Title: Hubble asteroid hunter. I. Identification of asteroid trails in Hubble Space Telescope images
Authors: Sandor Kurk + 13 others
Institution of the first author: European Center for Space Research and Technology (ESA), Keplerlaan 1, 2201 AZ Noordwijk, The Netherlands
Status: Posted in A&A [open access]
Strong incentives for small targets
Although none of them were there to see it, astronomers are quite convinced that the early days of the solar system’s life were chaotic and violent. Dozens of newly formed asteroids, planetesimals, and a few true giant planets all buzzed around the sun in a tight disk: collisions were inevitable, though their consequences varied. Sometimes two clumsy objects merged together, and sometimes one or both broke into smaller pieces.
To understand how lawless this time in our history was, astronomers would like to perform forensic analysis on asteroids that have survived to the relatively quiet present; if they could measure the current ratio of small asteroids to larger ones, they could limit the frequency of destructive collisions in the past. This in turn would inform the models of where objects were and how fast they were moving in the early days around the sun.
Unfortunately, the most valuable asteroids for such study – the smallest remnants – are also the hardest to find. We can only see asteroids when they reflect some sunlight back to Earth, and small rocks just don’t reflect much light, making them very faint.
Enter the Hubble Space Telescope. Hubble is a very capable and busy space telescope that is able to see these dark asteroid remnants. However, although Hubble is able to image objects in the solar system, it spends most of its time looking much further afield, gazing longingly at distant galaxies, quasars and other targets at cosmic distances.
But, sometimes would-be asteroid hunters get lucky, and even when Hubble is trying to measure something else, a local space rock strays into the field of view. As the asteroid and Earth move around the sun, the photobombing asteroid appears as a curved trail in the image, a hairline fracture against the otherwise dark background of the universe.
Today’s authors aimed to extract as much information as possible from these happy coincidences and ambitiously sought to search the entire relevant Hubble Image Archive for chance sequences caused by small, secret asteroids.
Citizen Science + Deep Learning
Every photo taken by Hubble eventually becomes public, free to download for anyone who wants to see a corner of the universe. The archive of these images is huge, containing more than 37,000 images taken with the instruments and filters the authors deemed most likely to catch their targets. Database scale requires automation, and to address this need, the authors turned to deep learning, specifically Google’s Cloud AutoML Vision model. When fed with an image, this algorithm reports what is in the image (in this case an asteroid, while in others a dog for example). Although they do not detail the specifics of the architecture in this article, they do share that the model consists of several intertwined machine learning components: they use a convolutional neural network to actually find the arcs of asteroids in images, but this network itself was designed. by a reinforcement learning algorithm, an artificial intelligence paradigm that trains a computer to find an optimal solution through trial and error and feedback from its own actions.
Such a machine learning model must be trained, and training requires a catalog of known examples for the model to be studied. Since such a catalog did not yet exist, the authors had to build their own, and to do so they called on citizen scientists. They set up a project on Zooniverse called Hubble Asteroid Hunter, and for about a year more than 11,000 volunteers logged on to scour the data and search for asteroid arcs with the naked eye. Each volunteer was shown several pictures from Hubble, to the question “Is there an asteroid in this picture?” for each, then asked to ignore images without streaks and flag images containing the telltale curves. These volunteers collectively submitted more than 2 million yes/no responses to the query, and in total, this extensive effort uncovered asteroid contrails in approximately 1% of all images.
After processing all the carefully labeled images from the public and integrating them into their model, the authors released their code to the full dataset. How did that happen ? In the end, the algorithm achieved 73.6% accuracy (meaning 73.6% of its identifications were correct) and 58.2% recall (meaning it managed to retrieve 58 .2% of all asteroids found by volunteers). Although this may seem slightly below average, it was more than enough to make new scientific discoveries.
By combining the tracks found by the volunteers and those found by the model, the authors assembled a stack of 2,487 possible asteroid arcs. They then manually reviewed each of these candidates, and after removing duplicates and accounting for false positives caused by cosmic rays, gravitational lenses, or Earth satellites, they narrowed the list down to 1,701 reliable asteroid detections. .
After checking whether any of these contrails could be attributed to any of the more than 1.2 million known asteroids, the authors concluded that 670 of the contrails corresponded to previously discovered sources and that the remaining 1,031 were caused by asteroids never seen before. They also found that these newly discovered asteroids were consistently fainter than known bodies, which they expected: the brighter an asteroid, the more likely it has already been detected by a ground-based survey. . This general swoon also hinted that many of their new discoveries are exactly the type of small asteroid that we have struggled to count in other surveys.
The authors are also beginning to explore other properties of their sample of new asteroids, including their spatial distribution and variability in brightness. While they ignore the biases of Hubble’s preferential pointing and leave much of this deeper analysis for future work, their presentation of this new sample and demonstration of the power of merging citizen science and machine learning is an exciting step forward in the small business of asteroid accounting. The more confidently we can count small asteroids, the closer we can get to understanding the ancient history of our solar system: now, if more drift into our view, we’ll be ready for them.
Astrobite edited by Ryan Golant
Featured image credit: ESA/Hubble & NASA, M. Thévenot (@AstroMelina); DC BY 4.0
About Ben Cassese
I am a second year Ph.D. in astronomy. student at Columbia University working on simulated observations of exomoons. Before joining the Cool Worlds Lab, I studied planetary science and history at Caltech, and before that I grew up in Rhode Island. In my free time, I like to hike, put too much effort into making coffee, and dream of adopting a dog from my New York apartment.