TL;DR

Researchers from the Alan Turing Institute have developed an AI model that could transform maritime surveillance by running directly on satellites. The technology is 2,500 times more computationally efficient than current systems, enabling vessel detection in minutes rather than hours.

A Breakthrough in Efficiency

The new model, unveiled at the Symposium on Edge Computing in Washington D.C., demonstrates for the first time that high-accuracy Synthetic Aperture Radar (SAR) vessel detection can run autonomously on low-power satellite hardware.

Current satellite surveillance faces significant delays because images must be transmitted to ground stations for analysis—a process that can take hours. The Turing team adapted an object detection model to run on specialised chips using under 10 watts of power, meeting strict satellite constraints.

“Satellites using this technology could scan an area of sea twice the size of Wales in under a minute, using less power than a lightbulb,” explained Colin Laganier, Research Engineer at the Defence AI Research Centre.

Combating Illicit Activity

The technology is designed to detect “dark” ships—vessels that disable their Automatic Identification Systems to avoid tracking. This capability is critical for monitoring illegal fishing, smuggling, trafficking and piracy.

The model maintains detection accuracy within 2-3% of larger, power-hungry systems whilst enabling “tip-and-cue” operations, where one satellite can automatically trigger follow-up observations from another.

Looking Forward

This UK research positions Britain at the forefront of AI-enabled space capability. Dr Victoria Nockles, Head of the Defence AI Research Centre, noted that the technology brings satellites closer to working autonomously, capturing and integrating multiple data types for real-time insights on remote regions. The team now hopes to test the model with maritime agencies for real-world deployment.

Source: Alan Turing Institute