EXXA - Denoising Astronomical Observations of Protoplanetary Disks

Astronomical observations of protoplanetary disks often contain noise that obscures important structures such as rings and gaps that may indicate planet formation. This project proposes to develop a machine learning pipeline for denoising astronomical observations using deep learning techniques including convolutional autoencoders, U-Net architectures, and diffusion models. The project will implement a complete workflow for loading FITS datasets, preprocessing observations, training denoising models, and evaluating performance using PSNR, SSIM, and MSE metrics. The final deliverables will include a reproducible open source pipeline, trained models, and evaluation tools for improving the quality of protoplanetary disk observations.