Identification of 4876 Bent-tail Radio Galaxies in the FIRST Survey Using Deep Learning Combined with Visual Inspection

Baoqiang Lao, H. Andernach, Xiaolong Yang, X. Zhang, Ru-Shuang Zhao, Zhen Zhao, Yun Yu, Xiaohui Sun, Sheng‐Li QinThe Astrophysical Journal Supplement Series 2025 · DOI

TL;DR

This paper presents a new catalog of 4876 Bent-Tail Radio Galaxies (BTRGs) identified from the FIRST survey, with 3871 being newly discovered. By combining a deep learning model (RGCMT) with visual inspection, the authors have created the largest and most comprehensive BTRG catalog to date, significantly advancing the study of these unique radio sources and their environments.

What Problem Does This Paper Address?

Bent-tail radio galaxies (BTRGs) are a fascinating subclass of radio galaxies characterized by their bent radio lobes. This distinctive morphology is thought to be caused by the galaxy moving through the dense intra-cluster medium, deflecting the radio jets. Studying BTRGs provides crucial insights into jet launching mechanisms, the effects of relativistic motion, magnetic fields, and the interaction of galaxies with their environment. However, identifying these objects is challenging due to their complex structures and the vastness of radio survey data. Previous catalogs relied heavily on manual visual inspection, which is time-consuming and limits the sample size. This paper aims to overcome these limitations by developing a more efficient and scalable method for identifying BTRGs, leading to a significantly larger and more reliable catalog.

Methods

The core of this work is a hybrid approach combining deep learning with human expertise.

  • Deep Learning for Initial Detection: The authors employed a deep learning model called RGCMT (Radio Galaxies Classification with Mask Transfiner). RGCMT builds upon the Mask R-CNN architecture, incorporating an ambiguity areas detector and a transformer block. This allows it to not only detect radio sources but also generate precise masks, bounding boxes, and confidence scores. RGCMT was trained to classify radio galaxies into five categories: Compact Sources (CS), straight FR-I (sFR-I), straight FR-II (SFR-II), Bent-Tail (BT), and One-sided Straight Extended diffuse sources (OSE).
  • Visual Inspection for Refinement: While RGCMT is highly accurate, the authors acknowledge that artificial intelligence methods can still produce false positives. Therefore, a crucial step involved visual inspection of the candidates identified by RGCMT. This manual review allowed for the filtering out of non-interesting candidates, such as those with incorrect bending angles, artifacts, or those not truly fitting the BTRG definition.

Datasets

  • Training and Evaluation Data: The RGCMT model was trained on 3172 FIRST images containing 3601 sources, and evaluated on a separate dataset of 1946 images. The labels for these datasets were generated through visual inspection by three individuals.
  • FIRST Survey Data: The primary dataset used for the search was the Faint Images of the Radio Sky at Twenty-cm (FIRST) survey, specifically the December 17, 2014 version. This survey provides 1.4 GHz continuum observations over a large portion of the sky.
  • Optical/Infrared Data: For host galaxy identification and redshift determination, the Dark Energy Spectroscopic Instrument (DESI) Legacy Surveys (LS) data release 10 (DR10) was utilized. This dataset provides multi-color imaging and improved depth.
  • Spectroscopic and Photometric Redshift Catalogs: To obtain redshifts for host galaxies, the authors queried the NASA/IPAC Extragalactic Database (NED) and several other spectroscopic catalogs (SDSS DR16, GAMA DR4, 2dFGRS, 6dFGS, WiggleZ, 2MRS). For hosts without spectroscopic redshifts, photometric redshifts from DESI LS DR10 were used.
  • Cluster Catalogs: To associate BTRGs with galaxy clusters, the authors cross-matched their catalog with 41 previously published cluster catalogs, including the one by Wen & Han (2024).

