Linear and nonlinear unmixing in hyperspectral imaging
Résumé
This chapter introduced spectral unmixing as a powerful analysis tool able to reveal latent and unobserved spectral and spatial structures in hyperspectral images acquired through various modalities, from long-range remote sensors to microscopy imagers. By identifying the spectral signatures of the main components present in the imaged scene while quantifying their respective spatial distributions over the scene, SU provides a compact, comprehensive and meaningful (i.e., physically interpretable) description of the whole set of measurements. In an unsupervised scenario, i.e., when both the endmember spectra and abundance vectors are unknown, SU can be formulated as a blind source separation problem.