Imaging through diffuse media using multi-mode vortex beams and deep learning

Area of Science:

• Optics
• Photonics
• Machine Learning

Background:

• Optical imaging through diffuse media is hindered by photon scattering, degrading image quality.
• Applications include biomedical imaging, non-destructive testing, and computer-assisted surgery.
• Existing methods struggle with severe scattering, limiting practical use.

Purpose of the Study:

• To develop a novel method for enhanced optical imaging through diffuse media.
• To investigate the efficacy of using multiple modes of vortex beams.
• To introduce and validate a new deep learning network, LGDiffNet, for image reconstruction.

Main Methods:

• Utilized multiple modes of Gaussian and Laguerre-Gaussian (vortex) beams to illuminate a dataset.
• Propagated beams through a scattering medium (diffuser).
• Employed the custom deep learning network, LGDiffNet, for image reconstruction.

Main Results:

• Vortex beams and LGDiffNet demonstrated superior image reconstruction compared to conventional methods.
• Achieved a ~1 dB enhancement in Peak Signal-to-Noise Ratio (PSNR) with a highly scattering diffuser.
• The system showed robustness without additional optimizations or reference beams.

Conclusions:

• The proposed method using vortex beams and LGDiffNet significantly enhances imaging capability through diffuse media.
• LGDiffNet demonstrates robustness and adaptability for practical applications, particularly in medical imaging.
• This approach offers a promising solution for overcoming scattering challenges in optical imaging.