Seminar – Nonlinear Spectral Unmixing: An Overview with Applications. Paul Gader. Univ. of Florida
; "Gader Seminar") |
IUMA Series Seminar Viernes 21 de Septiembre de 2012, 11:45 horas Abstract. Hyperspectral images are characterized by high spectral resolution measurements of light at each pixel. These measurements are referred to as spectra and are functions of the light reflected or emitted from materials in the scene. The spectrum at each pixel may consist of anywhere from around fifty to hundreds of measurements made in consecutive spectral bands that range from one to tens of nanometers wide. They can cover the Ultra-Violet, Visible, as well as Near, Short-Wave, and Long Wave Infra-Red regions of the electro-magnetic spectrum. The spatial extent of individual pixels can range from centimeters, to meters, to tens of meters. |
|
Hyperspectral image processing differs from standard image processing in that spectral pixels can be used to identify individual materials in a scene. Methodologies that identify materials from their spectral response to light are part of the area of spectroscopy. Processing hyperspectral image data to identify materials in a scene is referred to as imaging spectroscopy. Although chemists achieve great precision in laboratory spectroscopy, imaging spectroscopy is more difficult. Indeed, the limits of applicability of imaging spectroscopy are not known at this time. The unique goals of imaging spectroscopy are determining what materials are present in a scene and detecting and classifying materials in a scene at the sub-pixel level. A consequence of spatially large pixels is that the observed spectrum at a pixel is often a mixture of the spectral response from multiple materials. Although linear mixture models have been thoroughly investigated, nonlinear models have not. This talk focuses on techniques for unmixing nonlinearly mixed hyperspectral data and gives applications to terrestrial remote sensing, explosives detection, and planetary science. Biographical Sketch. |