JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 33, NO. 23, DECEMBER 1, 2015 4885
A High-Efficiency Multiple Events Discrimination
Method in Optical Fiber Perimeter Security System
Kun Liu, Miao Tian, Tiegen Liu, Junfeng Jiang, Zhenyang Ding, Qinnan Chen, Chunyu Ma,
Chang He, Haofeng Hu, and Xuezhi Zhang
Abstract—This paper proposes an integrated scheme to distin-
guish invasive events in optical fiber dual Mach–Zehnder Interfer-
ometry based perimeter security system. This algorithm combined
empirical mode decomposition, kurtosis characteristics with radial
basis function neural network, which can improve the recognition
rate of event discrimination and increase the variety of intrusion
events. Several experiments demonstrate that the proposed scheme
can discriminate four common invasive events (climbing the fence,
knocking the cable, cutting the fence, and waggling the fence) with
an average recognition rate above 85.75%, which can satisfy actual
application requirements.
Index Terms—Feature extraction, optical fibers, pattern
recognition, signal analysis.
I. INTRODUCTION
D
UAL Mach–Zehnder interferometry (DMZI) vibration
system has received increasingattention for high-accuracy
event detection, emerging in many applications of submarine ca-
ble security, pipeline leakage detection, perimeter security, etc.
[1]–[8]. The typical DMZI system adopts the phase modulate
fiber sensing technique, thus it has the advantages of high sen-
sitivity and fast response. The critical issues in improving the
efficiency of the DMZI system are feature description of invasive
signals and pattern recognition for accurately discriminating the
type of invasive events. Many attempts have been made toward
improving the DMZI. One study by Vries et al. [9] proposed
a method of intrusion classification in perimeter system based
on acoustic. The system employed a neural network classifier
with frequency domain features which can detect intrusion, such
as climbing, cutting and jumping around fences. However, the
system performance decayed when the quality of the sound (or
Manuscript received July 30, 2015; revised September 27, 2015 and October
19, 2015; accepted October 20, 2015. Date of publication October 25, 2015;
date of current version November 7, 2015. This work was supported in part
by the National Natural Science Foundation of China under Grant 61475114,
61405139, 61227011, 61378043, and 61505138, in part by National Instrument
Program under Grant 2013YQ030915, in part by the National Basic Research
Program of China under Grant 2010CB327806, in part by the Tianjin Science
and Technology Support Key Project under Grant 11ZCKFGX01900. K. Liu
and M. Tian contributed equally to this work. (Corresponding authors: Junfeng
Jiang and Tiegen Liu).
The authors are with the College of Precision Instrument & Opto-electronics
Engineering, Tianjin University, Tianjin 300072, China, and also with the Key
Laboratory of Opto-electronics Information Technology, Ministry of Educa-
tion, Tianjin 300072, China (e-mail: beiyangkl@tju.edu.cn; 675515611@qq.
com; tgliu@tju.edu.cn; jiangjfjxu@163.com).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/JLT.2015.2494158
signal to noise ratio) generated by the intruders and surrounding
environment decreased. Moreover, in order to locate the suspect,
this system requires more than one sensor which makes the sys-
tem complex and very expensive. Another study by Yousefi et al.
[10] reported a fence breach detection system that was based on
utilization of a three-axis accelerometer and a RISC micropro-
cessor. This system is capable of recognizing whether the breach
was due to rattling caused by strong wind or a person climbing
on the fence based on the change in energy of the signal through
two band-pass filters. However, this method can only distinguish
the two specified types but not events with similarities. In ad-
dition to these studies, Mahmoud et al. [11] proposed a feature
extraction method based on level crossing features in the time
domain, which combined with artificial neural networks [12]
to realize the pattern recognition. Unfortunately the threshold
is selected without an accurate basis, leading to deviation of
feature extraction.
In order to solve problems such as low recognition rate and
limit the kinds of intrusion events which exist in current event
discrimination, this paper proposes an integrated scheme of
event discrimination which consists of empirical mode decom-
position (EMD) and feature extraction based on the kurtosis and
radial basis function (RBF) neural network classification. The
EMD method was first proposed in 1998 by Huang et al. [13],
after which it was successfully applied to aerospace, biomed-
ical and other natural sciences with its excellent performance
in processing non-stationary and non-linear signal [14]–[16].
Kurtosis [17] is very sensitive to pulse signal and kurtosis fea-
ture is extracted in different intrinsic mode function (IMF) do-
mains. RBF neural network [18], [19] has many advantages
such as high convergence, global optimum approximation, ex-
cellent self-adaption and most importantly, approximation of
any nonlinear function.
By using the methods highlighted above, this proposed in-
tegrated scheme can be carried out in three steps. First, the
invasive signals are decomposed into a collection of IMF as a
pre-processing step based on EMD. Second, the kurtosis charac-
teristics are extracted from the IMFs which contain the necessary
information. Finally, the RBF neural network classification is
used to obtain the pattern recognition. To the best of our knowl-
edge, this is the first report on discriminating the four intrusion
events of dual DMZI system in practice (i.e., climbing the fence,
knocking the cable, cutting the fence and waggling the fence).
Experimental results also verify that a higher discrimination rate
and the ability to discriminate concrete types can be achieved,
which leads to a low false alarm rate of the system. Therefore,
it can satisfy the actual application requirements.
0733-8724 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications
standards/publications/rights/index.html for more information.