Boffins use machine learning tricks to create a starlight dimmer switch for Webb
Why starlight hides planets
Directly photographing planets beyond our solar system is challenging because the bright star they orbit can drown out the planets’ much fainter light. Existing image-processing methods estimate the starlight using models or reference images, but small optical changes can leave glare behind.
Those leftover signals can obscure a planet. The researchers’ approach is designed to account for imperfections in the telescope’s optics while estimating the light that needs to be removed.
How FLIP processes images
The team named its framework Physics-based Learning for Imaging Planets, or FLIP. It models the path light takes through a telescope, adjusts for optical imperfections and simulates the star’s glare for subtraction.
Researchers applied FLIP to data from Webb’s Near-Infrared Camera, known as NIRCam. In tests reported in a Nature paper, it improved sensitivity by between two and 15 times over conventional processing.
A planet emerges in Webb data
Using FLIP, the team created an image of 51 Eridani b, a known giant planet, at an infrared wavelength of 4.3 micrometers. Conventional processing had not revealed the planet at that wavelength.
The result demonstrates how modeling telescope behavior may help researchers detect faint planets in observations already collected. The team says it plans to reduce the framework’s computational demands and adapt it to the much larger volumes of ground-based telescope data.
Key points
- FLIP models telescope optics and subtracts simulated starlight from images.
- Tests found sensitivity gains of two to 15 times over conventional techniques.
- The framework imaged 51 Eridani b at 4.3 micrometers.
- Future work will target computing demands and large ground-based datasets.