CN107180521A - Optical fiber perimeter security protection intrusion event recognition methods and device based on comprehensive characteristics - Google Patents
Optical fiber perimeter security protection intrusion event recognition methods and device based on comprehensive characteristics Download PDFInfo
- Publication number
- CN107180521A CN107180521A CN201710258029.3A CN201710258029A CN107180521A CN 107180521 A CN107180521 A CN 107180521A CN 201710258029 A CN201710258029 A CN 201710258029A CN 107180521 A CN107180521 A CN 107180521A
- Authority
- CN
- China
- Prior art keywords
- characteristic vector
- signal
- optical fiber
- recognition methods
- intrusion event
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B29/00—Checking or monitoring of signalling or alarm systems; Prevention or correction of operating errors, e.g. preventing unauthorised operation
- G08B29/18—Prevention or correction of operating errors
- G08B29/185—Signal analysis techniques for reducing or preventing false alarms or for enhancing the reliability of the system
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
- G06F3/06—Digital input from, or digital output to, record carriers, e.g. RAID, emulated record carriers or networked record carriers
- G06F3/08—Digital input from, or digital output to, record carriers, e.g. RAID, emulated record carriers or networked record carriers from or to individual record carriers, e.g. punched card, memory card, integrated circuit [IC] card or smart card
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B13/00—Burglar, theft or intruder alarms
- G08B13/02—Mechanical actuation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Abstract
The invention discloses a kind of optical fiber perimeter security protection intrusion event recognition methods based on comprehensive characteristics and device, the recognition methods includes：It will determine that each passage of the signal feed-in all phase DFT filter group of disturbance starting point carries out frequency domain separating treatment, and calculate the normalized power value of each multi-channel output signal, multiple normalized power values are parallel output；The time domain zero-crossing rate generation characteristic vector of combined normalized performance number and whole section of disturbing signal, i.e., when described characteristic vector contains, the aspect information of frequency domain two；It is that invasion action quick high accuracy identification can be achieved by characteristic vector feed-in radial basis function neural network.Described device includes：Analog-to-digital conversion device and DSP devices.The present invention can distinguish four class intrusion events exactly；Compared to existing high-precision intrusion event recognition classifier, DMZI proposed by the invention invades action recognition device and had a clear superiority in terms of operating efficiency.
Description
Technical field
The present invention relates to digital signal processing technique field, more particularly to a kind of optical fiber perimeter security protection based on comprehensive characteristics
Intrusion event recognition methods and device.
Background technology
With optical fiber and the continuous maturation of Fibre Optical Communication Technology, fibre optical sensor is developed rapidly.Based on light
The perimeter security system of fine sensing technology[1][2][3]Also gradually it is taken seriously in safety-security area.With traditional infrared ray[4], electricity
Sub- fence[5]Compared Deng safety-protection system, optical fiber sensing system can to directly touch or be indirectly transferred to optical fiber it is various disturb into
Row monitoring in real time, sensitivity is higher thus more with practical value.It is used as a kind of fibre optical sensor of phase-modulation, double Mach-
Zehnder interferometers (dual Mach-Zehnder interferometer, DMZI) distributed optical fiber sensing system[6][7]
There is high sensitivity and fast response time in terms of detection disturbance event, but also can position in real time.At present, DMZI points
Cloth fibre optical sensor has been widely used for all kinds of safety-security areas[8][9][10][11]。
In all kinds of security protections application, the problem of being badly in need of solving accurately and accurately to recognize intrusion event, because still not having at present
There is a kind of intrusion event recognition methods to take into account the Accuracy and high efficiency of classification, and the solution key of the problem is signal
The design of Processing Algorithm.Particularly, seek to complete end-point detection[12][13]Afterwards, further design a kind of terse, appropriate
Invasion signal characteristic method described, and combine corresponding pattern classification measure, effectively identify that all kinds of invasions are moved
Make[14][15][16]。
Wherein, document [14] is although the different frequencies that its characteristic vector of the wavelet recognition method of proposition can be obtained by multi-level decomposition
The energy of band is characterized[17], but its amount of calculation can become big with the increase of the wavelet decomposition number of plies, in addition, this feature vector is only
Frequency domain character is considered, lacks time-domain information abundant enough, therefore recognition accuracy is relatively low, is only capable of recognizing 3 class intrusion events；
And empirical mode decomposition (Empirical Mode Decomposition, EMD) method that document [16] is proposed is needed
The kurtosis value of the intrinsic mode function (Intrinsic Mode Function, IMF) in decomposable process is asked for one by one, to these
The obtained characteristic vector of kurtosis value combination is classified, you can accurately recognize 4 class common actions, however these IMF need through
Going through the iteration of multiple complexity could obtain, therefore recognition efficiency is not high and influences practicality.
The content of the invention
The invention provides a kind of optical fiber perimeter security protection intrusion event recognition methods based on comprehensive characteristics and device, this hair
It is bright to distinguish four class intrusion events exactly；Compared to existing high-precision intrusion event recognition classifier, the present invention is carried
The DMZI gone out invades action recognition device and had a clear superiority in terms of operating efficiency, described below：
A kind of optical fiber perimeter security protection intrusion event recognition methods based on comprehensive characteristics, the recognition methods includes following step
Suddenly：
It will determine that each passage of the signal feed-in all phase DFT filter group of disturbance starting point carries out frequency domain separating treatment, and
The normalized power value of each multi-channel output signal is calculated, each multi-channel output signal is parallel output；
The time domain zero-crossing rate generation characteristic vector of combined normalized performance number and whole section of disturbing signal, i.e., described feature to
Amount is when containing, the aspect information of frequency domain two；
It is that invasion action quick high accuracy identification can be achieved by characteristic vector feed-in radial basis function neural network.
Wherein, each passage of the signal feed-in all phase DFT filter group that will determine to disturb starting point carries out frequency domain point
From processing, and calculate the normalized power value of each multi-channel output signal and be specially：
The signal parallel for determining disturbance starting point is fed into all phase DFT filter group, all phase DFT filter group includes Q
Individual sub- FIR filter g0,...,gQ-1, calculate filtering output yq(n) Q normalized power value Eq。
Wherein, the time domain zero-crossing rate of the combined normalized performance number and whole section of disturbing signal generates characteristic vector
Step is specially：
Signal x (n) overall zero-crossing rate is calculated, by the value and Q normalized power value EqThe comprehensive length that obtains is done for Q+1
Multi-feature vector F=[E0,E1,...,EQ-1,ZCR]。
Wherein, all phase DFT filter group is specially：
gq(n)=ωc(n)hq(n), q=0 ..., Q-1.
