Retrieving geological units with unsupervised clustering of gamma-ray spectrometry data

Abstract

Airborne gamma-ray spectrometry (AGRS) provides valuable insights for geological mapping, mineral exploration, geomorphological studies, and lithological differentiation. The interpretation of AGRS data is usually performed by skilled interpreters, requires visual acuity and is time-consuming. In this study, we used unsupervised clustering and object-based classification on the primary variables potassium (K), uranium (eU), and thorium (eTh) to automatically retrieve geological units in Mara Rosa Magmatic Arc – Goiás, Brazil, a well-studied area known by a gold district. We first cluster a color image composition (Red = K; Green = eTh; Blue = eU) using gaussian mixture models. We estimate the ideal number of classes to partition the image with the bayesian information criterion. Then, we performed a class separability analysis to identify and merge classes with low statistical separability. Finally, we used an object-based approach to reduce inconsistencies in the unsupervised classification map. To assess our approach, we compute the structural similarity index between the classification map and a reference geological map of the study area. The results show that our approach retrieved geologically meaningful regions in the AGRS data, reaching 0.94 of overall structural similarity with the geological map. Combining unsupervised clustering and AGRS data allows a better interpretation of the geological units into the same Complex or Groups.