Vector Segmentation Using Multiband Spatial Subpixel Analysis for Object Extraction

Many remote sensing applications require not only the land cover of extracted objects but also shape parameters of and structural relationships between areas of speciic landcover. Conventional sensors such as Landsat TM ooer suucient spectral resolution for land cover classiication. Their pixel size, however, is relatively high in relation to the observed objects, leading to a high percentage of mixed pixels. New sensors with adequate spatial resolution, on the other hand, lack the necessary spectral information. In this contribution, a segmentation method is introduced resulting in image objects with high spectral and accurate spatial information. It is a two-step process that applies multiband spatial subpixel analysis to alleviate the mixed pixel problem and to provide subpixel edge information in the rst step. In the second step, the subpixel information is incorporated into a vector segmentation method. The borders of the resultant objects are coded in vector format so that the information can be directly transferred to geographical information systems. werden kann.