All-phase Filtering Based Intrusion Recognition Algorithm Research In Dual Mach-Zehnder Interference Perimeter Security System

In recent years, the demand of event detection emerges widely in many application fields such as submarine cable security, pipeline leakage detection, and perimeter security. The dual Mach-Zehnder interference(DMZI) phase modulation sensor possesses the superiority of high sensitivity and fast response, and thus receives increasing attention. Nowadays, DMZI system has been successfully employed in event alarm and event positioning. However, the existing schemes of classifying and recognizing invasion events still need to be improved. Therefore, it is urgent to develop a high accuracy, low cost, high flexibility, and real-time event invasion discrimination method based on DMZI vibration system, which requires the fusion of new signal processing theory and other disciplines, such as pattern recognition.This paper proposes a novel scheme to distinguish invasion events, based on all-phase filtering theory in DMZI vibration system, which consists of 3 stages: endpoint detection, feature extraction and pattern recognition. The first two stages incorpoate the original all-phase filtering theory as the core component. During the endpoint detection stage, for the purpose of detecting the vibration starting point accurately, some parameters(such as boundary frequency) related to the surrounding undisturbed signal's low frequency spectrum distribution characteristics are extracted through FFT spectral analysis, which can be used to design one all-phase high-pass filter. During the feature extraction stage, we utilize the characteristics of all-phase filter bank, such as dividing frequency band easily, small passband fluctuation, and little interference between different passbands, to realize an accurate feature description of the invasion evens. Moreover, we also derive the closed-form formula of the filter coefficients and design one pipeline structure for the filter bank, which brings the merits of high accuracy, high flexibility, great rapidity and low cost, thus the filter bank has good applicability. During the pattern recognition stage, we just need to feed the feature extracted by the proposed all-phase filter bank into a radial basis function(RBF) neural network classification to accomplish the invasion event identification. Considering that all-phase filtering theory is the core of the proposed scheme, this paper also elaborates the derivation from classical frequency sampling method to all-phase filter bank and gives simulation experiments to vertify its theoretic characteristics.Field experiments show that, by properly setting the frequency vector, the proposed event discrimination scheme not only can eliminate a lot of false alarms caused by non-invasion events such as rain, wind, non-contact perimeter vibration, but also can discriminate two common invasion events(climbing the fence and kicking the cable) with a high recognition rate, thus it may possess high academic and engineering value.