An Orbital Error Correction Model Based on Triplet Network and Shrunken Estimation

Orbital error, one of the major error sources of InSAR observations, is characterized by long-wavelength artifacts which can downgrade the monitoring accuracy, especially for wide-swath SAR missions. In this letter, we present a novel approach for time-series orbital error correction, with an emphasis on the computational and estimation efficiency of orbital error parameters over wide-area scale scenes. The proposed method combines the temporal triplet network and shrunken estimator, which integrates
$L^{2}$
- with
$L^{1}$
-Norm regularization, also known as Lasso regularization. The rationale behind it is to first determine the initial orbital parameters by utilizing the traditional polynomial-based method in the spatial domain. Next, in order to weaken the interference of phase unwrapping errors and other undesired phase contributions, an additional correction procedure is implemented by building up a redundant triplet network in the time domain and further a shrunken estimation method. Experiments on synthetic data and real Sentinel-1 datasets covering Eastern California confirm that the presented method can better balance the accuracy and computational efficiency. The proposed approach may, therefore, be useful for the processing of emerging big data InSAR.