A binary sampling Res2net reconstruction network for single-pixel imaging.

Area of Science:

• Computer Vision
• Signal Processing
• Machine Learning

Background:

• Traditional compressive reconstruction algorithms exhibit high computational complexity, limiting their practical application.
• Deep learning networks offer a promising approach to accelerate the reconstruction process in compressive sensing.

Purpose of the Study:

• To develop a novel deep learning-based single-pixel imaging system.
• To design an efficient reconstruction network, Bsr2-Net, optimized for binary matrix sampling.

Main Methods:

• Implementation of a single-pixel imaging system leveraging deep learning.
• Development of the binary sampling Res2Net reconstruction network (Bsr2-Net).
• Comparative experimental analysis of Bsr2-Net against other reconstruction methods using metrics like structural similarity, peak signal-to-noise ratio, and reconstruction time.

Main Results:

• Bsr2-Net demonstrates superior performance compared to several recently reported deep learning networks.
• The proposed Bsr2-Net achieves reconstruction quality close to state-of-the-art algorithms.
• Significant reduction in reconstruction time was observed with the Bsr2-Net model.

Conclusions:

• The Bsr2-Net model provides an effective and efficient solution for compressive sensing reconstruction.
• Deep learning, particularly the Bsr2-Net architecture, significantly enhances the speed and quality of single-pixel imaging reconstruction.
• This work advances the field of compressive sensing by offering a computationally efficient and high-performance reconstruction method.