Related Experiment Video
Updated: Sep 22, 2025
09:43
Transmission of Multiple Signals through an Optical Fiber Using Wavefront Shaping
Published on: March 20, 2017
10.0K
Classification and reconstruction of spatially overlapping phase images using diffractive optical networks.
Deniz Mengu1,2,3, Muhammed Veli1,2,3, Yair Rivenson1,2,3
1Electrical and Computer Engineering Department, University of California, Los Angeles, Los Angeles, CA, 90095, USA.
Scientific Reports
|May 19, 2022
Summary
Diffractive optical networks can now classify and reconstruct overlapping phase images using deep learning and wave optics. This technology achieves over 85.8% accuracy for handwritten digits, enabling new computational imaging applications.
Area of Science:
- Optics and Photonics
- Machine Learning
- Computational Imaging
Background:
- Diffractive optical networks integrate wave optics and deep learning for all-optical computation.
- Classifying and reconstructing spatially overlapping, phase-encoded objects is a complex challenge due to phase ambiguity.
Purpose of the Study:
- To design and demonstrate diffractive optical networks for all-optical classification and reconstruction of overlapping phase images.
- To achieve high accuracy in classifying phase-encoded handwritten digits from the MNIST dataset.
Main Methods:
- Utilizing passive diffractive optical networks with successive transmissive layers.
- Training the network with approximately 550 million unique combinations of phase-encoded handwritten digits.
- Employing a shallow electronic neural network with compressed diffractive optical network output for phase image reconstruction.
Main Results:
- Achieved >85.8% accuracy in blind testing for all-optical classification of two overlapping phase images.
- Demonstrated simultaneous classification of two different, randomly selected, spatially overlapping phase images.
- Successfully reconstructed phase images using a compressed representation, reducing data by 20-65 times.
Conclusions:
- Diffractive optical networks offer a powerful framework for all-optical classification and reconstruction of complex phase-encoded objects.
- The developed method overcomes challenges associated with spatial overlap and phase ambiguity.
- Potential applications include advanced computational imaging, microscopy, and quantitative phase imaging.
