A multiple events recognition scheme based on improved feature vectors for fiber optic perimeter security system

Abstract

In this paper, we propose an improved feature extraction based multiple events recognition scheme for fiber optic perimeter security system. In the scheme, four common types of security sensing events, namely, background noises, waggling the fence, cutting the fence and climbing the fence are collected based on a dual Mach-Zehnder interferometry vibration sensor. Variational mode decomposition in frequency domain, sample entropy in irregularity and zero crossing rate in time domain are considered as the feature description of the given security sensing events. A series of experiments have been implemented by a radial basis foundation neural network, which shows that the proposed recognition scheme can accurately discriminate the three kinds of man-made intrusions from the background noises. The average identification rates of 98.42% and 100% are achieved for the three types of intrusions and background noises, respectively, which can fully satisfy the field application requirements, the recognition response time is also good of real time performance, which can be controlled less than 1.6 s. Therefore, the proposed events recognition scheme can provide a quite promising field application prospect in the fiber optic perimeter security system.