Identifying mode coupling wavelengths in doubly-clad optical fibers with deep learning

Abstract

Understanding the transmission of light waves in optical fibers and accurately determining the locations of mode coupling are crucial for enhancing the efficiency of optical devices and advancing innovative technologies such as fiber optic sensors, lasers, and modulators. This study utilizes deep learning and image recognition techniques to identify the wavelengths at which mode coupling occurs in optical fibers. Our research findings show that using the ResNet-18 model allows for the rapid and accurate identification of the wavelengths at which mode coupling occurs in optical fibers, as well as the modes involved, achieving an accuracy close to 100 %. We experimented with sampling the dataset at 5 nm and 10 nm intervals to create smaller training and validation sets. Despite the reduced data volume, high accuracy rates were maintained, exceeding 99 % and 97 % respectively. This study provides new insights into the use of deep learning for precise localization of mode coupling points and tracking of transmission modes in optical fibers.