RGC-Bent: A Novel Dataset for Bent Radio Galaxy Classification

Mir Sazzat Hossain 1,⋆ Khan Muhammad Bin Asad 2,† Payaswini Saikia 3,† Adrita Khan 1,2 Md Akil Raihan Iftee 1 Rakibul Hasan Rajib 1 Arshad Momen 1 Md Ashraful Amin 1 Amin Ahsan Ali 1 AKM Mahbubur Rahman 1 · 2025

Read RGC-Bent: A Novel Dataset for Bent Radio Galaxy Classification by Mir Sazzat Hossain 1,⋆ Khan Muhammad Bin Asad 2,† Payaswini Saikia 3,† Adrita Khan 1,2 Md Akil Raihan Iftee 1 Rakibul Hasan Rajib 1 Arshad Momen 1 Md Ashraful Amin 1 Amin Ahsan Ali 1 AKM Mahbubur Rahman 1 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-19249.txt.

Introduction

Mir Sazzat Hossain and co-authors share a new data set in 2025 for sorting bent radio galaxies by image. Bent galaxies shoot out curved jets of radio waves as they move through gas in galaxy clusters. Their bent shape tells scholars about the gas and about how clusters grow and change.

Few shared data sets exist for this task, so progress has been slow. The new set, called RGC-Bent, sorts galaxies into narrow-angle tail and wide-angle tail kinds. It comes from a known radio survey and ships with clear prep steps. The team tested top deep learning models on it, from classic image nets to transformer forms. A model called ConvNeXT scored best on both kinds. The set and scores are shared to speed work in the field.

Catalog record

CREATORMir Sazzat Hossain 1,⋆ Khan Muhammad Bin Asad 2,† Payaswini Saikia 3,† Adrita Khan 1,2 Md Akil Raihan Iftee 1 Rakibul Hasan Rajib 1 Arshad Momen 1 Md Ashraful Amin 1 Amin Ahsan Ali 1 AKM Mahbubur Rahman 1
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE20 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-19249

