Deep-learning empowered super-resolution plankton imaging
Acronym
SuperPI
Titel
Deep-learning empowered super-resolution plankton imaging
Kurzbeschreibung
The goal of SuperPI is to develop AI-based methods for pushing the boundaries of
optical microscopy and plankton imaging technology. Plankton imaging systems, like all
microscopic systems, are limited by the trade-off between magnification and depth-of-field
(DOF) that restricts analysis to very small sample volumes. Recent technical advances in
opto-electronics, such as electrically tunable lenses (ETL), show great promise to overcome
this challenge by substantially enhancing the DOF. However, acquired images still suffer
from sub-optimal optical resolution due to the interference of in-focus and out-of-focus
objects during image formation. SuperPI will explore the potential of AI-based solutions for
maximizing the optical resolution of enhanced-DOF images acquired with modern ETL
technology. We will develop a physics-informed neural network (Pi-NN) to achieve
“super-resolution imaging” in unprecedentedly large sample volumes. Furthermore, the
results from this deep-learning based optical model will be used to facilitate object
recognition in the high-resolution EDOF images. Altogether, SuperPI will represent a
significant step forward in enhanced-DOF imaging that will not only advance plankton
imaging technology, but also benefit a wide range of other applications in optical microscopy.
Start
Januar 2022
Ende
Dezember 2024
Bewilligungssumme (gesamt)
200000
Bewilligungssumme (GEOMAR)
133000
Zuwendungsgeber / Programm
-
Helmholtz-Gemeinschaft
/ Helmholtz AI - Artificial Intelligence Cooperation Unit (Impuls- und Vernetzungsfonds)
Koordination
Helmholtz-Zentrum für Ozeanforschung Kiel (GEOMAR), Germany
Kontakt
Partner
Helmholtz-Zentrum München ( HMGU), Germany
