Historical GeoAI: The Promise of Unlocking Analog Spatial Data

Abstract Much of the spatial data held within geospatial library collections exists only in analog form. So long as the processes of georeferencing and feature digitization remain labor-intensive and slow, this will remain the case. Recent years have seen substantial advances in AI-driven automated recognition of map text. This paper introduces a project designed to leverage these advances to speed the creation of digital orthomosaics using historical aerial photographs. The project utilizes the mapKurator system for text recognition as well as a Mask R-CNN model for further text detection. These tools serve as part of a workflow to create center points and attendant metadata for individual aerial photos, which are necessary for creating orthomosaics using ArcGIS Pro software. The results gesture toward a future where Historical GeoAI methods enable more efficient digitization of the analog spatial data contained within maps and aerial photographs.