Ergänzung und Fortführung eines digitalen Landschaftsmodelles mit multispektralen und hochauflösenden Fernerkundungsaufnahmen

Abstract

For a landuse-classification we combined high resolution multispectral data with high ground resolution imagery to get good ground resolution on the one hand, and to improve supervised multispectral classification with texture analyses of panchromatic data on the other hand. The program for multispectral classification takes into consideration that class- signatures are statistically very different, and probabilities vary for each pixel for every class. We calculate with the maximum likelihood algorithm the probability for every pixel for every class, to decide, whether a pixel is within a class with high probability, or if it is very uncertain for a pixel to belong to any class. Texture analyses in form of point and edge detection with the Foerstner Operator are used to extract information of the high resolution data. So homogeneous areas, edges and points are detected. This information can be analyzed and gives us an idea of highly textured areas, where villages and single buildings are. Forest areas can be detected by image-segmentation. A meanvalue and a standard-deviation is specified, within pixels in a focal window belonging to the forest in the panchromatic image. Furthermore, all extracted information is linked together. The high spatial resolution of the panchromatic analyses is linked with the multispectral information. Finally, our results are connected with the digital-landscape-model of the Federal Office of Metrology and Surveying and classes of certainty are generated. We get an actual landuse-layer with a ground resolution of about 15 x 15 meters.