Lightweight unfolding reconstruction for scalable single-pixel imaging.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Signal Processing

Background:

• Large-scale single-pixel imaging (SPI) faces significant computational challenges in image reconstruction.
• Existing methods often require substantial computational resources, limiting scalability and practical application.

Purpose of the Study:

• To introduce a novel, computationally efficient framework for large-scale single-pixel imaging.
• To address the computational burden associated with SPI reconstruction.
• To enable high-resolution SPI on resource-limited hardware.

Main Methods:

• Developed a physics-guided lightweight unfolding framework with approximately 2.1 × 10^5 parameters.
• Integrated Kronecker compressive sensing-based partitioned reduction to decrease computational dimensions.
• Maintained physical interpretability throughout the reconstruction process.

Main Results:

• Achieved 1024 × 1024-pixel image reconstruction in simulations at sampling ratios as low as 1.5%.
• Experimentally confirmed high-quality image recovery for 1024 × 768-pixel images.
• Demonstrated significant reduction in computational complexity while preserving image fidelity.

Conclusions:

• The proposed framework offers a scalable and efficient solution for high-resolution single-pixel imaging.
• Enables practical implementation of SPI in environments with limited computational power.
• Advances the feasibility of large-scale, high-resolution imaging applications using SPI technology.