AI for marine anomaly detection
The marine anomaly detection app identifies threats to the marine ecosystem, such as oil spills, harmful algae blooms, and heavy sediment discharges.
This application assigns an anomaly score to each maritime image captured by Φsat-2. The score increases as the content of the image deviates from a normal water state. If this score exceeds a certain threshold, an alert is triggered, and a specialised model immediately attempts to determine the nature and severity of the event.
This system enables multiple operational modes, including:
- Image prioritisation: images with the highest anomaly scores are downloaded first by the satellite. This allows for monitoring vast maritime areas while retaining only the most important information, thereby reducing the volume of data downlinked and optimising ground operators' time.
- Alerting: in the event of onboard detection of a major incident such as an oil spill, an alert is immediately sent to enable a faster response from the relevant authorities.
The innovation of this approach lies in its frugal use of AI, requiring minimal annotated data for training. This allows the application to be deployed on satellites with low computational power like Φsat-2, and to be quickly retrained to adapt to new situations observed from orbit.
Furthermore, the frugality of this solution makes it a versatile application that can be adapted to anomaly detection on other satellites or for other environments such as land, forests, etc., establishing it as a valuable tool for monitoring and protecting various ecosystems.
The marine anomaly detection application is developed by IRT Saint Exupéry Technical Research Institute and its partners in the Image processing for a Responsive Mission with AI (IRMA) project.
The app was a winner in ESA’s OrbitalAI Challenge.
