Unified mixing model for hyperspectral imagery

Abstract

Multiple mixing models for hyperspectral imagery have been developed over the years. The most common is the linear mixing model which states that each pixel's spectral signature is a linear combination of the unique materials (or endmembers) in the scene. For mixtures of particulates, an intimate mixture model exists that nonlinearly combine the spectra. For mixtures where multiple paths are involved such as in tree canopies, a bilinear mixture model exists. Previous research has combined the linear with the intimate mixing model or the linear with the bilinear mixing model, but not all three. This paper describes a unified mixture model (UMM) that combines all three models and adaptively identifies the correct mixture to calculate the corresponding amounts of each endmember. The algorithm is demonstrated on a well-known quantitative spectral data set and a hyperspectral image containing a complex, coastal scene.