Enhancing Astronomical Imaging with AI
How AI Can Help Humanity Better Understand the Universe
The field of astronomical imaging or astrophotography is an ever-growing field. The universe holds an infinite number of mysteries and marvels that are yet to be discovered. Astronomical imaging has evolved in major ways in recent years with the development of modern satellites and telescopes such as the JWST, the Planck Telescope, the WMAP that are able to take stunning visuals of celestial bodies in our universe.
History of Astronomical Imaging
For millennia, ancients used to ponder on the depths of the universe, sketching images of the moon, the sun and the planets visible to the naked eye. Unfortunately, as astronomical observation tools were not invented yet, the observations made by the Babylonians, Chinese, Greeks, Indians, etc. were limited to the human eye’s capability as well as some rudimentary observational tools such as the sundial, which was used to tell time for ancients but also to mark the position of the Sun in the sky.
In the Medieval period, astronomical imaging was also mostly relied on the naked eye and the previously made observations by ancient civilisations such as the Greeks, Romans, the Islamic Civilisation, etc. They translated the work made by great astronomers like Aristotle and Ptolemy. One of the most notable figures in medieval astronomy is Alhazen (Ibn al-Haytham), whose work on optics laid the foundation for developments in telescopic imaging. His book, “Book of Optics,” was influential in understanding how lenses could be used to magnify objects.
Then, in the 16th-17th century, the world was revolutionised by the invention of the telescope. It is unclear who exactly invented the first telescope; however, the credit mostly revolves around two figures: Hans Lippershey (the first person to build a telescope and file a patent for it) and Zacharias Janssen (credited to be the first person to build an optical telescope). Regardless, the invention of the telescope was a significant contribution to astronomical imaging. Galileo Galilei, a polymath, enhanced the magnifying quality of the telescope and became the first person to study the skies with it. He made detailed observations and sketches of the moon, the Phases of Venus, Jupiter’s moons, and much more. Marking a major leap in astronomical imaging. He disproved Copernicus’s Heliocentric Model and was unfortunately put in house-arrest by the Catholic Church for his contribution (the Catholic Church believed Copernicus’s Heliocentric Model that the Earth was the centre of the Solar System and the planets and the Sun revolved around it).
Another pivotal figure in astronomical imaging was Johannes Kepler, who published a book named ‘Astronomia Nova’ (translating to ‘New Astronomy’), which contained detailed information and sketches of the moon.
During 17th and 18th centuries, telescopic advancements continued to improve. Isaac Newton’s development of the reflecting telescope was pivotal in the development of high-accuracy images. This innovation allowed for clearer and more accurate images of distant celestial bodies.
Observatories continued to be built in the 17th, 18th, and 19th centuries, with the Paris Observatory in 1667, the Greenwich Observatory in 1675, the Berlin Observatory in 1711, the Radcliffe Observatory in 1773, and many more.
The first ever successful photograph of space was taken by John William Draper on March 23rd 1840, marking the birth of astrophotography. The photograph was a daguerreotype image or the moon, which required Draper to focus his telescope on the moon for 20 minutes.
10 years later, in 1850, John Adams Whipple and William Cranch Bond captured the first-ever photograph of a star (Vega) other than the Sun and was captured at Harvard College Observatory.
In mid-1851, the first-ever image of a solar eclipse was captured by a daguerreotypist named Johann Julius Friedrich Berkowski in Koenigsberg (currently known as Kaliningrad), Russia. To capture this image, Berkowski required to expose the daguerreotype plate for 84 seconds in the focus of the telescope from the totality of the solar eclipse, capturing the first-ever daguerreotype image of a solar eclipse.
In 1857, Warren De La Rue, a pioneer in astrophotography captured detailed images of the moon using a photography process called the collodion process. Using this process, De La Rue also photographed detailed and accurate images of the Sun’s Corona during a Solar Eclipse.
In 1889, the first-ever image of the Milky Way Galaxy was captured by Edward Emerson Barnard and revealed details in the Milky Way such as its structure and star clouds.
Images and daguerreotypes like this continued to be captured for many years, from Draper’s photograph of the Orion Nebula to the M31 Andromeda Galaxy by Isaac Roberts.
However, on October 24, 1946, a long-awaited dream was achieved as scientists at White Sands Missile Range launched a missile with a small motion picture camera into 104 kilometers into the sky (4 kilometers above what is considered outer space), where the camera captured the first-ever picture in space.
