Imagine a satellite that doesn't just take photos and send them back to Earth, but actually 'thinks' and analyzes the images while still in space. Instead of sending massive amounts of raw data that clog up the connection, it only sends the important bits, like where plastic pollution is in the ocean. It's like having a smart filter in the sky that cleans up the data before it ever reaches your computer.
Satellites capture more data than current ground infrastructures can process, creating a bottleneck. This leads to high latency and wasted bandwidth when sending raw data to Earth for analysis.
An AI middleware for SoC-FPGAs and a specific Deep Neural Network (DNN) for detecting marine plastic litter, optimized for low-power satellite hardware.
If you are an ocean cleanup organization dealing with the slow identification of plastic waste zones — this project developed a DNN for marine plastic litter detection that achieves an 80% F1 score. This allows for faster targeting of cleanup fleets by processing data directly on the satellite.
If you are a satellite manufacturer dealing with high power consumption and limited hardware space — this project developed AI middleware for SoC-FPGAs that enables 10-100M parameter models to run efficiently. This allows your hardware to handle complex AI without draining the battery.
If you are a port authority dealing with massive data latency from Earth Observation satellites — this project developed an edge-AI solution that reduces data payload and latency. This ensures you get critical environmental alerts in near real-time rather than waiting for ground processing.
Based on available project data, specific pricing or implementation costs are not provided.
Yes, the project develops AI middleware for SoC-FPGAs applicable to European AI accelerators, suggesting the underlying technology can be adapted for various Earth Observation cases.
Based on available project data, the IP and licensing terms are not specified in the provided summary.
The solution is designed for compatibility with on-board hardware, specifically targeting SoC-FPGAs like the NanoXplore NG-Ultra and the BALKAN-1 satellite.
The project period runs from 2023-12-01 to 2026-11-30, with evaluation planned on the BALKAN-1 satellite.
The consortium is lean and highly industry-focused, consisting of 4 partners across 4 countries (BG, CH, EL, FR). With a 50% industry ratio (2 SMEs and 2 other partners, including one university and one research center), the project is well-balanced between academic research and commercial application, led by an SME (AGENIUM SPACE).
Contact AGENIUM SPACE in France for partnership opportunities.
Contact us to find out how to integrate this edge-AI middleware into your satellite fleet.
