A Novel Building Extraction Method for Airborne <i>L</i> -Band CSAR Images Based on Fusion of Scattering Statistical Characteristics

Circular synthetic aperture radar (CSAR), as a novel imaging mode, enables multi-aspect imaging of the same scene, capturing scattering information from different azimuth angles, thereby significantly enhancing the performance of SAR image interpretation. In CSAR images, building targets exhibit strong scattering characteristics at certain angles due to their special shapes, reflecting the properties of double-bounce scattering. Additionally, buildings demonstrate anisotropic scattering properties, with their scattering amplitudes varying significantly across different angles compared to natural terrain. To address this, this paper proposes a novel building extraction method that comprehensively characterizes the scattering properties of CSAR images by fusing multiple statistical features. First, since natural targets also exhibit minor feature variations in azimuth, this paper introduces the differential mode variance (DMV), which highlights the anisotropy of buildings in sub-aperture sequences by incorporating contextual information from adjacent sub-apertures through differential processing. Subsequently, the dispersion of scattering across angles is calculated based on mode variance. Furthermore, as some natural targets also exhibit weak anisotropy in aspect, and man-made structures like buildings only show strong scattering at specific angles, this paper proposes the scattering energy aspect concentration (SEAC). This new feature dimension describes the distribution patterns of scattering power in sub-aperture sequences, effectively distinguishing buildings from natural targets with weak anisotropy. Finally, using real data from an L-band airborne CSAR system, the proposed method is compared with mainstream CSAR image man-made structure extraction methods in recent years. The results demonstrate that the proposed method exhibits superior extraction performance and better aligns with the requirements of practical applications. Additionally, the extraction results show a certain level of robustness to variations in the number of sub-aperture images.