Interferometric phase restoration using biquaternion neural networks in PolInSAR

Singular points (SPs) in interferometric synthetic aperture radar (InSAR) data hinder the generation of accurate digital elevation models (DEMs). Since one of the main causes of SPs is polarization-and-phase distortion due to scattering, it is necessary to process polarimetric information and phase information consistently to reduce SPs. This paper proposes an interferometric phase restoration method using biquaternion neural networks (BQNNs). The dynamics of BQNNs enable processing polarimetric information and phase information as a single entity. Therefore, it can restore interferometric phase information by utilizing pixel-by-pixel polarimetric information effectively. Experimental results show that, in a certain area, a DEM generated from a BQNN-filtered interferogram is more accurate than that generated from an interferogram filtered by our previously proposed method.