What does this research mean for the field?

Convolutional Neural Networks can accurately identify galaxy-galaxy strong gravitational lenses without false positives, successfully detecting faint lenses that are missed by human visual inspection. Novelty: ClaimNovelty.INCREMENTAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

The central goal is to evaluate how effective convolutional neural networks are in identifying galaxy-galaxy strong gravitational lenses compared to traditional methods.

May 19, 2026Open Access

Evaluation of the Effectiveness of Convolutional Neural Networks in Identifying Galaxy-Galaxy Strong Gravitational Lenses

Key Points

The central goal is to evaluate how effective convolutional neural networks are in identifying galaxy-galaxy strong gravitational lenses compared to traditional methods.
Sample objects included CNN-identified SGLs and visually inspected non-SGLs.
Fluxes were estimated using aperture photometry across grizy bands.
Lens light distribution was modelled and analysed to infer dark matter content and galaxy classification.
CNNs identified SGLs with no false positives and detected cases missed by astronomers.
The objects missed by astronomers were 3.5 to 6 times dimmer than those classified as SGLs.
Most lens galaxies identified were red, early-type, massive, and evolved.

Abstract

Observing strong gravitational lenses (SGLs) requires the almost precise alignment of a massive foreground object, typically a galaxy or cluster, with a distant (high redshift) background source and a telescope. Together with sufficient foreground object mass densities to significantly curve spacetime, multiple images, arcs, or rings of the source are observed. Machine learning, most commonly via Convolutional Neural Networks (CNNs), offers the most viable option in identifying an expected 105 SGLs among ~109 galaxies. CNNs provide fast object classifications and can be used on small datasets. The accuracy of CNNs is an ongoing area of research and this study evaluated their effectiveness in identifying galaxy-galaxy SGLs. Methods included sample objects from CNN identified and astronomer visual inspection confirmed SGLs, as well as visual inspection not SGLs. Fluxes were estimated via aperture photometry across grizy bands and modelled with Bagpipes to infer the lens galaxy’s stellar mass and photometric redshift. Lenstronomy modelled the lens light distribution and subtracted it from the observed data to isolate the source image, examined for SGL characteristics. Aperture and cosmology corrections and the Bagpipes derived stellar masses were used to analyse dark matter content and galaxy classification. Results demonstrated that CNNs identified SGLs with no false positives and successfully detected cases not recognized by astronomer visual inspection The faintness of visual inspection missed objects, 3.5 to 6 times dimmer than those confidently classified as SGLs, was possibly the main reason for differences between CNN and astronomer classifications. The lens colour-colour distribution indicated that most lens galaxies are populations of red, early-type, massive, evolved galaxies. False negatives were not investigated but research shows these occur when CNNs are trained on galaxies or galaxy clusters thus missing the other type. The rationales used by CNNs in identifying SGLs needs further investigation and is part of current research.

Ask AI

Helpful

Bookmark

View Full Paper