CN104634460A - Multi-peak self-adaption accurate peak searching method for distributed FBG (Fiber Bragg Grating) sensing network - Google Patents

Abstract

The invention relates to a multi-peak self-adaption accurate peak searching method for a distributed FBG (Fiber Bragg Grating) sensing network and belongs to the technical field of signal processing of optical fiber sensing systems. The method comprises the following steps: smoothening a spectral signal by using a five-point sliding mean filtering method to eliminate the influence of signal noise on peak searching accuracy; processing the smoothened spectral signal by utilizing Hilbert conversion to obtain an initial peak positioning point of a multi-peak spectral signal; performing Gabor filtering on the smoothened signal to obtain a division point of a left sideband of a spectrum, and symmetrically obtaining a right division point of a spectrum peak by virtue of the initial peak positioning point; performing peak value area division on a multi-peak spectrum by taking the left and right division points of the spectrum as a boundary; integrating the divided spectrum peak by taking an initial positioning peak value point as the center to obtain the areas of a left half peak and a right half peak, and judging the offset situation of the spectrum peak by comparing the sizes of the left half peak and the right half peak; performing accurate peak searching on each spectrum peak by using an index correction Gauss fitting algorithm to obtain a multi-peak accurate peak value point by virtue of the peak shape of the spectrum peak. The method can be used for self-adaption peak value area division of the multi-peak spectral signal of the distributed sensing network and high-accurate peak value positioning.

Description

The accurate Peak Search Method of a kind of Distributed FBG sensing network multimodal self-adaptation

Technical field

The invention belongs to the signal processing technology field of optical fiber sensing system, relate to the accurate Peak Search Method of a kind of Distributed FBG sensing network multimodal self-adaptation.

Background technology

Fiber Bragg Grating FBG (Fiber Bragg grating, FBG) sensor is as a kind of fiber optic passive device, there is the advantages such as volume is little, electromagnetism interference, the multiplexing formation distributed sensor of corrosion-resistant, high temperature resistant, highly sensitive, easy serial connection, be widely used in the fields such as civil engineering work, Aero-Space, petrochemical complex and engineering in medicine.Fiber-optic grating sensor obtains the change of parameter to be measured indirectly by detection of reflected spectral centroid wavelength shift, and reflectance spectrum centre wavelength its peak corresponding, therefore high-precision peak-seeking algorithm is most important to the measuring accuracy improving sensor-based system.

At present, peak-seeking algorithm mainly contains direct peak-seeking algorithm, half-peak detection algorithm, Monte carlo algorithm, probability statistics algorithm, Gauss curve fitting algorithm, fitting of a polynomial algorithm, genetic algorithm, ant group algorithm, 3 peak-seeking algorithms and the peak-seeking algorithm based on Steger.Direct peak-seeking algorithm and half-peak detection algorithm computation complexity is low, the response time is short, but its noiseproof feature is poor, is not suitable for the FBG spectrum peak-seeking under complex engineering environment; Monte carlo algorithm, the probability statistics algorithm linearity are poor, peak-seeking limited precision; Gauss curve fitting algorithm and fitting of a polynomial algorithm peak-seeking precision higher, but require strict to spectral pattern; Although genetic algorithm can improve peak-seeking accuracy, need the longer training time to determine parameter in parameter, be not suitable for real-time operation; Ant group algorithm computing cost is large, and solving speed is slow; 3 peak-seeking algorithms comparatively its peak-seeking precision of traditional algorithm have had and have increased substantially, but do not take into full account the impact of asymmetry on demodulation accuracy of spectrum; Peak Search Method based on Steger image algorithm quotes the Steger algorithm extracting gradation of image peak of curve, the peak-seeking of dissymmetric peak type in conjunction with super-Gaussian model realization, but the selection of model parameter is comparatively large to peak-seeking Accuracy, makes the application of algorithm be restricted.

Above-mentioned peak-seeking algorithm is all to improve unimodal peak-seeking precision for target, but do not consider the peak-seeking problem at multiple peak in distributed sensor reflectance spectrum, and the higher approximating method of peak-seeking precision can not carry out matching to the spectrum containing multiple peak, therefore the method can only be applied to the peak-seeking at single peak.Be at present all under the prerequisite of known FBG archicenter wavelength for FBG spectrum multimodal Peak Search Method, carry out peak-seeking operation to unimodal within the scope of the specific frequency spectrum intercepted, these class methods are inapplicable for the large-scale distributed FBG sensing network of complexity.Therefore, the present invention proposes the accurate peak-seeking algorithm of a kind of Distributed FBG sensing network multimodal self-adaptation.

