USING MACHINE LEARNING ALGORITHMS FOR LAND COVER CLASSIFICATION OF IKONOS IMAGERY

The opportunity to use high spatial resolution satellite images allows remote sensing scientists to experiment several approaches in order to generate always best classification results and final thematic maps. At present, it is not yet defined, among classification methods and available satellite data, which are the best ones to create an enough accurate land cover map and, above all, no automatic processing method has been proved to be so reliable. From this point of view machine learning algorithms seems to be really interesting respect to traditional methods and remote sensing literature of the last 5 years refers high quality in the achieved results and advantages in terms of costs. Satellite platforms like IKONOS and Quickbird provide lots of information in spectral and spatial context, and machine learning algorithms have the capability to take advantage of both components. The method combines results coming from a standard supervised classification based on pixel spectral characteristics with a second process in which spatial information of each class is involved to refine result. In this study the first results obtained exploring Foveal Vision technique on IKONOS multispectral data are illustrated