AMPHITRITE
Semi-Automated annotated HarMFul Algal Blooms Dataset and lIghtweighT CNN foR ubIquitous open waTer Harmful Algal Blooms dEtection
Transforming Harmful Algal Blooms Monitoring with Edge AI
Vision.
To enable real-time, scalable, and globally deployable detection of Harmful Algal Blooms (HABs) by leveraging edge AI processing on satellite platforms. AMFITRITE aims to transform environmental monitoring by creating a homogeneous annotated dataset and a lightweight, edge-optimized convolutional neural network (CNN) that runs onboard satellite hardware, thus reducing data transmission needs and enabling rapid environmental response.
The challenge.
Harmful Algal Blooms (HABs) are increasingly frequent and severe phenomena that threaten marine and freshwater ecosystems, compromise water quality, damage aquaculture and fisheries, and pose significant risks to public health and coastal economies. Existing detection methods are either slow, expensive, geographically limited, or lack the accuracy and scalability required for timely intervention. Statistical prediction models are often unreliable, in situ monitoring is costly and sparse, and satellite-based detection typically depends on ground processing, causing delays that undermine early-warning capabilities.
Rising HAB incidents
Costly and limited
water monitoring
Slow satellite-based detection systems
Approach & solutions.
To address this challenge, AMPHITRITE employs an innovative combination of satellite-based data processing and edge-optimized artificial intelligence. Its methodology enables continuous monitoring and early warning without the delays and limitations of traditional ground-dependent systems.
Key elements of this approach include :
- Semi-automated annotation pipeline that uses satellite imagery, in situ measurements, and machine-learning predictions to reduce manual effort and improve dataset accuracy.
- Progressive dataset creation, starting with inland waters and expanding to open-water environments to ensure wide applicability and robust model performance.
- Lightweight convolutional neural network (CNN) trained on the annotated datasets and optimized for efficient execution on the Intel Myriad X VPU.
- Deployment on the CogniSAT6 platform, enabling in-orbit, real-time HAB detection that eliminates the need for large-scale data transmission to Earth.
By integrating scalable edge AI with a streamlined data annotation process, AMPHITRITE lays the foundation for a new generation of rapid, accurate, and global HAB early-warning systems.
Environmental Monitoring
Ensures continuous observation of inland and coastal waters to detect changes that threaten ecosystem balance.
Aquaculture & Fisheries
Provides early alerts that help protect fish farms and marine food production from algae-related toxicity and stock loss.
Public Health & Water Safety
Supports authorities by identifying harmful blooms that can contaminate drinking water sources and recreational beaches.
Climate Change & Ocean Resilience
Offers data-driven insights into the increasing frequency and intensity of blooms linked to warming waters and nutrient run-off.
Satellite-Based Earth Observation
Demonstrates how space-based sensors coupled with AI replace costly local sampling and expand monitoring coverage globally.
Artificial Intelligence & Edge Computing
Deploys lightweight neural networks directly on orbiting devices to reduce processing delays and dependency on ground stations.
Our role.
INLECOM serves as the sole beneficiary and executing partner of AMPHITRITE, responsible for designing and implementing the project’s full technical pathway. This includes developing the semi-automated annotation methodology for HAB datasets, training and optimizing the lightweight CNN model, and ensuring its deployment and validation on the Myriad X VPU via the CogniSAT6 platform, using the dAIEDGE Virtual Lab for real-time edge execution.
Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or European Commission. Neither the European Union nor the granting authority can be held responsible for them.