Wherein, gq(n) it is sub- FIR filter coefficient；ωc(n) it is double window convolution window；hq(n) it is filter coefficient；Q is change
Amount, represents q-th of subfilter；Q is subfilter number contained by wave filter group.
Wherein, described is that invasion action quick high accuracy can be achieved to know by characteristic vector feed-in radial basis function neural network
It is not specially：
By the radial basis function neural network for training the characteristic vector feed-in of unmarked classification, after having optimized
Network parameter, calculate output valve Z1,...,ZP, the maximum of output valve is taken to determine affiliated point of current input feature vectors
Class.
Wherein, methods described also includes：
Pass through 4 output valve Z1,...,ZPRealize the accurate identification to climbing, tapping, rock and stealing the event cut.
A kind of identifying device of the optical fiber perimeter security protection intrusion event recognition methods based on comprehensive characteristics, described device bag
Include：Analog-to-digital conversion device and DSP devices,
Signal feeding analog-to-digital conversion device sampling is obtained into sample sequence x (n), enters DSP in the form of parallel data is inputted
Device, concurrently set in filter order N, frequency vector H 0 and 1 number e and m；
Characteristic vector is obtained by the processing of DSP devices, is finally acted by the different invasion of neural network recognization.
The beneficial effect for the technical scheme that the present invention is provided is：
1st, the implementation pattern identification classification in DMZI optical fiber sensing systems, and with the very high degree of accuracy；
By verification experimental verification, the present invention has high accuracy and efficient feature, compared to the pattern-recognition side based on EMD
Method has higher superiority.
2nd, the multi-feature vector simplified is constructed, realization is comprehensively described to all kinds of intrusion events；
Spy of the present invention in combination with the frequency domain and temporal signatures, the comprehensively every class intrusion event of accurate description of intrusion event
Point；Meanwhile, characteristic vector proposed by the invention is extremely simplified, and only just 4 class events accurately can be retouched with 4 elements
State, this lays the foundation for the pattern classification of follow-up high-accuracy high-efficiency rate.
3rd, it can avoid disturbing because of surrounding enviroment false-alarm caused by (rain, blow, the non-property invaded vibration) naturally.
Because, the power that surrounding enviroment are disturbed naturally is mainly distributed on low frequency region, and the present invention is configuring full phase
When the channelized frequencies of position wave filter group are vectorial, low frequency vector element is set to zero, therefore surrounding enviroment are disturbed pair naturally
Final composite character vector is not contributed, so as to avoid the generation of false-alarm phenomenon.
Brief description of the drawings
Fig. 1 is the schematic diagram of DMZI distributed optical fiber sensing systems；
Fig. 2 is DMZI distributed optical fiber sensing system intrusion event identification process figures；
Fig. 3 is intrusion event identifier design flow diagram；
Fig. 4 is wave filter group attenuation curve schematic diagram；
Wherein, (a) is the attenuation curve of wave filter group of the present invention；(b) it is the wave filter group based on classical Frequency Sampling Method
Attenuation curve.
Fig. 5 is RBF neural network structure figure；
Fig. 6 is wave filter group result schematic diagram；
Wherein, (a) is climbing；(b) it is percussion；(c) it is to rock；(d) it is shearing.
Fig. 7 is the averaged feature vector schematic diagram of four kinds of intrusion events；
Wherein, (a) is climbing fence；(b) it is percussion optical cable；(c) it is to rock optical cable；(d) cut for robber.
Fig. 8 implements figure for the hardware of the present invention；
Fig. 9 is DSP internal processes flow graphs.
Table 1 is the experimental precision comparison diagram of two methods；Table 2 is the processing time comparison diagram of two methods.
Embodiment
To make the object, technical solutions and advantages of the present invention clearer, further is made to embodiment of the present invention below
It is described in detail on ground.
Embodiment 1
The embodiment of the present invention is proposed a kind of based on synthesis using double Mach-Zehnder distributed optical fiber sensing systems as background
The intrusion event recognition methods of feature, referring to Fig. 1, this method comprises the following steps：
101：Disturbing signal to input carries out end-point detection, judges disturbance starting point；
102：The signal feed-in all phase DFT filter group of disturbance starting point will be determined[18][19][20]Each passage enter line frequency
Domain separating treatment, and calculate the normalized power value of each multi-channel output signal；
103：Combined normalized performance number and whole section of disturbing signal time domain zero-crossing rate (Zero-Crossing Rate,
ZCR characteristic vector) is generated；
104：By characteristic vector feed-in RBF (Radial basis function, RBF) neutral net[21][22]
Invasion action quick high accuracy identification can be achieved.
In summary, the embodiment of the present invention due to using multi-feature vector simultaneously consider intrusion event frequency domain, when
Domain information, four class intrusion events can be distinguished exactly, compared to existing high-precision intrusion event recognition classifier,
Had a clear superiority in terms of operating efficiency；Using all phase DFT filter group use parallel pipeline by the way of be operated, make
The interchannel interference very little between each subfilter is obtained, possibility is provided for accurate characteristic vector of extracting.
Embodiment 2
The scheme in embodiment 1 is further introduced with reference to specific calculation formula, example, it is as detailed below
Description：
The structural principle of distributed optical fiber sensing system based on DMZI principles is as shown in figure 1, P points are disturbance point, sensing
Cable length is L.The light that laser is sent by after isolator through being equally divided into two-beam line after coupler C1, this two beam
Light is injected in the double Mach-Zehnder interferometers being made up of C2, C3, and two-beam is respectively with clockwise and counterclockwise afterwards
Propagated in sensing loop, and interfered on the coupler (C3 or C2) of opposite end and be output to detector PD1 and PD2
On.Optical signal is changed into electric signal by detector, by corresponding high-speed collection card (Data after stopping direct current
Acquisition, DAQ) collect.Difference according to the actual requirements, capture card DAQ is set to different sample rates.
Particularly, the DAQ1 in Fig. 1 is used for end-point detection and intrusion event is classified, and DAQ2 is positioned for intrusion event.
Finally, needed for by the way that (Industrial Personal Computer, IPC) execution related algorithm is realized in industrial computer
Function (such as end-point detection, furnace-incoming coal and pattern classification).
