Accurate Monitoring of the Danube Delta Dynamics using Copernicus Data

摘要

In the following, we describe highly-automated image analysis approaches that help us classify satellite images, andallow us to monitor dynamical changes in image time series. We concentrated on flooding events within the DanubeDelta as seen by the European Sentinel-1 and Sentinel-2 satellites, and describe systematic processing approaches toextract pre-defined categories from the image data (being either Synthetic Aperture Radar or multispectral images). Onebasic tool to monitor dynamical changes is to analyze and compare the compressibility of image patches using theirNormalized Compression Distances. These distances can be converted into similarity matrices providing reliable maps ofsurface changes. The accuracy of these change maps was quantified for several typical test cases. In addition, weanalyzed the performance of an alternative active learning approach, where Gabor filters and Weber local descriptorswere used to extract features from image patches that were classified and semantically annotated. Then one can performdata analytics and generate maps based on the extracted semantic annotations; again, we used several representative testcases for benchmarking.