A Machine Vision-Based Fiber Profile Image Recognition Method for Alignment of FBG Inscribing

Abstract

The axial alignment of fiber core before fiber Bragg grating (FBG) inscription is time-consuming and laborious with naked eye, which requires autonomous alignment technology urgently. The image recognition and correction of optical fiber profiles are the primary breakthrough point and has been elevated to a more important position. This article employed a coaxial imaging device configured with an FBG inscribing system to obtain optical fiber images and proposed image recognition for alignment of FBG inscribing based on machine vision. First, a global image tilt detection algorithm based on improved Radon algorithm was proposed to detect horizontal tilt angle of fiber, and then, adaptive moment estimation (ADAM)-optimized U-Net was proposed to accurately segment the fiber core, achieving pixel accuracy of 98.82%. Finally, the coordinates of the midpoint of the fiber core were provided. Through this research, the strong technical support will be provided for the high flexibility, stability, and efficiency of FBG inscription, paving the road for the research of FBG automated inscription, and further serving the application of fiber optic sensing in a wider range of scenarios.