Drones against meteorites

At least 500 meteorites fall to Earth every year, but less than two per cent of them are ever discovered by humans. Researchers have now trained drones to search for impacted asteroids. It may seem like a surprising number, but up to 500 meteorites fall to Earth’s surface every year – we just don’t detect them all. Two percent of them remain undiscovered.

Researchers have now developed a new way to find asteroids that touch Earth: they have trained drones to find them. The drones fly autonomously and in a grid formation, taking pictures of large areas – and using artificial intelligence to find potential meteorites in the photos.

In most cases, meteorite impacts go unnoticed because they fall in places that are inaccessible – for example, in the ocean. In other cases they are not found in time. The drone, developed at the University of California, Davis, was tested in the vicinity of Lake Walker in Vegas, where a meteorite hit in 2019. “The images can be analysed using a machine learning classification system – so meteorites can be identified, among many other things in the area,” said Robert Citron, a doctoral researcher at the university.

The concept-aware meteorite classifier “uses a combination of different convolutional neural networks to recognise meteorites from images of drones in the field,” the research team wrote in an article published in the journal Meteoritics & Planetary Science.

However, the software has not yet been perfected. While it could identify test meteorites placed in dry lake beds, it also had several false detections. But experts are optimistic about the results already achieved. As Robert Citron explains, with each field test, more data is acquired that can be incorporated into the dataset and used to retrain the object detection network and improve precision.

They will therefore continue to improve detection accuracy. What they need now is a better drone with a higher resolution camera. It takes approximately 100 hours to find a meteorite using conventional methods. In other words, the technology could save researchers a huge amount of work.