HLA-Net: A Physics-Guided Skeleton-Based Framework for Automatic Gravitational Arc Detection in Astronomical Images

arXiv:2611.05438 · astro-ph.IM, astro-ph.GA, cs.LG · Submitted 2026-11-03 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Astrophysics. The day in astrophysics: observations, models and what the data forces.

Vera: I'm Vera.

Jocelyn: And I'm Jocelyn, here with Subrahmanyan.

Vera: We're looking at the paper: "HLA-Net: A Physics-Guided Skeleton-Based Framework for Automatic Gravitational Arc Detection in Astronomical Images".

Jocelyn: Here's the gist of the paper. Gravitational arcs produced by strong gravitational lensing provide a unique approach for investigating dark matter distribution, galaxy cluster mass reconstruction and the structure of the Universe The proposed framework is an interpretable multi-stage framework for automatic gravitational arc detection that explicitly exploits both the geometric morphology and physical characteristics of gravitational lensing without requiring manually annotated training datasets Framework Overview The proposed Hybrid Lens Arc Network (HLA-Net) is a modular physics-guided framework developed for automatic gravitational arc detection in astronomical FITS images The framework combines classical image processing, geometric reasoning and astrophysical constraints within a unified detection pipeline It progressively transforms raw astronomical observations into a ranked list of gravitational lens candidates through successive verification stages Image Preprocessing and Candidate Generation The process begins with FITS image preprocessing to suppress background noise while preserving faint elongated structures that may correspond to gravitational lensing arcs This involves intensity normalization, Gaussian smoothing, and adaptive histogram equalization Following this, candidate extraction is achieved through skeletonization and graph-based skeleton ordering to preserve geometric continuity Geometric Verification Stages Multiple geometric verification stages evaluate tortuosity, turning angle, curvature consistency and robust circle fitting to reject non-lensing objects These stages include computing geometric descriptors such as arc length, width, eccentricity, curvature, compactness and circularity Tangential orientation analysis evaluates the angular relationship between each skeleton and the estimated galaxy cluster centre Physics-Guided Ranking Strategy The final stage integrates geometry score, tangential alignment, radial consistency and curvature stability into a unified confidence measure This weighted scoring strategy balances multiple complementary measurements rather than relying on any individual feature The results demonstrate that HLA-Net effectively suppresses false-positive detections while preserving physically reasonable lens candidates through an interpretable and modular verification strategy Computational Efficiency and Interpretability The computational complexity of the proposed HLA-Net framework is approximated as O(N) + O(X C i=1 P i!) + O(C log C) This approach is computationally efficient because it operates through deterministic image processing and geometric analysis rather than iterative network training Furthermore, the framework offers complete interpretability because each intermediate processing stage produces measurable quantities that can be directly analyzed by astronomers The modular architecture allows for rapid candidate screening in large-scale sky surveys Limitations and Future Directions Limitations include the assumption that the approximate galaxy cluster centre is known prior to tangential verification Extremely faint arcs may be partially fragmented during binary segmentation, affecting the computed confidence score Highly irregular foreground galaxies occasionally satisfy several geometric constraints, producing isolated false-positive detections despite the progressive verification strategy Future research will focus on evaluating HLA-Net using larger multi-band astronomical datasets and multiple galaxy clusters to further assess its generalization capability References

1: A.

Vera: Einstein.

Jocelyn: Lens-like action of a star by the deviation of light in the gravitational field

2: F.

Vera: Zwicky.

