asteroids in archives | stars

Title: Hubble asteroid hunter. 1. Determine asteroid tracks in Hubble Space Telescope images

Authors: Sandor Cork + 13 others

First Author Foundation: European Center for Space Research and Technology (ESA), Keplerlaan 1, 2201 AZ Noordwijk, The Netherlands

condition: Posted in A&A [open access]


Strong motivation for small goals

Although there is none to see, astronomers are quite confident that the early days of the solar system’s life were chaotic and violent. Dozens of newly formed asteroids, planets, and a few well-meaning giant planets circled the sun in an airtight disk: collisions were inevitable, though their repercussions varied. Sometimes two clumsy bodies fuse together, and sometimes one or both of them disintegrate into smaller pieces.

To get a sense of just how lawless this era of our history was, astronomers like to do forensic analysis of asteroids that have survived to the relatively quiet present; If they could measure the current ratio of smaller asteroids to larger ones, they could limit the prevalence of devastating collisions in the past. That, in turn, would tell us models of where things were and how fast they moved in the early days around the sun.

Unfortunately, the most valuable asteroid for such a study – the smallest remnants – are also the most difficult to find. We can only see asteroids when they reflect some sunlight back toward Earth, and small rocks don’t reflect much light, which makes them very faint.

Entry Hubble Space Telescope. Hubble is a very capable and very busy space telescope capable of seeing the dim asteroid remnants. However, although Hubble is able to image solar system objects, it spends most of its time gazing at much farther away, staring longingly at distant galaxies, quasarsand other goals in cosmic distances.

But, sometimes, would-be asteroid hunters get lucky, and even when Hubble tries to measure something else, a local space rock accidentally wanders into the field of view. As both the asteroid and the Earth move around the sun, the detonated asteroid appears as a curved line in the image, a hairline fracture in the dark background of the universe.

Figure 1: Example of an asteroid trajectory imaged by Hubble and its successful AutoML model retrieval. The large galaxy is HCG007. Source: Figure 4a in the paper.

Today’s authors sought to extract as much information as possible from these happy coincidences, and ambitiously sought to rummage through the entire archive of relevant Hubble images in search of latitudinal streaks from secret small asteroids.

Citizen Science + Deep Learning

Every image that Hubble eventually takes becomes public, and is free to download for anyone who wants to see some corner of the universe. The archive of these images is enormous, containing more than 37,000 images taken with tools and filters that the authors believe are most likely to hunt their targets. The size of the database necessitates automation, and to meet this need the authors resorted to deep learning, specifically Google’s Cloud AutoML Vision model. When feeding an image, this algorithm reports what is in the image (in this case, an asteroid, while in others, dog for example). Although they do not detail the architecture in this article, they do share that the model consists of several interlocking components of machine learning: they use convolutional neural network To actually find asteroid arcs in images, but that same network is modeled by a reinforcement learning algorithm, an artificial intelligence model that trains a computer to find the optimal solution by trial and error and feedback from its actions.

This machine learning model needs training, and training requires an index of known examples of the model to be studied. Since such a catalog did not exist yet, the authors had to create their own, and to do this they sought the help of citizen scientists. They set up a project on Animal World Call Hubble Asteroid HunterOver the course of about a year, more than 11,000 volunteers logged on to comb the data and search for asteroid arcs by eye. Each volunteer was shown several images from Hubble, and asked, “Is there an asteroid in this image?” For each photo, you were then asked to discard the photos without lines and mark the photos with telltale curves. These volunteers collectively provided more than 2 million yes/no answers to the query, and in total, this massive effort revealed asteroid streaks in about 1% of all images.

model performance

After all the images carefully categorized by the audience were processed and entered into their model, the authors unleashed their code on the full data set. How did you do? In the end, the algorithm achieved 73.6%. judgments (i.e. 73.6% of his identities were correct) and 58.2% Re-Call (This means that it succeeded in recovering 58.2% of all asteroids found by volunteers.) While this may seem below average, it was more than enough to make some new scientific discoveries.

By combining the tracks found by volunteers with those found by the model, the authors assembled a stack of 2,487 possible asteroid arcs. Then they manually reviewed each of these candidates, and after removing the duplicates and calculating the false positives they caused cosmic raysAnd the Gravity lensesor satellites associated with Earth, they shortened the list to 1,701 reliable asteroid detections.

Figure 2: The apparent size or perceived brightness distribution of asteroids in the Hubble images. In blue are objects the authors can trace back to previously recorded asteroids, while in orange are the authors’ new candidate discoveries. Note that their new asteroids are systematically fainter than previous discoveries due to the challenges of detecting fainter objects from Earth’s surface. Source: Figure 9a in the paper.

After checking if any of these lines can be attributed to any of the more than 1.2 million Known asteroids, the authors concluded that 670 of the lines are consistent with 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 systematically fainter than known objects, which they expected: the brighter an asteroid, the greater the chance that it would actually be detected by a ground survey. This general weakness also hinted that many of their new discoveries are exactly the kind of small asteroid we’ve struggled to count in other surveys.

The authors also began to explore other properties of their sample of new asteroids, including spatial distribution and brightness diversity. Although they do not take into account the preferential signaling biases of Hubble and leave much of this additional analysis for future work, their presentation of this new sample and the demonstration of the power of integrating citizen science and machine learning is an exciting step forward in Asteroid Business Accounting. The more confident we are in our ability to count small asteroids, the closer we can get to understanding the early history of our solar system: Now, if more drift is in our view, we’ll be prepared for it.

Astrobite Edited by Ryan Golant

Featured image credit: ESA/Hubble & NASA, M. Thévenot (AstroMelina); CC BY 4.0

About Ben Cassese

I have a PhD in Astronomy in my second year. A student at Columbia University working on a simulation of external meteorology. Before joining Cool Worlds Lab, she studied planetary science and history at Caltech, and before that she grew up in Rhode Island. In my spare time, I enjoy backpacking, putting in a lot of effort making coffee, and daydreaming about adopting a dog in my New York City apartment.

Leave a Comment