Among the virtues of bees you may not be aware of is their knack for detecting bombs.
Thanks to the fact that they can pick up the scent of explosives with their antennae, researchers in countries such as Croatia have spent years perfecting how to use bees as landmine locators. But there’s a problem. As the insects whizz merrily about a mine-contaminated area, it’s extremely difficult for humans to keep track of where they go, not least because chasing bees across a minefield is not a great idea.
That’s where the drones come in. A team from Bosnia and Herzegovina and Croatia have come up with a way of using drones to monitor the bees while they work. The unmanned aerial vehicles fly around, capturing footage of the insects, which is later analysed by computers to reveal where landmines may be hidden in the ground.
Landmines buried during wars that happened decades ago continue to present a deadly threat in many parts of the world. Many thousands were planted during the Balkans war of the 1990s and many persist today. There are an estimated 80,000 landmines in Bosnia and Herzegovina and a further 30,000 or so in Croatia. Clearing the devices is seen as a long-term, arduous project with no easy solutions. But technological innovations could still make a difference.
A “danger of death” sign is seen at a minefield in a woodland in Sarajevo, Bosnia and Herzegovina on November 20, 2017. “We wanted to try to exclude humans from potential danger… and try to use drones,” says Vladimir Risojević from the University of Banja Luka in Bosnia and Herzegovina.
Previously, another team of fellow researchers had honed a method for training bees in landmine detection. They achieved this by getting the bees to associate the smell of TNT with food – a sugar solution.
In the field, the trained bees tend to cluster near to places where mines are buried, in the hope of finding food. Such efforts have been active for many years but Prof Risojević says he and his team realised that computers could help by automatically analysing footage of the mine-seeking bees, in order to plot their activity and more easily locate the mines.
Even this proved tricky.
“It’s very difficult for human observers to find these flying bees in this video footage let alone computer vision systems,” he says. “There were moments when I thought that we are outright crazy for trying to do that but I am pleasantly surprised with the results that we obtained.” The team began with drone-captured footage of an outdoor area, onto which they superimposed “synthetic bees” – fuzzy grey blobs zooming about the scene.
When they managed to get the synthetic bees to look indistinguishable from footage of real bees, the team turned to a machine-learning algorithm, and trained it to accurately detect and follow the blobs on screen.
In tests described in a recently published paper, the algorithm proved more than 80% accurate at tracking these digital bees.
The researchers then took to a minefield, a safe one with real but defused mines buried in undisclosed locations at the Croatia Mine Action Center, to see how the system performed under authentic conditions.