The ever-present lattice defects form a complex three-dimensional pattern that contains information about the crystal and its origin. Our method, based on a combination of spectroscopy and deep machine learning, enables the extraction of this information and allows for the accurate, non-invasive identification of crystal samples.