RGC-1

The Radio Galaxy Classifier (RGC) is a deep-learning system that sorts radio-loud active galactic nuclei (radio AGN) by morphology. It is named in honour of Radha Gobinda Chandra (1878–1975), the pioneering Bengali variable-star observer from Bagchar, Jessore, whose meticulous records became part of the international AAVSO archive.

RGC-1 is the first model in the series, presented in “RGC: a radio AGN classifier based on deep learning. I. A semi-supervised multiclass model for VLA images” — accepted with minor revision at Astronomy & Astrophysics. It introduces FIRST-2060, a hand-labelled catalogue of 2,060 sources from the VLA FIRST survey spanning four classes: wide-angle tails and narrow-angle tails (the bent sources) alongside the straight Fanaroff–Riley FR-I and FR-II types.

To make the most of the small labelled set, RGC-1 is semi-supervised: a Bootstrap-Your-Own-Latent (BYOL) self-supervised backbone with an E(2)-equivariant steerable CNN (E2CNN) encoder learns from ~20,000 unlabelled FIRST sources, and is benchmarked against supervised baselines including ConvNeXt. Because morphology probes environment — bent tails trace a galaxy’s motion through dense intracluster gas — an automated classifier at survey scale feeds directly into CASSA’s related MIMIC and GAZE efforts.