Methods of Hyperspectral Detection Based on a Single Signature Sample

Accurate radiance spectra of manmade materials are rarely available to remote sensing detection algorithms. Even when known instances of a material spectrum are available from a prior image, some form of signature adjustment is usually required by altered environmental conditions, if a useful signature is to be created. The translation of a reflectivity spectrum into the radiance space in which a remote sensing system operates presents an even more difficult problem. Here, two methods are described for exploiting prior signatures. First, a physics-informed statistical method for evolving in-scene signatures over time is derived. Second, a detection method is developed by growing a reflectance signature into an affine radiance subspace that is meant to capture prediction uncertainty.