Answers are grounded in this paper and cite their sources.
Complex topographic and climatic conditions in mountainous canyon regions often introduce severe decorrelation noise and atmospheric delay errors in interferometric synthetic aperture radar (InSAR) monitoring, thereby restricting the accuracy of landslide identification and deformation inversion. To address this challenge, we propose a deep learning based multisource noise suppression framework that integrates a U-Net architecture with residual connections to construct an end-to-end supervised network. The core of the technique is that training samples were generated from high-quality unwrapped phases and real noise images, enabling the model to automatically recognize and correct complex interference signals. A typical reservoir area in southwestern China was selected to build a training dataset containing multiple temporal baselines and diverse noise types. After training, the model was applied to real unwrapped InSAR phases. It reduced the phase standard deviation to 15% of the original value and lowered the phase-topography correlation to nearly zero. The method effectively suppresses noise induced by low coherence and atmospheric delays and maintains stable performance under complex noise conditions. The corrected phases greatly enhanced the clarity of landslide deformation boundaries and reduced background noise in small baseline subset-InSAR time-series analysis. As a result, the derived deformation estimates became smoother and more reliable, showing strong consistency with UAV LiDAR and field observations. The model also demonstrates strong generalization capability in untrained areas, indicating its potential for large-scale InSAR deformation monitoring in complex mountainous terrains. Overall, the proposed approach offers an efficient and scalable solution for enhancing InSAR deformation interpretation accuracy in mountainous canyon regions.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: