NFI, NCSA Win Second Place in AI Competition for Critical Mineral Assessment December 21, 2022 Honors and Awards Artificial IntelligenceCAIIData AnalyticsEarth and EnvironmentHPC OperationsInstitutional PartnershipsModeling and Simulation Share this page: Twitter Facebook LinkedIn Email By Andrew Helregel A team of researchers from the New Frontiers Initiative, the National Center for Supercomputing Applications and Center for Artificial Intelligence Innovation won second place in a prestigious artificial intelligence (AI) competition co-sponsored by the United States Geological Survey and the Defense Advanced Research Projects Agency, a research and development unit of the Department of Defense. The AI for Critical Mineral Assessment Competition, a collaboration between USGS, DARPA, The Mitre Corporation and NASA’s Jet Propulsion Laboratory, solicits innovative solutions for automatically georeferencing scanned or raster maps and extracting their features. NCSA was awarded second place in the Map Feature Extraction challenge of this year’s competition. “We were thrilled to learn that our approach earned a second-place finish and the highest score in the polygon feature,” said NCSA Research Scientist Shirui Luo. “Our creative, deep-learning and computer vision-based solution combined optical character recognition, adaptive histogram equalization, and a modified “query segmentation” U-Net to extract various polygon, point and line features. “The model performs astonishingly well on irregular and discontinuous geometries and sparse regions. NCSA is hoping that the progress made on this challenge does, in fact, increase USGS’ ability to complete critical mineral assessments. The findings in this study may provide useful guidance in designing other automated models to accelerate the critical mineral assessment process.” Examples of polygon and line features that can be extracted from the map are seen here. These features include subtle difficulties, such as disjoint polygon and line features, and basemap features that divide and cross over multiple polygons. The team, dubbed Illinois Central Modeling, included Luo, NCSA Assistant Data Engineer Albert Bode, Graduate Research Assistant Priyam Mazumdar, NFI Executive Director William Kramer, CAII Director Volodymyr Kindratenko and competition team leader and NCSA Data Engineer Aaron Saxton. “This was a truly exciting and unique problem. Because each map is different, there was no global harmonized label set. It didn’t fit the standard classification models that are so popular,” Saxton said of the challenge. “But it was clear that we needed to use the ability of modern machine learning and artificial intelligence to infer more abstractly. The body of research to solve this type of problem is pretty thin and the turnaround to find a solution was very fast. NFI was able to assemble top-rate students and staff, drawing from NCSA and across (the University of Illinois Urbana-Champaign) campus, to come together and rapidly act on a team solution. “This was a great opportunity for NFI, NCSA and CAII to make more connections between government entities and our academic communities and showcase our talents and expertise to those who may be unfamiliar with what we do.” Read more on this outstanding achievement by NFI, NCSA and CAII here.