A Hybrid Optimization Framework for Multi-Baseline InSAR Elevation Reconstruction Based on Sparse-Grid Quadrature Information Filtering

To mitigate the degradation of multi-baseline InSAR elevation reconstruction caused by interferometric phase noise, a hybrid optimization framework integrating fringe-aware feature extraction, statistical estimation, and sparse-grid quadrature information filtering, termed HOF-SGQIF, is proposed. In the proposed framework, the deep learning-based GAUNet detection network is first employed to extract fringe boundary information from the interferograms, which are then partitioned into fringe boundary and non-fringe boundary regions. For the fringe boundary regions, elevation information is estimated using the maximum likelihood estimation (MLE) method. For the non-fringe boundary regions, an SGQIF based procedure integrates a fast local phase-gradient estimator with a path-following strategy to recover elevation information. By combining fringe-edge information with information filtering, the proposed hybrid framework improves both the accuracy and efficiency of multi-baseline InSAR elevation reconstruction. Quantitative elevation reconstruction experiments show that the proposed method can effectively enhance reconstruction accuracy compared with conventional methods. The proposed method effectively reconstructs scene elevation information while maintaining strong noise robustness and high computational efficiency.