Event Discrimination of Fiber Disturbance Based on Filter Bank in DMZI Sensing System

To meet the growing demand of event discrimination in a dual Mach-Zehnder Interferometry (DMZI) vibration sensing system, this paper proposes a novel scheme of distinguishing invasion events. This scheme consists of three stages: endpoint detection, filter-bank-based feature extraction, and radial-basis-function neural network classification. As the hard core, the proposed filter bank, which is derived from the classical frequency-sampling-based filter design method, greatly suppresses the interchannel interference and, thus, provides accurate feature description for the radial basis function (RBF)-based classifier. Moreover, we also derive the closed-form formula of the filter co-efficients and, thus, design a pipeline structure for the filter bank, which brings the advantages of high accuracy, high flexibility, great rapidity, and low cost. Simulation verifies the proposed filter bank's superiority in separating different frequency bands. Field experiments also show that the proposed event discrimination scheme cannot only eliminate some false alarms caused by noninvasion events but can discriminate two common invasion events (climbing the fence and knocking the cable) with high recognition accuracy as well.