Invariant mixture recognition in hyperspectral images

We present an algorithm for identifying linear mixtures of a specified set of materials in 0.4-2.5 μm airborne imaging spectrometer data. The algorithm is invariant to the illumination and atmospheric conditions and the relative amounts of the specified materials within a pixel. Only the spectral reflectance functions for the specified materials are required by the algorithm. Invariance over illumination and atmosphere conditions is achieved by incorporating a physical model for scene variability in the constrained optimization formulation. The algorithm also computes estimates of the amounts of the specified materials in identified mixtures. We demonstrate the effectiveness of the algorithm using real and synthetic HYDICE imagery acquired over a range of conditions and altitudes