Deep Learning-Based Fiber Bending Recognition for Sensor Applications

• Published in: IEEE sensors journal Vol. 23; no. 7; pp. 6956 - 6962
• Main Authors: Bender, Deniz, Cakir, Ugur, Yuce, Emre
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.04.2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Accuracy; Bending; Convolutional neural networks; Curvature; curvature sensor; Deep learning; deep learning (DL); multimode fiber (MMF) specklegram sensing; Optical fiber networks; Optical fiber sensors; Optical fibers; residual network (ResNet); Sensors; Speckle
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2023.3249049

The sensitivity of multimode fibers (MMFs) to mechanical deformations has led to their widespread use in various fields, such as structural monitoring and healthcare. However, traditional optical fiber sensing techniques often involve complex equipment and analysis procedures. In this work, we demonstrate the use of deep learning (DL) to accurately detect both the curvature and location of a bent MMF under external force. The DL model is trained using intensity-only speckle images as input, which corresponds to the bending curvature and location. Our results show that the network can detect the bending location with an accuracy of 1.39 cm and the curvature with an accuracy of 0.158 <inline-formula> <tex-math notation="LaTeX">\text{m}^{-{1}} </tex-math></inline-formula>.