Optimized PCA Based Feature Extraction from Multi-look/Multi-resolution TerraSAR-X Data

With the launch of the German TerraSAR-X system in June 2007, a new generation of high-resolution spaceborne Synthetic Aperture Radar (SAR) data is available; which facilitates a spatially and thematically detailed SAR scene analysis. In fact, the high resolution of TerraSAR-X enables scene on land cover, such as urban areas, deserts, forests and fields, to be accurately mapped. Among the several feature extraction tools available in the literature, we choose in this paper to use Principal Components Analysis (PCA). In fact, based on covariance analysis, the PCA represents the input data in a linear subspace with minimum information loss. Our first objective in this article is to provide an optimal processing for finer PCA based feature extraction from SAR images. Then, since the presence of speckle in SAR images changes the radiometric and textural aspects of the different structures that may exist in the scene, our second objective is to study the sensitivity of the PCA performance, with regards to the amount of speckle that might be included in the SAR image. To control more accurately this amount, a speckle reduction was carried out by means of multi-looking.