I have been working with drones since 2014, but the outbreak of war in Ukraine marked a turning point in my career. Since 2022, my focus has shifted to exploring how to use drones to automate the automation of humanitarian magic – what capabilities they need, and how technology can make these efforts safer and more efficient. As part of this work, I closely follow the Geneva International Humanitarian Centre (GICHD), attend their events, and interact with experts regularly.
Given that drone-based solutions paired with AI, it is really only helpful at the non-technical survey (NTS) phase of the humanitarian reduction process. This means that drones scan large areas and collect data. The machine learning model then analyzes this data into the flag region possible Contains landmines. Not the exact place of the mine.
The technical survey (TS) of the confirmation and map of contaminated areas remains dependent on personnel with metal detectors, trained dogs and mechanical threshers. They enter the mining area to find out the exact location of the hazard.
The process has been long, risky and expensive:
The mines also continue to pose a threat to civilians – at least 5,757 mines/ERW casualties in 2023.
In this post, I explain why current drone-based solutions are not currently available for technical surveys (the most expensive and time-consuming stages at the moment), and share my opinion on the best way to solve the problem.
Discovering mines under soil or vegetation is nearly impossible
Drones with standard optical cameras or thermal cameras often capture images from a single downward angle. This method works well when surface level abnormalities are found, but cannot detect buried or hidden mines. Therefore, drones are mainly used in non-technical investigations in humanitarian magic.
One of the frontline solutions – Security Pro AI – reports that in trees and bushes, they have only a 5% detection rate.
Even though it is related to most of Ukraine’s mines scattered on the ground rather than buried Ukraine, the situation in Cambodia is very different:
- During the conflict between the 1970s and 1990s, 40-6 million landmines still exist
- Since 1979, more than 64,000 casualties have been reported, with children as the main victims
Non-metal and old metal mines are difficult to detect even on the surface
Non-metallic mines account for a large proportion of mines in current and former conflict areas. They are intentionally bypassed detection by traditional metal detectors.
Visually, it is difficult to detect non-metallic ores. They don’t glow, stand out in the image, and they don’t show up well on the thermal camera. Metal detectors and magnetometers either miss them or trigger too many false alarms.
Therefore, current drone-based detection tools often miss non-metallic mines altogether.
When it comes to old metal ores, corrosion changes their appearance and behavior, so they blend into the ground and react poorly to detection tools. It is even harder for people with deformities to recognize in images.
And because these mines are difficult to find, they take longer to find and remove, or they hide and put Deminers and Deminers and civilians at risk.
Weather and daytime dependence
If we are talking about drones with RGB and multispectral cameras, daylight is needed. In clouds, shallow or shadowed areas (forests, ruins), image quality and object detection also decline.
The temperatures on the ground are different from mine, and the thermal detection works best at dawn or dusk in turn. At noon, the sun heats everything on average, reducing contrast.
While the blurred surface details of rainwater and wet soil change the color and temperature of the soil and can hide soil interference or thermal abnormalities. Snow simply covers visual markers and equalizes ground temperatures, making mines undetectable.
Flying drones also significantly slowed down the NTS shedding phase only at some time, especially in areas where the weather is unpredictable.
This technology is very expensive
Among the seven affected countries, the estimated anti-someone mine pollution zones reached more than 100 kilometers.
According to Ukraine’s test, using new technology can cut $3000-5000 to $600-800 per hectare and $70,000 per square kilometer. In some areas, it may exceed the land price itself.
The main reason for high costs is to see multiple false alarms as real threats. On average, a team cleared more than 50 suspected mines and could only find one actual mine.
The most polluted area in developing countries. Without funding from international organizations or governments, they are unable to devalue them.
The cost is too high for companies to join. Once Demining becomes cheap enough, companies may lease mine contaminated land on conditions for clearing mine clearance. In return, they will be used for long-term use at a token price and some tax deductions.
Solution?
In my team, we explored ways to collect more data, view it through leaves and soil, and still maintain adequate resolution.
An example of a promising direction is a project by researchers at Oviedo University. They are testing an array-based grounded synthetic aperture (GPR-SAR) system installed on drones.
In reality, their on-board verification proved that the technology solved the following problems:
1) The radar accurately points out the location of the mine, only manual disarmament or damage is retained.
By using all possible radar paths (completely multi-static configurations), they obtain high-resolution images where the buried target appears brighter and clearer. And can detect with precise and challenging targets such as small, non-metallic and shallow buried objects such as plastic anti-human mines, wooden pressure panels and PVC pipes.
2) The solution can be worked day or night, in different weather, and even moderate vegetation.
How it works:
- Send radar pulses to the ground.
- Detect reflections of underground changes (such as plastic, metal, voids).
- 3D underground images are constructed with centimeter-level accuracy by combining radar signals from multiple transmitter receivers (TX-RX) pairs and flight positions.
This solution still has its limitations, but based on my background, it is the most relevant R&D direction at present.
One of the main advantages of GPR is how much data it can collect. More data means that researchers can improve the accuracy of the identification/classification phase through AI. This leads to more effective investigation and clearance efforts and reduces overall costs by 50% or more.
