The Halo Trust is tapping into Amazon Web Services’ (AWS’s) portfolio of artificial intelligence (AI) tools for a pilot project to test if deploying the technology will speed up the time it takes to clear landmines in war-torn Ukraine.
The Trust is the largest humanitarian landmine clearance organisation in the world, and has secured a $4m package of support from AWS to fund the pilot, which will see it use a mix of AI and machine learning tools to detect and identify war debris from drone images of Ukraine battlefields.
To date, the Halo Trust has flown 542 drones over minefields and battlefields in the Ukraine, generating 11 terabytes of data in the process, that all needs to be sifted through so the organisation knows where to prioritise the placement of its landmine-clearing resources.
“It’s a painstaking process and it costs money, and for us to do our job efficiently, we really have to be able to identify where the danger areas are … delineate the boundaries of them and map them out, before we engage in the expensive and time-consuming process of clearing,” Halo Trust’s head of program development Simon Conway told Computer Weekly. “And we want to be able to map those danger areas as quickly as we possibly can, so that we can say to people, ‘You can’t even go into this area or that area,’ and keep them safe.”
The project is currently in the “data discovery phase”, he confirmed, with the aim of entering “active implementation” in September.
As Conway acknowledged, training an AI to identify war debris from drone imagery will take time, as it requires thousands of image details to be manually identified and labelled, and the AI’s results will need to be vetted by the Trust’s trained personnel to improve the model’s accuracy.
“Being able to crunch through [the image] data as effectively as possible and have the means [to] use AI to identify if mines are exposed or there’s disturbed earth in certain patterns that suggests there might be mines in that location … will help us speed up our processes,” he added.
While the Trust currently has landmine clearing projects underway in places such as Mozambique, Sri Lanka and Zimbabwe, it opted to use Ukraine as the location for the AI pilot because of how easy it is to get drone footage to test the technology on.
In other parts of the world, where there are conflicts in areas where there is thick vegetation, for example, this kind of work would be harder to do using drones and satellite imagery, whereas the Ukrainian terrain is more accessible in that regard.
“There is a lot of [war debris] on the surface in Ukraine … we are seeing large lines of mines that are laying on the ground, and underground, too … so we will find more evidence and data points, and as that gets more sophisticated, we’ll be able to extend the technology globally,” said Conway.
Land survey data migration
The Halo Trust is also planning to migrate its land survey data to the AWS cloud in time, said Conway, but the company’s technologies are already hosting a number of the organisation’s other business-critical systems.
Speaking to Computer Weekly, Liam Maxwell, government transformation director at AWS, said there were a number of reasons why Amazon decided to embark on the collaboration with The Halo Trust.
“We’ve already done a lot of work in Ukraine to help, but there was a document which we saw from Halo quite early on, which said, ‘We want to rid the world of landmines,’ and that’s a big idea which really attracted us,” he said. “That’s an incredibly difficult problem to try and solve, it combines that idea of scale and speed … and those two components meant this felt like a really good match.”
“There was a gap between the ambitions to do this and the capability to do so, and do it at the scale, the speed and with the security that Halo Trust needed,” said Maxwell. “We felt very keenly this is something we could contribute to and help bring … normality back to areas that have been heavily disrupted.”
Once the pilot is complete, Conway said he would know it has been a success if by the end of it he is able to create a map of a confirmed hazardous area in the region that he knows the exact size of, and – in turn – can accurately calculate how much it would cost to clear.
“If I can do that off the back of a drone flight, and know enough about the terrain to figure out whether we’re going to [clear] it manually or mechanically, and the costs involved, that’s a good result,” he said.