Hyper-spectral imaging through a multi-mode fiber

TL;DR

This paper demonstrates that deep learning can effectively reconstruct hyper-spectral images from speckle patterns generated by multi-mode fibers, enabling high-resolution imaging with reduced computational effort.

Contribution

The study introduces a deep learning approach to recover hyper-spectral images from speckle patterns, achieving accurate reconstruction with limited data and reducing computational load.

Findings

Successfully identified 26 wavelengths and letters from speckle patterns

Reconstructed complete images using only partial speckle data

Potential applications in biomedical imaging and photonic computing

Abstract

Multi-mode fibers provide an increased amount of data transfer rates given a large number of transmission modes. Unfortunately, the increased number of modes in a multi-mode fiber hinders the accurate transfer of information due to interference of these modes which results in a random speckle pattern. The complexity of the system impedes the analytical expression of the system thereby the information is lost. However, deep learning algorithms can be used to recover the information efficiently. In this study, we utilize deep learning architecture to reconstruct input colored images from the output speckle patterns at telecommunication wavelength (C-band). Our model successfully identifies hyper-spectral speckle patterns at twenty-six separate wavelengths and twenty-six distinct letters. Remarkably, we can reconstruct the complete input images only by analyzing a small portion of the output speckle pattern. Thereby, we manage to decrease the computational load without sacrificing the accuracy of the classification. We believe that this study will show a transformative impact in many fields: biomedical imaging, communication, sensing, and photonic computing.