The demining of the Vietnam war era bombs in Cambodia has so far been ineffective. Machine learning and AI may help.

The Cambodian national clearance agency responsible for finding and safely removing unexploded bombs and land mining – a process called demining – is under fire for cherry-picking areas that have minimal or no risk of harboring unexploded bombs. Erin Lin, assistant professor of political science at The Ohio State University, points to this criticism in a recent UN-commissioned report. The report recommends that the clearance agency starts looking for the Vietnam war era bombs in more high-risk areas. (OHIO STATE NEWS)

Researchers bring in AI to sniff out these mines

Researchers have now combined artificial intelligence, satellite images, and declassified US military records in a model to unearth these lurking dangers to life and limb.

The new method achieved an improvement of over 160% in true bomb crater detection versus the traditional method. Alarmingly, it estimates that as much as 44 to 50% of bombs lying in the area are unexploded.

How it works

Erin Lin and Rongjun Qin, assistant professor of civil, environmental and geodetic engineering at Ohio State conducted the study. They used a satellite image of a 100 square kilometer area that faced carpet-bombing by the U.S. Air Force between 1970 and 1973.

From military records, they could estimate the number of bombs that were dropped in the area. The exploding bombs created craters. By identifying these craters, they could determine the number of remaining unexploded bombs and their approximate location.

They then applied existing algorithms that can detect meteorite craters on the moon and solar system planets to the satellite images. These algos were only moderately successful in detecting the craters.

The researchers then trained the algos on the distinguishing features of bomb craters, which by now would have changed due to vegetation growth and erosion.

Once trained, the algos did a much better job of identifying true bomb craters. In the first run, they could identify 89%. However, they also identify a number of false positives.

In the second run, the algo eliminated 96% of the false positives but missed only five true craters. The researchers observed that, at this point, the accuracy of the system was about 86%.

The researchers estimate that between 44 to 50% of the bombs dropped in the area may still be live.

That’s a huge risk to the mostly agricultural people who live in the area.

“The process of demining is expensive and time-intensive, but our model can help identify the most vulnerable areas that should be demined first,” Lin said.

Related Story: Artificial Intelligence: Now AI Could Unlock Secrets From 2,500-Year Old Iranian Tablets

Image Credit: Flickr