ESA-backed AI detects satellite anomalies with 96% accuracy
Award-winning AI pipeline spots subtle satellite glitches before they escalate
Deep Dive
A hierarchical ensemble pipeline by Lorenzo Riccardo Allegrini and Geremia Pompei detects anomalies in ESA satellite telemetry. It integrates shapelet-based and statistical feature extraction with cross-channel aggregation, using time-series cross-validation and masking to prevent data leakage. Tested on the ESA Anomaly Detection Benchmark, it earned 2nd place in the Spacecraft Anomaly Challenge.
Key Points
- Built by University of Perugia researchers, the pipeline combines shapelet-based and statistical feature extraction for anomaly detection
- Achieved 96% accuracy on ESA's Anomaly Detection Benchmark (ESA-ADB) using hierarchical modeling
- Ranked 2nd in the Spacecraft Anomaly Challenge, topping the public Kaggle leaderboard
Why It Matters
Enables proactive satellite maintenance by detecting anomalies 10x faster than traditional methods, reducing mission risks