Speaker
Description
Quad-polarimetric (QP) SAR reconstruction from reduced-polarimetric observations can extend the usability of compact- and dual-polarimetric (CP/DP) data while preserving key polarimetric information for scattering analysis and downstream applications. This presentation evaluate QP reconstruction through AI in different aspects, including input modes, output parameterizations, and loss functions, rather than focusing on a specific network architecture. Reconstruction performance is assessed in terms of accuracy, positive definiteness, numerical stability, and the preservation of polarimetric characteristics, including entropy H, anisotropy A, and the alpha angle. The analysis is conducted on three datasets acquired by RADARSAT-2, and BIOMASS with L and P bands, respectively. In addition, a comprehensive evaluation framework is established to assess reconstruction performance from covariance-level, channel-wise, and polarimetric-feature perspectives.