Radio Galaxy Morphology Generation using Residual Convolutional Autoencoder and Gaussian Mixture Models

The morphology of a radio galaxy is related to its center active galactic nuclei (AGN), which can reveal the evolution of the super massive black hole (SMBH). In this work, we propose a morphology generation framework for two typical radio galaxies namely Fanaroff- Riley type- I (FRI) and type-II (FRII) using residual convolutional autoencoder (RCAE) and Gaussian mixture models (GMMs). The encoder and decoder subnets in the RCAE are symmetric aside a fully-connected layer namely code layer hosting the extracted feature vectors, which can be randomly generated later by a three-component Gaussian Mixture models for simulating new FRI or FRII radio galaxy images. Experiments are demonstrated on real radio galaxy images, where we discuss the length of feature vectors, selection of lost functions, and make comparisons among proposed approach with the other networks. The results suggest a high efficiency and performance of our RCAE network. Code is available at: https://github.com/myinxd/rcae-gmm.