Scientists create a neural network that searches for ultrafine particles in space waste

European astronomers and mathematicians have found that computer vision algorithms can be used to analyze radar images and search for small, inconspicuous particles of space junk.

Scientists published the results of the study in the scientific journal IET Radar Sonar & Navigation. 

Frederica Massimi, a researcher at the Italian University of Rome-Tre, said that computer vision methods will not only allow in the future to monitor small-sized space waste, but will greatly increase the effectiveness of combating this danger. She added, "These methods will help us track extremely small objects that cannot be recorded in other ways."

Scientists reached this conclusion as part of an experiment in which they tried to take advantage of existing neural networks used in computer vision systems in order to analyze data collected by the European TIRA radar.

TIRA is a “radio dish” with a diameter of 47 meters that monitors near space in low Earth orbit and takes pictures that will later be used to search for space junk.

Scientists wondered whether the algorithms developed to analyze images captured by the TIRA radar could be replaced by neural networks from the YOLO family, which are often used to search for moving objects on photographs.

The scientists trained the YOLOv5 and YOLOv8 neural networks to analyze 3,000 images of near space, then verified their accuracy using 600 radar images that contained a limited amount of small-sized space debris.

The examination showed that both neural networks successfully identified between 85% and 97% of space waste particles with a length of 1 cm or more, and at the same time a small number of false positives were observed in their work. In this area, they far outperformed specialized algorithms designed to analyze data from the TIRA radar.

The researchers consider that these results of the experiment can be considered in the interest of using computer vision systems when mapping space waste in low Earth orbit, as well as to track its movements in real time (online). This will reduce the possibility of space devices colliding with space waste.