AI Scans Drone Photos to Find Landmines Hidden Across Ukraine

Ukraine is now the most heavily mined country in the world. World Bank assessments estimate that roughly 17 million hectares of land, an area larger than England and Wales combined, may be contaminated with landmines and unexploded ordnance. Hidden explosives wait in farm fields, schoolyards, forests, and along roads. They kill and maim civilians long after the fighting moves on, and they keep farmers off their land at a moment when Ukraine needs every harvest it can get. A tool called SpotlightAI, built by Safe Pro AI, is speeding up the search by letting drones and artificial intelligence do the first and most dangerous pass.

SpotlightAI works on a simple premise: instead of sending people with metal detectors across unknown terrain, fly a drone over the area first, photograph everything, and let a computer vision model trained on explosive ordnance scan every image. The platform processes each image in 0.2 seconds on Amazon Web Services, flagging suspected mines and unexploded bombs and pinning their GPS coordinates on a map that demining teams can prioritize. As of late 2025 it had analyzed more than 1.66 million drone images in Ukraine, identified more than 28,000 explosive threats, and surveyed 6,705 hectares, an expanse roughly comparable in size to Manhattan.

A Problem the Size of a Country

The scale of contamination in Ukraine has no modern precedent. An area of roughly 139,300 square kilometers, about a quarter of the country, may be contaminated with landmines and unexploded ordnance. Ukraine already ranks among the countries with the highest civilian casualties from mines in the world. Farms are especially affected, and every season that fields go unplanted adds to the economic damage of the war.

The traditional approach cannot keep up. Manual clearance with metal detectors and probe rods is slow, expensive, and dangerous. World Bank assessments estimate that fully clearing the country could take more than a decade and cost over 29.8 billion dollars. Humanitarian demining organizations have also faced a chronic shortage of trained surveyors, which means large areas wait months or years before anyone even looks at them. That is the bottleneck SpotlightAI was built to relieve: the initial survey, the first look at a piece of land to decide where the explosives probably are and where clearance teams should go first.

How the Detection Works

Demining teams fly commercial drones in systematic patterns over suspected contaminated areas, capturing high resolution images of the ground below. The images are uploaded to SpotlightAI, where computer vision models trained on vast datasets of explosive remnants of war look for the visual signatures of mines, cluster munitions, shells, and rockets. Some of these objects are only a few centimeters across, partially buried, or hidden in grass, which is exactly the kind of detail tireless machine analysis is good at catching.

Each image is processed in 0.2 seconds using AWS hyperscale computing, and every detection is geolocated so the output is a map, not just a list. Humanitarian operators in Ukraine have described it as one of the most effective tools available for visual detection of explosive ordnance, using it both for primary surveys of potentially contaminated areas and for quality control after clearance work is done. The AI does not replace deminers. It tells them where to dig, and just as importantly, where not to.

From Startup to United Nations Contract

Safe Pro AI, the company behind SpotlightAI, developed the platform with support from Amazon Web Services, which featured the work in a partner case study. The tool is used in Ukraine by humanitarian demining operators, and its results reached the point where the United Nations Development Programme placed a purchase order for AI minefield analysis to support its mine action program in the country. In 2025 Safe Pro Group also entered a strategic partnership with Ukraine's national mine action training center to modernize demining with AI powered threat detection.

The milestones have compounded quickly. By October 2024 the platform had confirmed more than 6,100 explosive remnants of war across 3,135 hectares. By December 2024 the count passed 16,500 detected items. By late 2025 it exceeded 28,000 threats identified across 6,705 hectares, with more than 1.66 million images analyzed. Each growth step came from the same fly, upload, and analyze loop that any demining organization can adopt without new hardware beyond a drone.

Why Speed Saves Lives

Landmine contamination is not only a safety problem. It is a poverty trap. Families cannot farm, children walk long detours to school, and displaced people cannot return home until land is declared safe. Every hectare that moves faster from suspected to cleared returns livelihoods and housing years earlier than the old timelines allowed. Faster surveys also make the enormous clearance budget go further, because machines scanning imagery are far cheaper per hectare than teams walking grid patterns with detectors.

The Ukrainian experience is becoming a template. Contaminated countries from Cambodia to Iraq watch what works there, and drone based AI survey is proving cheaper, faster, and safer than the methods that preceded it. A field that once waited a year for a survey team can now be flown in a day and analyzed before the drone lands. In the slow, dangerous work of removing the debris of war, that difference is measured in lives and in harvests.