I. Introduction
Hyperspectral imaging (also called imaging spectroscopy) is the process of dividing the electromagnetic spectrum into several narrow and contiguous wavelengths, and simultaneously acquiring an image for each wavelength. All those single-band images are then stacked in a 3-D data cube to produce the resulting hyperspectral image (HSI), where and correspond to the number of rows and columns of the single-band images, respectively, and is the number of wavelengths (also called spectral bands). To each pixel of the image is, therefore, associated an dimensional vector (or spectrum), which depicts the way the pixel site has interacted with the incident light and can be viewed as a function of the spectral wavelength . This spectrum depends on the materials composing the pixels, since each physical material can be uniquely defined by its spectral signature [1]. When analyzing a scene in the visible and near-infrared domain, this signature is called reflectance and corresponds to the light that was reflected by the scene. When working in the longwave infrared (LWIR) domain, the signature is expressed in terms of emissivity, being the ratio of the energy emitted by the scene with respect to the incident energy. Hyperspectral imagery, by acquiring detailed spectral properties of the scene, finds an always-increasing number of real-life applications in various remote sensing fields, such as vegetation mapping [2], geological [3], and hydrological sciences [4], as well as food quality inspection [5], [6] and medical imagery [7], [8], among others. However, this wealth of spectral information comes with several drawbacks, such as the high-dimensional nature of the data to be handled or the computational burden due to the large amount of data to process, making hyperspectral imagery a very dynamic and quickly evolving field of research [9].