Boffins use machine-learning tricks to create a starlight dimmer switch

Scientists have developed a way to remove starlight from telescope images that improved sensitivity to faint exoplanets by up to 15 times in tests on James Webb Space Telescope data.

A team led by Northwestern University's Rodrigo Ferrer-Chavez combined a physics-based telescope model with optimization techniques used in machine learning to subtract unwanted starlight.

Directly imaging exoplanets remains difficult because their host stars' glare can overwhelm their faint light. Existing techniques subtract modeled or reference starlight, but tiny changes in telescope optics can leave residual glare that hides planets.

The team's Physics-based Learning for Imaging Planets (FLIP) framework models how light passes through the telescope, adjusts for optical imperfections, and simulates the glare to be subtracted.

Applied to observations from Webb's Near-Infrared Camera (NIRCam), FLIP produced the first image of the known giant planet 51 Eridani b at an infrared wavelength of 4.3 micrometers, where conventional processing had failed. …