Position Characteristics Removing Algorithm for Pattern Recognition in Interferometric DOFS

Abstract

In distributed optical fiber sensing system, pattern recognition is employed to identify vibration signals. This letter introduces a novel algorithm for removing position characteristics to enhance the pattern recognition accuracy based on single core feedback interference system. The detrimental effects of position characteristics on signal identification are analyzed and simulated. Experiments using a hybrid Mach-Zehnder and Sagnac interferometer combined with a convolutional neural network show a classification accuracy of 98.1% across seven types of position-independent vibration signals. This represents a 1.1% improvement over original signals. Focused on accurately classifying vibration signals from various locations, the position characteristics removing algorithm allows classifier trained solely on data collected from a single point, with enhanced accuracy and the ability to generalize across different signal types and locations.