Peer-reviewed optical-fibre sensor applies ResNet–LSTM inference to two-dimensional underwater flow

Marine roboticsMachine learningPeer reviewed16 August 202672-hour extension

Authors: Haiqing Li, Boxue Sun, Xiaonan Bian, Yujia Wang and Xianping Fu. Status: journal article; performance remains author-reported.

A biomimetic U-shaped plastic optical-fibre sensor uses a cilium-like cantilever and a hybrid ResNet–LSTM. The authors report mean velocity error of 0.0105 m/s and direction error of 1.51°.

Why it matters and impact: The design could support underwater navigation and inspection, but field robustness, fouling, calibration drift, turbulence transfer and independent reproduction remain open.