Tutorial 1 2018
Linear and Nonlinear Hyperspectral Unmixing
Abstract: This course will discuss many algorithms for hyperspectral unmixing with in-‐depth discussions of several representative techniques. The first part of the course will cover linear unmixing. Geometric and Optimization based algorithms will be described first and will be followed by presentation of signature library based algorithms. Probabilistic approaches will be covered in some depth with a focus on sparsity and representations of natural materials by probability distributions including Beta, Gaussian, and Gaussian Mixture Models. A comparison of signature library and probabilistic methods will discussed in terms of representing natural variability. Piecewise linear unmixing will then be covered and stand as a bridge between linear and nonlinear approaches. Model-‐driven and data driven approaches to nonlinear unmixing will be discussed and compared. Matlab code will be used to present live demos of many concepts and code will be available for many of the examples on github.