Nonlinear unmixing in hyperspectral remote sensing.

Spectral unmixing of hyperspectral imagery is a vibrant field of research, yielding numerous powerful applications in remote sensing of mineral formations, and agricultural and vegetation scenes. Unfortunately, most unmixing algorithms are based upon the naive linear mixing assumption, where it is easily observed in real-life data sets that spectral mixing interactions can display highly complex behavior. Further improvements in spectral unmixing algorithms and applications hence require insights into the sources of these non-linear spectral interactions, and techniques to cope with them. This project is focused on the analysis of the underlying mechanisms of non-linearity in spectral mixing interactions. By assessment and further advancement of existing and recently developed techniques for non-linear unmixing, new techniques will be explored, and innovative spectral unmixing methods that are capable of dealing with these subtle non-linear spectral interactions will be developed. The experience present in both research groups, the Visionlab (University of Antwerp) and M3BIORES (KULeuven), along with the individual data sets, code base, and used algorithms and techniques, will be leveraged in a cooperation and fusion of techniques. This will lead to new unmixing applications with improved accuracy, susceptible to the subtle spectral information currently ignored in applications based on the linear mixing assumption.

·