DEMINE Foundation's AI Drone Is Scanning Ukraine's 139,000 km² Minefield

The problem is measured in square kilometers and centuries. Ukraine’s territory holds an estimated 139,000 km² of land potentially contaminated with mines and unexploded ordnance, an area larger than England [Landmines Cover 139,000 km² of Ukraine, 2026]. At the current pace of conventional demining, returning that land to safe use would take more than 750 years [DEMINE Foundation]. The DEMINE Foundation, a UK-registered charity founded in 2023, is building a system to compress that timeline. Its bet is that a low-cost drone, equipped with off-the-shelf sensors and purpose-trained AI, can detect, classify, and map threats from the air, turning a slow, dangerous ground survey into a scalable aerial scan.

The hardware wedge

DEMINE’s technical approach is defined by its constraints. Instead of developing custom military-grade drones, the foundation’s system is built on commercially available hardware. The core platform is an inexpensive, off-the-shelf drone paired with standard sensors and an onboard computer [PERPLEXITY SONAR PRO BRIEF]. The proprietary layer is the software: computer vision models trained to recognize mines and unexploded ordnance (UXO) from RGB imagery, running inference at the edge to process data in near real-time [DEMINE Foundation]. This creates a specific tradeoff. The system sacrifices the extreme precision of ground-penetrating radar or specialized multispectral sensors for affordability and deployment speed. The goal is not to replace final, hands-on clearance but to rapidly generate accurate hazard maps that tell human demining teams exactly where to focus their dangerous work.

A non-profit's path to deployment

As a charity, DEMINE operates on a fundamentally different axis than a venture-backed startup. It is registered in England and Wales as a private company limited by guarantee, a structure common for non-profits [PERPLEXITY SONAR PRO BRIEF]. Funding comes from donations and grants, not equity sales; its website lists beneficiary banking details for donor contributions [DEMINE Foundation]. The co-founders, James Phipps and David Phipps, established the organization with the primary objective of advancing R&D for locating UXO in civilian areas, initially within Ukraine [PERPLEXITY SONAR PRO BRIEF]. David Phipps also serves as President of Global Operations at NextPlat Corp, a publicly traded company [Bloomberg Markets, 2026]. The lack of a traditional sales motion shifts the adoption challenge. Success depends on convincing humanitarian NGOs, civil defense units, and international aid organizations that the drone-generated data is reliable enough to integrate into their existing clearance workflows.

The data challenge at scale

For an AI system, performance is a function of data. DEMINE’s blog openly discusses the difficulty of acquiring the training imagery needed for robust object recognition, noting the scarcity of labeled datasets for mines and UXO in varied terrain and conditions [DEMINE Foundation]. This is the central technical hurdle. The model must achieve a very high recall rate,missing a mine is unacceptable,while managing a false-positive rate that doesn’t render the maps useless. In practice, this means the initial “detection” is likely a probability score attached to a geographic coordinate, requiring human verification. The system’s real value emerges when it can scan hundreds of hectares in a day and output a prioritized list of locations for ground teams to inspect, dramatically increasing their effective coverage.

A short technical breakdown illustrates the operational flow. A drone flies a pre-programmed grid, capturing standard RGB video. Frames are processed on the drone’s computer by a convolutional neural network trained to identify objects of interest. Each detection is tagged with GPS coordinates and a confidence score, then compiled into a geospatial map. This map is the deliverable. The technical risk lies not in the individual components, which are commodity items, but in the model’s accuracy across the endless variations of a real-world landscape: a mine partially buried, obscured by vegetation, or rusted beyond its textbook shape.

Scaling this system introduces sobering challenges. Edge computing on a small drone limits model complexity and battery life. Weather, light conditions, and terrain will degrade camera-based detection. Furthermore, the step from a promising research prototype to an operationally trusted tool used by multiple NGOs in active conflict zones is a vast leap. It requires not just technical reliability but also rigorous validation protocols, training for local operators, and smooth data integration with legacy demining information systems. The foundation’s partnership with mapping firm T-Kartor, announced in a press release, is a step toward that last requirement, aiming to embed detection data into professional GIS platforms [DEMINE Foundation]. The 750-year timeline is a stark measure of the need. DEMINE’s bet is that a pragmatic, low-cost AI scanner can make the first meaningful dent in it.

Sources

1. [DEMINE Foundation] Autonomous Mine Detection | https://www.deminefoundation.com/
2. [PERPLEXITY SONAR PRO BRIEF] DEMINE Foundation Brief |
3. [Bloomberg Markets, 2026] David Phipps, NextPlat Corp Profile | https://www.bloomberg.com/profile/person/19077873
4. [Landmines Cover 139,000 km² of Ukraine, 2026] United24 Media Article | https://united24media.com/latest-news/landmines-cover-139000-km2-of-ukraine-an-area-larger-than-england-11112
5. [DEMINE Foundation] Press Release: T-Kartor partnership | https://www.deminefoundation.com/blog/press-release-t-kartor-partnership
6. [DEMINE Foundation] Understanding the AI training challenge | https://www.deminefoundation.com/blog/data-we-need-it-understanding-the-challenge