AI can predict the location of landmines with up to 92 per cent accuracy, making removal of the deadly devices faster and more efficient, say researchers.
Some around the world are contaminated by landmines and they are being buried in current war zones like Ukraine faster than they are disarmed elsewhere. Landmines are cheap to make, extremely long-lasting and also very time-consuming to find and make safe, which means they are a uniquely dangerous legacy of conflict.
At least 4710 people were injured or killed by landmines and other abandoned explosives in 2022, with civilians making up 85 per cent of victims 鈥� half of which were children.
Advertisement
Martin Jebens at the International Committee of the Red Cross and his colleagues have created an AI tool they call DeskAId that can use satellite images to pinpoint the likely locations of landmines based on knowledge of where they have been found in previous conflicts, as well as the locations of roads, buildings and medical facilities. The AI learns patterns in the placement of mines in relation to these sites which it uses to predict where they may have been laid when shown maps of new areas.
In a paper, the researchers say the system is already being tested in Cambodia and that they are in talks to begin trialling it in other regions.
Those behind the project didn鈥檛 respond to a request for comment, but at The HALO Trust, an NGO that works on landmine clearance, says AI has the potential to bolster removal efforts 鈥� although the final step will always involve workers on the ground.
鈥淭here is massive, massive potential for efficiencies: time-saving, money-saving and, ultimately, hopefully, life-saving,鈥� he says. 鈥淸Landmines are] the hidden killer. You don鈥檛 know where they鈥檝e been placed. Some accidents can be as simple as a family in their car driving over an unpaved road, and that road was mined a few years ago. It鈥檚 a landmine that鈥檚 invented to take out a 20-tonne tank 鈥� you can imagine what it does to a personal vehicle.鈥�
Mathewson says clearing landmines starts with accumulating data, which could be accurate military information on mine locations, roughly sketched maps made by people who were present when they were laid or verbal interviews with locals, as well as data on where and when accidents have occurred. Sometimes this information can be decades old, as in The HALO Trust鈥檚 current work in Cambodia, which is reliant on declassified spy satellite imagery taken by the US in the 1980s.
arXiv
