Answers are grounded in this paper and cite their sources.
Context . Bent radio active galactic nuclei (RAGNs) – wide-angle tails (WATs) and narrow-angle tails (NATs) – trace dense environments in galaxy groups and clusters, yet no multi-class classifier simultaneously separates them from straight Fanaroff-Riley types (sFRI, sFRII) using visually inspected labels and unlabelled data. Aims . We release FIRST-2060, a four-class labelled dataset of 2060 RAGNs (sFRI, sFRII, WAT, NAT) constructed from three publicly available catalogues through multi-tier visual inspection, together with the semi-supervised RGC 1.0 model that leverages 20 000 unlabelled radio sources. We benchmarked RGC 1.0 against five supervised baselines spanning convolutional neural networks (CNNs) and vision transformers. Methods . FIRST-2060 is provided in two preprocessing variants: R L 1 , which retains spurious sources, and R L 2 , from which they are removed. The RGC model integrates the self-supervised framework Bootstrap Your Own Latent (BYOL) with an E (2)-equivariant steerable CNN (E2CNN) encoder, pre-trained on the unlabelled data and fine-tuned on the labelled sets. All six models were evaluated with five-fold cross-validation, gradient-weighted class activation mapping (Grad-CAM) attention analysis, and controlled class-imbalance experiments. Results . ConvNeXT ( M 1 ) and RGC ( M 2 ) form a top tier at macro- F 1 0.80 ± 0.02 and 0.79 ± 0.02, respectively, representing a difference within one standard deviation. The RGC model is the only model whose Grad-CAM contours consistently trace the morphological structure of RAGNs – lobes, jets, and bends – rather than defaulting to compact blobs or diffuse patterns. Conclusions . The four-class scheme introduced here enables WAT and NAT resolved catalogues that can serve as environment probes and progenitor classifications for diffuse cluster radio emission. The complementary strengths of M 1 and M 2 – in cross-type and within-type discrimination, respectively – suggest that an ensemble approach may offer a practical framework for survey-scale morphological catalogues.
No takes yet. Share an insight, caveat, or question.
Hossain et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: