The following are suggestions for possible thesis. Furthermore, we are open to suggestions and own ideas. In this case, it is best to contact Prof. Iwaszczuk and/or Mona Goebel. However, other topics can also be found on the German version of the website.
|Topic||Short description||Type of degree||Contact person|
|Influence of image reconstruction methods on semantic segmentation quality||In optical satellite remote sensing, clouds often interfere with image evaluation. Therefore, clouds and their shadows are detected in advance with specific methods. Image reconstruction methods can help to perform a complete semantic segmentation despite cloud coverage. In this work, the influence of clouds in the training data as well as the influence of different image reconstruction methods on the quality of the classification will be determined.||Master||Lina Budde|
|Detecting low forest vegetation based on point clouds||Within the framework of our project, an overview should first be developed of which Deep Learning methods exist or are suitable for extracting low forest vegetation (vegetation =< 2m). From this, a preferred method can be chosen and tested and analysed on a given dataset. You are also welcome to develop your own methods. This work can be adapted as a theoretically or praxis oriented work. Keywords: low forest vegetation, extraction, segmentation DeepForest||Bachelor / Master||Mona Goebel|
|Analysis of seasonal ground motions in southern Hesse||The Ground Motion Service Germany shows ground motions recorded by Persistent Scatterer Interferometry (PSI). This service shows that seasonal ground uplift and subsidence occurs in southern Hesse in the town of Crumstadt. PSI is only suitable for urban areas and not for rural areas. Thus, the information on ground motion refers only to the city and not to the surrounding area. This project will use at least one other method to fill this data gap, as well as investigate possible causes of seasonal movement (groundwater fluctuations, gas storage, etc.). (BBD)||Master||Katrin Krzepek|
|Classification of scanned building fragments for the reconstruction of historical buildings||The fragments of a Jewish Bima were captured by laser scanning. From these fragments the whole building was reconstructed manually. In this work, point clouds of the laser scanned fragments will be classified as building parts using neural networks (e.g. PointNet++). This serves as a first step to simplify the puzzling together of the fragments.||Master||Jakob Schmidt|
|Hyperspectral data for vegetation mapping||The German environmental satellite EnMAP (Environmental Mapping and Analysis Program) is the first hyperspectral satellite developed and built in Germany. With its spectrometers, it analyzes the solar radiation reflected from the Earth's surface from visible light to the short-wave infrared at a spectral resolution not previously available. From this, precise statements about the condition and changes of the earth's surface can be derived. This work will address the possibilities of hyperspectral remote sensing in relation to peatland ecosystems and the mapping of various plant species.||Bachelor||Katrin Krzepek|
|Investigating statistical methods for correlating in situ point measurements with remote sensing raster data||It is common practice to evaluate remote sensing data with “in situ” data measured on the ground. A well-known example is precipitation measurements from radar data that are matched with precipitation measurements from weather stations. However, since remote sensing data usually have pixel sizes ranging from a few meters to kilometers, the question here is how best to reconcile point measurements of in situ locations with raster remote sensing data, for example, when the point measurement is at the edge between two pixels.||Bachelor / Master||Katrin Krzepek|