Processing Hyperspectral Images

Abstract: Hyperspectral imaging is a technology that has been used in remote sensing to study planetary surfaces for over 20 years. Hyperspectral cameras (or imaging spectrometers) have a high spatial resolution (~ 1–30 m) coupled with regular sampling (every ~ 4–15 nm) of a broad spectral range, which can cover wavelengths ranging from ultraviolet (~ 0.35 μm) to thermal infrared (~ 12 μm). With this dense spectral sampling, imaging spectrometers provide images with an extra dimension to conventional imaging: the spectral dimension. Indeed, each point of a hyperspectral image is associated with a vector called a “spectrum”, which is formed by a few hundred spectrally adjacent values with a physical meaning (for example, radiance measured by the sensor or surface reflectance after an atmospheric correction step). In this chapter, we are primarily concerned with the processing of remote-sensing hyperspectral images to study surface materials, which is one of the most common applications. In this context, analysis of land scenes imaged by their associated spectra (not by their pixels) becomes interesting because radiance or reflectance spectra contain a wealth of information on the chemical and physical state of surface materials.