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

Don SarXiv 2026 · arXiv

TL;DR

HLA-Net is a physics-informed, modular framework designed to detect gravitational arcs in astronomical images without the need for massive, manually labeled training datasets. By combining skeleton-based geometric analysis with astrophysical constraints—such as tangential alignment to galaxy cluster centers—it provides an interpretable and computationally efficient alternative to "black-box" deep learning models.

What Problem Does This Paper Address?

Gravitational lensing is a cornerstone of modern cosmology, where the mass of a foreground galaxy cluster bends light from distant background galaxies into spectacular arcs and rings. These structures are vital for mapping dark matter and understanding the expansion of the universe. However, finding them is a "needle in a haystack" problem. Arcs are rare, often extremely faint, and morphologically similar to ordinary elongated galaxies or imaging artifacts like diffraction spikes.

With the advent of massive surveys from the James Webb Space Telescope (JWST) and the upcoming Vera C. Rubin Observatory (LSST), astronomers are facing millions of images. Manual inspection is no longer feasible. While deep learning has shown promise, it requires vast amounts of accurately labeled data—which is scarce for these rare phenomena—and often lacks the interpretability required for rigorous scientific validation. HLA-Net aims to bridge this gap by using the laws of physics to guide the detection process.

Methods

HLA-Net (Hybrid Lens Arc Network) employs a multi-stage pipeline that moves from raw pixels to physically verified candidates:

  1. Adaptive Enhancement: The framework uses Contrast Limited Adaptive Histogram Equalization (CLAHE) and multi-scale ridge enhancement to make faint, curvilinear structures stand out against the background noise of the galaxy cluster.
  2. Skeleton Extraction: Connected components are reduced to a one-pixel-wide "skeleton" (medial axis). This simplifies complex shapes into mathematical trajectories, allowing for precise geometric analysis.
  3. Geometric Filtering: Candidates are initially screened based on morphology—evaluating features like length, eccentricity, and circularity to discard compact galaxies.
  4. Physics-Guided Verification: This is the core innovation. The system evaluates two primary astrophysical constraints:
  5. Tangential Orientation: Genuine arcs should be oriented perpendicularly to the radial line connecting the arc to the cluster's center of mass.
  6. Radial Consistency: The framework checks if the arc follows the expected gravitational curvature (approximately concentric circles around the lens center).
  7. Weighted Scoring: Finally, all geometric and physical metrics are integrated into a unified confidence score to rank the most plausible candidates.

Datasets

The framework was evaluated using publicly available, high-resolution FITS (Flexible Image Transport System) images of the Abell 1689 galaxy cluster, captured by the Hubble Space Telescope (HST). Abell 1689 is a well-known benchmark in the field because it contains a high density of strong lensing features alongside many "distractor" objects like edge-on galaxies.

Key Insights and Ideas

  • Physics as a Prior: Instead of asking a neural network to "learn" what an arc looks like from thousands of examples, HLA-Net uses the known geometry of gravitational lensing as a hard constraint. This eliminates the need for a labeled training set.
  • Interpretability is a Feature: Unlike deep learning models where the decision-making process is opaque, HLA-Net provides "transparent intermediate outputs." An astronomer can look at a candidate's tangential score or skeleton continuity to understand exactly why the system flagged it.
  • Computational Efficiency: The algorithm operates with roughly linear complexity $O(N)$ relative to the number of pixels. Because it only performs intensive calculations on small candidate regions rather than the whole image, it is highly scalable for large-scale sky surveys.

Key Results

The experimental results on the Abell 1689 dataset demonstrate the framework's ability to prune false positives effectively:

  • Search Space Reduction: The pipeline started with 169 initial arc candidates and refined them down to 83 geometrically consistent candidates.
  • Modeling Success: Robust circle fitting successfully modeled 113 candidate arc structures, showing strong agreement with expected lensing theory.
  • High-Confidence Detections: The top-ranked candidate (ID 120) achieved a final confidence score of 0.8205 and was classified as an "Excellent" lens candidate.
  • Comparative Advantage: As shown in the paper's comparison table, HLA-Net is one of the few methods that offers high interpretability and physics-guided verification without the high computational cost or data dependency of deep learning.
Table 1: Comparison of representative gravitational lens detection approaches with the proposed HLA-Net framework.

Limitations and Future Work

Despite its strengths, the authors note a few areas for improvement:

  • Center Dependency: The current version assumes the approximate center of the galaxy cluster is already known to calculate tangential alignment.
  • Fragmentation: Extremely faint arcs can sometimes break apart during the initial segmentation phase, which leads to lower skeleton continuity scores and potential missed detections.
  • False Positives: Highly irregular foreground galaxies can occasionally mimic the geometry of a lens, requiring further refinement of the scoring weights.

Future research will focus on testing HLA-Net on multi-band (color) astronomical data and across a wider variety of galaxy clusters to further prove its generalization capabilities.

Summary updated

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 \textbf{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.
BibTeX
@misc{s2026hla,
  author = {Don S},
  title = {{HLA-Net: A Physics-Guided Skeleton-Based Framework for Automatic Gravitational Arc Detection in Astronomical Images}},
  year = {2026},
  eprint = {2611.05438},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.05438},
}
Figure