Title and authors: Vera: Let's talk about the title and authors of this paper, "HLA-Net: A Physics-Guided Skeleton-Based Framework for Automatic Gravitational Arc Detection in Astronomical Images." The core idea here is that they’re using a hybrid approach combining skeleton representations with physics guidance to automatically find gravitational arcs >

Jocelyn: And the authors are from Amrita Vishwa Vidyapeetham Kerala, India. They seem to have a strong background in computational astrophysics and image analysis, which fits the kind of complex framework they've proposed >

Subrahmanyan: For me, I look at this title and I see a clear move toward making detection robust by anchoring it in established physical principles rather than just chasing high accuracy on training sets that might not generalize well to new data >

Vera: Right. The authors state their goal is to build something interpretable that can work without needing those manually annotated training datasets, which is a big deal because most advanced methods rely heavily on labeled examples >

Jocelyn: It’s about creating a system where you can see *why* the network picked a candidate, not just that it picked one, so they emphasize the interpretability of this multi-stage pipeline >

The paper's summary: Vera: The abstract of this paper explains that gravitational arcs are rare and often faint, which makes automatic detection very difficult because they can easily be mistaken for other things like edge-on galaxies or diffraction spikes >

Jocelyn: They lay out the problem clearly—traditional methods like thresholding or Hough transforms are sensitive to noise and produce too many false detections, while deep learning models exist but often lack the necessary physical grounding >

Subrahmanyan: The summary highlights that this paper proposes HLA-Net as an interpretable multi-stage framework that explicitly uses both geometric morphology and physical characteristics to overcome these detection challenges >

Vera: It goes on to say they started with a certain number of initial arc candidates extracted from ordered skeleton representations, and through the pipeline, they reduced the search space down to eighty-three geometrically consistent candidates for an Abell one thousand six hundred eighty-nine cluster example >

Jocelyn: That reduction from hundreds of potential starting points down to eighty-three consistent ones shows that this progressive verification process is quite effective at culling the noise early on >

The paper's improvements: Vera: The main improvement they highlight is this physics guided ranking strategy where they don't just look at one feature like arc length or curvature by itself >

Jocelyn: Instead, they combine a unified confidence measure that weighs geometric scores, tangential alignment, radial consistency, and curvature stability together instead of relying on any single measurement >

Subrahmanyan: That integration is the crucial part because it acknowledges that no single geometric descriptor is perfect on its own; it has to be the whole physical picture that confirms a candidate >

Vera: They demonstrate that this weighted scoring strategy successfully suppresses false positive detections while still keeping candidates that are physically reasonable, which is what they call preserving physically reasonable lens candidates >

Jocelyn: It’s about moving beyond just finding *something* elongated and moving toward finding something that actually behaves like a gravitational arc in the context of the cluster structure >

Conclusion: Vera: So, to wrap up, this paper on "HLA-Net: A Physics-Guided Skeleton-Based Framework for Automatic Gravitational Arc Detection in Astronomical Images" shows how you can use a structured pipeline to automatically find these rare and faint arcs without needing massive manual labels >

Jocelyn: The main point is that by combining geometry with physical constraints like tangential alignment and curvature, you get a much more reliable way to rank candidates than relying on just one visual feature >

Subrahmanyan: For the bigger picture, this approach suggests that for large-scale surveys, we need tools that are not just fast but also inherently physically informed so we can better reconstruct the dark matter distribution and cluster mass structures across the universe >

Vera: It’s a very concrete method because every stage of the process produces measurable quantities that astronomers can actually analyze directly, which is a big step toward real scientific discovery >

Jocelyn: We'll keep an eye on how this framework performs when applied to larger, multi-band datasets in the future to really test its robustness beyond what they did with Abell one thousand six hundred eighty-nine > <ref:2611.05438#pg3>

Don S.,

School of Computing Amrita Vishwa Vidyapeetham

astro-ph.IM, astro-ph.GA, cs.LG

Submitted: 2026-11-03

Updated: 2026-11-03

Comments: 16 pages, 2 figures, planning to submit: The Open Journal of Astrophysics

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 83/100

The gist: Gravitational arcs produced by strong gravitational lensing provide a unique approach for investigating dark matter distribution…

