Monday Webinar – Anomaly Detection in Hyperspectral Imaging (Neal Gallagher)

Hyperspectral imaging is uniquely suited to detection of minor signals in heterogeneous mixtures. The reason is that although signal of interest may be small on a volume basis, it may dominate signal in individual pixels. Three methods for detecting minor target signal of interest will be discussed. The first method employs generalized least squares (GLS) target detection; this is a weighted classical least squares model. The second method uses GLS and extended least squares (ELS) iteratively to improve sensitivity and selectivity of the detections. The third approach is also iterative and uses a whitened principal components analysis (WPCA) targeted anomaly detection. The mathematics of the three approaches are similar and demonstrate how clutter (interferences in the images) can be modeled locally resulting in a flexible ‘adaptive’ model. Targeted anomaly detection expands on the adaptive concept by allowing the target signal to also be characterized locally. Examples for the three approaches will be shown.