Enhancing Man-Made Structure Extraction Using PolSAR Data

Man-made structure extraction is crucial for urban planning, environmental monitoring, and disaster management. While optical sensors are affected by weather and lighting conditions, synthetic aperture radar (SAR) provides consistent imaging capabilities. This study utilizes Polarimetric Synthetic Aperture Radar (PolSAR) data and advanced scattering decomposition to enhance classification accuracy. Microwave scattering data from concrete blocks at various angles were collected in an anechoic chamber to train machine learning models, which were then applied to the Advanced Land Observing Satellite-2/Phased Array type L-band Synthetic Aperture Radar-2 (ALOS-2/PALSAR-2) satellite imagery. To address misclassification between man-made structures and natural areas, we implemented a three-step refinement: (1) confidence-based Polarimetric Orientation Angle (POA) correction and (2) adaptive scattering decomposition to redistribute power between double-bounce (Pd) and volume (Pv) scattering, and (3) region of interest (ROI)-based statistical refinements further improved class separation. The method significantly reduced misclassification errors, demonstrating its effectiveness in extracting man-made structures.