Early Fully-Convolutional Approach to Wavefront Imaging on Solar Adaptive Optics Simulations

摘要

Aberrations are presented in the wave-front images from celestial objects taken with large ground-based telescopes, due to the effects of the atmospheric turbulence. Therefore, different techniques, known as adaptive optics techniques, have been developed to correct those effects and obtain new images clearer in real time. One part of an adaptive optics system is the Reconstructor System, it receives information of the wavefront given by the wavefront sensor and calculates the correction that will be performed by the Deformable Mirrors. Typically, only a small part of the information received by the wave-front sensors is used by the Reconstructor System. In this work, a new Reconstructor System based on the use of Fully-Convolutional Neural Networks is proposed. Due to the features of Convolutional Neural Networks, all the information received by the wavefront sensor is then used to calculate the correction, allowing for obtaining more quality reconstructions than traditional methods. This is proved in the results of the research, where the most common reconstruction algorithm (the Least-Squares method) and our new method are compared for the same atmospheric turbulence conditions. The new algorithm is used for Solar Single Conjugated Adaptive Optics (Solar SCAO) with the aim of simplifying the system since all the needed calculations are performed with the network. The found improvements can be stated around at 0.4 rad of mean WFE over the recovered wavefront.