Groß, M.M.GroßSeibert, F.F.Seibert2022-03-082022-03-081992https://publica.fraunhofer.de/handle/publica/320051In the following paper, new approaches are described for the classification and the cluster analysis of multispectral Landsat TM datas using neural networks. First, the fundamental aspects of the neural networks used for this purpose and their advantages for nongaussian distributed density functions in feature space are outlined. Furthermore, the explored network topologies and models are presented. For classification, back propagation netwoks under supervised training are used at the pixel and texture level. For cluster analysis, however, a generalized self-organizing Kohonen Map has been chosen. The resulting information can be visualized by directly displaying the neural activity mapped onto the RGB colour space. Due to the topological ordering, the similarity of pixel colours identifies similar prperties in the feature space.enclassification performancecluster analysisLandsat TMneural networkRGB-Farbwürfel006conference paper