The Small Deformation Assessment of the Integrated Fiber Optic by the Speckle Field Changes Using Siamese Neural Network

Optical fibers sensors have various advantages for precise measurements of wide spectra of parameters. Currently, the deep-learning usage enhances the morphological analysis of the specklegram images. In this study, the design of the full-length optical fiber integration with firm pigtails fixation was used. The Siamese neural network with four convolutional and two fullyconnected layers per each tower was built for assessment of the deformation. The non-loaded specimen specklegram as the first input and image of the specklegram at the loaded state as the second input were used. The accuracy of the trained network with three sets of hyperparameters ranged from 77% to 82.8%.