Principal Component Analysis-Based Dense Target Discrimination Method for Satellite Synthetic Aperture Radar Images

Abstract

When detecting targets of interest in satellite synthetic aperture radar (SAR) images, clustering is often required to construct the candidate pixels of an object. In this paper, we propose a new feature using principal component analysis to address the problem of dense targets that cause performance degradation when using the density-based spatial clustering of noise algorithm, which is suitable for the clustering process. Further, simulations are used to analyze the proposed features for different target and clutter clusters extracted from real TerraSAR-X (TSX) images, and the proposed dense target discrimination technique is applied to the TSX images. The simulation results reveal that the proposed feature is effective in distinguishing clusters composed of dense targets, and the target detection performance can be improved by re-separating dense target clusters from actual SAR images.