Adaptive Localization of Multiple Vibrations for Interferometric Optical Fiber Sensing System Using Pulse Identification With Deep Learning

An adaptive pulse detection and identification approach with deep learning (DL) is proposed for multipoint localization in interferometric distributed optical fiber vibration sensing system. In comparison to traditional localization methods, the proposed approach significantly enhances the generalization capability for vibration localization through pulse sequence identification. Localization of multiple simultaneous arbitrary vibrations can be enabled by the proposed approach. The principle of pulse sequences carrying the characteristics of vibration is elucidated. A multimodal feature fusion dual-branch parallel network (MFF-DBPNet) is constructed to detect characteristic changes in subpulses. Experimental verification of vibration signal localization on a 45-km fiber is demonstrated. The results indicate that the localization error for multiple vibrations is less than 15 m. The relative localization error ranges from 0.02% to 0.04% for periodic vibration signals and from 0.02% to 0.07% for transient vibration signals. Furthermore, the generalization ability of the method is validated through variations in the types and frequencies of vibration signals. The results indicate that such variations have a negligible impact on the localization accuracy of the proposed approach.