Fast Endmember Extraction Based on the Geometry of the Datacube

doi:10.1117/3.1000981.ch14

Book: Optical Satellite Signal Processing and Enhancement

Author(s): Shen-En Qian

Published: 2013

https://doi.org/10.1117/3.1000981.ch14

Author Affiliations +

Abstract

Hyperspectral sensors produce hundreds of narrow spectral bands for each pixel within the field of view (FOV) and provide the capability to identify Earth’s surface constituent. In hyperspectral remote sensing, the spatial resolution of the images is constrained by the sensors’ instantaneous field of view (IFOV), which is determined by the sensor’s optical system. Due to geographic features on the Earth’s surface, the scene of an acquired datacube is heterogeneous by nature. More than one land-cover type or feature may be included in an IFOV of the sensor. As a result, reflected radiation from a ground footprint as observed as a pixel in the hyperspectral imagery (often referred to as datacube) has rarely interacted with a volume composed of a single homogeneous material. These pixels usually consist of a combination of one or more surface materials (mixed pixels) because the radiance values detected by the detector pixels come from a heterogeneous ground surface. The larger the ground footprint is, the less likely the observed pixel consists of a pure material. Even when the IFOV of a sensor is filled by a single target material, mixing of radiance from adjacent pixels can occur because of atmospheric scattering. The signatures of the pure materials are usually referred to as “endmembers” (EMs) in hyperspectral analysis terminology.