Self-Supervised Contrastive Learning for Satellite Cloud & Shadow Removal
Multi-modal model reconstructing cloud-obscured optical satellite pixels by fusing contemporaneous Synthetic Aperture Radar (SAR) signals.
Project Overview
Solves perpetual cloud cover issues in Earth observation. Fuses cloud-penetrating Sentinel-1 SAR microwave radar with clouded optical Sentinel-2 imagery using self-supervised contrastive representations, synthesizing seamless cloud-free terrain maps.
Solves perpetual cloud cover issues in Earth observation. Fuses cloud-penetrating Sentinel-1 SAR microwave radar with clouded optical Sentinel-2 imagery using self-supervised contrastive representations, synthesizing seamless cloud-free terrain maps.
Multi-modal model reconstructing cloud-obscured optical satellite pixels by fusing contemporaneous Synthetic Aperture Radar (SAR) signals.
Core Project Objectives
Capture continuous analog/digital sensor readings with robust noise filtering and hardware calibration.