Mapping landslides with Deep Learning algorithms applied to EO data

7th GeoHazards Challenge (GH7)


Challenge Description

In all parts of the world landslides occur regularly, which can often change the appearance of landscapes. By mapping landslides, this phenomenon can be documented and valuable insights can be gained. Despite the fact that automation is being used in more and more areas, this area is still largely manual work, which is carried out by humans. Due to this labor-intensive activity and the associated costs, landslide mapping is only available for very few areas. This problem could be counteracted by using automated methods. Nevertheless, even today this is still a challenging activity. The availability of high-resolution EO data is growing exponentially and it is a desirable goal to exploit this data source. Thus, landslide inventory can be generated quickly. In this challenge we aim to use a modified U-Net model for semantic landslide segmentation on a regional scale. This approach has been successfully applied and validated by ETHZ. The promoter of this challenge is also the Swiss Federal Institute of Technology Zurich (ETHZ).



Image: Mapping landslides with Deep Learning strategies. Validation of the results in the test site Oregon, USA.

This site uses cookies and Google Analytics to enhance your experience. When you continue to browse, you are accepting You are using an outdated browser. Please, upgrade your browser to improve your experience.