Javier Rodriguez de rivera

Adaptive optics play a crucial role in acquiring high-quality images in ground-based telescopes. The primary challenge lies in mitigating the impact of atmospheric turbulence on captured light, which induces aberrations in the wavefront, resulting in distorted images. Over the past decade, several studies have been carried out exploring the application of predictive models to this technology, aiming to increase the efficiency of the process, reduce costs as well as improve the response time of the system. In particular, neural networks have been proven to be a powerful tool toward this end. This study assesses the potential of novel training and data processing techniques, while also evaluating the influence of specific parameters such as input data length and system frequency. In this study, the ideal training parameters are first analyzed through several low-complexity models, followed by the integration of more sophisticated neural networks. The feasibility of predicting slopes from sensor images has been demonstrated, validating the use of image-based sensor data. These results also open the path for exploring multi-scale sensors, like pyramid sensors, which could further improve predictions by capturing data at different spatial resolutions.