between de-weighting and orthogonalization methods.
Anomaly Detection in Hyperspectral Imaging - Neal Gallagher
Chemometrics & Machine Learning in Copenhagen via YouTube
Overview
Syllabus
Intro
Why Hyperspectral Imaging? • Useful for detection of small signals in heterogenous samples where quantities of contaminants may be low on a volume basi but may dominate signal in a single pixel, and for
PCA Anomaly Detection
Anomaly Detection Summary
T² is a Weighting
Example of GLS Weighting
Detection Algorithms
GLS Target Detection Example Signal from the unadulterated wheat gluten is highly variable and
56 ppm Example
Iterative De-Weighting
De-Weight Target by Clutter
200 ppm Melamine in Wheat Gluter
GLS Target Detection Summary
De-Weight vs Orthogonalize
Targeted Anomaly Detection
Hidden Watermark
Section 9
Conclusions
Taught by
Chemometrics & Machine Learning in Copenhagen
