Dense-U-net: Dense encoder–decoder network for holographic imaging of 3D particle fields

Digital holographic imaging is able to reconstruct phase and three-dimensional (3D) information of an object from a one-shot two-dimensional (2D) lensless hologram. A dense encoder–decoder network, called Dense-U-net, is proposed to realize the reconstruction of a 3D particle field. The radii and 3D...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Optics communications Jg. 493; S. 126970
Hauptverfasser: Wu, Yufeng, Wu, Jiachen, Jin, Shangzhong, Cao, Liangcai, Jin, Guofan
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier B.V 15.08.2021
Schlagworte:
Deep learning
Dense-U-net
Digital holography
Image reconstruction
Deep learning
Digital holography
Dense-U-net
Image reconstruction
ISSN:0030-4018, 1873-0310
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Digital holographic imaging is able to reconstruct phase and three-dimensional (3D) information of an object from a one-shot two-dimensional (2D) lensless hologram. A dense encoder–decoder network, called Dense-U-net, is proposed to realize the reconstruction of a 3D particle field. The radii and 3D coordinates of the particles are encoded into a 2D grayscale image as the ground truth. The 2D hologram was served as the input of the network. A Dense_Block is designed and added to the encoder and decoder of the traditional U-net model. Each layer takes the inputs from all previous layers and passes its own feature map to all subsequent layers. Thus, as the diffraction is distributed throughout the entire hologram, both the features with high signal-to-noise ratio at the center of the particle and the diffraction features with low signal-to-noise ratio around the particle can be learned, thereby facilitating the full characterization of the particles. Using the specially designed dense connection module, network training can be accomplished with fewer parameters and faster speeds. •A dense encoder–decoder network is proposed to realize reconstruction of 3D particle field.•The Dense-U-net network needs few training parameters and short training time.•A new way is proposed to generate the dataset for digital holographic networks.
ISSN:0030-4018
1873-0310
DOI:10.1016/j.optcom.2021.126970