Abstract
Abstract
Multimode fibers (MMFs) offer promising prospects in applications such as medical endoscopic imaging owing to their compact footprints. However, modal coupling and scattering effects during propagation severely constrain their practical deployment. Although existing approaches—such as electronic deep learning and diffractive deep neural networks—partially address this issue, they rely heavily on electronic computation or static 3D-printed diffractive layers, suffering from high hardware dependency and limited operational flexibility. In contrast, multi-plane light conversion (MPLC) technology efficiently transforms spatial light fields in free space and enables dynamic mode control via spatial light modulators, providing an effective approach to optical inverse problems with high adaptability across diverse application scenarios. To address these challenges, we propose an MPLC-based image restoration method for MMF transmission, where cascaded phase masks are constructed via a wavefront matching technique to achieve direct, high-efficiency all-optical inverse scattering for image recovery. During numerical modeling, an adaptive weight adjustment mechanism for mask updating and a mean structural similarity index (MSSIM) threshold module were integrated to enhance system stability and recovery performance. Furthermore, phase mask noise, propagation channel noise, and photodetector noise were incorporated into the image recovery process to realistically reflect external perturbations during transmission. Simulation results indicate that for handwritten digit images from the test set transmitted through a 1 m-long MMF, the restored light field achieves a low insertion loss down to 1 dB, an MSSIM of
0.72
±
0.04
, and a reduced SSIM standard deviation of 0.044. The proposed architecture provides an effective all-optical solution for MMF image recovery, holding significant potential for applications such as medical endoscopic imaging.