Meteorites are important for the study of the Solar System and the universe, in general, and engineers have found a new way to find them here on Earth.
Engineers have developed a method to search for meteorites by flares before they fall, based on the use of drones with a camera. Using the algorithm, they calculated the area of the likely fall of meteorite fragments in the US state of Nevada, and the drone flew to this place and took many detailed images, in which the neural network then found objects similar to most meteorites.
Thousands of meteoric bodies enter the Earth’s atmosphere every day. The vast majority of them burn up over oceans and sparsely populated land areas, as well as during the day when flares from them are almost invisible. Moreover, only a small part of the meteoric bodies that burn in the atmosphere do not completely burn up and fall to the surface of the Earth, as a rule, they produce flares of magnitude -8 and brighter.
Potentially, by shooting a falling meteorite, it is possible to calculate the trajectory of its flight and the approximate area of falling of its debris. This is done by professional and amateur meteorite search networks, consisting of cameras in different parts of the Earth.
But in fact, only about three dozen meteorites were found in this way. This is partly due to the fact that such searches are labor-intensive – the search areas are tens of square kilometers and, as a result, they often do not give a result.
A group of engineers led by Jim Albers of the SETI Institute and the Ames Research Center at NASA decided to simplify and reduce the cost of finding meteorites by flares from their fall and developed a system based on drones for this.
They used a 3DR Solo quadcopter, which was equipped with a laser altimeter, GoPro camera, PixHawk GreenCube flight controller, and Arducopter software. The engineers wrote a program that divides the search area into a grid of places where the drone should take a shot.
The size of the grid is selected in accordance with the shooting height, which the authors varied from two to six meters. And since the drone is equipped with an altimeter, it maintains a set altitude, regardless of the terrain.
After flying over the terrain, the drone returns, and data is downloaded from it for analysis. For this, the developers used the RetinaNet neural network. They took a network pre-trained on the ImageNet dataset and is good at detecting objects, and retrained it on their own dataset.
To do this, they used eight fragments of the Mbale meteorite, which fell in Uganda in 1992. They placed fragments in different locations and filmed them with a camera handheld and with a drone from the air. They also added photos of meteorites from the Internet to the dataset for the drones to use.
The result was a dataset of 762 images, which were additionally reflected vertically and horizontally to obtain 2448 frames on which the authors trained the neural network to search for candidate objects.
The authors tested the method on a meteorite fall in the US state of Nevada on July 14, 2019. It was tracked by NASA meteorite search stations. Their calculations show that the mass that could reach the ground was 35.3 ± 3.7 kilograms.
Engineers took the density of the meteorite at 3.2 grams per cubic centimeter and, using the wind model, calculated the area of the fall, the most likely points (small areas), and the approximate distribution of the mass of possible fragments over them.
They chose two points where the model predicted the possibility of fragments with a mass of about 10 and 100 grams, respectively. They sent drones into them, and at one of the points, they scattered all eight meteorites from their collection to test their recognition in a new area.
Join the discussion and participate in awesome giveaways in our mobile Telegram group. Join Curiosmos on Telegram Today. t.me/Curiosmos
• Citron, R. I. (n.d.). Recovery of Meteorites Using an Autonomous Drone and Machine Learning. PDF
• Citron, R. I., Jenniskens, P., Watkins, C., Sinha, S., Shah, A., Raissi, C., Devillepoix, H., & Albers, J. (2021, June 9). Recovery of meteorites using an autonomous drone and machine learning. Wiley Online Library.
• Gater, W. (2021, June 22). Meteorite-hunting drones could help find freshly fallen space rocks. New Scientist.