Generally, the intrusion event identification of DMZI distributed optical fiber sensing systems need to generally undergo as shown in Figure 2
Flow：First, strengthen the quality of fiber-optic vibration signal by the measure of the data prediction such as noise reduction, high-pass filtering and extract event
The time endpoint location of generation；Secondly, huge pretreated sample is rightly described as letter by operation characteristic extraction algorithm
The characteristic vector of short refining；Finally, suitable grader is selected to train and test characteristic vector, the knowledge of output invasion action
Other result.In above step, feature extraction algorithm is the main factor of influence invasion action recognition performance.
201：Initialization and pretreatment；
Fast Fourier analysis (Fast Fourier Transform, FFT) is carried out to multigroup non-intrusive sample action to obtain
To the lower-cut-off frequency f of idle environmente, further use feFor cut-off frequency high-pass filter (High Pass Filter,
HPF endpoint location corresponding with intrusion event in fiber-optic vibration signal) is detected.
In addition, using feWith action upper cut-off frequency fu(statistical spectral analysis can be done to common invasion vibration signal in advance to obtain
To) to Q subfilter g0,...,gQ-1Coefficient configured.
Wherein, above-mentioned initialization and pretreatment the step of it is known to those skilled in the art, the embodiment of the present invention is to this
Do not repeat.
202：Feature extraction；
Pretreated signal x (n) is fed into all phase DFT filter group (comprising Q sub- FIR filters parallel
g0,...,gQ-1), calculate its filtering output yq(n) the Q normalized power value E of (q=0 ..., Q-1)q。
In addition, signal x (n) overall zero-crossing rate (ZCR) is directly calculated, by the value and EqThe comprehensive length that obtains is done for Q+1
Multi-feature vector F=[E0,E1,...,EQ-1,ZCR]。
Wherein, the normalized power value of q-th of output signal can be calculated by following formula and obtained：
Zero-crossing rate can be calculated by following formula and obtained：
Wherein, " sign " represents to take symbol manipulation, is calculated by following formula：
203：RBF neural pattern-recognition.
The multi-feature vector F and corresponding label that feature extraction is obtained (are used to recognize all kinds of actions) feed-in RBF god
It is trained through network, then the network is tested by the characteristic vector feed-in of unknown species invasion signal, and export invasion action
Pattern recognition result.
As shown in figure 3, by all phase DFT filter group, input signal x (n) can be divided into and occupy different frequency passage
Q subsignal y0(n),...,yQ-1(n), and then be convenient to therefrom extract specific features, as subsequent action pattern-recognition according to
According to.Document [18] points out that all phase FIR filter coefficient can be obtained by following 3 easy steps.
1st, it is GENERALIZED DISCRETE LINEAR RANDOM SYSTEM inverse Fourier transform (Inverse for some frequency vector H for meeting H (k)=H (N-k)
Discrete Fourier Transform, IDFT) and obtain vectorial h=[h (n-N+k) ..., h (- 1), h (0), h
(1) ..., h (N-1)], i.e.,
Wherein, h (n) is filter coefficient；H (k) is filter frequencies vector；N is frequency vector length；N is variable；k
For variable.
2nd, by the window f (n) that some length is N, convolution is carried out with the tilt window of itself, generation length is rolled up for 2N-1 double window
Product window, i.e.,
wc(n)=f (n) * f (- n) ,-N+1≤n≤N-1, (2)
Wherein, wc(n) it is double window convolution window；F (- n) is window function.
3rd, by h (n) and wc(n) corresponding element is multiplied, and produces the filter coefficient g (n) that final length is 2N-1.
In above step, the free transmission range of wave filter is realized by changing frequency vector H in step 1.To obtain
The subfilter that Q passband is located at different range is obtained, might as well be by its frequency vector HqThe expectation passband of (q=0 ..., Q-1)
Element value is set to 1, and other elements are set to 0.Specifically, frequency vector HqForm it is as follows：
Wherein, e is resistance band；M is expectation passband width；Q represents q-th of subfilter.
The surrounding enviroment of DMZI systems disturb (the cut-off frequency f of the interference based on low-frequency component during without invasioneIt can pass through
FFT statistical analyses are done to multigroup invasion sample to obtain).Obviously, frequency f ∈ (0, fe) in environmental disturbances to should not be used as action special
Levy, therefore the H of formula (3)qIn low-frequency range there is provided e 0 suppresses to it.
The sampling rate for making DAQ1 is fs, then frequency vector HqThe analog bandwidth that occupies of each element be Δ f=fs/ N,
Therefore numeric parameter e should configure as follows：
E=[fe/ Δ f]=[Nfe/fs] (4)
Wherein, symbol " [] " represents the operation that rounds up, and Δ f is the analog bandwidth that each element is occupied.
So as to the subfilter g corresponding to formula (3)qIdeal passband scope be：
F ∈ [(qm+e) Δ f, (qm+e+m) Δ f], q=0 ..., Q-1. (5)
From formula (3), (5) can be seen that, the passband width of each subfilter is m Δs f, it is assumed that the upper limit of common intrusion event
Frequency is fu(can carry out FFT statistical analyses to multigroup action in advance to obtain), then the pass band width of whole wave filter group is B=fu-
fe, each subfilter passband width is then B/Q, therefore parameter m configurations are as follows：
Further, the H (k) in wushu (1) replaces with the H of formula (3)q(k) it, can obtain its broad sense IDFT result of calculation
hq(n)
Finally, according to step 3, the coefficient that can obtain all sub- FIR filters of Q all phase is：
gq(n)=ωc(n)hq(n), q=0 ..., Q-1. (8)
Derive and can be seen that more than, only need to use formula (4) respectively, (6) determine environmental disturbances parameter e and passband parameter m, by it
Substitution formula (7), (8) can directly obtain all coefficient g of all phase FIR filter group0(n),...,gQ-1(n), i.e., by three steps
Rapid all phase DFT filter design method is reduced to substitute into the realization of the step of analytic formula one, therefore proposed by the present invention for the complete of DMZI systems
The method for parameter configuration of phase filter has taken into account high flexibility and high efficiency.
It is well known that FIR filtering is the continuous convolution process of feedforward, Fig. 3 Q subfilter is actually with parallel stream
Pipeline mode is worked, therefore the feature extraction based on all phase DFT filter group that this method is proposed is compared without any iterative processing
In EMD decomposition methods, operating efficiency is substantially increased.
