Researchers from the University of Maryland are employing deep learning, an AI technique, along with satellite imagery to identify unexploded munitions in Ukraine. The team has successfully mapped approximately 2.5 million artillery strike craters spanning a 500-mile area in the country’s agricultural regions. This data can be used by demining organizations to prioritize cleanup efforts in the most hazardous areas. It is estimated that failure rates for Soviet-era artillery shells, used by both sides of the conflict, range from 10% to 30%. This suggests that nearly one million unexploded shells may be scattered throughout the surveyed region.
Associate Professor Sergii Skakun, a co-author of the research paper published in the Science of Remote Sensing journal, emphasized the peril faced by farmers in the affected region. They are forced to choose between bankruptcy and the risk of injury or death. Skakun explained that farmers cannot delay planting until the area is safe, as they depend on the income generated from their crops. However, the presence of unexploded munitions puts them in danger, especially when operating tractors.
Russia’s Artillery Doctrine and the Risk of Unexploded Ordnance
Russia’s artillery strategy involves saturating target areas with a large number of unguided munitions. During heavy offensives, it is estimated that the Russian side expends up to 60,000 shells per day. Anecdotal evidence suggests that incidents involving unexploded ordnance are frequent. Erik Duncan, a faculty specialist in the Department of Geographical Sciences and a co-author of the study, shared an example of a tractor hitting a mine in a district where he spoke to a demining official that morning.
Failure to adequately clear leftover ordnance after the fighting ceases would leave Ukraine facing the same dire situation as other countries in former war zones. Annual global estimates of civilian deaths caused by unexploded ordnance range from 10,000 to 20,000, with a significant number of victims being children, according to UNICEF. Eastern Ukraine has been classified as one of the most mine-affected regions on Earth since 2017, even before the full-scale invasion by Russian President Vladimir Putin in 2022.
Using Deep Learning and Satellite Imagery for Efficient Detection
To develop the AI system capable of identifying artillery craters, Erik Duncan manually labeled 18,000 craters in high-resolution satellite imagery. The researchers then utilized commercial data from Planet Labs’ SkySat and Maxar’s WorldView satellites, which can capture details as small as 30 to 50 centimeters. This data enables officials responsible for monitoring large areas to identify fields, villages, and other locations that are likely to contain unexploded ordnance. By eliminating the need for labor-intensive visual analysis of satellite data, the system offers efficiency and statistical reliability.
Although the AI system cannot directly detect mines, it has been trained to identify military vehicle tracks on open ground that were not made by farm equipment. These tracks serve as indicators of military operations and suggest the presence of nearby anti-tank mines. The researchers obtained satellite data through UMD’s participation in the NASA Harvest Program, which monitors global croplands. They also benefited from the NASA Commercial Smallsat Data Acquisition (CSDA) Program.
Future plans for the system include expanding its geographic coverage, obtaining more recent imagery, and refining it into a user-friendly platform. This platform can be utilized not only in Ukraine but also in other regions recently affected by shelling. The researchers even suggest the possibility of using a modified version of the system to detect historical artillery craters, although this would require further research to identify features obscured by the passage of time.
Leave a Reply