Developing an Object - based Hyperspatial Image Classifi er with a Case Study Using WorldView - 2 Data

N o v e m b e r 2 0 1 3 1027 Abstract Recent advancements in remote sensing technology have provided a plethora of very high spatial resolution images. From pixel-based processing designed for low spatial resolution data, image processing has shifted towards object-based analysis in order to adapt to the hyperspatial nature of currently available remote sensing data. However, standard object-based classifi ers work with only object-level summary statistics of the refl ectance values and do not suffi ciently exploit within-object refl ectance pattern. In this research, a novel approach of utilizing the object-level distribution of refl ectance values is presented. A fuzzy Kolmogorov-Smirnov based classifi er is proposed to provide an object-to-object matching of the empirical distribution of the refl ectance values of each object and derive a fuzzy membership grade to each class. This objectbased classifi er is tested for urban objects recognition from WorldView-2 data. Results indicate at least 10 percent increase in overall classifi cation accuracy using the proposed classifi er in comparison to various popular objectand pixel-based classifi ers.