Semantic Segmentation-based approach to the analysis of old maps

Archives around the world contain millions of digitised old maps from the late Middle Ages to the recent past.They are invaluable data sources but complicated to work with individually or systematically, e.g., single rasterised maps used in historical GIS or set of maps prepared at smaller scales and frequently over multiple generations (van Wissen et al, 2024).These maps are mostly only accessible as raster maps and require much manual work to prepare them to construct a georeferenced raster map or vectorised map suitable for spatial analysis.Both technical and conceptual issues and potentials bar more effective use of these data sources.This presentation describes an innovative project that creates an accessible interface with semantic segmentation computational capacities for the first time to analyse old maps, apply interpreted categories/classes to additional maps, work with the same regions on different map series, and produce vectorised data sets for presentation and further research.Recognising the centrality of researchers in interpretative work with digitised old maps (Uhl et al, 2019) and the confluences between technical and conceptual issues, our goal in the continuing project (Annanias, 2023) is to create a human-in-the-loop process for data acquisition, preparation and guided analysis that is more efficient using rasterised old map sources and that can works with user-defined categories for different kinds of map features.