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

RGC-Bent: A Novel Dataset for Bent Radio Galaxy Classification Mir Sazzat Hossain 1,⋆ Khan Muhammad Bin Asad 2,† Payaswini Saikia 3,† Adrita Khan 1,2 Md Akil Raihan Iftee 1 Rakibul Hasan Rajib 1 Arshad Momen 1 Md Ashraful Amin 1 Amin Ahsan Ali 1 AKM Mahbubur Rahman 1 Abstract We introduce a novel machine learning dataset tailored for the classification of bent radio active galactic nuclei (AGN) in astronomical observations. Bent radio AGN, distinguished by their curved jet structures, provide critical insights into galaxy cluster dynamics, interactions within the intracluster medium, and the broader physics of AGN. Despite their astrophysical significance, the classification of bent radio AGN remains a challenge due to the scarcity of specialized datasets and benchmarks. To address this, we present a dataset, derived from a well-recognized radio astronomy survey, that is designed to support the classification of NAT (Narrow-Angle Tail) and WAT (Wide-Angle Tail) categories, along with detailed data processing steps. We further evaluate the performance of state-of-the-art deep learning models on the dataset, including Convolutional Neural Networks (CNNs) and transformer-based architectures. Our results demonstrate the effectiveness of advanced machine learning models in classifying bent radio AGN, with ConvNeXT achieving the highest F1-scores for both NAT and WAT sources. By sharing this dataset and benchmarks, we aim to facilitate the advancement of research in AGN classification, galaxy cluster environments and galaxy evolution. The source code is available at: https://github.com/mirsazzathossain/RGC-Bent † † address: 1 Center for Computational & Data Sciences, Independent University, Bangladesh 2 Center for Astronomy, Space Science and Astrophysics, Independent University, Bangladesh 3 Center for Astrophysics and Space Science, New York University Abu Dhabi 1 Introduction Active Galactic Nuclei (AGNs) are some of the most luminous objects in the universe, powered by the accretion of matter onto supermassive black holes. The classification of AGNs is important for understanding the evolution of galaxies in clusters, and the growth of supermassive black holes [ 1 ] . Bent radio AGN (sometimes also called bent radio galaxies, head-tail galaxies, or bent-tailed galaxies) are a subset of radio-loud AGN, characterized by distinctly curved jet structures caused by interactions with the intergalactic medium [ 2 ] . These structures are commonly observed in galaxy clusters, where the interaction between the galaxy’s jets and the intracluster medium (ICM) plays a significant role in shaping their appearance. The bending of the jets is often linked to the relative motion between the host galaxy and the surrounding ICM, which results in ram pressure that distorts the jets’ trajectories [ 3 ] . (a) WAT (b) NAT Figure 1 : Examples of Wide-Angle Tailed (WAT) and Narrow-Angle Tailed (NAT) radio galaxies. (a) WAT galaxies exhibit a wider ‘C’-shaped structure with jets extending over 90 degrees, while (b) NAT galaxies show a narrower ‘V’-shaped structure with jets within a 90-degree angle. Two notable subtypes are identified within this category based on the angle between their jets: Wide-Angle Tailed (WAT) Galaxies and Narrow-Angle Tailed (NAT) Galaxies, as shown in Figure 1 . WAT galaxies feature two-sided jets that expand outward, creating a wide ‘C’ shape with an opening angle exceeding 90 degrees [ 4 ] . As the galaxy moves through the dense intracluster medium, interactions with it are believed to cause the bending of its jets. In contrast, NAT galaxies, also called head-tail galaxies, have closely aligned jets forming a narrow ‘V’ shape with an opening angle under 90 degrees and are usually located in galaxy clusters, moving at high velocities through the intracluster medium [ 4 , 5 ] . The study of bent radio AGN, including WATs and NATs, is crucial for understanding the dynamics of galaxy clusters, the behavior of the intracluster medium, the interactions between galaxies and their environments, and the dynamics of AGN jets and their surrounding conditions [ 1 , 6 ] . In this work, we make the following contributions: • We present a dataset of bent radio AGN images, captured by the Very Large Array (VLA), specifically compiled to facilitate machine learning-based classification. • We evaluate the performance of five machine learning models on this dataset, establishing a baseline for automated bent radio AGN classification. • We release all code as open source for key data preprocessing tasks, including background estimation, source identification, mask generation, and background removal via masking. 2 Existing Datasets and Limitations Machine learning has been widely applied in astronomy, but the lack of specialized datasets remains a challenge. Existing datasets like Galaxy Zoo [ 7 ] focus on general galaxy classification and radio galaxy Zoo deals with radio sources [ 8 ] . The MiraBest dataset [ 9 ] provides labeled Fanaroff-Riley Type I and II sources [ 10 ] but lacks the labeling of bent AGNs into WATs and NATs. While no dedicated machine learning dataset exists for bent radio AGNs, several catalogs provide relevant information. The Proctor et al. [ 11 ] catalog was among the first, but it contained automated annotations which cannot be relied upon for high-precision tasks without manual verification. In contrast, Sasmal et al. [ 12 ] introduced a manually curated catalog, offering more accurate source identification. This catalog serves as the primary data source for our dataset creation. Converting it into a machine learning-ready dataset requires careful data acquisition, preprocessing, and expert validation. In this paper, we document this process in detail to facilitate future research in bent radio AGN classification. 3 Dataset Creation 3.1 Data Acquisition The raw data is obtained from the Faint Images of the Radio Sky at Twenty-cm (FIRST) survey using source coordinates from the catalog proposed by Sasmal et.al. [ 12 ] . The FIRST survey, conducted at 1.4 GHz with the Karl G. Jansky Very Large Array (VLA) in New Mexico, USA, covers the sky north of \delta=-40^{\circ} with a 5-arcsecond resolution and a 0.15 mJy sensitivity. The data is stored in Flexible Image Transport System (FITS) format and includes flux density measurements at different frequencies. We retrieve the images from NASA SkyView 1 1 1 https://skyview.gsfc.nasa.gov , a virtual observatory providing access to multi-survey astronomical data. Using the provided source coordinates, we query SkyView to download the corresponding FIRST survey radio images, which are then stored in FITS format for subsequent processing. 3.2 Data Processing The first few steps (up to and including Section 3.2.3 ) of data processing were performed using the Python package PyBDSF [ 13 ] . 3.2.1 Background Estimation The first step in processing is estimating the background noise level for each image. We compute the background level and its standard deviation by segmenting the image into smaller subregions and calculating the local mean ( \mu_{\text{local}} ) and standard deviation ( \sigma_{\text{local}} ) of the pixel intensities within each subregion. The background estimation is modeled as: B(x,y)=\mu_{\text{local}}+k\cdot\sigma_{\text{local}},\vskip-3.99994pt (1) where B(x,y) denotes the local background intensity, and k is a scaling factor that accounts for typical noise fluctuations. 3.2.2 Source Identification After background estimation, sources are identified by detecting contiguous regions (or “islands”) where the intensity significantly exceeds the local background. Two thresholds are applied during this process: • Island Threshold ( T_{\text{isl}} ): This parameter, typically set to 3, defines the minimum signal level for considering a group of contiguous pixels as part of an

Read the complete text