A subpixel target detection algorithm for hyperspectral imagery

The goal of this research is to develop a new algorithm for the detection of subpixel scale target materials on the hyperspectral imagery. The signal decision theory is typically to decide the existence of a target signal embedded in the random noise. This implies that the detection problem can be mathematically formalized by signal decision theory based on the statistical hypothesis test. In particular, since any target signature provided by airborne/spaceborne sensors is embedded in a structured noise such as background or clutter signatures as well as broad band unstructured noise, the problem becomes more complicated, and particularly much more under the unknown noise structure. The approach is based on the statistical hypothesis method known as Generalized Likelihood Ratio Test (GLRT). The use of GLRT requires estimating the unknown parameters, and assumes the prior information of two subspaces describing target variation and background variation respectively. Therefore, this research consists of two parts, the implementation of GLRT and the characterization of two subspaces through new approaches. Results obtained from computer simulation, HYDICE image and AVI RIS image show that this approach is feasible