An interview with Shahab Moeini, Head of Strategy at SubSphere, shows how drones and artificial intelligence (AI) could revolutionize humanitarian demining. The system is available to our partner organizations free of charge.

What motivated you personally to get involved in AI-supported demining?
I bring a rather unusual combination of experience to the table: I am a drone and remote sensing expert as well as AI and machine learning enthusiastic, but I have also worked as a deminer myself in the military and have been a humanitarian aid worker with the International Committee of the Red Cross ICRC in various crisis areas. Of all the weapons systems I have come to know, mines are the most insidious. These weapons are cheap to produce, easy to deploy, but extremely difficult to remove and neutralize. During my humanitarian work, I have witnessed how devastating these weapons are—not only for the individuals who step on them and suffer life-changing injuries or death, but also for the economic development of entire regions. We have a moral obligation to clear these mines.

How does AI-assisted mine detection work? What data does the AI use to distinguish mines from other objects?
At Subsphere, we deploy drones over suspected minefields to collect high-resolution aerial data. These drones can be simple commercial models, equipped with various advanced sensors—including radar, infrared, thermography, LiDAR, and ground-penetrating radar (GPR).

The collected data is then uploaded to our secure server by a partner organization operating in the field. Our AI-enhanced system, the Deminer™, processes this data and analyzes it for patterns that may indicate the presence of landmines. Using machine learning algorithms trained on real-world data, the system identifies probable mine locations with high accuracy.

It’s important to note that our system does not perform real-time detection. The data volumes are substantial, and many mine-contaminated areas lack reliable internet connectivity. The drones must first complete their full survey of the area, after which the complete dataset is analyzed as a whole.

How reliable are the current AI models? Are there tests or certifications to guarantee their safety in practice?
This is an important question. To be honest - there are many researchers and startups who claim that AI can solve all problems in mine clearance. This is not true. AI can help, but we need to know the limits. Our system currently gets it right 85 percent of the time when it comes to surfaced or semi-surfaced or abundant mines. We're not quite there yet with buried mines, but I'm hoping for a breakthrough by the fall of 2025. Reliability depends on various factors, such as humidity, temperature and soil conditions.

Critics warn of so-called “AI hallucinations”, i.e. false but plausible-sounding results. How do you deal with this issue?
At Subsphere, we believe in transparency. We openly share our error logs and communicate honestly about the limitations of our technology. From the outset, we’ve made it clear: AI is not the ultimate solution—but it is a powerful tool. This open approach has significantly strengthened the trust of our partners, especially in Ukraine—a success we’re proud of.

We are currently collaborating closely with several key organizations in Ukraine, including the Ukrainian Demining Service (UDS), the International Humanitarian Demining Academy, the Ukrainian National Humanitarian Demining Association (UNHDA), and a national aviation university.

One of the major challenges we face is the lack of comprehensive validation for AI-based mine detection. To address this, we work with real-world data provided by our global partners and cross-reference it with mines physically uncovered during clearance operations. This allows us to continuously refine and validate our models.

While false positives — cases where the AI mistakenly flags a mine—can slow down operations, false negatives, where a mine is missed, pose a potentially lethal risk. That’s why our team combines world-class AI scientists with experienced demining professionals to identify and reduce sources of error. Sometimes, it’s something as subtle as a clump of grass mimicking the shape or signature of a mine. In this work, precision is everything—lives depend on it.

Many mines are deeply buried or covered with vegetation. Can AI overcome these obstacles, or are manual inspections still necessary?
Buried landmines require a combination of advanced sensing technologies — ground-penetrating radar (GPR), thermal imaging, and magnetometers—to be effectively detected. Each sensor contributes a unique layer of information: GPR identifies subsurface anomalies, magnetometers detect metallic components, and thermal imaging captures temperature differences between mines and surrounding soil. In regions with significant daily temperature shifts, thermal images taken in the morning and evening can reveal mines, as they tend to heat up faster and retain warmth longer than the ground around them. Our AI integrates data from all these sources to highlight likely mine locations, serving as a powerful analytical tool. However, it does not replace human expertise—it enhances the capabilities of trained demining professionals, helping them make faster, safer, and more informed decisions.

Why do you have a branch in Switzerland in addition to your headquarters in Canada?

Switzerland is a global center for humanitarian demining, hosting key international organizations such as the UN Mine Action Service (UNMAS), the Geneva International Centre for Humanitarian Demining (GICHD), the International Committee of the Red Cross (ICRC), and many others. This unique concentration of expertise makes Switzerland a driving force in shaping global mine action policy and practice.

