31 May 1994 Hartmann wavefront sensing with an artificial neural network processor
Author Affiliations +
An artificial neural network has been developed for wavefront reconstruction by processing the centroid displacements of the subaperture images of a 4 X 4 Hartmann sensor. Training was accomplished by computing the first differentials of random but known mixes of aberration polynomials, presenting these at the network input and adjusting the interconnection weights to minimize the Zernike coefficient output errors. After training, the residual rms wavefront errors for noise-free systems were typically within 2%. Similar results were obtained with a 5 X 5 system. As simulated noise was added, the errors increased slowly but were similar in magnitude to those obtained analytically provided that the network had sufficient complexity. At very low SNR values the neural network outperformed the analytic method provided that it had been trained for the same noise level.
© (1994) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Eli Ettedgui-Atad, Eli Ettedgui-Atad, G. Catalan, G. Catalan, John W. Harris, John W. Harris, Colin M. Humphries, Colin M. Humphries, A. M. Smillie, A. M. Smillie, Alistair E. Armitage, Alistair E. Armitage, G. S. Hanspal, G. S. Hanspal, R. J. McKeating, R. J. McKeating, } "Hartmann wavefront sensing with an artificial neural network processor", Proc. SPIE 2201, Adaptive Optics in Astronomy, (31 May 1994); doi: 10.1117/12.176082; https://doi.org/10.1117/12.176082

Back to Top