ISSLIDE: A new InSAR dataset for Slow SLIding area DEtection with machine learning

Remote sensing images have been leveraged in multiple fields of applications: land-cover mapping, boat detections, flood and landslides delineations or post-disaster building damage assessment. Nevertheless, their wide spatial coverage has a cost in terms of resolution so that slow ground displacements and deformations are not visible. This issue is tackled by using the phase information of Synthetic Aperture Radar (SAR) images through Interferometric SAR (InSAR) approaches. Therefore, studying ground displacements requires both geomorphological and SAR interferometry knowledge. This dual-expertise requirement may explain the lack of InSAR datasets in the literature. The few available datasets based on InSAR are targeting volcanoes unresting period detection. But to the best of our knowledge, there is no dataset adapted to machine learning approaches and targeting slow gravitary movement detections. With this dataset, we propose a new InSAR dataset  for Slow SLIding areas DEtections (ISSLIDE) with machine learning. The dataset is composed of standardly processed interferograms and manual annotations created following geomorphologist strategies. A total of 200 independent moves are delineated on the each of the 54 interferograms. Extracting patches surrounding these moves ends up with 13,230 samples for training purpose. The dataset is provided for future user to develop new machine and deep learning methodologies in detecting and localizing slow moving areas with standard, multi-temporal or multi-interferogram studies.