applicability of library-based SU techniques in situations where spectral libraries are not available or cannot be built. Such methods are reviewed in the " How to Construct Spectral Libraries " section. Moreover, library pruning techniques, which were originally devised to reduce the size of libraries and lessen the computational complexity of SU, have evolved to consider the quality of the unmixing results. Recent library pruning methods aim at removing, before unmixing, entire EM classes or individual spectral signatures that are unlikely to be present in an observed image. This reduces the ill-posedness of the SU problem and can improve abundance estimation. These techniques are discussed in the " Library Pruning Techniques " section. Table 2 summarizes the key ideas involved in library TABLE 1. CHARACTERISTICS OF EACH GROUP OF SU TECHNIQUES AND WHERE THEY ARE REVIEWED IN THE ARTICLE. Amount of user supervision Computational cost Requires spectral libraries? Estimates pixeldependent EMs? Section in the article MESMA AND VARIANTS * * * * ü ü " MultipleEndmember Spectral Mixture Analysis and Its Variants for Small Spectral Libraries " Illustration of the key ideas Figure 10 Sparse SU: SU with sparsity constraints; EM-model-free: SU without explicit EM models. FUZZY SU * * * * ü ü " MultipleEndmember Spectral Mixture Analysis and Its Variants for Small Spectral Libraries " SPARSE SU * * * ü ü " Sparse Unmixing " MACHINE LEARNING LOCAL SU * * * * * * * * ü û " Machine Learning Algorithms " * û ü " Local Unmixing Methods " PARAMETRIC EM MODELS EM-MODEL-FREE BAYESIAN * * * * * * * û ü " Parametric Models " * * û û " EndmemberModel-Free Methods " * * * * û û " Bayesian Methods " Figure 11 Figure 12 Figure 13 Figure 14 TABLE 2. CHARACTERISTICS OF SPECTRAL LIBRARY EXTRACTION AND PRUNING TECHNIQUES AND WHERE THEY ARE REVIEWED IN THE ARTICLE. LIBRARY EXTRACTION TECHNIQUES Key idea IMAGE-BASED LIBRARY EXTRACTION Extracts multiple EM signatures from the observed image and cluster them to construct a library Adapted to the HI? Amount of user supervision Depends on the existence of pure pixels? Section in the article LIBRARY PRUNING TECHNIQUES Key idea ü * * ü " Image-Based Library Construction " LIBRARY REDUCTION Removes redundant signatures from an existing library to reduce the computational complexity of SU Adapted to the HI? Amount of user supervision Improves the computational cost of SU? Improves SU quality? Section in the article HI: hyperspectral image. û * ü û " Library Reduction Techniques " LIBRARY GENERATION FROM PHYSICS MODELS Create synthetic EM signatures using physicochemical mathematical models describing EM variability û * * * û " Generating Spectral Libraries From Physics Models " EM SELECTION Removes entire EM classes (e.g., water and trees) not present in the observed image from the library ü * * ü ü " Endmember Selection Methods " SPATIAL INTERPOLATION OF EM SIGNATURES Estimate EM signatures for each pixel by interpolating pure pixels at known spatial locations ü * * ü " Spatial Interpolation of Endmember Signatures " SAME-CLASS EM PRUNING Selects the signatures from each EM class most closely related to the observed image before SU ü * * ü ü " Pruning Libraries Within the Same Class " DECEMBER 2021 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE 227