Onboard Marine Anomaly Detection on sat-2: From Simulation-Based Development to In-Orbit Demonstration
This work tested an onboard Artificial Intelligence pipeline designed to detect anomalies in marine imagery directly on a satellite, aiming to improve data efficiency. The system was initially trained using simulated data, but real-world testing showed a performance mismatch. Retraining the model with better annotation strategies on actual Level-1 imagery significantly boosted detection accuracy, proving the feasibility of onboard AI for Earth Observation.