Tensor Decomposition-Based Compression of Wide-Area Long Time Series InSAR Images and Application to Surface Deformation Monitoring

Objectives With the continuous accumulation of synthetic aperture radar (SAR) images, interferometric synthetic aperture radar (InSAR) technology is facing a computational bottleneck caused by the surge in the number of SAR images when processing wide area and long-term surface deformation monitoring tasks. Especially for the distributed scatterers InSAR (DS InSAR) method, the full combination strategy of interferometric pairs results in an extremely time-consuming solution process, limiting its widespread application in large-scale deformation monitoring. However, temporal InSAR data often contains a large amount of redundant information in the spatiotemporal dimension. Removing redundant information through image compression has become an effective solution.

Methods This paper proposes a tensor decomposition⁃based method for wide area long time series InSAR image compression. The proposed method utilizes the consistency of spatial pixel statistical properties to reconstruct the covariance matrix into a third-order tensor, and uses Tucker decomposition to achieve spatiotemporal data compression.

Results To verify its effectiveness, Sentinel-1A images from the main urban area of Kunming city are selected for the experiment. The results show that under the compression subspace settings of 2×3 and 2×5, the processing efficiency is improved by about 24 times and 40 times, respectively, and the deformation inversion accuracy still meets the monitoring requirement. When the subspace is expanded to 2×10 and 2×15, although some information is lost, the main deformation area can still be identified, and the efficiency is improved by about 80 times and 120 times , respectively.

Conclusions This paper provides a new approach to address the computational challenges in wide area long-term InSAR deformation monitoring, and has good engineering application prospects.