Finding one meteorite usually requires researchers to walk at least a few million square metres. Seamus Anderson from Curtin University has a way to speed up his meteorite hunts.
By sending out drones with mounted cameras to photograph different sites, Seamus can collect over 10,000 images in a day. Then, the images are digitally divided into 100 million smaller chunks. These chunks are processed by artificial intelligence (AI).
By training the AI with photos of real meteorites and spray-painted rocks, Seamus’s AI can remove 99% of image chunks that don’t include meteorites. Then, his team checks approximately 50,000 remaining image chunks for non-meteorite objects. After this, they’re left with the most meteorite-like objects.
Next, a second drone that can fly closer to the ground is sent out to take clearer photos of the locations with any meteorite-like objects. After another round of manually checking drone images, researchers will go inspect the most promising sites.
Using drones and AI, Seamus and his team has found at least 5 meteorites so far – all without too much walking!
“You’re going from about 300 days of human effort down to a dozen or so,” says Seamus. His team plans to train their AI to better identify and remove non-meteorite image chunks to improve their success.