As it was previously stated, the degree of accuracy of event recognition is determined by the transmission performance of all phase DFT filter.Particularly, to some
For subfilter, it should be tried one's best greatly in adjacent subfilter frequency band attenuation, the coupling between characteristic vector each element could be reduced
Close；In turn, in the case where event recognition precision is certain, the attenuation outside a channel of wave filter is bigger, required subfilter number
Mesh is fewer, and corresponding characteristic vector length is shorter.Document [17] points out, the interpolating function of all phase DFT filter frequency response
Traditional rectangular window Fourier spectrum is instead of with convolution window Fourier spectrum, therefore very big attenuation outside a channel can be ensured, below with one
Instantiation illustrates the problem.
Make frequency vector length N=256, data collecting card DAQ1 sample rate fs=10KHz, the son filtering of wave filter group
Device number Q=5.Analyzed by field statistics, the cut-off frequency for extrapolating environmental disturbances is fe=250Hz, general intrusion event
Upper limiting frequency is fu=3500Hz, according to formula (4), (6) can determine that filter parameter e=6, m=17, window function f (n) selection
Hamming windows.According to formula (7), (8) obtain 5 subfilter g0(n),...,g4(n), corresponding attenuation curve is 20lg | Gq
(j2 π f) |, shown in (a) in q=0 ..., 4, such as Fig. 4；To be compared, provide using declining that classical Frequency Sampling Method is obtained
Subtract curve 20lg | Hq(j2 π f) |, shown in (b) in q=0 ..., 4, such as Fig. 4.
As can be seen from Figure 4, the wave filter group transmission curve figure obtained using classical Frequency Sampling Method is had substantially in passband
Ripple, the 1st side lobe attenuation of each subfilter only has -20dB；By contrast, all phase DFT filter group that this method is used is several
Passband ripple is not present, the 1st side lobe attenuation of each subfilter approaches -70dB, it means that compared to Frequency Sampling Method,
Degree of coupling between each subfilter of all phase DFT filter group is greatly lowered；In addition, being less than f ∈ (0,250) in frequency
Hz region, the attenuation curve of all phase DFT filter group is still close to -70dB, it means that environmental disturbances can more thoroughly be disappeared
Remove.These superperformances of all phase DFT filter group ensure that the high accuracy of subsequent action identification.
It can be appreciated that all phase DFT filter group parallel output subsignal y0(n),...,yQ-1(n) Q performance number reflection
Energy distributions of the input signal x (n) in each frequency range, if the zero-crossing rate ZCR of these performance numbers and former whole section of input x (n) is combined,
Can construct when containing, the multi-feature vector of the aspect information of frequency domain two.
1) power：It is assumed that output sample length is L, then all phase DFT filter group q-th of output signal yq(n) be averaged
Power EqIt can be calculated by following formula：
Wherein, EqFor mean power.
However, due to zero-crossing rate ZCR be always at (0,1) it is interval in, and mean power may far beyond the interval, because
And be ensure characteristic vector inside each element span harmony, normalized need to be done to mean power, that is, calculate as
Lower performance number EqPerformance number after can finally being normalized：
2) zero-crossing rate：Zero-crossing rate ZCR is the performance indications integrally counted in time domain, and reflection is signal intensity speed journey
Degree.ZCR can be calculated by following formula：
Here operator " sign " represents to take symbol manipulation, is calculated by following formula：
Both the above feature is combined, you can construct multi-feature vector F=[E0,E1,...,EQ-1, ZCR], it is used for
Feed-in RBF neural carries out movement recognition.
In recent years, artificial neural network is realized by wide because having the advantages that powerful self-learning capability and being easy to hardware
General concern.RBF neural has the advantages that learning algorithm fast convergence rate as a kind of feedforward neural network.This method is adopted
Invasion movement recognition is carried out with RBF neural.
Most basic RBF neural is main to be made up of input layer, single hidden layer and the part of output layer three.Input layer is used as spy
The input of vector is levied, its neuron number should be identical with the dimension of training sample；Single hidden layer is used as god using RBF
It is connected entirely with certain weights implementation through first activation primitive, and between output layer；Output layer is then that hidden neuron is exported
Linear combination, output neuron number determines by the model number of required differentiation.For the action recognition feelings of DMZI systems
For condition, as shown in figure 5, needing feed-in (Q+1) dimensional vector F=[E0,E1,...,EQ-1, ZCR], it is output as the decision value of M action
Z1,...,ZP, these decision values calculate by following formula：
Wherein, h is hidden neuron number, ciAnd ωi,pIt is the center corresponding to i-th of hidden neuron and power respectively
Weight, ρ (F, ci) it is RBF, sample data F is normally defined to data center ciBetween Euclidean distance monotone decreasing letter
Number, P is identification species number.This method is used as RBF from Gaussian function.Z is calculated according to formula (13)1,...,ZPAfterwards,
Its maximum is taken to judge the classification (can recognize that P kinds invasion action altogether) of invasion action.
The training process of Fig. 5 RBF networks is as follows：For known intrusion event classification (be assumed to be pth kind, p=1 ...,
P characteristic vector), Z is exported by correspondencepLabeled as 1, remaining is labeled as 0, with actual Z in training processpValue is equal with mark value
Square error (means quare error, MSE) is guide, and multiple characteristic vectors for being used to train are processed, constantly optimized
Each parameter (including average value c of the RBF of hidden layer1,...,ch, variances sigma1,...,σh) and hidden layer and output layer it
Between weights ωi,p(i∈[1,h],p∈[1,P])。
The test process of Fig. 5 RBF networks is as follows：The network that the characteristic vector feed-in of unmarked classification is trained, is borrowed
The network parameter helped after having optimized, calculates output valve Z1,...,ZP, its maximum is taken to determine the institute of current input feature vectors
Category classification.
In summary, the embodiment of the present invention due to using multi-feature vector simultaneously consider intrusion event frequency domain, when
Domain information, four class intrusion events can be distinguished exactly, compared to existing high-precision intrusion event recognition classifier,
Had a clear superiority in terms of operating efficiency；Using all phase DFT filter group use parallel pipeline by the way of be operated, make
The interchannel interference very little between each subfilter is obtained, possibility is provided for accurate characteristic vector of extracting.
Embodiment 3
Feasibility checking is carried out to the scheme in Examples 1 and 2 with reference to specific experiment, accompanying drawing and form, in detail
See below description：
Test in the distributed feedback laser that lasing light emitter is 1550nm, light intensity is 3.5mV DMZI distributing optical fiber sensings
Carried out in system.Sensing optic cable total length is 2.25km, DAQ1 sample rate fss=10kHz, the record time is 3s.It is common to 4 classes
Intrusion event：Climbing fence, tap optical cable, rock and robber cuts 480 data acquisitions of progress, being per class intrusion event number of repetition
120 times.