Key concepts

Hybrid Lens Arc Network (HLA-Net)
A physics-guided framework that automatically detects gravitational arcs in astronomical images. It integrates classical image processing with geometric reasoning and astrophysical constraints to create a unified detection pipeline. It works by progressively transforming raw data into a ranked list of lens candidates.
Skeletonization and Graph-Based Ordering
These are initial steps in image preprocessing used to extract the geometric structure of potential arcs. Skeletonization reduces the image to its essential centerline, while graph-based ordering helps preserve the continuity and shape of these elongated structures, which is crucial for identifying arc-like features.
Physics-Guided Ranking Strategy
The final stage where a unified confidence score is calculated by weighting multiple complementary measurements. This score combines geometric scores, tangential alignment, radial consistency, and curvature stability. This holistic approach ensures that only candidates exhibiting consistent physical behavior across all verification modules receive the highest confidence.
Interpretability and Transparency
The framework provides transparent intermediate outputs at every stage of the detection process. Each module produces measurable quantities that astronomers can directly analyze. This contrasts with black-box deep learning methods, making it easy to understand why a specific candidate was selected or rejected.

Terminology

Summary

Gravitational arcs produced by strong gravitational lensing provide a unique approach for investigating dark matter distribution, galaxy cluster mass reconstruction and the structure of the Universe The proposed framework is an interpretable multi-stage framework for automatic gravitational arc detection that explicitly exploits both the geometric morphology and physical characteristics of gravitational lensing without requiring manually annotated training datasets

Framework Overview

The proposed Hybrid Lens Arc Network (HLA-Net) is a modular physics-guided framework developed for automatic gravitational arc detection in astronomical FITS images The framework combines classical image processing, geometric reasoning and astrophysical constraints within a unified detection pipeline It progressively transforms raw astronomical observations into a ranked list of gravitational lens candidates through successive verification stages

Image Preprocessing and Candidate Generation

The process begins with FITS image preprocessing to suppress background noise while preserving faint elongated structures that may correspond to gravitational lensing arcs This involves intensity normalization, Gaussian smoothing, and adaptive histogram equalization Following this, candidate extraction is achieved through skeletonization and graph-based skeleton ordering to preserve geometric continuity

Geometric Verification Stages

Multiple geometric verification stages evaluate tortuosity, turning angle, curvature consistency and robust circle fitting to reject non-lensing objects These stages include computing geometric descriptors such as arc length, width, eccentricity, curvature, compactness and circularity Tangential orientation analysis evaluates the angular relationship between each skeleton and the estimated galaxy cluster centre

Physics-Guided Ranking Strategy

The final stage integrates geometry score, tangential alignment, radial consistency and curvature stability into a unified confidence measure This weighted scoring strategy balances multiple complementary measurements rather than relying on any individual feature The results demonstrate that HLA-Net effectively suppresses false-positive detections while preserving physically reasonable lens candidates through an interpretable and modular verification strategy

Computational Efficiency and Interpretability

The computational complexity of the proposed HLA-Net framework is approximated as O(N) + O(X C i=1 P i!) + O(C log C) This approach is computationally efficient because it operates through deterministic image processing and geometric analysis rather than iterative network training Furthermore, the framework offers complete interpretability because each intermediate processing stage produces measurable quantities that can be directly analyzed by astronomers The modular architecture allows for rapid candidate screening in large-scale sky surveys

Limitations and Future Directions

Limitations include the assumption that the approximate galaxy cluster centre is known prior to tangential verification Extremely faint arcs may be partially fragmented during binary segmentation, affecting the computed confidence score Highly irregular foreground galaxies occasionally satisfy several geometric constraints, producing isolated false-positive detections despite the progressive verification strategy Future research will focus on evaluating HLA-Net using larger multi-band astronomical datasets and multiple galaxy clusters to further assess its generalization capability

References

[1] A. Einstein. Lens-like action of a star by the deviation of light in the gravitational field

[2] F. Zwicky. On the probability of detecting nebulae which act as gravitational lenses

[3] M. Bartelmann. p. 2

Where the paper says it

However, the method remained sensitive to crowded fields where edge on galaxies and diffraction artifacts often produced elongated structures similar to genuine gravitational arcs. Seidel and Bartelmann [9] further investigated curvature based arc detection using geometric measurements derived from connected astronomical objects.

