The predictive power of mineral prospectivity analysis depends on high quality, spatially accurate, analysis-ready datasets. Of paramount importance are geologic maps and mineral site data, but the state of readiness for utilizing these datasets remains sub-optimal for advanced computational techniques. As the U.S. Geological Survey (USGS) fulfils its mission to map the distribution of critical mineral commodities, non-georeferenced maps held within historical collections represent rich sources of input data. Through a series of machine learning challenges organized by the Defense Advanced Research Projects Agency (DARPA) in collaboration with the USGS, significant progress has been made in accelerating data ingestion, processing, and preparation tasks that enable mineral prospectivity mapping and mineral resource assessment workflows. Specifically, two tasks that previously required time-intensive human effort, 1) georeferencing map images, and 2) legend-based feature extraction from map images, are discussed.
|Title||Automated georeferencing and feature extraction of geologic maps and mineral sites|
|Authors||Graham W. Lederer, Joshua Mark Rosera, Margaret A. Goldman, Garth E. Graham, Asitang Mishra, Amanda Towler, Brian Wilson, Dustin Graf, Michael Milano, Elizabeth Roberts, Gabrielle Hedrick, Carsten Oertel, Anastassios Dardas, Thomas McEntee|
|Publication Type||Conference Paper|
|Publication Subtype||Conference Paper|
|Record Source||USGS Publications Warehouse|
|USGS Organization||Geology, Energy & Minerals Science Center|