Since the beginning of the war in Ukraine, large amounts of satellite and aerial imagery have been collected, offering unique opportunities to analyze the impact of the conflict on civilian infrastructure. This thesis focuses on developing a deep learning model to automatically detect and map destroyed buildings over time, helping to assess damage patterns and support humanitarian or reconstruction efforts.

Possible tasks include:

• Preprocessing multi-temporal satellite data. (Collection is already done)

• Designing and training deep learning models (e.g., CNNs, transformers) for building damage detection.

• Evaluating the model against reference datasets and assessing temporal changes.

• Visualizing and interpreting results to identify trends in infrastructure destruction.

Requirements for students:

• Background in machine learning or deep learning (preferably PyTorch).

• Experience with geospatial data or willingness to learn remote sensing concepts.

• Strong programming skills in Python.

• Motivation to work with real-world, large-scale datasets.
   

In case you are interested in working on this topic, please contact jan.pauls@uni-muenster.de