Non-pre-trained complex-valued neural network-enhanced Fourier single-pixel imaging

View at publisher Summarise with VIRAT
Authors: Hao-Ran Shi, Yifan Liu, Qi Wang
Cited by: 0
References: 34

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Single-pixel imaging (SPI) is a computational imaging architecture that utilizes non-resolution detectors to acquire two-dimensional and three-dimensional image information under structured illumination. This method possesses strong anti-interference capability under low-light conditions and has great potential for applications. However, the imaging quality and reconstruction efficiency of SPI remain significant constraints on its practical applications under low sampling conditions. Here, we propose a non-pre-trained complex-valued network-based compressive Fourier SPI technique to obtain a high-performance reconstructed image at a low sampling rate. This technique utilizes a physical model as prior information to achieve network-driven optimization and thus does not need to be pre-trained on any dataset. The performance is demonstrated simultaneously by simulation analysis and experimental validation, showing a peak signal-to-noise ratio of 23.68 dB at a sampling rate of 0.05 for 64×64pixel images. The proposed complex domain network-based compressive SPI imaging scheme paves the way for practical applications.

Fields of Study & Subjects

Random lasers and scattering media Sparse and Compressive Sensing Techniques Adaptive optics and wavefront sensing

Publication details

Journal / source
Applied Optics
Volume
65
Issue
12
Pages
3944–3944
Publication date
14 April 2026
Language
EN