Improving Astronomy Image Quality Through Real-Time Wavefront Estimation
We present a new framework for detecting telescope optics aberrations in real-time. The framework divides the problem into two subproblems that are highly amenable to machine learning and optimization. The first involves making local wavefront estimates with a convolutional neural network. The second involves interpolating the optics wavefront from all the local estimates by minimizing a convex loss function. We test our framework with simulations of the Vera Rubin Observatory. In a realistic mini-survey, the algorithm reduces the total magnitude of the optics wavefront by 66%, the optics PSF FWHM by 27%, and increases the Strehl ratio by a factor of 6. The resulting sharper images have the potential to boost the scientific payload for astrophysics and cosmology.