Data-Driven Approaches to Detection in Complex Spectral Imagery

• Activity Number: 146
• Type: Invited
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): David Messinger*+
• Companies: Rochester Institute of Technology
• Keywords: image analysis ; spectral imagery ; multispectral ; hyperspectral

Abstract

Quantitative information extraction from remotely sensed spectral imagery requires models for the "background" and "foreground" image signatures.  Given these models, one can then make decisions per-pixel as to the likelihood of the presence of a signature of interest.  Traditional processing schemes rely on statistical or linear subspace models and have been successful in several applications.  However, the new generation of sensors has a significant improvement in spatial and spectral coverage and resolution: the Worldview-2 sensor has spatial resolution of ~2m in 8 spectral bands in the visible - near infrared spectrum, while airborne hyperspectral imagery has similar spatial resolution with ~200 contiguous, narrow spectral channels in the visible - short wave infrared.  These sensors image the Earth's surface at ever-greater levels of detail making it simple to show that assumptions of normality, or that the data are well-defined by linear subspaces, are not well-met.  We will describe new data-driven algorithms for applications such as spectral clustering and detection.  Results will be shown for space-based multispectral imagery as well as airborne hyperspectral imagery.