An Improved H/ $\alpha$ Unsupervised Classification Method for Circular PolSAR Images

In conventional polarimetric synthetic aperture radar, targets are usually assumed isotropic, and potential polarimetric variations in azimuth are ignored. As to polarimetric circular SAR (CSAR), the azimuthal aperture is much larger, and polarimetric variations in azimuth are no longer negligible. Moreover, whether a target has changed polarimetric properties in azimuth, i.e., anisotropic, is an important feature. H/α classification scheme is a famous and effective unsupervised classification method. However, when applying H/α classification scheme against polarimetric CSAR data, problems occur. First, during the formation of a large aperture, polarimetric properties from different angles of view are combined, which affects the estimation of H and α. Second, H/α method cannot distinguish anisotropic and isotropic targets. In this paper, a pixel-wise H/α calculation method and an unsupervised classification scheme against polarimetric CSAR images are proposed to solve the two problems. With the pixel-wise H/α calculation method, H and α are more accurately calculated. Meanwhile, anisotropic and isotropic targets which have same scattering mechanism can be distinguished by the proposed classification method. The effectiveness of the pixel-wise H/α calculation method is demonstrated by both the simulated and real data. The unsupervised classification method is demonstrated based on real data, acquired by airborne CSAR system at P-band, the Institute of Electronics, Chinese Academy of Sciences, Beijing, China.