High performance few-mode fiber-based light field direction sensing system using deep convolutional neural network: fiber speckle demodulation network (FSDNET)

• DOI: 10.1364/oe.524755
• Publication Date: 2024-05-21T07:00:16Z
• Authors: Ya Wen; Xing Zhao; Zhixiang Jiang; Haoran Li; Da Li
• Keywords: Optical time-domain reflectometer

Precisely sensing the light field direction information plays essential role in fields of three-dimensional (3D) imaging, sensing, target positioning and tracking, remote etc. It is thrilling to find that optical fiber can be used as a component due its high sensitivity, compact size, strong resistance electromagnetic interference. According core principle few-mode output speckle pattern sensitive change incident direction, variation characteristics further investigated this research study. Based on simulation analysis transmission characteristics, corresponding with range ±6° horizontally vertically are calculated. Furthermore, deep convolutional neural network (CNN): demodulation (FSDNET) proposed constructed establish what we believe novel way reveal identify mapping relationship between speckle. The theoretical shows mean absolute error (MAE) perceived directions true 0.01°. Then, system based developed. Regarding performance system, MAE FSDNET for have appeared training set 0.0389°, testing unknown not set, 0.0570°. Therefore, experimental results prove achieved by FSDNET.