In the 20th century, satellites were newly invented. The first-ever satellite launched was the Sputnik 1 on 4 October 1957. However, the Sputnik did not take any pictures of Earth while it orbited. The first satellite to take a picture was U.S. Explorer 6 on August 14, 1959. The US Explorer 6 captured an image of the Earth while it was in orbit, 27,000 km above the ground.
Satellite Imagery continued to advance throughout the 20th and 21st centuries. New satellites were launched such as the Wilkinson Microwave Astronomy Probe, the James Webb Space Telescope, the Planck Telescope, etc. The development of the Global Positioning System (GPS), which relies on satellite imagery, has provided a useful navigation tool for humanity. Modern Space Telescopes, probes, and satellites have developed imaging for light that is not visible to the human eye such as infrared light, microwaves, etc.
AI in Astronomical Imaging
In the modern world, where AI serves as a promising assistant to humanity, AI has been used in many cases in Astronomical Imaging.
Enhancement
AI algorithms can effectively reduce noise, disruptions and disturbances in astronomical images. Deep Learning Techniques such as Denoising Convolutional Neural Networks (DnCNNs) are specifically trained for noise reduction in images. The DnCNNs are trained on two sets of data — noisy data and clean data, they then identify patterns in the clean data and noisy data. Once noisy images are input into them, they apply the clean data patterns to the noisy data and output the clearer, more accurate images.
Denoising Autoencoders (DAEs) are models in machine learning that are designed to reconstruct clean data from noisy inputs. If distorted images and clean images are input as their training set, they analyse the patterns, similarities, and differences between these images. They can enhance a noisy image by applying the patterns in clean images to the distorted images and create an accurate one. VAEs are primarily built on a component called an Autoencoder. An autoencoder has two separate components itself; an encoder and a decoder. The encoder compresses the input into a lower dimensional latent space (a simplified and compact representation of the input data with less dimensions)and a decoder that reconstructs this information to identify the most important features of the input.
DAEs are primarily built on a component called an autoencoder, which in turn is composed of two other components- an encoder and a decoder. An encoder takes in the input given and converts it into lower dimensional latent space (a simplified and compact representation of the input data with less dimensions), the decoder reconstructs the information and preserves the most important features while removing the noise and distortions.
Object Detection
Because AI can detect patterns in images and data, it excels in object detection. Just recently, scientists from the Asteroid Institute trained an AI named THOR to identify asteroids in images. The team then provided THOR with 400,000 images from the institute’s archive and surprisingly, the AI discovered 27,500 previously unknown asteroids!
YOLO (You Only Look Once) models are object-detection systems that can identify and classify objects. The YOLO algorithm divides the input image into a grid of cells, and for each cell, it predicts the probability of the presence of an object. It also predicts the class of the object. Unlike two-stage object detectors , YOLO processes the entire image in one pass, making it faster and more efficient.
Via object detection and pattern recognition, AI can quickly detect distant cosmic events such as supernovae and gamma-ray bursts.
Data Processing & Storage
The amount of data in even one astronomical image can be massive. With stars, galaxies, and many more celestial bodies in the backgrounds or in the focus, astronomical images tend to be full of data and require time for analysis and processing. AI can help with this. Using pattern recognition algorithms, AI can identify specific traits in large amounts of data. Aside from this, AI can process and analyse this data in a considerably low amount of time, prompting quick, efficient, and easy data processing.
Challenges with AI in Astronomical Imaging
One of the main challenges that AI faces in astronomical imaging is the process of training the AI itself. Some astronomical events like supernovae or gamma ray bursts are rare and recorded data of them are not very common, because of this, if an AI identifies any of these uncommon astronomical events, it might lead to an imbalance with the data and creating labelled datasets for AI training requires significant human effort and expertise.
As mentioned before, AI models can process large amounts of data, however, to process the data, AI models need computational resources such as large data storage and memory and computational power and capacity, especially deep learning models, which tend to be computationally intensive.
Conclusion & The Future for AI in Astronomical Imaging
As AI continues to evolve, it will not only enhance our perception and understanding of astronomical imaging, but also provide a path to the future for other fields. The use of AI into astronomical imaging represents a significant leap forward, and as Neil Armstrong once said — “One small step for a man, one giant leap for mankind.”
