A Surveillance System of Fiber-Optic Cables With Multi-Channel DAS and CNN

Abstract

We propose a surveillance system of fiber optic cables with multi-channel distributed acoustic sensing (DAS) interrogator equipped with optical rotary switch. By switching the optical connection between the interrogator and multiple optical fiber-under-test, we take samples of the sensed vibration signal from the fibers with single pulse generator and optical receiver. To identify the source of detected vibration signal, we separated the monitoring process as 2-phases. In the phase-1, we monitor the fibers periodically in a 3 second period, and simply classify the vibration signal as threat or non-threat to the fiber-optic cable route using 1-dimensional convolutional neural network. When the result is threat, the optical switch is fixed to the port and we intentionally monitor the fiber for 10 second to acquire more vibration signal. We adopted 2-D convolutional neural network to classify the source of vibration signal. We trained the two classification models with vibration signal from the deployed fiber-optic cables and the results show the accuracy of 94.5 % for 1-D classification and 98.2 % for 2-D classification.