Single-pixel image classification via optical compression encoding and all-optical diffraction decoding.

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Computer Vision

Background:

• Single-pixel sensing offers efficient perception by bypassing image reconstruction.
• Traditional electrical neural network methods for single-pixel sensing face latency and high power consumption issues.

Purpose of the Study:

• To propose and demonstrate a single-pixel image classification framework using optical compressive encoding and all-optical diffraction decoding.
• To enhance perception efficiency and address privacy concerns in single-pixel sensing.

Main Methods:

• Developed an architecture combining compressive sensing with a diffraction optical neural network (DONN).
• Implemented all-optical compressive feature computation and diffractive decoding for classification.
• Integrated compressed encoding to overcome modulation limitations in conventional DONNs.

Main Results:

• The proposed framework demonstrated superior performance compared to traditional DONNs, especially in noisy or low-light environments.
• Achieved classification outcomes directly from optically computed compressed features using a single diffractive layer.
• Showcased strong privacy protection by transmitting compressed features in the optical domain.

Conclusions:

• The optical compressive encoding and all-optical diffraction decoding framework offers an efficient and private solution for single-pixel sensing.
• This method holds significant potential for privacy-sensitive applications such as medical imaging and biometric authentication.
• The integration of compressive sensing with DONNs overcomes inherent limitations, paving the way for advanced optical perception systems.