Semi-supervised Learning for Classification of Polarimetric Sar-data

In the last decades Synthetic Aperture Radar (SAR) technology gained more and more importance in remote sensing. Although often more difficult to interpret than optical data, SAR data has certain advantages, like independence from daylight or less influence of weather conditions. Furthermore the data contain information, which cannot be provided by other remote sensing technologies. Today, many modern SAR sensors like TerraSAR-X are available and provide a huge amount of data to the scientific and commercial communities. This data need to be analysed and interpreted. However, manual inspection is impracticable due to the detail, size and number of contemporary SAR images. That is why there is a great need of automatic algorithms, which are able to interpret SAR images with high accuracy. These methods should be robust regarding slight changes of acquisition circumstances, flexible and adaptable to problem specific tasks. Modern algorithms of machine learning have shown these properties in many areas and are therefore very promising.