Classification of Land Cover Types in TerraSAR-X Images Using Copula and Speckle Statistics

Land cover classification using high resolution Synthetic Aperture Radar (SAR) images requires well-designed features, as well as suitable classification models. This paper addresses a pixel-based land cover classification problem in high resolution SAR images using speckle statistic and image intensity as features, copula function for joint probability modeling, and Bayesian classifier for classification. The performance of these techniques were analyzed and tested on a three-class database collected from TerraSAR-X High Resolution Spotlight Mode (HS), Geocoded Ellipsoid Corrected (GEC) images, over different cities of North Rhine-Westphalia (NRW), Germany.