Project Detail

Integrated Spectral Dimensionality Reduction

Investigate dimensionality reduction techniques in combination with clustering and classification.

Lead: Ye, Jieping
Collaborators: Razdan, Anshuman;Wonka, Peter
Sponsor: National Geospatial Intelligence Agency
Date: 08/05/2008  - 08/04/2010

Abstract

The aim of this project is to present a mathematical toolset to process multispectral or hyperspectral images. Multispectral (MS) and hyperspectral (HS) images are typically acquired by a sensing device that captures the earth's surface from the air. The overall objective of the research is to provide new tools that allow analysts to extract meaningful information from MS and HS images. The proposed methodology builds on existing concepts including dimensionality reduction, unsupervised learning (clustering), and supervised learning (classification). The main research question is the following: How can clustering, classification, and dimensionality reduction be combined to improve automatic and semi-automatic geo-spatial analysis? An excellent and balanced team has been put together that combines experience in theoretical and applied mathematics, as well as experience in geo-spatial image analysis to tackle this problem and work toward significant contributions in this area.