Adaptive Polarimetric Orientation Angle Correction and Scattering Decomposition for Robust Man-Made Structure Extraction in Polarimetric Synthetic Aperture Radar

Yasumin Siriprathan*, Junichi Susaki, Yoshie Ishii

Abstract

Accurate extraction of man-made structures is essential for urban planning and disaster management, yet conventional Polarimetric Synthetic Aperture Radar (PolSAR) methods often suffer from orientation-induced scattering ambiguity and confusion between buildings and vegetation in complex urban environments. This study presents a physics-informed framework for detecting vertical structures using ALOS-2/PALSAR-2 data. The method integrates experimental backscattering knowledge with satellite-scale analysis through three refinements: (1) confidence-based Polarimetric Orientation Angle (POA) correction, (2) adaptive scattering decomposition to rebalance double-bounce and volume components, and (3) Region of Interest–based statistical refinement to improve class separability. Machine learning models trained on microwave backscattering measurements from concrete blocks were applied to satellite data using Random Forest classification. Validation across Tokyo, Bangkok, and Manila achieved over 94%, representing 15–20 percentage point improvements over baseline methods. The refined approach significantly reduces misclassification in mixed-scattering regions and demonstrates strong robustness and transferability across diverse urban environments.

Keywords: Man-made structure extraction, Polarimetric synthetic aperture radar, Scattering

Full Text: