Regularization in Hyperspectral Unmixing

Abstract

Remote sensing involves acquiring data from the Earth’s surface without physical contact with the area being sensed. Classically, the satellite-based sensors capture the data in four to six different regions in the electromagnetic spectrum covering the visible, infrared, and thermal infrared wavelength bands and are known as multispectral (MS) sensors. Recently, hyperspectral imaging has emerged as a powerful, passive remote sensing technology. The hyperspectral imagers (HySI, also known as imaging spectrometers) acquire a set of coregistered images of a scene with a relatively large instantaneous field of view (IFOV) (about 4m × 4m to 20m × 20 m) and much finer spectral resolution (10 nm within more than 200 contiguous wavelength bands). This has enabled quantitative analysis of an area within an IFOV of the sensor. The unprecedented capability of the hyperspectral sensors enables the remote acquisition of images where each pixel is a vector with a high spectral resolution that enables better analysis of the contents in an area.