Deep Learning Based Dinsar Data Process Chain for Deformation Retrieval

Land deformation poses significant risks to urban infrastructure, natural landscapes, and human safety. Differential Synthetic Aperture Radar Interferometry (DInSAR) has proven to be an effective tool for monitoring ground deformation at high spatial and temporal resolutions. However, traditional approaches often suffer from computational inefficiencies due to the complexity of data processing. This study explores the integration of deep learning (DL) techniques into the Permanent Scatterers and Distributed Scatterers (PSDS) processing chain to enhance efficiency and scalability. The methodology involves creating a robust dataset using PSDS protocols and incorporating a Tomographic SAR Neural Network (TSNN) for ground deformation speed estimation. Preliminary results from Sentinel-1 satellite data for the Campi Flegrei region demonstrate the feasibility of this approach, with the DL model achieving an accuracy of 70% in capturing deformation trends. The findings suggest that DL methods can complement and potentially replace traditional analytic workflows, enabling automated and efficient ground deformation monitoring, facilitating progress towards the optimization of DL models into the PSDS chain and assessing their robustness against diverse data and environmental conditions.