At the same time, Switzerland is at the forefront of drone and sensor innovation. Leading companies like SwissDrones and XER Technologies are pioneering the development of high-endurance, high-precision UAV platforms that are ideal for humanitarian applications, including mine detection. Many of the components and technologies we use in our operations are developed and manufactured in Switzerland—demonstrating the country’s dual leadership in both humanitarian action and advanced technology.

How much does Deminer™ technology cost?
Our system currently runs on secure servers based in US and Canada. Due to the vast volumes of data processed, computing power accounts for nearly half of our operational costs. Ultimately, our goal is to transfer the system to a major humanitarian organization, with our team remaining in a support role only.

Maintaining 24/7 operations with a dedicated team of six people costs approximately $300,000 annually. While we’ve received interest from large corporations, many of them aim to adapt the system for military or weapons-related purposes. This is not aligned with our mission. For that reason, we remain cautious about certain tech-driven initiatives—where investments are tied to weapons development.

We are committed to keeping our work purely humanitarian. Since March 2025, we have offered humanitarian organizations free access to Deminer™, along with full support and training. Sharing this technology can directly save lives—and for us, that is what matters most.

How could AI change demining in the next five to ten years? Do you see a long-term possibility of fully automating the process?
AI has the potential to fundamentally transform demining over the next five to ten years — making it faster, safer, and more scalable. In the immediate future, our focus is on expanding the capabilities of our AI system to detect a broader range of explosive remnants of war (ERW), including cluster munitions. We expect this capability to be operational by early 2025 or 2026.

Beyond detection, we're exploring the integration of AI with robotics to enable precise, remote interventions — similar to how the DaVinci surgical robot allows doctors to perform complex procedures at a distance. Applied to demining, this approach could allow trained operators to investigate and neutralize suspicious objects without ever entering the danger zone themselves. The result: significantly improved safety for demining personnel.

As for full automation — it's a long-term goal, but not an immediate reality. The complexity and variability of terrain, vegetation, and the unpredictability of buried mines make full autonomy extremely challenging. However, AI can and will take on more of the dangerous, data-intensive work, acting as a force multiplier for human teams.

In short, we don’t see AI replacing humans—but we do see it empowering them, accelerating clearance timelines, and most importantly, saving lives.

What are the biggest ethical challenges in this area?
One of the biggest ethical challenges in AI-based demining is the tension between profit and principle—money over morals. As this technology proves its effectiveness, it inevitably attracts the attention of powerful private actors, including defense contractors and tech companies, many of whom see its potential for weaponization or military surveillance rather than humanitarian use.

The core mission of AI-based demining is to save lives, restore land for civilian use, and support post-conflict recovery. But when commercial interests enter the picture, the focus can shift away from these goals. Some companies express interest in the technology not to help vulnerable communities, but to repurpose it for battlefield intelligence, autonomous targeting, or border control systems. This raises deep ethical concerns: the very tools designed to prevent harm could end up being used to cause it.

There’s also a risk of exclusivity and access. If AI-based demining becomes dominated by private firms, access may be limited to clients who can afford it—leaving the poorest, most mine-affected countries behind. The life-saving potential of this technology must not be gated by commercial licensing fees or proprietary restrictions.

What would you say to a deminer who is afraid that this technology will take their job?
I completely understand your concern—it's a natural reaction when new technology enters any field. But let me be very clear: AI-based mine detection is not here to replace deminers. It’s here to support you, protect you, and make your work safer and more efficient.

The reality is, no AI system can fully replace the expertise, experience, and judgment of a trained deminer in the field. What this technology does is analyze aerial and sensor data to identify high-risk areas—so when you go in, you already have a clearer idea of where to focus. That means less time in dangerous zones, fewer false alarms, and a reduced chance of encountering hidden threats unexpectedly.

Think of it as an extra layer of safety—like a smart tool that helps you work smarter, not harder. It helps speed up operations and lowers the overall risk, but it still depends entirely on human decision-making and field knowledge.

In fact, as this technology grows, it will create more opportunities for skilled deminers to lead operations, train others, and guide how these systems are used on the ground. You’re not being replaced — you’re being empowered.

Your work is still at the heart of the mission. The goal is not to take your job — it's to make sure you can go home safe every day after doing it.

Further information: www.subsphere.ca

Photo: Freepik