Researchers Develop AI Capable of Detecting and Classifying Galaxies
Researchers at UC Santa Cruz have developed Morpheus, a computer program that is capable of analyzing the pixels in astronomical image data. It can then identify and classify all of the galaxies and stars that exist in large data sets that come from astronomy surveys.
What is Morpheus
Morpheus is a deep-learning framework that consists of various different artificial intelligence (AI) technologies. The AI technologies focus on certain applications like image and speech recognition.
Brant Robertson is a professor of astronomy and astrophysics. He is in charge of the Computational Astrophysics Research Group at UC Santa Cruz. According to Robertson, certain tasks that were traditionally done by astronomers need to be automated. This is because the sizes of astronomy data sets are constantly increasing.
“There are some things we simply cannot do as humans, so we have to find ways to use computers to deal with the huge amount of data that will be coming in over the next few years from large astronomical survey projects,” he said.
Ryan Hausen is a computer science graduate student at UCSC’s Baskin School of Engineering. He collaborated on Morpheus with Anderson over the past two years.
Their results were published on May 12 in the Astrophysical Journal Supplement Series. The Morpheus code will also be released to the public and there will be online demonstrations.
Morphologies of Galaxies
Astronomers are able to learn how galaxies form and evolve through time by observing the morphologies of galaxies.
There are some large-scale surveys that are set to take place which will generate massive amounts of image data that can be used. One of those surveys is the Legacy Survey of Space and Time (LSST), and it will be conducted at the Vera Rubin Observatory in Chile.
Robertson has been actively working on ways to use the data to better understand the formation and evolution of galaxies.
When the LSST is conducted, it will take over 800 panoramic images per night with a 3.2 billion pixel camera. Two times each week, the LSST will also record the entire visible sky.
“Imagine if you went to astronomers and asked them to classify billions of objects — how could they possibly do that? Now we'll be able to automatically classify those objects and use that information to learn about galaxy evolution,” Robertson said.
Deep-Learning Technology for Galaxies
Deep-learning technology has been used by some astronomers to classify galaxies, but it usually requires existing image recognition algorithms to be adapted. The algorithms are traditionally fed curated images of galaxies.
Morpheus was developed specifically for astronomical image data. It uses the original image data, which is in the standard digital format used by astronomers.
According to Robertson, one of the main points of Morpheus is pixel-level classification.
“With other models, you have to know something is there and feed the model an image, and it classifies the entire galaxy at once,” he said. “Morpheus discovers the galaxies for you, and does it pixel by pixel, so it can handle very complicated images, where you might have a spheroidal right next to a disk. For a disk with a central bulge, it classifies the bulge separately. So it's very powerful.”
The researchers utilized information from a 2015 study in order to train the deep-learning algorithm. The study collected data and classified around 10,000 galaxies in Hubble Space Telescope images from the CANDELS survey. Morpheus was then applied to image data from the Hubble Legacy Fields.
After processing an image of a part of the sky, Morpheus then generates a new set of images of that same area, and it color-codes all objects based on their morphology. Astronomical objects are separated from the background, and it identifies stars and different types of galaxies. The program runs on USCS’s lux supercomputer, where a pixel-by-pixel analysis for the entire data set is quickly generated.
“Morpheus provides detection and morphological classification of astronomical objects at a level of granularity that doesn't currently exist,” Hausen said.
The work completed by the researchers was supported by NASA and the National Science Foundation.
- Tech Leaders Highlighting the Risks of AI & the Urgency of Robust AI Regulation
- NVIDIA: From Chipmaker to Trillion-Dollar AI Powerhouse
- Laura Petrich, PhD Student in Robotics & Machine Learning – Interview Series
- Liquid Neural Networks: Definition, Applications, & Challenges
- Patrick M. Pilarski, Ph.D. Canada CIFAR AI Chair (Amii) – Interview Series