Gravitational lensing

[4] J. P. Gardner, J. C. Mather, M. Clampin, R. Doyon, et al., The james webb space telescope

[5] T. Broadhurst et al., Strong-lensing analysis of a1689 from deep advanced camera images

[6] Euclid Collaboration, Euclid preparation: Vii. forecast validation for euclid cosmological probes

[7] Ž. Ivezić et al., Lsst: From science drivers to reference design and anticipated data products

[8] F. Lenzen and H. p. 2

Where the paper says it

Since gravitational arcs generally appear as elongated, curved, thin structures early researchers focused on morphological measurements, curvature estimation, connected component analysis and adaptive filtering techniques to separate arc-like objects from ordinary galaxies and background noise. Lenzen et al.

Schindler, A new method for the detection of gravitational arcs

[9] G. Seidel and M. p. 2

Where the paper says it

However, the method remained sensitive to crowded fields where edge on galaxies and diffraction artifacts often produced elongated structures similar to genuine gravitational arcs. Seidel and Bartelmann [9] further investigated curvature based arc detection using geometric measurements derived from connected astronomical objects.

Bartelmann, Arc identification based on local elongation

[10] Colin Jacobs, Thomas Collett, Karl Glazebrook, Christopher McCarthy, A. K. Qin, T. M. C., Abbott et al., Finding high-redshift strong lenses in des using convolutional neural networks

[11] François Lanusse et al., Cmu deeplens: Deep learning for automatic image-based galaxy-galaxy strong lens finding

[12] C. Alard, Automated detection of gravitational arcs

[13] P. J. Marshall et al. p. 3

Where the paper says it

These features were subsequently classified using Support Vector Machines (SVM), Random Forests, Decision Trees, Artificial Neural Networks and ensemble learning techniques. Marshall et al.

p. 3
Where the paper says it

These features were subsequently classified using Support Vector Machines (SVM), Random Forests, Decision Trees, Artificial Neural Networks and ensemble learning techniques. Marshall et al.

p. 3
Where the paper says it

These features were subsequently classified using Support Vector Machines (SVM), Random Forests, Decision Trees, Artificial Neural Networks and ensemble learning techniques. Marshall et al.

p. 3
Where the paper says it

These features were subsequently classified using Support Vector Machines (SVM), Random Forests, Decision Trees, Artificial Neural Networks and ensemble learning techniques. Marshall et al.

p. 3
Where the paper says it

These features were subsequently classified using Support Vector Machines (SVM), Random Forests, Decision Trees, Artificial Neural Networks and ensemble learning techniques. Marshall et al.

p. 3
Where the paper says it

These features were subsequently classified using Support Vector Machines (SVM), Random Forests, Decision Trees, Artificial Neural Networks and ensemble learning techniques. Marshall et al.

p. 3
Where the paper says it

These features were subsequently classified using Support Vector Machines (SVM), Random Forests, Decision Trees, Artificial Neural Networks and ensemble learning techniques. Marshall et al.

, Automated detection of galaxy-scale gravitational lenses in high-resolution imaging data

[14] C. E. Petrillo et al. p. 4

Where the paper says it

Petrillo et al. [14] successfully applied deep convolutional neural networks to the Kilo Degree Survey (KiDS), enabling automated identification of strong gravitational lens candidates from widefield astronomical observations.

p. 4
Where the paper says it

Petrillo et al. [14] successfully applied deep convolutional neural networks to the Kilo Degree Survey (KiDS), enabling automated identification of strong gravitational lens candidates from widefield astronomical observations.

p. 4
Where the paper says it

Petrillo et al. [14] successfully applied deep convolutional neural networks to the Kilo Degree Survey (KiDS), enabling automated identification of strong gravitational lens candidates from widefield astronomical observations.

p. 4
Where the paper says it

Petrillo et al. [14] successfully applied deep convolutional neural networks to the Kilo Degree Survey (KiDS), enabling automated identification of strong gravitational lens candidates from widefield astronomical observations.

p. 4
Where the paper says it

Petrillo et al. [14] successfully applied deep convolutional neural networks to the Kilo Degree Survey (KiDS), enabling automated identification of strong gravitational lens candidates from widefield astronomical observations.

p. 4
Where the paper says it

Petrillo et al. [14] successfully applied deep convolutional neural networks to the Kilo Degree Survey (KiDS), enabling automated identification of strong gravitational lens candidates from widefield astronomical observations.

p. 4
Where the paper says it

Petrillo et al. [14] successfully applied deep convolutional neural networks to the Kilo Degree Survey (KiDS), enabling automated identification of strong gravitational lens candidates from widefield astronomical observations.