The selection of characteristic vector length is primarily for from the aspect of two：On the one hand, it is contemplated that to recognize that P=4 classes invade thing
Part, is the degree of accuracy of Assured Mode identification classification, and characteristic vector length Q+1 should be greater than or equal to the event number 4 to be recognized；
On the other hand, because the computation complexity that can cause feature extraction using long characteristic vector increases, Q values again should not mistake
Greatly.Based on considerations above, all phase subfilter number is set to Q=3 by this experiment, so that characteristic vector F length is 4 (texts
The characteristic vector length of [15] is offered for 6).
Filter frequencies vector length N=256 is set, then frequency resolution Δ f=fs/N=39.0625Hz.It is right respectively
Non-intrusive sample and invasion sample do FFT statistical analyses, obtain the cut-off frequency f of wave filtere=220Hz, fu=4000Hz.
Therefore, the passband width of all phase DFT filter group is B=fu-fe=3780Hz, by formula (4), (6) can determine that parameter e=6, m=
32, the free transmission range that can estimate subfilter by formula (5) is respectively f ∈ [234.375,1484.375] Hz, f ∈
[1484.375,2734.375] Hz, f ∈ [2734.375,3984.375] Hz.4 classes invade signal and pass through full phase filtering
Each way signal waveform after the processing of device group is as shown in Figure 6.
As can be seen from Figure 6, primary signal x (n) shape differences of 4 classes action are not obvious；At all phase DFT filter group
After reason, its output signal y0(n),y1(n),y2(n) apparent difference is but showed, it ensure that the standard of follow-up mode identification
Exactness.
According to formula (9)~formula (12), multi-feature vector F=[E can be calculated0,E1,...,EQ-1,ZCR].For more prominent table
Up to the feature of all kinds of intrusion event characteristic vectors, the multi-feature vector of all samples contained to every class event package is averaged, and is obtained
To average aggregate characteristic vector as shown in Figure 7.
As can be seen from Figure 7, the multi-feature vector of all kinds of events has significant difference：Climb signal intensity most fast, have
Maximum zero-crossing rate；Rock the performance number E of signal0Maximum, E1, E2It is then far less, i.e. E0~E2Change is the most violent；Shearing letter
Number E0E of the value higher than knocking0Value, E0~E2Change more acutely, and zero-crossing rate minimum.
This experiment carries out pattern-recognition using Fig. 5 RBF neural, and design parameter sets as follows：Minimum MSE values are preset
For 0.0442289；Maximum neuron number h is 30.Neuron number will gradually increase, until MSE meets preset requirement.Will
4 output Z=[Z of neutral net1,Z2,Z3,Z4] represent climb, tap, rocking, shearing intrusion event respectively.Although with
More training sample can improve the precision of identification, but this can increase processing time, and for more than balance 2 points, this experiment is gathered altogether
480 groups of data, wherein, the number of training of every kind of invasion signal is set to 50, and test sample number is set to 70.Table 1 gives finally
Recognition success rate, and be compared with EMD methods.
From table 1 it follows that this method has greatly improved compared to EMD methods in precision aspect tool, this aspect
It is due to accurate division of the wave filter group to frequency so that the degree of coupling reduction between each characteristic element so that feature is described
It is more accurate；On the other hand it is due to the introducing of zero-crossing rate so that characteristic vector is while have time domain, frequency domain and statistical nature concurrently, more
Plus the feature of all kinds of events is comprehensively described.
In addition, table 2 also list the processing time of EMD methods and this method.This also presents this method in terms of efficiency
Great advantage.
The experimental precision contrast of the two methods of table 1
As can be seen from Table 1, this method average recognition rate is that the average recognition rate of 88.5725%, EMD methods is 85.75%,
Therefore this method overall recognition accuracy is higher than EMD methods.Specifically, the accuracy of identification of this method reaches as high as 100% and (rocks thing
Part), the discrimination of climbing event and shear event is above EMD methods, taps the discrimination of event close to EMD methods.Its
Main reason is that：
1) this method employs multi-feature vector (EMD methods have only used single kurtosis characteristic vector) to intrusion event
It is described more fully；
2) method introduces the degree of coupling of each subchannel very small all phase DFT filter group, (each time of EMD methods are repeatedly
Coupled between the IMF that generation obtains still larger).
The processing time contrast of the two methods of table 2
Embodiment 4
The embodiment of the present invention 1 is corresponding with the optical fiber perimeter security protection intrusion event recognition methods based on comprehensive characteristics that 2 provide
Identifying device, referring to Fig. 8.
Signal x (t) to be filtered is first passed around into end-point detection, the starting point that outgoing event occurs is judged, from the off, will
Signal feeding A/D (analog-to-digital conversion device) samplings of a period of time obtain sample sequence x (n) below, the shape inputted with parallel data
Formula enters DSP devices, in filter order N, frequency vector H 0 and 1 number e and m is concurrently set, by the inside of DSP devices
Algorithm process (including input data x (n) all phase pretreatment and convolution window, Q filter coefficient vector g0,...,
gQ-1, obtain the filtering output y of signal0,...,yQ-1, and then characteristic vector F is obtained, finally recognize 4 kinds not by RBF neural
Same invasion action).
Wherein, Fig. 8 DSP (Digital Signal Processor, digital signal processor) is core devices, in letter
In number parameter estimation procedure, following major function is completed：
(1) core algorithm is called, the filtering of input signal is completed；
(2) filter order N, filter bandwidht λ, filter passband original position p are adjusted according to actual needs, are come with this
Build subfilter g.
(3) each subchannel filter result is exported respectively.
(4) calculate and obtain characteristic vector F.
(5) intrusion species is differentiated according to RBF neural.
It may be noted that as a result of digitized method of estimation, thus determine the complexity of system, real-time levels and steady
Surely the principal element spent not is the periphery connection of DSP devices in Fig. 8, but the core that DSP internal program memories are stored
Algorithm for estimating.
The internal processes flow of DSP devices is as shown in Figure 9.
Fig. 9 flows are divided into following several steps：
(1) need to be required (signal passband bandwidth such as to be filtered) according to concrete application first, all phase DFT filter parameter is set
N, e and m.The step is the proposition real needs in terms of engineering, to cause follow-up process targetedly to be handled.
(2) according to formula (18), filter filtering coefficient g is generated.
(3) then, CPU main controllers read sampled data from I/O ports, into internal RAM.
(4) input signal is filtered using constructed all phase DFT filter group, obtains exporting yq(n), q=
0,...,Q-1。
(5) by exporting yq(n) and energy calculation formula, obtain characteristic vector F.
