Diffusion Geometry Based Nonlinear Methods for Hyperspectral Change Detection

We propose a suite of nonlinear image processing algorithms derived from the analysis of the underlying diffusion geometry of a collection of hyperspectral images of interest. These tools enable the comparison of spatio-spectral features of hyperspectral images acquired under different conditions, for the purposes of target detection, change detection and anomaly assessment. This methodology also automatically extracts independent components of the spectrum and builds an empirical model of the constituents of the scene. It is precisely through this model that the most efficient target search and change detection can be performed. We will integrate these tools into an existing hyperspectral image toolbox, and validate the methods on Air Force data as well as that from our proprietary hyperspectral acquisition hardware. BENEFIT: The eventual development of a commercial suite of efficient hyperspectral image processing algorithms deployable in on- and off-line applications including image acquisition systems