Spectral Analysis of Geodata for Disaster Management


In today's world, where more and more natural disasters are occurring, crisis management is a burning issue for the remote sensing community. Appropriate response times and high-quality data for crisis detection of such events can make the difference between a well-organized crisis response and a disaster.
Current practice includes heuristic methods of image acquisition by UAVs and mostly heuristic image analysis. These insights are used to tailor the response to areas most in need of assistance. Using a combination of proven GIS solutions, publicly available research satellite imagery, and emerging machine learning approaches to segmentation and object recognition, the quality and speed of crisis intelligence could be greatly improved.
The goal of this work is to assess the damage of the increasingly frequent floods in Germany using near-infrared imagery. For this purpose, a possible pipeline has to be investigated, evaluated and determined. The development of automated image processing and analysis algorithms for georeferenced satellite images is a crucial milestone in the aforementioned task.
These results, obtained in a classical way of image processing, will then be used to train a machine learning model for segmentation and object recognition, leading to the classification of impact severity in different regions.