Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Video

Updated: Jun 12, 2026

End-To-End Deep Neural Network for Salient Object Detection in Complex Environments
03:31

End-To-End Deep Neural Network for Salient Object Detection in Complex Environments

Published on: December 15, 2023

Optical model compression learning for galaxy morphology classification with diffractive deep neural networks.

Zongyu Lu, Jinye Li, Mingxuan Li

    Optics Express
    |June 11, 2026
    PubMed
    Summary
    This summary is machine-generated.

    Related Concept Videos

    You might also read

    Related Articles

    Articles linked to this work by shared authors, journal, and citation graph.

    Sort by
    Same author

    SiN<sub><i>x</i></sub> Waveguide-Integrated Carbon Nanotube Photodetector with Bandwidth over 67 GHz Operating at 1.55 and 2 μm Band.

    ACS nano·2026
    Same author

    Cross-domain knowledge distillation for integrated optical neural networks: toward robust and compact learning with pseudo-zero-shot capability.

    Optics express·2025
    Same author

    Broadband shortwave infrared GeSn light-emitting diode at room temperature fabricated by exploring the Sn content gradient method.

    Optics letters·2025
    Same author

    Optimizing optical neural network design for enhanced compatibility with analog computation.

    Optics express·2025
    Same author

    High-speed waveguide lateral Ge/Si avalanche photodetector for C-band and L-band.

    Optics express·2024
    Same author

    Lateral GeSn p-i-n photodetectors on insulator prepared by the rapid melting growth method.

    Optics letters·2024
    Same journal

    Denoising algorithm of Φ-OTDR systems based on adaptive fractional wavelet transform denoising.

    Optics express·2026
    Same journal

    Millisecond photon-to-photon latency and high-speed volumetric projection system for optogenetics.

    Optics express·2026
    Same journal

    Polarization-encoded coaxial structured light for high-precision 3D surface profilometry.

    Optics express·2026
    Same journal

    Discrete freeform optical design based on collaborative optimization of point cloud and local normals.

    Optics express·2026
    Same journal

    Ultrafast ghost imaging with 25 GHz speckle switching and wavelength-division multiplexing.

    Optics express·2026
    Same journal

    Atomic vapor cells fabricated by femtosecond laser welding of standard-optical-quality glass.

    Optics express·2026
    See all related articles

    Optical computing using diffractive deep neural networks (D2NNs) can analyze galaxy images efficiently. A new optical model compression learning framework improves shallow D2NNs, enhancing galaxy classification accuracy and robustness for astronomical data analysis.

    Area of Science:

    • Astronomy and Astrophysics
    • Computer Science
    • Optical Computing

    Background:

    • Galaxy morphology is crucial for understanding cosmic evolution.
    • Large astronomical surveys generate vast datasets, posing challenges for energy-efficient processing.
    • Diffractive deep neural networks (D2NNs) offer high-throughput, low-power optical computing but face limitations with depth (attenuation) and shallow networks (representation).

    Purpose of the Study:

    • To address the trade-off between depth and representation in D2NNs for astronomical data analysis.
    • To develop an optical model compression learning (OMCL) framework for D2NNs.
    • To improve the performance and practicality of optical neural networks for large-scale galaxy surveys.

    Main Methods:

    • Extended knowledge distillation with optical-domain considerations to create the OMCL framework.

    More Related Videos

    Lensless Fluorescent Microscopy on a Chip
    11:23

    Lensless Fluorescent Microscopy on a Chip

    Published on: August 17, 2011

    Related Experiment Videos

    Last Updated: Jun 12, 2026

    End-To-End Deep Neural Network for Salient Object Detection in Complex Environments
    03:31

    End-To-End Deep Neural Network for Salient Object Detection in Complex Environments

    Published on: December 15, 2023

    Lensless Fluorescent Microscopy on a Chip
    11:23

    Lensless Fluorescent Microscopy on a Chip

    Published on: August 17, 2011

  • Applied OMCL to compress deep D2NNs into shallower models.
  • Conducted numerical simulations using galaxy images from the Sloan Digital Sky Survey (SDSS).

    • Main Results:

    • The compressed D2NN model achieved 69.25% classification accuracy on SDSS galaxy images, a 27.5% improvement over conventional training.
    • The model demonstrated partial recognition of unseen galaxy categories.
    • Compressed models exhibited smoother, sparser phase patterns, enhancing noise robustness and tolerance to low-precision quantization (3-bit).

    Conclusions:

    • Transferring optical-domain features from deeper networks effectively mitigates limitations in shallow D2NNs.
    • The OMCL framework provides a practical approach for developing fabrication-friendly, high-performance optical neural networks.
    • This research paves the way for efficient optical analysis of large astronomical datasets.