Numerical study of adaptive optics compensation based on Convolutional Neural Networks

Abstract

Abstract Adaptive optics compensation technique is needed to improve the image quality when laser is propagating in the atmosphere. A novel compensation technique based on Convolutional Neural Networks (CNN) is proposed in the study. In general, CNN accepts an input image or patch, applies a sequence of transforming layers in order to extract features and finally classifies the inputs. We have tried to extract the features from the intensity images of distortions and obtain the corresponding Zernike coefficients by CNN. The network is trained by using large data sets of two intensity images which are measured in-focal and out-of-focus and Zernike coefficients for the distortions as inputs. The results indicate that the Zernike coefficients of distortions can be predicted by CNN which is mostly used to classify in recent researches. The compensation performance of adaptive optics based on CNN is tested for different strength turbulence scenarios and the robustness of the system is considered under different signal-to-noise ratio conditions.