Source: cnBeta
Go outside on a clear night and if you're very lucky you'll see meteorites falling. NASA estimates that 50,000 meteorites from space have been found on Earth. Meteors or fireballs that form when they enter the atmosphere can be beautiful, but they are difficult to track. Of those 50,000 meteorites, astronomers were only able to map the orbits of about 40 meteorites.
Simers-Anderson and his colleagues at Curtin University in Australia report that they found a meteorite in remote Australian inland that once orbited along an elliptical orbit between Venus and Jupiter, and they found the meteorite with two drones and machine learning methods.
Curtin University's Space Science and Technology Centre, located in the city of Perth, operates the Desert Fireball Net, a system of 50 automated cameras that monitor Australia's night sky for the arrival of meteors. One night last year, two of the cameras tracked a streak in the sky, and the system calculated that a small rock could have fallen into the desert bush of Western Australia in an area known as The Narab.
The observations were not ideal, and the researchers estimated that the meteorite weighed between 150 and 700 grams and fell in an area of 5 square kilometers, but Anderson and two colleagues decided to conduct a field trip. In December, they drove more than 1,000 kilometres from Perth in search of a needle in a haystack: a blackened rock on the desert floor, 50 kilometres from the nearest paved road. In the past, such trips were meaningless. Meteorite hunters usually search the ground on foot, going back and forth in grid patterns, hoping they'll find something useful. Eighty percent of the time, they all fail. It turns out that humans are not very good at dealing with this repetitive work.
That's what technology does. They used off-the-shelf hardware, a quadcopter drone with a 44-megapixel camera, and a desktop computer with a good graphics card. The unusual part is the convolutional neural network they run on it. The machine learning system needed to be trained, so the researchers gave it a drone-taken picture of the terrain of Nyulab. Some of them include meteorite samples borrowed from local museums and placed on the ground. The images are scored on a 1-point scale, i.e. definitely a meteorite, even though each image is only shown as a black dot. Other images showing random terrain nearby are given a score of 0. Through repeated exercises, the machine algorithm learned to deal with false positives: bottles, jars, desert plant roots, and the occasional kangaroo bone.
In three days, the researchers conducted 43 drone flights, flying back and forth at an altitude of about 20 meters, recording 57,255 images. Back at camp, they start working on their images. In the first four flights alone, the algorithm gave at least 0.7 to 59,384 objects, in the range of 0 to 1, with many possible specimens. The researchers will soon be able to shrink them down to 259, then 38, which they recheck with a second, smaller drone. Soon, they narrowed the range to 4 and, guided by GPS, walked to look for them, eventually finding a soft stone, a common iron-rich meteorite. It is 5 cm long, about the size of an egg, and weighs 70 grams.