Single-pixel imaging method based on equivariant plug-and-play

Single-pixel imaging is an emerging imaging technology that decomposes the imaging process into modulation and reconstruction processes. Inspired by equivariant imaging, this study incorporates the theory of equivariant imaging into the plug-and-play-based single-pixel imaging framework, proposing equivariant plug-and-play-based single-pixel imaging (EPnP-SPI). By enforcing equivariance on a denoiser, the transformation consistency inherent in single-pixel imaging is introduced into the deep neural network to constrain algorithmic iterations. The key feature of EPnP-SPI lies in its ability to provide additional physical constraints for network iterations through transformation consistency, enabling the network to learn feature information beyond existing data pairs. Both simulation and optical experimental results demonstrate the superior performance of the proposed method at low sampling rates.