What are plastic landmines, and can AI help detect it? Researchers develop AI method to identify hidden explosives using drones

Hidden anti-personnel landmines continue to threaten millions of people living in conflict and post-conflict regions around the world. Many of these explosive devices are made from plastic, making them difficult to locate with traditional metal detectors and other common detection methods. Researchers are now exploring new technologies that can improve the search process and support humanitarian demining efforts. A team from Binghamton University has developed a system that combines drones, cameras, and machine-learning software to identify suspected landmines from aerial images. The research aims to help demining teams work more efficiently while reducing risks during field operations and early land surveys.

What are plastic landmines?

Plastic landmines are anti-personnel explosive devices made mainly from plastic instead of metal. They are designed to injure people who step on or disturb them. Because they contain very little metal, they are difficult to detect with standard metal detectors. Many plastic landmines, including the PFM-1 butterfly mine, are small in size and can remain hidden on the ground or become buried under soil and vegetation over time. They continue to pose risks to civilians and demining teams long after conflicts have ended.

Can AI help detect it?

Artificial intelligence can help detect plastic landmines by analyzing drone images with machine-learning algorithms. In this study, researchers used the YOLO object detection model to identify PFM-1 landmines from aerial photographs captured at low altitude. The AI system was trained using images of real and 3D-printed mines placed in different environments and lighting conditions. It is designed to identify suspected hazardous areas quickly, allowing trained demining teams to focus their search while working safely in the field, even without an internet connection.

Researchers introduce a new detection system

Researchers have developed a new artificial intelligence-based method to detect plastic landmines spread across large areas. The study, titled Deep Learning and Multiview-Based Detection of Scatterable PFM-1 Landmines: Performance, Out-of-Sample Evaluation, and Field Readiness, was published in the journal Geomatics. The project was led by geology graduate Sharifa Karwandyar, Associate Professor of Geography Thomas Pingel, and Associate Professor of Earth Sciences Alex Nikulin at Binghamton University.

The research focuses on plastic anti-personnel landmines, which are difficult to detect because they are small and often made almost entirely of plastic. Traditional metal detectors usually cannot identify these mines. Other geophysical methods, including ground-penetrating radar, magnetometry, and electromagnetic induction, also perform less effectively when searching for plastic landmines than metal ones.

Why plastic landmines remain difficult to locate?

Plastic anti-personnel landmines present a major challenge during humanitarian demining operations. Their small size and plastic casing make them difficult to identify with existing detection systems. One of the biggest concerns is the PFM-1 landmine, a scatterable mine first developed during the Soviet era. These mines are designed to spread over large areas after being released from the air. Their shape allows them to fall slowly, similar to maple seeds, helping them cover wide areas.

According to Alex Nikulin, these mines were designed to injure rather than kill. He explained that treating injured soldiers places a greater burden on military resources than fatalities. He also noted that every part of the mine's design helps it avoid detection. Because these mines can remain hidden for years, they continue to pose risks to civilians, demining workers, and communities returning after conflicts.

Why drones are important for finding hidden mines?

Researchers explained that the location of these mines depends on the conditions in the affected area. In active conflict zones such as Ukraine, scatterable landmines often remain close to the ground surface. In regions where conflicts ended years ago, the mines may become buried under soil or covered by vegetation over time.

Since each PFM-1 mine is roughly the size of a mobile phone, drones must fly close to the ground. Flights are generally conducted at heights of about 10 to 20 meters to collect images with enough detail for analysis. Sharifa Karwandyar said the technology is designed as a first-pass assessment. Instead of replacing manual demining, it helps determine whether an area should be classified as a suspected hazardous location requiring further investigation.

Researchers trained AI using real and replica landmines

The research began as part of Karwandyar's master's thesis. She used a drone-mounted camera to capture aerial photographs. Software combined the images into larger maps before they were analyzed using the You Only Look Once (YOLO) machine-learning algorithm. The research team trained the AI system using inactive PFM-1 landmines together with 3D-printed copies.

The mines were placed throughout Binghamton University's Nature Preserve in different locations and under different conditions. Images were collected from multiple viewing angles, lighting conditions, and environmental settings. This allowed researchers to build a large dataset showing how the mines appeared in real landscapes. The goal was to prepare the AI model for situations it would likely encounter during actual humanitarian operations.

Two AI models were tested during the study

Researchers created two different versions of the YOLO detection model. The first model was trained only to recognize PFM-1 landmines. The second model learned to identify both PFM-1 mines and many other common objects that could appear in outdoor environments.

The second system produced lower performance scores because it had to distinguish landmines from natural objects such as leaves and other materials visible in camera images. Researchers said these lower results may actually provide a more realistic picture of field performance because real environments always contain many distracting objects.

System is designed for field use without internet access

Most of the computing work takes place before deployment. Thomas Pingel explained that training the AI model can require several hours or up to one day, depending on the number of images used. Once training is complete, the system can operate with simple equipment. Researchers said only a consumer-grade laptop, drone, and camera are required in the field.

Karwandyar also focused on making the software process images in real time or near real time. This allows demining teams to examine results while still working in the field instead of returning to another location for analysis. The system can also function without an internet connection.

This feature is important because many conflict and post-conflict areas have damaged communication infrastructure. In places such as Ukraine, signal jamming and GPS interference can also make online systems unreliable.

Technology supports trained demining teams

Researchers emphasized that artificial intelligence is not intended to replace trained demining specialists. Landmine clearance continues to depend on professionals who understand safe demining procedures and local communities familiar with the terrain. The AI system is designed to make the search process more efficient by helping identify locations that deserve closer inspection.

Nikulin said there is often a gap between laboratory research and the practical challenges faced during humanitarian demining work. By working closely with non-governmental organizations, the research team ensured the technology addresses field requirements rather than remaining a laboratory project. If adopted more widely, the system could improve early landmine surveys, reduce search time, and support safer humanitarian operations in regions affected by landmine contamination.

FAQs

Q1. What are plastic landmines, and why are they difficult to detect?
Plastic landmines contain very little metal, making them hard to detect with metal detectors. Their small size and plastic casing also reduce the effectiveness of many traditional geophysical detection methods.

Q2. Can AI replace human landmine clearance teams?
No. AI helps identify suspected hazardous areas using drone images. Trained demining experts and local communities still perform verification, safe clearance, and disposal of landmines during humanitarian operations.