SURFCLASS

AI for satellite interferometry (InSAR) to simplify data exploitation and interpretation by adding a classification layer to large inSAR point cloud databases.

Over the years, InSAR has become a common approach to map and monitor ground displacement at different scales, from local to regional and national. At large scales, InSAR provides such a volume of data, which is dramatically demanding for final users to interpret. Additional layers that can fasten and support data interpretation are crucial to properly tackling users’ operations. AI can be employed to integrate InSAR data with other modalities to automatically predict new relations and extract ready-to-use information.

The project goal is to support the analysis of large databases of InSAR displacement measurements by identifying and classifying spatial patterns corresponding to driving phenomena (e.g. landslide, subsidence, local instabilities), using Machine Learning (ML) methodologies. SURFCLASS is built upon the results reached by MATTCH (Machine Learning Methods for SAR-derived Time Series Trend Change Detection), which has already confirmed the suitability of ML approaches to perform change trend detection in InSAR time series.  SURFCLASS addresses the design of a more powerful DL model, which can exploit diverse geographical layers (SAR, DEM, Land cover, Sentinel-2 images) and the spatiotemporal correlations among measurement points, searching for similarities and obtaining a “classification” of the points with respect to driving deformation phenomena.

TRE Altamira is a leading company providing InSAR services globally, with extensive experience in processing satellite radar (SAR) data. Polimi-DEIB contributes to the project with its significant expertise in Artificial Intelligence and Machine Learning methodologies.