Improving Land Use Classification in Rugged Terrain Using Radiometric Corrections and a Possibility Based Classification Approach

Abstract-- The integration of multi-source ancillary data is one of the most promising techniques for improved classification of remote sensing images. In this research a classification strategy based on possibility theory and fuzzy subsets was applied in order to combine spectral and ancillary information. The ancillary information was used to define expert rules on the geographic context of the different land use classes. The method was tested on a complex, rugged area in Central Switzerland. Special emphasis was laid on the pre-processing of the Landsat TM image. A new physical based radiometric correction was applied to the imagery in order to eliminate the impact of atmosphere and slope-aspect effects. Due to the integration of ancillary information by possibility theory a notable accuracy improvement was achieved in comparison to a Maximum-Likelihood classifier with nonparametric priors