The HALO Trust is working with AWS to clear mines faster and save lives in the world’s conflict zones

Amazon Web Services (AWS) is investing $4 million to support the work of the HALO Trust and trial the use of artificial intelligence (AI) with drone imagery to locate minefields and other explosive remnants of war in Ukraine. Innovating with AWS will enable HALO to make wider use of the high-resolution drone footage it collects, including testing machine learning (ML) models for identifying mines. AWS and HALO will explore new ways to map minefields and prioritize mine clearance—while saving lives and restoring communities.

If all conflicts were to cease tomorrow, the death and destruction they cause would still go on for generations. Landmines and unexploded ordnance (UXO) maim and kill indiscriminately and continue to do so for decades or more after conflicts end. In the process, these explosives cause wider psychological and economic damage, preventing the safe use of vast swathes of land, keeping people from their homes, and forcing them to live in fear while going about their lives. Not knowing where mines are, or which areas are safe, can constrain communities and prevent recovery from conflict. Clearing mines is a painstaking, dangerous, manual task that relies on the skill and bravery of individuals carefully feeling their way through fields to find pressure plates and tripwires. It can easily feel endless.

The task is all the more terrifying because, in reality, conflict is far from ceasing. HALO is the world’s largest humanitarian demining organization, and more than 90 percent of its 11,000-plus staff are local to countries affected by conflict. In Ukraine, one of the most mined countries on earth, it must deal with shifting frontlines covering Europe’s second largest country, occupying forces constantly laying new mines, and pressure to open up liberated land for farming to maintain food supplies. It’s impossible for people clearing mines on their hands and knees to keep up with the challenge without help. That’s why HALO’s partnership with AWS is so crucial.

“AWS support will be transformative,” said HALO Trust CEO James Cowan. “It will fundamentally change our ability to store, collaborate, process, analyze, and disseminate the vast quantity of real-time data we collect through our global operations in the cloud—including in active warzones like Ukraine.”

Providing mine-clearing efforts with sharper eyes in the sky

Some of the most valuable data available to HALO comes in the form of aerial images, which can guide search efforts and help to identify likely minefields and other areas contaminated by explosives. The sharper the imagery that’s available, the more valuable it can be, but the greater the computing capacity required to store and analyze it. The $4 million that AWS is investing in the partnership will provide HALO with the hosting and computing capacity it needs to store and process high-resolution aerial images at scale, and to do so securely.

“HALO has collected a vast quantity of real-time data on war zones, and it’s vital to prevent that data from being accessed by anyone outside our organization,” said Jennifer Hyman, the HALO Trust’s US head of communications. “Our impartiality is a critical part of our license to operate and helps to keep our people safe. The AWS Cloud provides the guarantees of security that we need.”

A HALO Trust employee operating a drone during a field survey.

The scalable computing capacity of AWS is enabling HALO to combine high-resolution satellite imagery with drone imagery, which it is using for the first time at scale in Ukraine. Images from drones deliver resolutions of 1 centimeter (cm) per pixel, which is sharp enough for imagery analysts to spot recently laid mines sitting near the surface and other debris of war, like craters, trenches and destroyed tanks. Already, HALO has uploaded 11 terabytes (TB) of drone footage from Ukraine to AWS, including visual representations of 288 minefields.

Training ML models to spot mines at scale

The most ambitious element of the partnership will involve using Amazon SageMaker to train ML models to spot mines and debris of war in drone imagery. It’s a strategy that AWS and HALO have previously used in a pilot to identify bomb-damaged buildings from satellite imagery, which can alert teams to where unexploded bombs and shells are likely to be. Doing so cut the time involved in identifying damage in such hotspots from many days to hours. However, the demands of using models to detect mines are far greater. It’s why AWS and HALO are using the human-in-the-loop capabilities of Amazon SageMaker Ground Truth, to harness the experience of demining experts and increase accuracy quickly.

A collection of mines and unexploded ordnance (UXO) from Chkalovske, Ukraine.

“There are many different models and variations of mines. For example, there are seven different variations of the TM-62 anti-vehicle mine, several of which we are currently finding in Ukraine,” said Jesse Hamlin, senior GIS officer at the HALO Trust. “That requires a huge amount of training for our ML models, which puts even more importance on the richness of the high-resolution imagery that we’re able to work on with AWS. If we make this work, we’ll be able to start scanning large swathes of territory and greatly speed up the process of finding mines on the surface.”

Turning high-resolution images into restored lives

HALO recently cleared its two millionth mine—in addition to millions more UXO—and the rate with which it removes these hazards will only need to increase as conflicts continue to spread. Every device that’s deactivated adds up to a life potentially saved, and every area that’s cleared affects thousands more.

In many countries, HALO has the arduous task of prioritizing the clearance of hundreds of minefields. By better understanding where clearance is most needed, it can pinpoint where to deploy its life-saving resources. Geospatial ML with Amazon SageMaker, which trains models on satellite imagery and the maps built up by HALO’s field teams, has a key role to play. AWS and HALO will develop ML models to identify signs of renewed human habitation and activity in current and previously mined areas. This provides a better understanding of where clearance demand is greatest. It also demonstrates how tackling the threat of explosives enables life to flourish once again and helps to maintain support for demining, by highlighting the ongoing value of HALO’s efforts.

“We can see the impact that successful demining has,” said Hyman. “The minefield in Huambo, Angola where Princess Diana walked decades ago alongside HALO deminers is now a bustling city. Our work makes that possible. It doesn’t just save lives. It restores lives – and that’s why it’s so important.”