Navigation auf


Department of Geography Geographic Information Systems

Computational Cartography

  • How can deep learning models learn to simplify buildings in map generalization?

    How can deep learning models learn to simplify buildings in map generalization?

  • Computational Cartography

    Recognition of group patterns in geological maps by building similarity networks.

  • Computational Cartography

    Energy Minimizing Methods for Feature Displacement in Map Generalization

One of our traditional research foci, with four decades of history, has been on methods for computational cartography, and more specifically automated map generalization. Map generalization is a cartographic process in which a target map is derived from an original map at a reduced scale, reducing the content and complexity of the map while taking care to preserve the structural features of the original map as much as possible. This process is of fundamental importance for digital cartography and the spatial sciences, as maps and data frequently need to be derived at smaller scales. Over the past decades, we have contributed to knowledge acquisition and representation methodological for map generalization, as well as constraint-based algorithms and agent-based computational models to carry out generalization tasks. However, in cartographic practice such approaches still require implicit expert knowledge from well-trained cartographers and some manual intervention.

Recently, deep learning has shown success in image processing and computer vision. Many new deep learning architectures are continuously being developed to integrate different data sources and solve problems in related domains. Pioneering work has also shown that it is possible to adapt deep learning models to the task of building an automated pipeline for map generalization, although many challenges remain.

Hence, in our SNSF project DeepGeneralization we seek to combine the traditional knowledge of map generalization in cartography with the new practices of deep learning models. Expert knowledge in map generalization, contributed by project partners swisstopo and IGN France, will help to build balanced and robust training datasets covering different scenarios of map generalization tasks. Deep learning architectures will be adapted to the specific needs of map generalization to integrate different map situations and eventually realize an automated end-to-end pipeline of the whole process.

Sample Publications

SwisstopoEDU Recognition Award 2023

For their MSc theses, which they carried out within the “DeepGeneralization” project, our former MSc students Nicolas Beglinger and Jan Winkler have been presented the swisstopoEDU Recognition Award 2023:

MSc thesis Nicolas Beglinger
MSc thesis Jan Winkler

Group members

Prof. Dr. Robert Weibel (Co-PI)
Dr. Cheng Fu (Co-PI)
Dr. Zhiyong Zhou (postdoc)
Joris Senn (undergrad research assistant)
Songlin Wang (undergrad research assistant) 

Belongs to the organizational unit
Geographic Information Systems