Euclid: Searches for strong gravitational lenses using convolutional neural nets in Early Release Observations of the Perseus field (2411.16808v1)

Abstract: The Euclid Wide Survey (EWS) is predicted to find approximately 170 000 galaxy-galaxy strong lenses from its lifetime observation of 14 000 deg^2 of the sky. Detecting this many lenses by visual inspection with professional astronomers and citizen scientists alone is infeasible. Machine learning algorithms, particularly convolutional neural networks (CNNs), have been used as an automated method of detecting strong lenses, and have proven fruitful in finding galaxy-galaxy strong lens candidates. We identify the major challenge to be the automatic detection of galaxy-galaxy strong lenses while simultaneously maintaining a low false positive rate. One aim of this research is to have a quantified starting point on the achieved purity and completeness with our current version of CNN-based detection pipelines for the VIS images of EWS. We select all sources with VIS IE < 23 mag from the Euclid Early Release Observation imaging of the Perseus field. We apply a range of CNN architectures to detect strong lenses in these cutouts. All our networks perform extremely well on simulated data sets and their respective validation sets. However, when applied to real Euclid imaging, the highest lens purity is just 11%. Among all our networks, the false positives are typically identifiable by human volunteers as, for example, spiral galaxies, multiple sources, and artefacts, implying that improvements are still possible, perhaps via a second, more interpretable lens selection filtering stage. There is currently no alternative to human classification of CNN-selected lens candidates. Given the expected 10^5 lensing systems in Euclid, this implies 10^6 objects for human classification, which while very large is not in principle intractable and not without precedent.

• The paper presents a CNN-driven method for automatically detecting strong gravitational lenses using Euclid’s early release data.
• It benchmarks the performance of various CNN architectures, revealing challenges with false positives and a maximum lens purity of about 11%.
• The study outlines pathways for refining network architectures and data pre-processing to enhance automated lens detection in large-scale surveys.

Overview of Machine Learning-Driven Gravitational Lens Detection in Euclid Early Release Observations

The field of cosmology and astrophysics continually seeks effective methods to probe the universe's dark matter and dark energy through various celestial phenomena. One such phenomenon, strong gravitational lensing, serves as a powerful tool for examining the distribution of mass in the universe, providing insights into cosmic scales influenced by dark matter and dark energy. The paper "Euclid: Searches for Strong Gravitational Lenses using Convolutional Neural Nets in Early Release Observations of the Perseus Field" explores the potential of leveraging convolutional neural networks (CNNs) for the identification of strong gravitational lenses within the observations made by the Euclid spacecraft.

Research Context and Objectives

The Euclid mission, undertaken by the European Space Agency (ESA), aims to map the universe's dark matter and dark energy by observing over 14,000 square degrees of the sky. The paper focuses on utilizing machine learning algorithms, specifically CNNs, to automate the detection of strong gravitational lenses, deemed necessary due to the infeasibility of manual identification given the expected large volume of lensing events.

Key aims outlined in the paper include:

• Developing CNN-based methods to automatically detect strong gravitational lenses while maintaining low false positive rates.
• Quantifying the precision and completeness of existing CNN-based detection pipelines for Euclid's VIS imaging data.
• Evaluating these methods on Euclid's Early Release Observations (ERO) in the Perseus field to establish a benchmark for future analysis during the mission.

Methodology and Results

In this paper, several CNN architectures were applied to data from the Euclid ERO of the Perseus field, focusing on sources with a VIS magnitude below 23. The networks demonstrated proficient performance on controlled datasets, showcasing the ability to recognize simulated lenses successfully. However, a discrepancy was noted when translating these results to actual Euclid data, where the highest achieved lens purity was approximately 11%, indicating room for improvement. The detections comprised false positives often characterized by spiral galaxies and other non-lens phenomena, suggesting potential improvements via additional lens selection filtering stages.

Strong numerical outcomes included:

• Projection of detecting about 170,000 strong gravitational lenses using Euclid’s survey data.
• Demonstration of CNNs' utility in classifying lens candidates, although currently dependent on supplementary human verification for final classification.

Implications and Future Directions

The result emphasizes the need for further refinement in network architectures and pre-processing techniques to enhance classification specificity and reduce false positives. This implies training using more diverse and representative datasets and potentially incorporating multi-band data from Euclid's NISP camera for better discrimination capabilities.

The practical implications are significant; the eventual goal is to process the expansive data generated by Euclid surveys with minimal human intervention, thereby accelerating the analysis of large-scale cosmic structures and improving the understanding of dark matter and dark energy distribution. Theoretically, this work underscores the potential of integrating advanced machine learning techniques into observational astrophysics, aligning with broader trends of automating large-scale data analysis in scientific research.

Conclusion

The paper offers insights into the challenges and potential pathways for improving the efficiency of gravitational lens detection using CNNs in astronomical surveys. It sets a foundational precedent for subsequent research and application in large cosmic surveys, paving the way for more reliable automated detection systems that can keep pace with the increasing data volumes expected from next-generation telescopes.