Space would look better if the Earth’s atmosphere wasn’t photobombing all the time.
Even the images obtained by the world’s best ground-based telescopes are blurry due to moving air pockets in the atmosphere. Seemingly harmless, this blur can obscure the shapes of objects in astronomical images, leading to error-ridden physical measurements that are essential for understanding the nature of the universe.
Now, researchers from Northwestern University and Beijing’s Tsinghua University have presented a new strategy to solve this problem. The team took a well-known computer vision algorithm used to sharpen photographs and applied it for the first time to astronomical images from ground-based telescopes. The researchers also trained an artificial intelligence (AI) algorithm on simulated data to match the Vera C. Rubin Observatory’s image parameters, so the tools will be immediately compatible when the observatory opens next year. .
Astrophysicists are already using the technology to remove blur, but the adapted AI-driven algorithms will work faster than current technology and produce more realistic images. The resulting image is less blurry and more true to reality. They’re also beautiful, but that’s not what technology is for.
“The purpose of photography is often to get a clean, good-looking image,” says Emma Alexander of Northwestern University, senior author of the study. “But images of celestial bodies are used in science. By cleaning up the images in the right way, we can get more accurate data. It allows us to take scientific measurements, after all, the images look better, too.”
The research will appear in the monthly notice of the Royal Astronomical Society on March 30.
“It’s like looking up from the bottom of a pool. Water pushes light away and distorts it. The atmosphere is much less dense, of course, but the concept is similar.”
— Emma Alexander, Computer Scientist
Alexander is an Assistant Professor of Computer Science at the McCormick School of Engineering at Northwestern University and runs the Bio-Inspired Vision Lab. She co-led her new research with her Tianao Li, an electrical engineering undergraduate at Tsinghua University and a research intern in Alexander’s lab.
When light emanates from distant stars, planets, and galaxies, it passes through the Earth’s atmosphere before it hits our eyes. Our atmosphere not only blocks certain wavelengths of light, it also distorts the light that reaches Earth. Even a clear night sky contains moving air that affects the light that passes through it. That’s why stars twinkle, and why the best ground-based telescopes are located at high altitudes where the atmosphere is thinnest.
“It’s like looking up from the bottom of a pool,” said Alexander. “Water pushes light away and distorts it. Of course, the atmosphere is much less dense, but the concept is similar.”
This blurriness becomes a problem when astrophysicists analyze images to extract cosmological data. By studying the apparent shape of galaxies, scientists can detect the gravitational effects of large-scale cosmic structures. This structure bends light to reach Earth. This can cause elliptical galaxies to appear more rounded or elongated than they actually are. But atmospheric blurring smudges the image in ways that distort the shape of the galaxy. Removing the blur allows scientists to collect accurate shape data.
“Small differences in geometry tell us about gravity in space,” said Alexander. “It’s already difficult to detect these differences. When looking at images from ground-based telescopes, the shapes can be distorted. Whether it’s due to gravitational effects or the atmosphere.” It is difficult to know.”
This tool has 38.6% less error than traditional methods for removing blur
To tackle this challenge, Alexander and Li combined an optimization algorithm with a deep learning network trained on celestial imagery. Among the training images, the team included simulated data that match the expected image parameters of the Rubin Observatory. The resulting tool produced images with 38.6% fewer errors compared to traditional methods for deblurring, and 7.4% fewer errors compared to modern methods.
When the Rubin Observatory officially opens next year, the telescope will begin a decade-long survey of the vast swaths of the night sky. The researchers trained a new tool on data specifically designed to simulate Rubin’s upcoming images, thus helping to analyze the survey’s highly anticipated data.
For astronomers interested in using this tool, the open-source, easy-to-use code and accompanying tutorial are available online.
“Now we are putting this tool into the hands of astronomy experts,” Alexander said. “We believe this will be a valuable resource for aerial surveys to obtain the most realistic data possible.”
The study, “Galaxy Image Deconvolution for Weak Gravitational Lensing Using an Deployed Plug-and-Play ADMM,” used computational resources from the Computational Photography Laboratory at Northwestern University.