An Approach to Spatial Classification of Hyperspatial Imagery

Our research investigated approaches to classification of imagery having ground spatial resolutions from 0.5m to 1mm, testing the relative importance on classification accuracy of spatial resolution and, secondarily, of small variations in view direction and sun direction. The classifications involved both spectral and spatial information and depended upon computation of the various Haralick texture measures for a 10x11 mosaic of wetlands images collected from an altitude of 2m in the nadir view direction. Each image represented a ground area approximately 1m x 1m. Our results underscore the increasing importance of spatial, as compared to spectral information as the spatial resolution in the imagery changes from 0.5m to 1mm. Our results support the hypothesis that both sun and view direction should be considered in the classification process.