Webb Telescope Captured It but Nobody Saw It: Scientists Discover an Overlooked Giant Planet in Old Data, New Algorithm Boosts Detection Sensitivity 15-Fold
IT之家 November 21 news — A research team at Northwestern University has developed a new exoplanet imaging processing technique that has successfully identified, from data previously captured by the James Webb Space Telescope (JWST), a giant planet that had not been observable through conventional image processing methods.
The researchers say the technique can boost the telescope's detection sensitivity for faint planets by 2 to 15 times. The findings were published on November 4 in Nature.

The planet, named 51 Eridani b, had already been discovered by ground-based telescopes, but due to interference from the intense glare of its host star, astronomers had been unable to pick it out from a set of JWST infrared observation data.
A major challenge in direct imaging of exoplanets is the enormous brightness difference between a star and its planet. Even a relatively bright young giant planet may be about 100,000 times fainter than its host star. After starlight passes through the telescope's optical system, it forms a bright halo in the image that masks the faint light signal of nearby planets.
Astronomers have long used coronagraphs to block starlight so they can observe surrounding planets. However, some starlight still leaks around the instrument, forming a residual halo.
For about the past 20 years, researchers have typically obtained a reference image by observing another star, then compared and subtracted it from the observation image of the target star to remove unwanted starlight. But telescope mirrors and other optical components cannot remain perfectly stable at all times; tiny displacements, tilts, and temperature changes can alter the light distribution, causing differences between the reference image and the actual observation and leaving residual light sufficient to mask a planet.
To address this, a research team led by Northwestern doctoral student Rodrigo Ferrer-Chavez developed a computational method called Physics-based Learning for Imaging Planets (FLIP). IT之家 notes that Jason J. Wang, an assistant professor of physics and astronomy at Northwestern, is also a co-author of the paper.
FLIP no longer relies entirely on actual photographs of reference stars. Instead, it builds a digital twin model of the telescope and reconstructs the starlight halo in the image by simulating the process of starlight passing through the telescope's mirrors and optical components and reaching the detector.

The researchers used computational methods such as automatic differentiation and gradient optimization to tune millions of parameters in the model so that the simulation results match the telescope's real observations as closely as possible. The model also accounts for optical distortions, tiny instrument displacements, and other imperfections, thereby reconstructing starlight more accurately and subtracting it from the raw image.
The research team particularly emphasizes that although FLIP uses computational optimization tools similar to those used to train artificial intelligence models, it does not itself use artificial intelligence to identify planets; instead, it builds its model based on the known physics of the telescope.
Wang said researchers already understand the physical process by which light travels through the telescope, so they do not need artificial intelligence to understand what is happening in the image.
The team tested FLIP using JWST data previously captured of three stars: HIP 65426, which has a known planet; 51 Eridani; and TYC 5899, which had no previously known companion planet.
Test results showed that FLIP outperformed conventional image processing methods across different observation datasets, improving detection sensitivity by about 2 to 15 times. The most notable result was the successful identification of the giant planet 51 Eridani b from JWST infrared observations of the 51 Eridani system.
The planet had previously been discovered by researchers using ground-based telescopes, but JWST observations at a wavelength of 4.3 micrometers had consistently failed to identify it through conventional processing methods. This wavelength band is especially important for studying carbon dioxide in planetary atmospheres, because carbon dioxide strongly absorbs infrared light at the corresponding wavelength, further weakening the planet's already faint signal.
After reprocessing the same set of observation data with FLIP, the researchers successfully isolated the light signal of 51 Eridani b. Paper data show that in the F430M filter observation, where conventional methods failed to detect the planet, the new method achieved a signal-to-noise ratio of about 6.3, reaching a significant detection level.
Wang said this result shows that some past telescope observation data may already contain valuable planetary information; what was missing was a sufficiently effective tool to extract it.
The technique could also influence future space telescope missions. NASA is advancing the planned Habitable Worlds Observatory (HWO), one of whose goals is to directly image Earth-like planets orbiting Sun-like stars.
Because Earth-like planets are far fainter than giant planets, future telescopes may need to keep the stability of their optical systems at the picometer level, i.e., one trillionth of a meter. Such requirements could mean the telescope would need to be about 100 to 1,000 times more stable than the Webb telescope.
The research team believes that physics-model-based image processing techniques like FLIP may in the future be able to compensate for some of the telescope's tiny instabilities, reducing reliance on extreme hardware stability. However, the method still needs further optimization of computational cost and validation on more telescopes and observation datasets.
Ferrer-Chavez, the paper's first author, said humanity has not yet found planetary systems around other stars similar to those in the Solar System, but that does not mean they do not exist — it may simply be that current observation techniques cannot yet identify them. As detection technology continues to improve, astronomers may discover more types of planets and further study the formation and evolution of planetary systems.
Related Articles
- HMD 2320 Flip 4G feature phone exposed: Unisoc T127 chip, 1.77-inch external screen
- 6499 → 2559 yuan: Xiaomi MIX Flip 2 small foldable phone 512G hits new low on JD.com after national subsidy
- Launched at 5999 yuan: Xiaomi MIX Flip 2 small foldable phone hits new low of 2271 yuan after national subsidy
- 5999 → 2339 yuan: Xiaomi MIX Flip 2 small foldable phone hits new low, Snapdragon 8 Elite + dual VC 3D cooling
- HMD Nokia 2360 Flip clamshell feature phone exposed: Unisoc T127 chip, 1.8-inch external screen
- Launched at 5999 yuan: Xiaomi MIX Flip 2 small foldable phone hits new low of 2378 yuan after national subsidy