Nonlinear Signature-Matched Hyperspectral Change Detection

Hyperspectral change detection provides significant potential to detect targets in foliage and urban clutter with enhanced performance relative to existing technologies. Change detection algorithms must discard non-relevant change while extracting relevant change. To date, linear prediction methods have been applied; however, the space-invariance affine models employed do not accurately model background changes that are often nonlinear in nature.  SET Corporation, in conjunction the Johns Hopkins University Applied Physics Laboratory (APL), proposes a novel hyperspectral Image (HSI) change detection algorithm that employs a kernel-based support vector machine approach to accurately model the underlying data and detect relevant change.  The system builds on the success of our nonlinear kernel-based Support Vector Data Description (SVDD) anomaly detector that has demonstrated superior modeling capabilities for non-Gaussian HSI data.  The system is robust to spatial misregistration and includes a novel nonlinear radiance remapping approach enabling automated, robust nonlinear, signature-matched hyperspectral change detection. Feasibility of the proposed approach will be demonstrated in Phase I, with Phase II focusing on development and evaluation of an end-to-end system prototype. BENEFIT: This technology will offer a new capability to perform change detection that is robust to image registration errors, with high accuracy nonlinear modeling to reliably detect relevant change. The resulting technology will have direct application to surveillance and security systems for military, law enforcement and commercial sectors.  Environmental applications including agricultural remote sensing, land use analyses and climate change monitoring will all be enhanced by this technology.  )