Forest classification with high resolution satellite images : research into the use of spatial information to improve the multi-spectral classification with SPOT images

Computerized classification of remote sensing imagery is an invaluable tool for management and inventory of land use. Many types of land use can be quickly and accurately classified with relatively simple and widely used programs, which analyse and classify each pixel based on its special reflection, or spectral signature. The spectral signature of a pixel is just one part of the information present in remote sensing imagery. Another part is the spatial distribution of pixels and more specifically the texture. The texture information incorporated in the computerized classification process was tested in Tunesia and in the Noordoostpolder (the Netherlands). Unfortunately no significant improvement of the classification results could be found in the forest classification test in Tunesia. The classification test of the Noordoostpolder showed that when texture information is included the classification results are substantially improved.