High-fidelity imaging through a perturbed multimode fiber using a u-architecture network with fully connected layers

The multimode fiber transmission is highly sensitive to perturbations, which imposes many limitations on its application in endoscopic imaging. Here, a u-architecture network with fully connected layers (FC-UANET) is designed to improve robustness to perturbations, which can efficiently extract hidden image information and fiber characteristics from local and global features of speckle patterns, thereby achieving high-definition imaging through a dynamically perturbed multimode fiber. In the comparative experiment, the FC-UANET demonstrated the highest reconstruction accuracy. Further research indicated that input images can be accurately reconstructed from local speckle patterns, which can significantly improve imaging speed and reduce computational complexity. Meanwhile, the FC-UANET exhibited generalization ability to cross-domain datasets, which is beneficial for endoscopic imaging that has difficulty obtaining large-scale datasets. For endoscopes, in most scenarios, the deformation of transmission media is inevitable, and the self-designed FC-UANET exhibits high robustness to fiber deformations caused by strong mechanical perturbations, which is expected to provide a feasible solution for the application of multimode fibers in endoscopes.