(6) characteristic vector F is sent into RBF neural, obtains final output Zp。
It may be noted that being realized as a result of DSP so that whole parameter estimation operation becomes more flexible, can be according to signal
Comprising various components concrete condition, pass through flexible in programming change algorithm inner parameter set.
The embodiment of the present invention is to the model of each device in addition to specified otherwise is done, and the model of other devices is not limited,
As long as the device of above-mentioned functions can be completed.
Bibliography
[1]Rangaswamy S,van Doorn E.Perimeter security system:U.S.Patent 8,
232,878[P].2012-7-31.
[2]P.M.B.S.Girao,O.A.Postolache,J.A.B.Faria,and J.M.C.D.Pereira,“An
overview and a contribution to the optical measurement of linear
displacement,”IEEE Sens.J.,vol.1,no.4,pp.322–331,Dec.2001.
[3]H.Gong,H.Song,S.Zhang,Y.Jin,and X.Dong,“Curvature sensor based on
hollow-core photonic crystal fiber sagnac interferometer,”IEEE Sens.J.,
vol.14,no.3,pp.777–780,Mar.2014.
[4] Yu Shengyun, Sun Shengli multiple wireless infrared acquisition design of Intelligent Security Guard System [J] laser with it is infrared, 2008,
38(4)：345-347.
[5] the quick tension type electronic fence peripheries of Fan Zhi take precautions against Guoan in Design of Alarm System principle and application present situation [J]
It is anti-, 2008 (3)：42-45.
[6]Q.Sun,D.Liu,and H.Liu,“Distributed disturbance sensor based on a
novel Mach–Zehnder interferometer with a fiber-loop,”Proc.SPIE,vol.6344,
pp.63440K(1)–63440K(7),2006.
[7]Q.Sun,D.Liu,and H.Liu,“Distributed fiber-optic sensor with a ring
Mach–Zehnder interferometer,”Proc.SPIE,vol.6781,pp.67814D(1)–67814D(8),2007.
[8]S.Xie,M.Zhang,S.Lai,and Y.Liao,“Positioning method for dual Mach-
Zehnder interferometric submarine cable security system,”Proc.SPIE,vol.7677,
pp.76770A–1–76770A–4,Apr.2010.
[9]L.Jiang and R.Yang,“Identification technique for the intrusion of
airport enclosure based on double mach-zehnder interferometer,”Journal of
Computers,vol.7,no.6,pp.1453–1459,2012.
[10]J.Juarez,E.Maier,K.Choi,and H.Taylor,“Distributed fiber-optic
intrusion sensor system,”IEEE J.Lightw.Technol.,vol.23,no.6,pp.2081–2087,
Jun.2005.
[11]S.Xie,Q.Zou,L.Wang,M.Zhang,Y.Li,and Y.Liao,“Positioning error
prediction theory for dual Mach-Zehnder interferometric vibration sensor,”
IEEE J.Lightw.Technol.,vol.29,no.3,pp.362–368,Feb.2011
[12]X.Huang,J.Yu,K.Liu,T.Liu,and Q.Chen,“Configurable filter-based
endpoint detection in DMZI vibration system,”IEEE Photon.Technol.Lett.,
vol.26,no.19,pp.1956–1959,Oct.2014.
[13]Huang X,Wang Y,Liu K,et al.High-Efficiency Endpoint Detection in
Optical Fiber Perimeter Security[J].2016,PP(99)：1-1.
[14]L.Liu,W.Sun,Y.Zhou,Y.Li,J.Zheng,and B.Ren,“Security event
classification method for fiber-optic perimeter security system based on
optimized incremental support vector machine,”in Pattern
Recognition.Springer,2014,pp.595–603.
[15]S.S.Mahmoud,Y.Visagathilagar,and J.Katsifolis,“Real-time
distributed fiber optic sensor for security systems：Performance,event
classification and nuisance mitigation,”Photonic sensors,vol.2,no.3,pp.225–
236,2012
[16]K.Liu,M.Tian,T.Liu,J.Jiang,Z.Ding,Q.Chen,C.Ma,C.He,H.Hu,and
X.Zhang,“A high-efficiency multiple events discrimination method in optical
fiber perimeter security system,”IEEE J.Lightw.Technol.,vol.33,no.23,pp.4885–
4890,2015.
[17] Li Kaiyan, Zhao Xingqun, Sun little Han, wait a kind of Regularizations for optical fiber link vibration signal pattern-recognition of
Compound characteristics extracting method [J] Acta Physica Sinicas, 2015,64 (5)：243-249.
[18]Huang X,Jing S,Wang Z,et al.Closed-Form FIR Filter Design Based
on Convolution Window Spectrum Interpolation[J].IEEE Transactions on Signal
Processing,2015,64(5)：1-1.
[19] data signal all phase FFT spectrum analysis and filtering technique [M] Electronic Industry Presses, 2009.
[20]X.Huang,Y.Wang,K.Liu,T.Liu,C.Ma,and Q.Chen,“Event discrimination
offiber disturbance based on filter bank in dmzi sensing system,”IEEE
Photonics Journal,vol.8,no.3,pp.1–14,2016.
[21]H.Huan,D.Hien,and H.Tue,“Efficient algorithm for training
interpolation RBF networks with equally spaced nodes,”IEEE Trans.Neural
Netw.,vol.22,no.6,pp.982–988,Jun.2011.
[22]W.Lyons,H.Ewald,and E.Lewis,“An optical fibre distributed sensor
based on pattern recognition,”J.Mater.Process.Technol,vol.127,no.1,pp.23–30,
2002.
It will be appreciated by those skilled in the art that accompanying drawing is the schematic diagram of a preferred embodiment, the embodiments of the present invention
Sequence number is for illustration only, and the quality of embodiment is not represented.
The foregoing is only presently preferred embodiments of the present invention, be not intended to limit the invention, it is all the present invention spirit and
Within principle, any modification, equivalent substitution and improvements made etc. should be included in the scope of the protection.
Claims (7)
1. a kind of optical fiber perimeter security protection intrusion event recognition methods based on comprehensive characteristics, it is characterised in that the recognition methods
Comprise the following steps：
It will determine that each passage of the signal feed-in all phase DFT filter group of disturbance starting point carries out frequency domain separating treatment, and calculate
The normalized power value of each multi-channel output signal, each multi-channel output signal is parallel output；
The time domain zero-crossing rate generation characteristic vector of combined normalized performance number and whole section of disturbing signal, i.e., described characteristic vector is accumulate
Containing when, the aspect information of frequency domain two；
It is that invasion action quick high accuracy identification can be achieved by characteristic vector feed-in radial basis function neural network.