, Testing convolutional neural networks for finding strong gravitational lenses in kids

[15] R. B. Metcalf et al. p. 4

Where the paper says it

Their work demonstrated the potential of deep learning models for large scale lens searches while substantially reducing the need for manual inspection. The Strong Gravitational Lens Finding Challenge organized by Metcalf et al.

p. 4
Where the paper says it

Their work demonstrated the potential of deep learning models for large scale lens searches while substantially reducing the need for manual inspection. The Strong Gravitational Lens Finding Challenge organized by Metcalf et al.

p. 4
Where the paper says it

Their work demonstrated the potential of deep learning models for large scale lens searches while substantially reducing the need for manual inspection. The Strong Gravitational Lens Finding Challenge organized by Metcalf et al.

p. 4
Where the paper says it

Their work demonstrated the potential of deep learning models for large scale lens searches while substantially reducing the need for manual inspection. The Strong Gravitational Lens Finding Challenge organized by Metcalf et al.

p. 4
Where the paper says it

Their work demonstrated the potential of deep learning models for large scale lens searches while substantially reducing the need for manual inspection. The Strong Gravitational Lens Finding Challenge organized by Metcalf et al.

p. 4
Where the paper says it

Their work demonstrated the potential of deep learning models for large scale lens searches while substantially reducing the need for manual inspection. The Strong Gravitational Lens Finding Challenge organized by Metcalf et al.

p. 4
Where the paper says it

Their work demonstrated the potential of deep learning models for large scale lens searches while substantially reducing the need for manual inspection. The Strong Gravitational Lens Finding Challenge organized by Metcalf et al.

, The strong gravitational lens finding challenge

[16] Elizabeth Buckley-Geer, David L. Burke, Andrina Carnero Rosell, Matias Carrasco Kind, Jordi Carretero, Luiz N. da Costa, C. Davis et al., Finding high-redshift strong lenses in des using convolutional neural networks

[17] A. Einstein. Lens-like action of a star by the deviation of light in the gravitational field

[18] F. Zwicky. On the probability of detecting nebulae which act as gravitational lenses

[19] M. Bartelmann, Gravitational lensing

[20] J. P. Gardner, J. C. Mather, M. Clampin, R. Doyon, et al., The james webb space telescope

[21] T. Broadhurst et al., Strong-lensing analysis of a1689 from deep advanced camera images

[22] Ž. Ivezić et al., Lsst: From science drivers to reference design and anticipated data products

[23] C. Alard, Automated detection of gravitational arcs

[24] P. J. Marshall et al., Automated detection of galaxy-scale gravitational lenses in high-resolution imaging data

[25] C. E. Petrillo et al., Testing convolutional neural networks for finding strong gravitational lenses in kids

[26] R. B. Metcalf et al., The strong gravitational lens finding challenge

[27] Elizabeth Buckley-Geer, David L. Burke, Andrina Carnero Rosell, Matias Carrasco Kind, Jordi Carretero, Luiz N. da Costa et al., Finding high-redshift strong lenses in des using convolutional neural networks

[28] A. Einstein. Lens-like action of a star by the deviation of light in the gravitational field

[29] F. Zwicky. On the probability of detecting nebulae which act as gravitational lenses

[30] M. Bartelmann, Gravitational lensing

[31] J. P. Gardner, J. C. Mather, M. Clampin, R. Doyon, et al., The james webb space telescope

[32] T. Broadhurst et al., Strong-lensing analysis of a1689 from deep advanced camera images

[33] Ž. Ivezić et al., Lsst: From science drivers to reference design and anticipated data products

