Valorisez vos réalisations phares sur People@HES-SO Plus d'infos
PEOPLE@HES-SO – Annuaire et Répertoire des compétences
PEOPLE@HES-SO – Annuaire et Répertoire des compétences

PEOPLE@HES-SO
Annuaire et Répertoire des compétences

Aide
language
  • fr
  • en
  • de
  • fr
  • en
  • de
  • SWITCH edu-ID
  • Administration
« Retour
Lavanchy David

Lavanchy David

Professeur HES associé

Compétences principales

Machine Learning

Digital Signal Processing

Digital image processing

Artificial Intelligence (AI)

  • Contact

  • Enseignement

  • Conférences

Contrat principal

Professeur HES associé

Téléphone: +41 24 557 63 41

Bureau: C31

Haute école d'Ingénierie et de Gestion du Canton de Vaud
Route de Cheseaux 1, 1400 Yverdon-les-Bains, CH
HEIG-VD
Institut
iAi - Institut d'Automatisation Industrielle
BSc HES-SO en Génie électrique - Haute école d'Ingénierie et de Gestion du Canton de Vaud
  • Machine Learning
  • Traitement du signal et d'images

2025

High-speed AI image space wavefront sensing using embedded computing :
Conférence ArODES
achieving 1000 frames per second

Gaston Baudat, David Lavanchy, Guillaume Müller

Proceedings of the SPIE Photonics West 2025

Lien vers la conférence

Résumé:

Adaptive optics (AO) is a crucial field in optics, requiring precise wavefront sensing. This paper presents a fast wavefront sensing solution based on AI4Wave, a patented technology for AI-based image-space wavefront sensing. AI4Wave eliminates dedicated sensors like Shack-Hartmann, using only a camera. This high-speed implementation leverages AI edge computing on an NVIDIA Jetson module. The system captures defocused images processed by a feedforward neural network (NN) trained exclusively on synthetic data. This enables real-time phase retrieval, overcoming Shack-Hartmann’s limitations and handling large wavefront errors. AI4Wave employs synthetic, normalized data, making it robust to optical layout changes. Using NVIDIA TensorRT and advanced AI edge computing, it achieves processing speeds of 1000 frames per second, supporting various optical setups. The deterministic feedforward NN approach ensures consistent results without iterative optimization. Preliminary tests show high accuracy and repeatability, with exposure times as short as 24 µs, capturing environmental perturbations. This provides a reliable, industrial-grade solution for high-speed wavefront sensing.

Réalisations

Médias et communication
Nous contacter
Suivez la HES-SO
linkedin instagram facebook twitter youtube rss
univ-unita.eu www.eua.be swissuniversities.ch
Mentions légales
© 2021 - HES-SO.

HES-SO Rectorat