| KEYWORD |
AI-based landslide monitoring from satellite and ground data
keywords WEAKLY ANNOTATED SEGMENTATION, INSAR, GNSS
Reference persons LIA MORRA, TATIANA TOMMASI
Research Groups DAUIN - GR-23 - VANDAL - Visual and Multimodal Applied Learning Lab
Thesis type RESEARCH / EXPERIMENTAL
Description The objective of this thesis is to investigate the integration of InSAR (Interferometric Synthetic Aperture Radar) and GNSS (Global Navigation Satellite System) data for landslide monitoring by exploiting deep learning models.
InSAR is a remote sensing modality for the analysis of radar images captured by satellites equipped with Synthetic Aperture Radar. The radar antenna continuously transmits microwaves toward the Earth’s surface and captures the waves reflected back to the antenna position. By measuring both the amplitude and phase of the backscattered electromagnetic waves it is possible to assess the magnitude and location of ground movements. On the other hand, GNSS receivers on the ground collect information from a constellation of satellites to determine location. By exploiting ad-hoc engineered stations, it is possible to monitor small ground displacements and deformations at selected points.
While InSAR provides information over a large area with spatial continuity but temporal discontinuity based on satellite overpasses, GNSS offers measurements with temporal continuity but only at limited spatial positions. The combination of these cues can be formalized as a weakly supervised semantic segmentation problem, where GNSS information is interpreted as sparse ground truth annotations for InSAR data.
Following an initial phase dedicated to literature review and data organization, the main focus of this thesis will be developing a deep learning model that given an InSAR map and a few annotated GNSS points, will produce as output complete map annotations.
Required skills The successful candidate should be strongly motivated and is expected to have good knowledge of Python programming and PyTorch deep learning framework.
Deadline 15/05/2024
PROPONI LA TUA CANDIDATURA