Urban Area SAR Image Man-Made Target Extraction Based on the Product Model and the Time–Frequency Analysis

Abstract

This paper proposed an innovative framework to almost automatically extract man-made target from a high-resolution (HR) polarimetric SAR (PolSAR) image of an urban area. The core part of this framework is a new PolSAR image feature extraction method, which is developed by combining the spherically invariant random vector (SIRV) product model with the time–frequency (TF) analysis technology. The SIRV product model can better characterize HR SAR images, and the TF analysis will assist the classification by taking advantages of the anisotropic property to avoid the confusion of natural and man-made targets. Therefore, using this kind of extracted features, man-made targets can be easily discriminated with a simple unsupervised K-means classifier. Experimental results demonstrate the effectiveness of the proposed framework, in which man-made targets are extracted with clear contours, and natural surfaces are very continuous and homogenous. In addition, plenty of interesting targets with special scattering performances are highlighted in several rare classes. Their features are worth studying. Above all, because of barely requiring prior knowledge, the framework should be promising in a wide spectrum of applications by providing the rapid man-made target information acquisition of urban areas.