Abstract
In the landcover classification of the Naze area, Amamioshima, using Landsat-TM and Spot-XS images, forest, urban and coral areas are categorized using core clusters based on sampling data covered by a single image for the whole projected area. However, there have been some problems with the classification accuracy of farmland due to the complex topography of the area.
The purpose of this study was to examine whether national land numerical information is useful for improvement of classification accuracy. We made a core cluster based on a clustering algorithm as supervised data for each mesh sampling, since 1/2 regional mesh data have 8 land use planning categories, such as forest, farmland and urban areas, in each mesh. By using these core clusters for the whole area, landcover classification accuracy was improved for farmland. The effective use of known information, such as national land numerical information, for object area pixels makes it possible to improve landcover classification accuracy.
