Home


Nonlinear wavefront reconstruction with Convolutional Neural Networks

Paper link

For my bachelor thesis I worked under the supervision of Dr. Sebastiaan Haffert, during which we tried to find an algorithm to do nonlinear wavefront reconstruction. We studied variants of the generalised Optical Differentiation Wavefront Sensor (g-ODWFS) (Haffert 2015), but our method should generally be applicable to all variants of Fourier-based wavefront sensors (including the Pyramid and Zernike). We ended up going with a data-driven approach in Convolutional Neural Networks, which can learn complex nonlinear input-output mappings. We showed that this improves the estimation of large wavefront aberrations, leading to increased convergence rate of the adaptive optics system and a higher Strehl ratio in the case of quickly changing atmospheric turbulence.

Other Research Projects

Adaptive optics control with Reinforcement Learning

Nonlinear wavefront reconstruction with Neural Networks

Bayesian Inference of Meteor Spectra