研究目的
To develop a novel multi-pass InSAR method for 3D reconstruction of complex mountainous areas using robust low rank tensor decomposition to achieve reliable elevation estimates with a limited number of SAR images.
研究成果
The proposed method robustly reconstructs elevation maps from complex mountainous areas using low rank tensor decomposition, outperforming state-of-the-art methods like SqueeSAR by a factor of two in accuracy with fewer SAR images. It is suitable for operational large-area processing due to parallelizability.
研究不足
The method may not correctly estimate elevations in areas where layover occurs, as indicated in the real data experiments. Computational costs for large tensors are high, but can be mitigated by parallel processing patch-wise.
1：Experimental Design and Method Selection:
The methodology involves exploiting the low rank prior knowledge in multi-pass InSAR stacks. It includes steps such as statistically homogeneous pixels selection based on Anderson-Darling test, phase history retrieval of Distributed Scatterers (DS) by the phase triangulation algorithm, applying robust low rank tensor decomposition to the filtered InSAR stack, and reconstructing elevations based on the outlier-free InSAR stack via periodogram.
2：Sample Selection and Data Sources:
Simulations used a multi-pass InSAR phase stack of 512x512 pixels by 7 interferograms with elevation patterns. Real data used one TanDEM-X InSAR stack with 7 images of a complex mountainous area.
3：List of Experimental Equipment and Materials:
SAR images from TanDEM-X, computational tools for tensor decomposition and optimization algorithms.
4：Experimental Procedures and Operational Workflow:
Simulate InSAR stacks with added sparse outliers, apply the proposed method and compare with PSI and SqueeSAR methods. For real data, process the TanDEM-X stack similarly.
5：Data Analysis Methods:
Use standard deviation and bias metrics for accuracy assessment, and visual comparison of elevation maps and profiles.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