[34] C. Alard, Automated detection of gravitational arcs

[35] P. J. Marshall et al., Automated detection of galaxy-scale gravitational lenses in high-resolution imaging data

[36] C. E. Petrillo et al., Testing convolutional neural networks for finding strong gravitational lenses in kids

[37] R. B. Metcalf et al., The strong gravitational lens finding challenge

[38] Elizabeth Buckley-Geer, David L. Burke, Andrina Carnero Rosell, Matias Carrasco Kind, Jordi Carretero et al., Finding high-redshift strong lenses in des using convolutional neural networks

[39] A. Einstein. Lens-like action of a star by the deviation of light in the gravitational field

[40] F. Zwicky. On the probability of detecting nebulae which act as gravitational lenses

[41] M. Bartelmann, Gravitational lensing

[42] J. P. Gardner, J. C. Mather, M. Clampin, R. Doyon, et al., The james webb space telescope

[43] T. Broadhurst et al., Strong-lensing analysis of a1689 from deep advanced camera images

[44] Ž. Ivezić et al., Lsst: From science drivers to reference design and anticipated data products

[45] C. Alard, Automated detection of gravitational arcs

[46] P. J. Marshall et al., Automated detection of galaxy-scale gravitational lenses in high-resolution imaging data

[47] C. E. Petrillo et al., Testing convolutional neural networks for finding strong gravitational lenses in kids

[48] R. B. Metcalf et al., The strong gravitational lens finding challenge

[49] Elizabeth Buckley-Geer, David L. Burke, Andrina Carnero Rosell, Matias Carrasco Kind, Jordi Carretero et al., Finding high-redshift strong lenses in des using convolutional neural networks

[50] A. Einstein. Lens-like action of a star by the deviation of light in the gravitational field

[51] F. Zwicky. On the probability of detecting nebulae which act as gravitational lenses

[52] M. Bartelmann, Gravitational lensing

[53] J. P. Gardner, J. C. Mather, M. Clampin, R. Doyon, et al., The james webb space telescope

[54] T. Broadhurst et al., Strong-lensing analysis of a1689 from deep advanced camera images

[55] Ž. Ivezić et al., Lsst: From science drivers to reference design and anticipated data products

[56] C. Alard, Automated detection of gravitational arcs

[57] P. J. Marshall et al., Automated detection of galaxy-scale gravitational lenses in high-resolution imaging data

[58] C. E. Petrillo et al., Testing convolutional neural networks for finding strong gravitational lenses in kids

[59] R. B. Metcalf et al., The strong gravitational lens finding challenge

[60] Elizabeth Buckley-Geer, David L. Burke, Andrina Carnero Rosell, Matias Carrasco Kind, Jordi Carretero et al., Finding high-redshift strong lenses in des using convolutional neural networks

[61] A. Einstein. Lens-like action of a star by the deviation of light in the gravitational field

[62] F. Zwicky. On the probability of detecting nebulae which act as gravitational lenses

[63] M. Bartelmann, Gravitational lensing

[64] J. P. Gardner, J. C. Mather, M. Clampin, R. Doyon, et al., The james webb space telescope

[65] T. Broadhurst et al., Strong-lensing analysis of a1689 from deep advanced camera images

[66] Ž. Ivezić et al., Lsst: From science drivers to reference design and anticipated data products

[67] C. Alard, Automated detection of gravitational arcs

[68] P. J. Marshall et al., Automated detection of galaxy-scale gravitational lenses in high-resolution imaging data

[69] C. E. Petrillo et al., Testing convolutional neural networks for finding strong gravitational lenses in kids

[70] R. B. Metcalf et al., The strong gravitational lens finding challenge

[71] Elizabeth Buckley-Geer, David L. Burke, Andrina Carnero Rosell, Matias Carrasco Kind, Jordi Carretero et al., Finding high-redshift strong lenses in des using convolutional neural networks

[72] A. Einstein. Lens-like action of a star by the deviation of light in the gravitational field

[73] F. Zwicky. On the probability of detecting nebulae which act as gravitational lenses

