Abstract:Application of machine learning techniques on fiber speckle images to infer fiber deformation allows the use of an unmodified multimode fiber to act as a shape sensor. This approach eliminates the need for complex fiber design or construction (e.g., Bragg gratings and time-of-flight). Prior work in shape determination using neural networks trained on a finite number of possible fiber shapes (formulated as a classification task), or trained on a few continuous degrees of freedom, has been limited to reconstruction of fiber shapes only one bend at a time. Furthermore, generalization to shapes that were not used in training is challenging. Our innovative approach improves generalization capabilities, using computer vision-assisted parameterization of the actual fiber shape to provide a ground truth, and multiple specklegrams per fiber shape obtained by controlling the input field. Results from experimenting with several neural network architectures, shape parameterization, number of inputs, and specklegram resolution show that fiber shapes with multiple bends can be accurately predicted. Our approach is able to generalize to new shapes that were not in the training set. This approach of end-to-end training on parameterized ground truth opens new avenues for fiber-optic sensor applications. We publish the datasets used for training and validation, as well as an out-of-distribution (OOD) test set, and encourage interested readers to access these datasets for their own model development.

Bending recognition based on the analysis of fiber specklegrams using deep learning

Deep learning based optical curvature sensor through specklegram detection of multimode fiber

Learning to sense three-dimensional shape deformation of a single multimode fiber

Averaging Speckle Patterns to Improve the Robustness of Compressive Multimode Fiber Imaging Against Fiber Bend.

Multipoint and Multi-physical Quantity Sensing by Multimode Fiber with Demodulation of Deep Learning

Fiber Bending Sensor with Turning Point in a Multimode Fiber Peanut-Like Structure

Flexible single multimode fiber imaging using white LED

Highly Sensitive Vector Bending Sensor Based on an Embedded Multimode D-shaped LPFG.

Deep Learning-Based Simultaneous Temperature- and Curvature-Sensitive Scatterplot Recognition

High-Performance Bending Sensor Based on Fiber Microspheres Array Michelson Interferometer

Bend Sensors with Direction Recognition Based on Long-Period Gratings Written in D-shaped Fiber

Bend-insensitive Distributed Sensing in Singlemode-Multimode-singlemode Optical Fiber Structure by Using Brillouin Optical Time-Domain Analysis

Femtosecond Laser Inscription of Fiber Bragg Grating in Twin-Core Few-Mode Fiber for Directional Bend Sensing

Pure Directional Bending Measurement With a Fiber Bragg Grating at the Connection Joint of Eccentric-Core and Single-Mode Fibers

Sensitivity-enhanced two-dimensional bending sensor based on single-mode fiber measurement

Robust real-time imaging through flexible multimode fibers

Directional Bending Sensor Based on a Dual Side-Hole Fiber Mach–Zehnder Interferometer

Fiber-Optic Shape Sensing Using Neural Networks Operating on Multispecklegrams

Image transmission through a flexible multimode fiber by deep learning

Deep Learning-Enabled Two-Directional Stretchable Strain Sensor Based on a Single Multimode Fiber

Bend-tolerant fiber sensor based on BOTDR system