Machine Learning for Aberration Correction with Spatial Light Modulators

Alexandre Rodrigues Lopes Ferreira

Key information

Authors:

Alexandre Rodrigues Lopes Ferreira (Alexandre Rodrigues Lopes Ferreira)

Supervisors:

Diogo Manuel Ribeiro Ferreira (Diogo Ribeiro Ferreira); Marta Leitão Mota Fajardo (Marta Leitão Mota Fajardo)

Published in

June 27, 2024

Abstract

This thesis explores the integration of machine learning techniques with the capabilities of spatial light modulators to enhance aberration correction in optical systems. The primary objectives include understanding the shaping of Laguerre-Gauss beams, identifying conditions under which aberrations occur, and formulating these aberrations using Zernike polynomials. The goal is to use deep learning models to predict the coefficients of those polynomials from intensity-profile images with such aberrations. A computational framework for simulating optical aberrations was developed to generate training data for the proposed models. The models were trained to recognize aberration patterns accurately and to allow for the application of corrective measures that are provided by spatial light modulators. Key aspects of this work involve the study of optical phenomena related to ultra-short laser pulses, the creation of a simulation framework for generating realistic aberration patterns, and the design and validation of machine learning models to detect these aberrations. Performance metrics such as computational cost, prediction accuracy, and correction efficiency were tested and evaluated. The results show that the proposed models succeed in their aberration detection task, significantly improving beam-focusing capabilities and overall optical performance when applying the corresponding corrections of space light modulators. Additionally, the proposed approach is adaptable to the broader scope of a larger project, ensuring seamless integration with existing equipment and operational conditions.