Lightweight unfolding reconstruction for scalable single-pixel imaging

Single-pixel imaging (SPI) at large scale is limited by the computational burden of reconstruction. We introduce a physics-guided lightweight unfolding framework (∼2.1 × 10⁵ parameters) combined with Kronecker compressive sensing-based partitioned reduction. The method reduces computational dimension while maintaining physical interpretability. Simulations demonstrate 1024 × 1024-pixel reconstruction at sampling ratios as low as ∼1.5%, and experiments confirm high-quality recovery at 1024 × 768 pixels. Owing to its scalability and efficiency, the approach enables practical high-resolution SPI on resource-limited hardware.