Key Insights and Ideas

  • Synergy of AI and Human Expertise: The paper strongly advocates for a combined approach. Deep learning efficiently sifts through massive datasets, while human visual inspection provides the necessary nuance to ensure high-quality scientific catalogs, especially for complex object classes like BTRGs.
  • BTRGs as Probes of Environment: The significant fraction of BTRGs found within or near galaxy clusters (1825 out of 4876) highlights their utility as tracers of dense environments and their sensitivity to the intra-cluster medium.
  • Challenging Classical Classifications: The study reinforces findings that the traditional Fanaroff-Riley (FR) luminosity break is not a definitive classifier for radio galaxies, and that BTRG morphology can be more significant than FR type.
  • Red Galaxies Dominate BTRG Hosts: The overwhelming majority of BTRG host galaxies are red, suggesting a strong association with older stellar populations, consistent with their presence in galaxy clusters.

Key Results

  • New Catalog: A catalog of 4876 BTRGs has been constructed, representing the largest such catalog to date.
  • High Discovery Rate: 3871 of the identified BTRGs are newly discovered, a testament to the effectiveness of the combined method.
  • Classification Breakdown: The catalog includes 4424 Wide-Angle-Tail (WAT) sources and 652 Narrow-Angle-Tail (NAT) sources.
  • Host Galaxy Identification: Optical counterparts were identified for 4193 BTRGs.
  • Redshift and Luminosity Range: The catalog covers redshifts from $z = 0.0023$ to $z = 3.43$ and 1.4 GHz luminosities from $1.91 \times 10^{20}$ to $1.45 \times 10^{28}$ W Hz$^{-1}$.
  • Cluster Association: 1825 BTRGs (approximately 43.5%) were found to be associated with known galaxy clusters.
  • RGCMT Performance: The RGCMT model achieved a mean average precision (mAP) of 98.4% on its evaluation dataset for radio galaxy classification.

Limitations and Future Work

  • Visual Inspection Subjectivity: While visual inspection enhances reliability, it can still introduce some level of subjectivity. The authors acknowledge that the classification scheme for opening angle and FR type is based on visual inspection.
  • Host Galaxy Identification Challenges: For some BTRGs, no reasonable host galaxy candidates were found in the DESI LS DR10 data.
  • Undetermined Cluster Membership: For a portion of BTRG hosts, their membership in associated galaxy clusters remains undetermined due to a lack of $r_{500}$ information.
  • Future Work: The authors suggest that this catalog will be invaluable for future statistical and scientific studies of BTRGs. The automated nature of the RGCMT model is expected to be crucial for large-scale statistical analyses of radio galaxies. The paper also notes that some BTRGs identified in previous studies were excluded from this catalog due to stricter selection criteria (e.g., OA > 170°), indicating potential for future comparative studies.

Summary updated

Abstract

Abstract Bent-tail radio galaxies (BTRGs) are characterized by bent radio lobes. This unique shape is mainly caused by the movement of the galaxy within a cluster, during which the radio jets are deflected by the intracluster medium. A combined method, which involves a deep learning-based radio source finder along with visual inspection, has been utilized to search for BTRGs from the Faint Images of the Radio Sky at Twenty cm survey images. Consequently, a catalog of 4876 BTRGs has been constructed, among which 3871 are newly discovered. Based on the classification scheme of the opening angle between the two jets of the galaxy, BTRGs are typically classified as either wide-angle-tail (WAT) sources or narrow-angle-tail (NAT) sources. Our catalog comprises 4424 WATs and 652 NATs. Among these, optical counterparts are identified for 4193 BTRGs. This catalog covers luminosities in the range of 1.91 × 10 20 ≤ L 1.4 GHz ≤ 1.45 × 10 28 W Hz −1 and redshifts from z = 0.0023 to z = 3.43. Various physical properties of these BTRGs and their statistics are presented. Particularly, by the nearest neighbor method, we found that 1825 BTRGs in this catalog belong to galaxy clusters reported in literature.
BibTeX
@article{lao2025identification,
  author = {Baoqiang Lao and H. Andernach and Xiaolong Yang and X. Zhang and Ru-Shuang Zhao and Zhen Zhao and Yun Yu and Xiaohui Sun and Sheng‐Li Qin},
  title = {{Identification of 4876 Bent-tail Radio Galaxies in the FIRST Survey Using Deep Learning Combined with Visual Inspection}},
  journal = {The Astrophysical Journal Supplement Series},
  year = {2025},
  doi = {10.3847/1538-4365/ad9c6d},
}
Figure