Scientists develop AI-backed koala spotting system

Chennai: Australian scientists have published an improved and innovative method for estimating the number of koalas in an area by using drones and an artificial intelligent algorithm.

Published in the journal Ecology and Evolution, the study explains the statistical method that uses the number of koalas automatically detected in infrared images of bushland as a starting point.

Queensland University of Technology researchers led by Associate Professor Grant Hamilton continue the quest of identifying surviving koala populations in bushfire areas.

A New South Wales parliament report in June says that the bushfires destroyed 24 per cent of koala habitat and killed at least 5,000 koalas.

Professor Hamilton, who co-authored the latest study with PhD student Evangeline Corcoran and Dr Simon Denman, said all methods for spotting koalas in heavy bushland faced challenges, whether spotters used traditional methods such as people looking up at the trees, dogs brought in to sniff out the koalas or high-tech tools such as infrared drones.

“All abundance estimation methods are at least a bit wrong—that’s why they’re called estimates,” Professor Hamilton said.

Evangeline Corcoran, the lead author on the article, said that finding wildlife in a complex environment could be very challenging.

“We never have perfect knowledge, so we never know exactly how many koalas were there when we do a count,” Ms Corcoran said.

“No matter how accurate the drone cameras, a koala could be hiding behind a branch when the drone flies over the area or perhaps one koala is counted twice in an aerial survey.That’s why we generally have a margin of error. We use different terminology, but for example in general terms our current count might have an error margin of plus or minus 10 per cent. That means we’re confident that the true number of koalas is somewhere within the margin of error”.


NT Bureau