Shape-adaptive neighborhood classification method for remote sensing image

Abstract

High spatial resolution remote sensing images are playing an increasing important role in various applications in the world. As the fundamental work, classification of remote sensing is significant in the applications. This paper proposed a new feature extraction approach based on the shape adaptive neighborhood (SAN) for the classification of high spatial resolution remote sensing images. The heterogeneity based on the color characteristics was employed to determine the SAN of each pixel. Then all the color features, texture features and shape features were extracted from each SAN, and were fused by the feature level data fusion methods to the final feature space of the RS image. Then the features were used to do the classification work. As the experiment results shown, the total precision of the classification was 0.9187, and the kappa coefficient was 0.7950. By analyzing different maximum size of the SAN and different threshold of the heterogeneity, the best maximum size of the SAN was 11*11 for the study area and the most suitable threshold was 0.5. © 2008 SPIE.