Study uses AI to find unexploded bombs from Vietnam War
At the Ohio State University, researchers are using AI to detect Vietnam War-era bomb craters in Cambodia from satellite images – with the hope of finding unexploded ordnance.
An assistant professor of electrical and computer engineering, as well as Civil, Environmental and Geodetical Engineering, Rongjun Qin is drawing together bits and pieces from many scientific disciplines (data analytics, satellite photogrammetry, remote sensing and geographic information systems) to find ways to benefit society as a whole.
Qin is working with fellow Buckeye, Erin Lin, an assistant professor of political science. Their study, funded by Ohio State's Translational Data Analytics Institute, was recently featured by Science Daily and Ohio State News.
“We had the idea of using remote sensing techniques to identify craters, which basically tells the number of exploded ordnance,” Qin said. “This could be used to estimate the unexploded ordnance by coupling the de-classified records of bombing.”
According to Lin, this new method increased true bomb crater detection by more than 160 percent over standard methods - useful data, considering 44 to 50 percent of the bombs left behind in Cambodia may remain unexploded. The study suggests anywhere from 1,405 to 1,618 unexploded carpet bombs are still unaccounted for in the area.
She said the danger is not hypothetical. In the six decades following the bombing of Cambodia, more than 64,000 people were killed or injured by unexploded bombs. Today, the injury count averages one person every week.
Meanwhile, despite the current social distancing guidelines being respected because of COVID-19, Qin and his team are still earning accolades in remote sensing research. Most recently, at the IEEE GRSS Data Fusion Contest.
“We won two tracks again,” Qin said. “This is the largest annual competition event in the IEEE geoscience and remote sensing community. Teams all over the world develop machine learning algorithms for remote sensing image analysis given a certain period, and compete for accuracy. The competition was just concluded last week, we had it all done through telework.”
- Track 1: Fourth place win: "Land cover classification with low-resolution labels," by Huijun Chen, Changlin Xiao, Wei Liu, Rongjun Qin, The Ohio State University for Automated label pre-processing, random forests, followed by classification refinement based on prior knowledge on class confusion.
- Track 2: First place win: "Land cover classification with low- and high-resolution labels," by Huijun Chen, Changlin Xiao, Wei Liu, Rongjun Qin, The Ohio State University. An ensemble of random forests trained on refined labels.
Qin said the IEEE contest goals are more of a world-wide focus, compared to the Cambodian ordnance detection research goals.
"The ordnance work detects only craters and the IEEE contest identifies 10 types of land classes. It looks into global level land-cover classification, and the ordnance paper is study specific," he said. "So the techniques used are related but the methodologies are crafted differently."
Qin and his team also acknowledge the Office of Naval Research (ONR), as much of their expertise developed for such an achievement is a result of its ongoing support.
Held annually since 2006, the IEEE Data Fusion Contest is a scientific challenge promoting innovations in analyzing multi-source big earth observation data. It fosters collaboration between the computer vision and earth observation communities, while advancing the automated interpretation of remotely sensed data. The GRSS overall goal is to provide updates in geospatial image analysis (e.g., machine learning, deep learning, image and signal processing, and big data) and data fusion (e.g., multi-sensor, multi-scale, and multi-temporal data integration). It aims to connect people and resources, while educating students and professionals.
Story by Ryan Horns Communications Specialist, adapted from The Ohio State News.