Automated Segmentation of Historical Cadastral Maps: A Machine Learning Approach to the Franziszeischer Kataster

The Franziszeischer Kataster, created between 1817 and 1861, represents one of Europe’s most eminent historical cadastralmapping projects, covering over 670,000 square kilometers across the Habsburg Monarchy. Despite its significant potential for historical research, ecology, geography, and spatial planning, the utilization of this valuable resource has been limited by accessibility challenges and the complex nature of the source material. This paper presents an automated approach to segment key structural elements from these historical cadastral maps using machine learning techniques. We developed a Unet-based model for identifying and extracting four primary features: roads, buildings, water bodies, and forests. For the evaluation, we selected the Styrian district of Murtal and achieved an overall Intersection over Union (IoU) score of 96.28% and maintained F1 scores consistently above 95% across all feature classes, despite significant class imbalance in the training data. The results demonstrate the potential for automating the analysis of historical cadastral maps, enabling their integration into modern Geographic Information Systems (GIS) and facilitating comparative studies of land use changes over the past two centuries.This approach represents a significant step forward in making historical spatial data more accessible and open for analysis for various research disciplines.