[74] M. Bartelmann, Gravitational lensing

[75] J. P. Gardner, J. C. Mather, M. Clampin, R. Doyon, et al., The james webb space telescope

[76] T. Broadhurst et al., Strong-lensing analysis of a1689 from deep advanced camera images

[77] Ž. Ivezić et al., Lsst: From science drivers to reference design and anticipated data products

[78] C. Alard, Automated detection of gravitational arcs

[79] P. J. Marshall et al., Automated detection of galaxy-scale gravitational lenses in high-resolution imaging data

[80] C. E. Petrillo et al., Testing convolutional neural networks for finding strong gravitational lenses in kids

[81] R. B. Metcalf et al., The strong gravitational lens finding challenge

[82] Elizabeth Buckley-Geer, David L. Burke, Andrina Carnero Rosell, Matias Carrasco Kind, Jordi Carretero et al., Finding high-redshift strong lenses in des using convolutional neural networks

Table 1: Comparison of representative gravitational lens detection approaches with the proposed HLA-Net framework.

Method Year Technique Dataset Accuracy Physics-Guided Interpretability Verification DL Learning Automation Cost Key Limitation

Table 2: Summary of processing modules in the proposed HLA-Net framework.

Module Primary Function

Table 3: Top candidates after tangential orientation analysis.

Rank Candidate ID Average Angle (◦) Tangential Score Skeleton Length Interpretation

Table 4: Highest-ranked candidates after geometry verification.

Rank Candidate Geometry Tangential Final Score Quality

Table 5: Final ranking of the highest-confidence gravitational lens candidates produced by the proposed HLA-Net framework.

Rank Candidate ID Geometry Tangential Radial Final Lens Score Quality

Table 6: Computational complexity of HLA-Net modules.

Module Complexity

**Section 4.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Physics Guided Verification Stage: The system can rank candidate gravitational arcs by integrating geometric, tangential alignment, curvature stability and radial symmetry into a unified lens confidence score to prioritize physically meaningful candidates over visually similar structures.

  2. Interpretability and Transparency: The framework provides transparent intermediate outputs throughout the detection process where each module produces measurable quantities that can be directly analyzed by astronomers, contrasting with deep learning methods where the internal decision-making process often remains difficult to interpret.

  3. Computational Efficiency for Large Surveys: HLA-Net achieves an overall complexity of O(N) + O(C Pi!) which approaches linear complexity for practical datasets because the number of candidate structures is considerably smaller than the total number of image pixels, making it suitable for rapid screening in large astronomical surveys.

  4. Progressive Candidate Refinement: The system performs progressive refinement through successive geometric and astrophysical verification stages where each module evaluates complementary physical characteristics, allowing false-positive candidates to be gradually eliminated while genuine arc-like structures receive increasingly higher confidence scores.

  5. Robustness Against False Positives: By utilizing a weighted lens confidence score that integrates geometric, tangential alignment, curvature and radial measurements, the system avoids relying on any individual feature and ensures that candidates exhibiting consistent behaviour across all verification modules receive higher confidence than candidates performing well in only a single stage.

Abstract

Gravitational arcs produced by strong gravitational lensing provide a unique approach for investigating dark matter distribution, galaxy cluster mass reconstruction and the structure of the Universe. Automated detection remains challenging because gravitational arcs are rare, often faint and morphologically similar to elongated galaxies.In this paper, we propose HLA-Net (Hybrid Lens Arc Network), an interpretable multi-stage framework for automatic gravitational arc detection that explicitly exploits both the geometric morphology and physical characteristics of gravitational lensing without requiring manually annotated training datasets.The proposed framework was evaluated on astronomical FITS images containing complex galaxy cluster structures. Starting from 169 initial arc candidates extracted from ordered skeleton representations, the multi-stage verification pipeline reduced the search space to 83 geometrically consistent candidates, while robust circle fitting successfully modelled 113 candidate arc structures. The results demonstrate that HLA-Net effectively suppresses false-positive detections while preserving physically reasonable lens candidates through an interpretable and modular verification strategy.

Related papers