University of AlbertaMultimedia Research Centre · Dept. of Computing Science
ROSSRemote Observation, Sensing & System
Research / R/01 Earth Observation / An AI denoiser with a per-pixel trust score
R/01 · Earth Observation

An AI denoiser with a per-pixel trust score

Clients don't just get a movement map: they get movement plus a 0–1 trust score for every spot. AI cleaning cut new-site turnaround from ~1 month to a single InSAR pair.

Earth ObservationInSARDeep LearningDeployment

InSAR scenes reach ≈600 million pixels (30k × 20k). The group’s AI denoising step turns raw interferograms into ground-movement signals with a per-pixel 0–1 trust score. That cleaning cut new-site turnaround from about a month to a single InSAR pair (two SAR acquisitions), and it is proven in the field on roadways, railways, dams and mining sites.

Wrapped and unwrapped interferogram pair with radian colorbars

Wrapped vs unwrapped interferogram pair.

Displacement-rate map of the Tulare Lake Basin with the California aqueduct contour

Displacement-rate map over a ~1,500 km² area south of Bakersfield, California — from the group’s InSAR work with 3vG.