Ausgewaehlte Algorithmen der Segmentierung und Klassifikation zur thematischen Bildverarbeitung in der Fernerkundung

The very fast evolution in remote sensing systems with increasing numbers of spectral bands and higher geometric resolution requires the development of new algorithms for image preprocessing and image recognition. These algorithms should be efficient for the processing of high data rates and large data volumes. A new approach to generalised framework for different steps of image segmentation will be used and developed further. The principal advantage of this method, called graph network, is the noise reduction and the parallel structure of the algorithms. For special tasks in remote sensing it is more and more necessary to produce high-level data products (e.g. for fire recognition) in 'real time' directly on board of the remote sensing platform. As an example a thematic data on-board processing system in form of a multi-spectral classification was developed for the small satellite BIRD (Bispectral InfraRed Detection). For this classification a special hardware based on the neural network processor NI1000 was designed. (orig.)Available from TIB Hannover: RA 437(1999-36) / FIZ - Fachinformationszzentrum Karlsruhe / TIB - Technische InformationsbibliothekSIGLEDEGerman