A Study on Extraction of Man-Made Targets Using SVM Method from High Resolution PolInSAR Data

Abstract

This paper studys a man-made target extraction technique using SVM (Support Vector Machine) method from high resolution PolInSAR(polarimetric Interferometric synthetic aperture radar ) data. Fully polarimetric interferometric data in X band was acquired by a China-made airborne SAR system in January 2010, over the Linshui area in Hainan province. This study focuses on man-made targets extraction in Linshui area, where the ground survey allows the validation of the final extraction results. Different PolSAR (polarimetric SAR) and PolInSAR(polarimetric interferometric SAR) feature indicators, such as H/A/α,based on the Pauli formalism or an optimal set of coherences are subsequently analyzed and the representatives of features are selected. In order to consider effect of man-made targets extraction with different feature indicators, the SVM method is investigated. The results based on PolSAR, PolInSAR and combined PolSAR and PolInSAR features are respectively compared, the best result is obtained via combining PolSAR and PolInSAR features.