A fast hyperspectral hit-or-miss transform with integrated projection-based dimensionality reduction

The Hit-or-Miss transform (HMT) is a common tool in Mathematical Morphology (MM) used in template matching and object detection and subsequent classification applications. The HMT probes a query image with a pair of structuring elements (SEs) which are designed to detect specific objects of interest. The relative size of hyperspectral image data in particular provides a wealth of information on a scene however, it also makes object detection via a HMT a computationally expensive process. We aim to solve this problem through employing both spatial and spectral dimensionality reduction (DR) techniques to transform a hyperspectral image and its associated SEs designed for the HMT into a reduced space.