Exploration
Mohammad Ebdali; Ardeshir Hezarkhani; Adel Shirazy; Amin Beiranvand Pour
Abstract
This research endeavor concentrates on minerals exploration within the context of a hydrothermal polymetallic vein deposit environment. Stream sediment sampling was executed to acquire geochemical signatures pertinent to mineralization zones. The mineralization nature is classified as epithermal, predominantly ...
Read More
This research endeavor concentrates on minerals exploration within the context of a hydrothermal polymetallic vein deposit environment. Stream sediment sampling was executed to acquire geochemical signatures pertinent to mineralization zones. The mineralization nature is classified as epithermal, predominantly involving polymetallic sulfides. The geochemical analyses yielded multi-element concentration maps, facilitating the identification of anomalies and the establishment of zoning. Although recent developments underscore the efficacy of machine learning, notably deep learning techniques, in the detection of geochemical anomalies, the majority of preceding studies were predicated on univariate statistical methodologies. To address this constraint, a multivariate approach was implemented, incorporating spatial characteristics such as shape, overlap, and zoning within anomalies and halos. Considering the limited availability of validated mineralized samples, unsupervised and semi-supervised methodologies—most notably Generative Adversarial Networks (GANs)—were employed. GANs were trained using multi-element geochemical maps, applying transfer learning to mitigate the challenges posed by restricted deposit data, thereby facilitating the delineation of prospective exploration zones. Quantitative analyses have indicated that the approach utilizing GANs attained an accuracy exceeding 92% alongside a minimal cross-entropy loss of approximately 0.07, thereby surpassing conventional methodologies in detecting of weak anomalies. The model effectively corroborated previously recognized anomalies while simultaneously detecting new prospective mineralization areas, thereby augmenting exploration opportunities. This investigation illustrates that GANs enable a more thorough utilization of geochemical datasets, integrating a wide range of variables and intricate spatial characteristics. Although GANs demonstrate superior proficiency in modeling weak anomalies, conventional techniques continue to be effective for more pronounced anomalies. The integration of both methodologies may enhance the efficiency of mineral exploration endeavors. In summary, the results emphasize the promise of GANs and sophisticated machine learning frameworks in enhancing anomaly detection and expanding mineral exploration within hydrothermal polymetallic systems.
