Accelerating Multiframe Blind Deconvolution via Deep Learning

Ground-based solar image restoration is a computationally expensive procedure that involves nonlinear optimization techniques. The presence of atmospheric turbulence produces perturbations in individual images make it necessary to apply blind deconvolution These techniques rely on the observation many short exposure frames are used simultaneously infer instantaneous state atmosphere and unperturbed object. We have recently explored use machine learning accelerate this process, with promising results. build upon previous work propose several interesting improvements lead better models. As well, we new method based algorithm unrolling. In method, problem solved gradient descent unrolled accelerated aided by few small neural networks. role networks correct estimation solution at each iterative step. model trained perform fixed number steps curated dataset. Our findings demonstrate both methods significantly reduce time compared standard procedure. Furthermore, showcase these models can be an unsupervised manner using observed from three different instruments. Remarkably, they also exhibit robust generalization capabilities when applied datasets. To foster further research collaboration, openly provide models, along corresponding training evaluation code, as well dataset, scientific community.