Summary of the invention

In view of this, the object of the present invention is to provide the accurate Peak Search Method of a kind of Distributed FBG sensing network multimodal self-adaptation, the method may be used for the multimodal spectral signal adaptive peak region segmentation of distributed sensor and high precision peak location.

For achieving the above object, the invention provides following technical scheme:

The accurate Peak Search Method of a kind of Distributed FBG sensing network multimodal self-adaptation, in the method, comprises the following steps:

Step one: adopt 5 level and smooth spectral signals of slip mean filter method; Adopt 5 slip mean filter methods to the smoothing process of original spectrum signal, to eliminate " burr " and " ghost peak " that caused by signal noise, and on this basis multimodal self-adaptation peak-seeking is carried out to the signal after level and smooth, improve precision and the accuracy of peak-seeking.

Step 2: peak region segmentation is carried out to multimodal spectrum in conjunction with Hilbert conversion and Gabor filtering algorithm; Adopt Hilbert conversion to realize multimodal spectrum peak and just locate, effectively can eliminate the signal noise do not eliminated completely in spectral signal to the impact of peak value positioning precision, improve the reliability that multi-peak detects; The Gabor filtering algorithm being widely used in picture signal Edge Gradient Feature is adopted to find the spectrum left side, peak band cut-point, more accurately can obtain the point that left side band variable quantity is maximum, this filtering algorithm only relates to convolution algorithm simultaneously, effectively reduces the computation complexity of algorithm;

Step 3: just to locate centered by peak point the left and right half-peak integration in spectrum peak after segmentation;

Step 4: judge peak type by the area of more left and right half-peak, adopts index correction Gauss curve fitting algorithm to each accurate peak-seeking in spectrum peak, with the accurate peak-seeking of the self-adaptation realizing multimodal on this basis.

Further, in step 2, adopt Hilbert to convert the method combined with Gabor filtering algorithm and realize spectrum multimodal adaptive peak region segmentation, for follow-up high precision matching peak-seeking algorithm provides possibility; Specifically comprise the following steps:

1) spectral signal after level and smooth converts through Hilbert, according to the odd function character of Hilbert conversion, obtains the first anchor point of original spectrum peak value by the zero crossing correspondence of signal after detecting conversion;

2) spectral signal after level and smooth, through Gabor filtering process, is extracted the character of signal edge unique point, is obtained the variable quantity maximum point of original spectrum left side band, namely part cutpoint on the left side by the zero crossing of signal after detection filter according to Gabor filtering;

3) centered by spectrum peak just anchor point, be spaced apart step-length just to locate between peak point and left side band cut-point, get a little on the right side of first anchor point, part cutpoint on the right side as spectrum;

4) with left and right cut-point for border, region segmentation is carried out to each peak in multimodal spectrum.

Further, in step 3, compose the asymmetrical characteristic at peak in conjunction with FBG, after spectrum peak region segmentation, take cut-point as border reconstruct original spectrum signal, and just to locate peak point as center, left and right half-peak is quadratured respectively, by more left and right half peak area S _lwith S _rsize judge the drift condition of peak type, for correction of peak value provides reference.

Further, in step 4, the accurate Peak Search Method of described index correction Gauss curve fitting is the correction of peak value method put forward for spectrum peak type asymmetrical characteristic; According to the spectrum peak drift condition judged, convolution index correction function on FBG standard gaussian fitting function basis, effectively improve peak-seeking precision, index correction function expression formula is as follows:

$H\left(\mathrm{\&lambda;}\right)=\left\{\begin{array}{c}\frac{1}{\mathrm{\&tau;}}{e}^{\left(\frac{\mathrm{\&lambda;}}{\mathrm{\&tau;}}\right)},S_L<S_R\\ \frac{1}{\mathrm{\&tau;}}{e}^{(-\frac{\mathrm{\&lambda;}}{\mathrm{\&tau;}})},S_L>S_R\end{array}\right.$

Wherein, τ is correction; By the definition of function, and in conjunction with the function model of FBG spectrum, correction can be derived Δ λ _bfor spectral bandwidth.

