The use of proximal sensing hyperspectral images facilitates the compositional mapping of samples from ornamental rocks, quickly characterizing areas with mineralogical compositions that may reduce their quality. In this work, images obtained in the laboratory with a SPECIM proximal sensor in the SWIR wavelength interval were used for a classification with the Spectral Angle Mapping (SAM) algorithm based on the selection of endmembers through the use of colour compositions generated from band ratio images. The resulting mineralogical map revealed spatially coherent spectral domains associated with variations in calcite–dolomite composition, Mg-bearing phases, iron-related spectral behaviour, and water-bearing or clay mineral components that were not distinguishable macroscopically. These results confirm that proximal imaging spectroscopy offers a substantial potential for obtaining a mineral classification that helps to quickly and objectively characterize these raw materials, saving costs and facilitating a more sustainable exploitation.