AI Identifies More Than 1,300 Unusual Objects in Hubble Space Telescope Images

AI has analysed decades of images from the Hubble Space Telescope and found unusual celestial objects that had gone unnoticed by astronomers. ESA scientists employed a neural network called AnomalyMatch to search 100 million image cutouts from the Hubble Legacy Archive in approximately 2.5 days. They identified more than 1,300 unusual objects, of which about 800 had never been recorded before, including colliding galaxies, gravitational lenses, and jellyfish galaxies. This shows how AI can speed up new discoveries in large astronomical archives.

AI searches Hubble's archive

According to NASA, Hubble's massive data archive is simply too extensive for astronomers to search through by hand. To address this, ESA scientists David O'Ryan and Pablo Gómez created an AI model named AnomalyMatch, which was trained to identify unusual patterns in images. In only 2-3 days, it searched through about 100 million small 'cutouts' from the Hubble Legacy Archive – a process that would have taken astronomers many years to accomplish. This was the first systematic search of the entire Hubble archive for anomalies. After the AI identified promising candidates, scientists verified about 1,300 anomalies in the archive.

Rare cosmic objects discovered

'For example, a lot of the oddities we saw were galaxies merging or interacting with each other. Some were strange gravitational lenses, where one galaxy's gravity bends the light from another behind it. Then there were galaxies with massive areas where stars are forming or those with 'jellyfish' gas tails. Some of the edge-on disks that are forming planets even resembled hamburgers. What's pretty incredible is that several dozen anomalies just don't seem to fit into any known category.