Earth observation systems play an important role in the geoscientific community, because they regularly provide remote sensing data with a spectrally, spatially and temporally high resolution. These characteristics enable various applications to accurately monitor the earth’s landcover and its changes. Beside the challenge to deal with large amounts of data and limited class label information, current and future challenges comprise the definition and the way of integration of prior knowledge and the automatic determination of sophisticated feature representations.
– Landcover classification methods
– (Deep) Representation/Feature learning
– Self-taught learning
– Sparse representation for big data
– Incremental/Sequential learning