Monolayer directional metasurface for all-optical image classifier doublet

Rui Xia , Lin Wu , Jin Tao

Optics Letters +

|

May 1, 2024

View abstract on PubMed

Summary

This study introduces a single-layer metasurface to improve diffractive deep neural networks for image classification. This approach enhances spatial efficiency and simplifies alignment for better performance in applications like target recognition.

Area Of Science

Optics and Photonics
Artificial Intelligence
Materials Science

Background

Diffractive deep neural networks (DDNNs) offer passive, scalable, and efficient computation for tasks like holographic imaging and object classification.
Previous DDNN implementations faced limitations due to spatial size and alignment challenges.

Purpose Of The Study

To address spatial constraints and alignment issues in diffractive deep neural networks.
To introduce a novel monolayer directional metasurface for improved DDNN performance.

Main Methods

A monolayer directional metasurface was designed and fabricated.
The metasurface was integrated with diffractive deep neural networks.
MNIST datasets were used to train and test the metasurface-based DDNN for digital image classification.

Main Results

The metasurface-based DDNN achieved excellent digital image classification results.
Classification accuracy reached 84.73% for ideal phase mask plates and 84.85% for the metasurface.
The single-layer metasurface demonstrated improved spatial utilization efficiency.

Conclusions

The monolayer directional metasurface effectively reduces spatial constraints and mitigates alignment issues in DDNNs.
This approach offers a more practical and efficient solution for implementing DDNNs in real-world applications.
The metasurface-based DDNN shows significant potential for advanced optical computing and AI tasks.