Nonlinear Unmixing for Imaging Spectroscopy

Instructors:	Paul Gader, University of Florida, pgader@ufl.edu
	Rob Heylen, University of Antwerp, Belgium, rob.heylen@uantwerpen.be

Tutorial Description:

This tutorial will focus on physical models and computational algorithms for nonlinearly unmixing pixels acquired by imaging spectrometers. Examples will be given on Earth Science and Planetary data. Some examples will be demonstrated with MATLAB scripts.

The tutorial will cover the following models / algorithms:

Linear / Convex (Review)
Linear with non-linear perturbations
Multi-linear
Piecewise Convex
Microscopic using Hapke and relate approximations
Ray Tracing and Radiosity
Kernel Methods
Neural Networks
Manifold Learning

Background:

Nonlinear mixing occurs in imaging spectroscopy in many ways, most notably in microscopic or intimate mixing of small particles and multiple scattering in urban and forested environments. While linear mixing is mainly due to finite resolution of optical systems and the need to create relatively large pixels, non-linear mixing is based on the direct interaction of light with matter in the field of view. Therefore, parameters of approximate physical models of the interaction are often estimated as a part of the unmixing process. Using a linear mixing model when a nonlinear mixing model is more accurate can result in very large errors in estimating characteristics of sub-pixel materials. Therefore, better models and algorithms are needed.

Prerequisites:

The instructors have co-authored a review paper:

R. Heylen, M. Parente, P. Gader, "A Review of Nonlinear Hyperspectral Unmixing Methods," in IEEE J. Selected Topics Applied Earth Observations and Remote Sensing, v. 7, n. 6, pp. 1844-1868, June 2014. doi: 10.1109/JSTARS.2014.2320576

that contains material on many methods(excluding the most recent) that will be presented. A laptop computer with MATLAB would be useful.

Keywords: Hapke models, Canopy models, Mineral and Planetary spectroscopy, hyperspectral mixing and unmixing

Number of Participants: max. 20

Tutorial Fee: 100 Euro

Please subscribe for the tutorial during Abstract Submission / Registration via ConfTool

Back to tutorials overview