OAM-basis underwater single-pixel imaging based on deep learning at a low sampling rate.

Area of Science:

• Optics and Photonics
• Computer Vision
• Artificial Intelligence

Background:

• Underwater imaging faces challenges from low sampling rates and high turbidity.
• Existing methods struggle to reconstruct clear images in adverse aquatic conditions.

Purpose of the Study:

• To develop a novel underwater single-pixel imaging technique.
• To enhance image reconstruction quality and robustness in turbid waters.

Main Methods:

• Utilized an orbital angular momentum (OAM) basis as the sampling scheme.
• Employed a dual-attention residual U-Net generative adversarial network (DARU-GAN) for image reconstruction.
• Tested performance under a 3.125% sampling rate and 128 NTU turbidity.

Main Results:

• Successfully restored underwater target images with high fidelity.
• Demonstrated robust generalization capabilities in challenging environments.
• Achieved superior reconstruction quality compared to conventional methods.

Conclusions:

• The OAM-basis sampling with DARU-GAN is an effective solution for high-turbid underwater imaging.
• This integrated approach overcomes limitations of low sampling rates and water turbidity.
• Presents a promising advancement for underwater optical sensing and imaging applications.