An Improved Man-Made Structure Detection Method for Multi-aspect Polarimetric SAR Data

Abstract

Multiaspect polarimetric synthetic aperture radar (SAR) captures the polarimetric properties of targets from various observational aspects. The comprehensive multiaspect scattering characteristics are valuable for man-made structure detection and classification. Typically, the anisotropic scattering of targets could be characterized by the differences in the statistical properties of polarimetric data across aspects. However, both the statistical similarities in man-made structures and variabilities in natural targets at different aspects can negatively impact the ability to distinguish between them. Consequently, relying solely on anisotropic analysis may not yield favorable man-made structure detection results. Since man-made structures usually include special shapes, such as dihedral angle, there are significant variations in scattering power across different aspects. Therefore, this article proposes an improved man-made structure detection method that integrates scattering power characteristics and anisotropic features. First, to highlight differences between aspects, this article introduces a similarity matrix to perform azimuth sequence filtering. Subsequently, anisotropic features are extracted through differences in statistical distribution, and scattering power characteristics at individual aspects, along with their variations, are extracted using the fuzzy C-means clustering combined with spatial neighborhood. Two different features are fused to distinguish man-made structures from natural targets. Finally, the most significant azimuth aspect is determined by comparing the scattering contributions of individual subapertures. Experimental verification with airborne circular polarimetric SAR data confirms that the multifeature fusion method, following azimuth sequence filtering, effectively improves the detection of man-made structures and their most anisotropic subapertures.