2. a kind of optical fiber perimeter security protection intrusion event recognition methods based on comprehensive characteristics according to claim 1, it is special
Levy and be, each passage of the signal feed-in all phase DFT filter group that will determine to disturb starting point is carried out at frequency domain separation
Manage, and calculate the normalized power value of each multi-channel output signal and be specially：
The signal parallel for determining disturbance starting point is fed into all phase DFT filter group, all phase DFT filter group includes Q son
FIR filter g0,...,gQ-1, calculate filtering output yq(n) Q normalized power value Eq。
3. a kind of optical fiber perimeter security protection intrusion event recognition methods based on comprehensive characteristics according to claim 1, it is special
Levy and be, the step of time domain zero-crossing rate of the combined normalized performance number and whole section of disturbing signal generates characteristic vector has
Body is：
Signal x (n) overall zero-crossing rate is calculated, by the value and Q normalized power value EqIt is the comprehensive of Q+1 to do the comprehensive length that obtains
Close characteristic vector F=[E0,E1,...,EQ-1,ZCR]。
4. a kind of optical fiber perimeter security protection intrusion event recognition methods based on comprehensive characteristics according to claim 1 or 2, its
It is characterised by, subfilter is specially in all phase DFT filter group：
gq(n)=ωc(n)hq(n), q=0 ..., Q-1.
Wherein, gq(n) it is sub- FIR filter coefficient；ωc(n) it is double window convolution window；hq(n) it is filter coefficient；Q is variable,
Represent q-th of subfilter；Q is subfilter number contained by wave filter group.
5. a kind of optical fiber perimeter security protection intrusion event recognition methods based on comprehensive characteristics according to claim 1, it is special
Levy and be, described is that the identification of invasion action quick high accuracy can be achieved specifically by characteristic vector feed-in radial basis function neural network
For：
By the radial basis function neural network for training the characteristic vector feed-in of unmarked classification, by the net after having optimized
Network parameter, calculates output valve Z1,...,ZP, the maximum of output valve is taken to determine the affiliated classification of current input feature vectors.
6. a kind of optical fiber perimeter security protection intrusion event recognition methods based on comprehensive characteristics according to claim 5, it is special
Levy and be, methods described also includes：
Pass through 4 output valve Z1,...,ZPRealize the accurate identification to climbing, tapping, rock and stealing the event cut.
7. a kind of identification of optical fiber perimeter security protection intrusion event recognition methods based on comprehensive characteristics for described in claim 1
Device, it is characterised in that described device includes：Analog-to-digital conversion device and DSP devices,
Signal feeding analog-to-digital conversion device sampling is obtained into sample sequence x (n), DSP devices are entered in the form of parallel data is inputted,
Concurrently set in filter order N, frequency vector H 0 and 1 number e and m；
Characteristic vector is obtained by the processing of DSP devices, is finally acted by the different invasion of neural network recognization.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710258029.3A CN107180521A (en) | 2017-04-19 | 2017-04-19 | Optical fiber perimeter security protection intrusion event recognition methods and device based on comprehensive characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710258029.3A CN107180521A (en) | 2017-04-19 | 2017-04-19 | Optical fiber perimeter security protection intrusion event recognition methods and device based on comprehensive characteristics |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107180521A true CN107180521A (en) | 2017-09-19 |
Family
ID=59831983
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710258029.3A Pending CN107180521A (en) | 2017-04-19 | 2017-04-19 | Optical fiber perimeter security protection intrusion event recognition methods and device based on comprehensive characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107180521A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107730800A (en) * | 2017-11-13 | 2018-02-23 | 浙江众盟通信技术有限公司 | Anti-Interference Analysis method based on fiber-optic vibration safety pre-warning system |
| CN108399696A (en) * | 2018-03-22 | 2018-08-14 | 中科润程(北京)物联科技有限责任公司 | Intrusion behavior recognition methods and device |
| CN108986363A (en) * | 2018-08-24 | 2018-12-11 | 天津大学 | Optical fiber security protection intrusion event recognition methods and device based on ARMA modeling |
| CN109003407A (en) * | 2018-07-16 | 2018-12-14 | 胡志雄 | A kind of intelligent-induction device and method invaded for detecting door and window |
| CN109064696A (en) * | 2018-08-17 | 2018-12-21 | 成都九洲电子信息系统股份有限公司 | The optical fiber perimeter security system realized based on deep learning |
| CN109523729A (en) * | 2018-10-31 | 2019-03-26 | 天津大学 | Based on the optical fiber perimeter security protection intrusion event recognition methods modeled entirely and identifier |
| CN110275896A (en) * | 2019-05-28 | 2019-09-24 | 深圳供电局有限公司 | Optical cable invasion construction event recognition method, device, equipment and readable storage medium storing program for executing |
| CN111160106A (en) * | 2019-12-03 | 2020-05-15 | 上海微波技术研究所（中国电子科技集团公司第五十研究所） | Method and system for extracting and classifying optical fiber vibration signal features based on GPU |
| CN111597994A (en) * | 2020-05-15 | 2020-08-28 | 华侨大学 | Optical fiber perimeter security intrusion event identification model construction method and security system |
| CN112309063A (en) * | 2020-10-30 | 2021-02-02 | 魏运 | Method and device for extracting hybrid fiber intrusion signal feature spectrum |
| WO2021207102A1 (en) * | 2020-04-07 | 2021-10-14 | Nec Laboratories America, Inc. | Object localization and threat classification for optical cable protection |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1134712B1 (en) * | 2000-03-15 | 2008-05-07 | Siemens Aktiengesellschaft | Method for the processing of the signal in a danger detector, and detector with means for the implementation of such method |
| CN104966076A (en) * | 2015-07-21 | 2015-10-07 | 北方工业大学 | Optical fiber intrusion signal classification and identification method based on support vector machine |
| CN105488935A (en) * | 2015-12-25 | 2016-04-13 | 天津大学 | Distributed optical fiber disturbance positioning system based on asymmetric double March-Zehnder interference and positioning method |
| CN106023499A (en) * | 2016-04-28 | 2016-10-12 | 北京北邮国安技术股份有限公司 | Fiber security signal dual identification method and system |
