Detection of underwater objects in hyperspectral imagery

One of the biggest challenges in detecting underwater objects in hyperspectral imagery is that, unlike the land-based case, the observed spectrum of an underwater target is highly dependent on the properties of the surrounding water, as well as the depth of the target. In this paper we present a very general framework for underwater detection. The framework uses physics-based models to create a target space — the set of all observed spectra that a given target could generate for a given image. We then exploit the geometrical structure that is present in the target space to perform a nonlinear dimensionality reduction that greatly simplifies the detection problem. We also illustrate the framework with examples that use simulated targets at various depths.