Spotlight TerraSAR-X Data Modeling using Spectral Space-Variant Measures, for scene Targets and Structure Indexing

Abstract:
The main objective of this paper is to define an analysis model for high resolution spotlight SAR imagery, which is able to integrate the radiometric, as well as geometric and texture properties of the SAR data, in order to facilitate large data-base queries by informational content indexing of the images. The model proposed in this paper uses the information contained in the spectra of the SAR signal. A scene is described by a number of parameters computed based on Short-Time Fourier Transform. Considering the properties of the spotlight imaging mode, a phase correction algorithm is applied to the images prior to feature extraction. The classification is done using a Bayesian Support Vector Machine classifier. The method allowed for the recognition of more than thirty targets and structures in the scenes. The method is extended on quadpol TerraSAR-X data, where a Clifford-Fourier Transform is employed. This gives a representation of the spectra in a multidimensional space.