Radio-Interferometric Image Reconstruction with Denoising Diffusion Restoration Models

2026/01/31 by Michel Morales, Emma Tolley, Remi Poitevineau · 1 citation
Physics and Astronomy · #astro-ph.IM #astro-ph.CO

paper · pdf

11 pages, 11 figures

arxiv created 2026/08/06 · arxiv updated 2026/08/07

Abstract

Reconstructing images of the radio sky from incomplete Fourier information is a key challenge in radio astronomy. In this work, we present a method for radio interferometric image reconstruction using a data-driven prior for the radio sky based on denoising diffusion probabilistic models (DDPMs). We train a DDPM on radio galaxy observations from the VLA FIRST survey, then create simulated VLBA, EHT, and ALMA observations of radio galaxies. We use an unsupervised posterior sampling method called Denoising Diffusion Restoration Models (DDRM) to reconstruct the corresponding images using our DDPM as a prior. Our approach naturally incorporates the PSF of the instrument. We are able to reconstruct images with very high fidelity and demonstrate a marked improvement over CLEAN and MS CLEAN. While DDRM naturally produces multiple samples, these are not calibrated and do not constitute reliable uncertainty estimates in the current implementation. The code for training and inference of the model is available at https://github.com/epfl-radio-astro/diffusionRI.git

Citations

• Simulating realistic radio continuum survey maps with diffusion models
• The R2D2 deep neural network series paradigm for fast precision imaging in radio astronomy
• A Conditional Denoising Diffusion Probabilistic Model for Radio Interferometric Image Reconstruction
• On Calibrating Diffusion Probabilistic Models
• MiraBest: A Dataset of Morphologically Classified Radio Galaxies for Machine Learning
• Diffusion Posterior Sampling for General Noisy Inverse Problems
• Convolutional Deep Denoising Autoencoders for Radio Astronomical Images
• Denoising Diffusion Probabilistic Models
• First M87 Event Horizon Telescope Results. II. Array and Instrumentation
• First M87 Event Horizon Telescope Results. IV. Imaging the Central Supermassive Black Hole
• The LOFAR Two-metre Sky Survey IV. First Data Release: Photometric redshifts and rest-frame magnitudes
• Image Domain Gridding: a fast method for convolutional resampling of visibilities
• FiLM: Visual Reasoning with a General Conditioning Layer
• An optimized algorithm for multi-scale wideband deconvolution of radio astronomical images
• SGDR: Stochastic Gradient Descent with Warm Restarts
• Layer Normalization
• Deep Residual Learning for Image Recognition
• A Generic and Efficient E-field Parallel Imaging Correlator for Next-Generation Radio Telescopes
• U-Net: Convolutional Networks for Biomedical Image Segmentation
• MORESANE: MOdel REconstruction by Synthesis-ANalysis Estimators
• WSClean: an implementation of a fast, generic wide-field imager for radio astronomy
• The application of compressive sampling to radio astronomy II: Faraday rotation measure synthesis
• Compressed sensing imaging techniques for radio interferometry
• The Noncoplanar Baselines Effect in Radio Interferometry: The W-Projection Algorithm
• Multiscale CLEAN Deconvolution of Radio Synthesis Images
• Correcting direction-dependent gains in the deconvolution of radio interferometric images