Machine learning helps image a planet that Webb data couldn't show before

[Darren51]

Astronomers have found a smarter way to remove the glare of a star and reveal the planet beside it. The Register reports that a Northwestern University team led by Rodrigo Ferrer-Chavez has developed a method called FLIP, short for Physics based Learning for Imaging Planets

FLIP combines a physical model of the telescope with machine learning optimisation to subtract starlight. The idea is to understand exactly how the telescope spreads light, then remove that pattern more accurately than standard processing can

Tested on James Webb Space Telescope NIRCam data, it produced the first image of the giant planet 51 Eridani b at a wavelength of 4.3 micrometres. Conventional processing had failed at that wavelength

Sensitivity improved by between 2 and 15 times compared with standard methods, and the work has been published in Nature. The team says the computing demands still need to come down, and ground based telescopes would need far more data to use it

Using physics to guide machine learning is a strong approach, because it keeps the results grounded in how the instrument works. Getting more out of existing Webb data is a bonus for everyone. Imaging Earth like planets is still a goal for the future, but tools like this move it closer

How far could methods like FLIP take planet hunting? And which kinds of planets would you most like to see imaged directly?

Boffins use machine-learning tricks to create a starlight dimmer switchwww.theregister.com