| CN106384463A (en) * | 2016-11-24 | 2017-02-08 | 天津大学 | Method for identifying opening fiber surrounding security invasion events based on mixed characteristic extraction |
-
2017
- 2017-04-19 CN CN201710258029.3A patent/CN107180521A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1134712B1 (en) * | 2000-03-15 | 2008-05-07 | Siemens Aktiengesellschaft | Method for the processing of the signal in a danger detector, and detector with means for the implementation of such method |
| CN104966076A (en) * | 2015-07-21 | 2015-10-07 | 北方工业大学 | Optical fiber intrusion signal classification and identification method based on support vector machine |
| CN105488935A (en) * | 2015-12-25 | 2016-04-13 | 天津大学 | Distributed optical fiber disturbance positioning system based on asymmetric double March-Zehnder interference and positioning method |
| CN106023499A (en) * | 2016-04-28 | 2016-10-12 | 北京北邮国安技术股份有限公司 | Fiber security signal dual identification method and system |
| CN106384463A (en) * | 2016-11-24 | 2017-02-08 | 天津大学 | Method for identifying opening fiber surrounding security invasion events based on mixed characteristic extraction |
Non-Patent Citations (1)
| Title |
|---|
| 王博: "分布式光纤周界安防系统定位和模式识别研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107730800A (en) * | 2017-11-13 | 2018-02-23 | 浙江众盟通信技术有限公司 | Anti-Interference Analysis method based on fiber-optic vibration safety pre-warning system |
| CN108399696A (en) * | 2018-03-22 | 2018-08-14 | 中科润程(北京)物联科技有限责任公司 | Intrusion behavior recognition methods and device |
| CN109003407A (en) * | 2018-07-16 | 2018-12-14 | 胡志雄 | A kind of intelligent-induction device and method invaded for detecting door and window |
| CN109064696A (en) * | 2018-08-17 | 2018-12-21 | 成都九洲电子信息系统股份有限公司 | The optical fiber perimeter security system realized based on deep learning |
| CN108986363A (en) * | 2018-08-24 | 2018-12-11 | 天津大学 | Optical fiber security protection intrusion event recognition methods and device based on ARMA modeling |
| CN109523729A (en) * | 2018-10-31 | 2019-03-26 | 天津大学 | Based on the optical fiber perimeter security protection intrusion event recognition methods modeled entirely and identifier |
| CN110275896A (en) * | 2019-05-28 | 2019-09-24 | 深圳供电局有限公司 | Optical cable invasion construction event recognition method, device, equipment and readable storage medium storing program for executing |
| CN110275896B (en) * | 2019-05-28 | 2021-07-20 | 深圳供电局有限公司 | Optical cable intrusion construction event identification method, device, equipment and readable storage medium |
| CN111160106A (en) * | 2019-12-03 | 2020-05-15 | 上海微波技术研究所（中国电子科技集团公司第五十研究所） | Method and system for extracting and classifying optical fiber vibration signal features based on GPU |
| CN111160106B (en) * | 2019-12-03 | 2023-12-12 | 上海微波技术研究所（中国电子科技集团公司第五十研究所） | GPU-based optical fiber vibration signal feature extraction and classification method and system |
| WO2021207102A1 (en) * | 2020-04-07 | 2021-10-14 | Nec Laboratories America, Inc. | Object localization and threat classification for optical cable protection |
| US20220316921A1 (en) * | 2020-04-07 | 2022-10-06 | Nec Laboratories America, Inc | Object localization and threat classification for optical cable protection |
| CN111597994A (en) * | 2020-05-15 | 2020-08-28 | 华侨大学 | Optical fiber perimeter security intrusion event identification model construction method and security system |
| CN111597994B (en) * | 2020-05-15 | 2023-03-07 | 华侨大学 | Optical fiber perimeter security intrusion event identification model construction method and security system |
| CN112309063A (en) * | 2020-10-30 | 2021-02-02 | 魏运 | Method and device for extracting hybrid fiber intrusion signal feature spectrum |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107180521A (en) | Optical fiber perimeter security protection intrusion event recognition methods and device based on comprehensive characteristics | |
| CN106384463B (en) | Optical fiber perimeter security protection intrusion event recognition methods based on hybrid feature extraction | |
| CN104729667B (en) | A kind of disturbance kind identification method in distributed optical fiber vibration sensing system | |
| CN104240455B (en) | A kind of disturbance event recognition methods in distribution type fiber-optic pipeline safety early warning system | |
| Liu et al. | A novel three-step classification approach based on time-dependent spectral features for complex power quality disturbances | |
| CN103995969B (en) | Configurable optical fiber invasion event occurring end point detecting method and detector | |
| CN109165670B (en) | TS-RBF fuzzy neural network robust fusion algorithm applied to infrared flame identification | |
| CN109489800A (en) | A kind of disturbance event recognition methods in distribution optic cable vibration safety pre-warning system | |
| Huang et al. | An event recognition scheme aiming to improve both accuracy and efficiency in optical fiber perimeter security system | |
| CN104376306A (en) | Optical fiber sensing system invasion identification and classification method and classifier based on filter bank | |
| CN108986363A (en) | Optical fiber security protection intrusion event recognition methods and device based on ARMA modeling | |
| Yan et al. | Bearing fault diagnosis via a parameter-optimized feature mode decomposition | |
| CN105023379A (en) | Signal identification method of fiber perimeter early-warning system of airport | |
| CN103116957A (en) | Method for optical fiber perimeter security and protection system shielding climate impact | |
| CN103956756A (en) | Electric system low-frequency oscillating mode identification method | |
| Huang et al. | Hybrid feature extraction-based intrusion discrimination in optical fiber perimeter security system | |
| CN111160106A (en) | Method and system for extracting and classifying optical fiber vibration signal features based on GPU | |
| Zhao et al. | Probabilistic principal component analysis assisted new optimal scale morphological top-hat filter for the fault diagnosis of rolling bearing | |
| Sun et al. | Variational mode decomposition-based event recognition in perimeter security monitoring with fiber optic vibration sensor | |
| CN111639583A (en) | Method and system for identifying power quality disturbance of power grid | |
| CN105549107B (en) | The optical fiber disturbance event end-point detecting method and detector of flexibly configurable | |
| CN109523729A (en) | Based on the optical fiber perimeter security protection intrusion event recognition methods modeled entirely and identifier | |
| CN109617051B (en) | New energy power system low-frequency oscillation parameter identification method | |
| CN114415056A (en) | Distributed power supply island fault detection method | |
| Karimian et al. | Novel method based on Teager Energy Operator for online tracking of power quality disturbances |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170919 |