Beneficial effect of the present invention is: first the method for the invention adopts 5 slip mean filter methods to the smoothing process of spectral signal, and effectively eliminating noise disturbance affects sensing accuracy; Adopt Hilbert conversion and Gabor filter method to process the signal after level and smooth, achieve the adaptive peak region segmentation of multimodal spectrum, the defect of peak ranges when solving traditional peak-seeking algorithm multimodal peak-seeking, must be pre-determined; Judge the drift condition of peak type to compose the size of half peak area about peak, propose the Gauss curve fitting method based on index correction, taken into full account FBG spectrum peak type asymmetry problem, effectively improve peak-seeking precision.The method easily extensible is applied in the spectral signal demodulation of the optical fiber sensing system such as long period fiber grating, Mach Zehnder interferometer, meanwhile, the multimodal self-adaptation peak-seeking requirement by selecting different functional parameters can meet the different physics waveforms such as electrocardio ripple, brain wave, sound wave, seismic event.

Accompanying drawing explanation

In order to make object of the present invention, technical scheme and beneficial effect clearly, the invention provides following accompanying drawing and being described:

Fig. 1 is FBG spectrum multimodal self-adaptation peak-seeking algorithm flow chart;

Fig. 2 is cut-point determination principle schematic;

Fig. 3 is index correction Gauss curve fitting peak-seeking algorithm flow chart.

Embodiment

This method adopts 5 slip mean filter methods to the smoothing filter preprocessing of spectral signal.Due to can certain noise be contained in the spectral signal that sensor-based system gathers, be present in FBG reflectance spectrum with the form of " burr " and " ghost peak ", if do not processed, peak-seeking precision can be had a strong impact on, can effective erasure signal noise by level and smooth pre-service, improve the peak-seeking precision of algorithm.

Hilbert transfer pair peak point is adopted to carry out just location to pretreated multimodal spectral signal.Hilbert conversion can effectively suppress low amplitude wave noise to be disturbed when processing low frequency sequence signal, burst after conversion has odd function character, Primary Location the peak of multimodal spectral signal can be gone out, for the process of follow-up peak value region segmentation provides reference based on this characteristic.

Multimodal spectral signal through smoothing processing carries out Gabor filtering process.Gabor filter method is often used to the marginal information detecting gray level image, namely the most precipitous on brightness change curve point.Odd component in one dimension Gabor filter function is acted on FBG spectral signal, may correspond to and detect that the maximum position of the spectrum left side, peak band variable quantity parts cutpoint on the left side as peak region, in conjunction with acquired peak value just anchor point realize splitting the peak region of spectrum.

Multimodal spectrum peak region segmentation is realized by the first anchor point of the multimodal of above-mentioned acquisition and the spectrum left side, peak band variable quantity maximum point.Specifically comprise:

1) through the spectrum wavelength value λ that just anchor point is corresponding of Hilbert conversion acquisition ₀, λ ₁λ _n, the wavelength value λ ' of the same some correspondence maximum through the left side band variable quantity of Gabor filtering process acquisition ₀, λ ' ₁λ ' _ndiffer from, obtain half peak wavelength interval.

2) the spectrum wavelength value λ that just anchor point is corresponding ₀, λ ₁λ _nand sue for peace in half peak wavelength interval, obtain parting wavelength value λ corresponding to cutpoint on the right side " ₀, λ " ₁λ " _n;

3) with wavelength value corresponding to left and right cut-point for border, multimodal spectrum is carried out peak region segmentation.

Because FBG spectrum peak is asymmetric peak type, in order to reach the requirement of accurate peak-seeking, based on the peak region be partitioned into, judging the drift condition of peak type, adopting following steps:

1) centered by spectrum just anchor point, with left and right cut-point for border, about the closed region form original spectrum and horizontal ordinate, difference integration, tries to achieve left and right half-peak area and is respectively S _lwith S _r;

2) according to S _lwith S _rmagnitude relationship, judge the drift condition of peak type, namely

According to the drift condition of peak type, the Gauss curve fitting algorithm proposed based on index correction carries out accurate peak-seeking.The mathematic(al) representation of index correction Gauss curve fitting function is as shown in formula (1):

H _EMG(λ)＝G(λ)*H(λ) (1)

Wherein, H _eMG(λ) be revised Gauss curve fitting function; The FBG that G (λ) is standard composes peak Gauss model; H (λ) is index correction function.FBG composes peak model can by formula (2) approximate representation:

$G\left(\mathrm{\&lambda;}\right)=\mathrm{Aexp}[-4\ln 2{\left(\frac{\mathrm{\&lambda;}-{\mathrm{\&lambda;}}_B}{\mathrm{\&Delta;}{\mathrm{\&lambda;}}_B}\right)}^2]---\left(2\right)$

Wherein, λ is FBG wavelength; λ _bfor FBG centre wavelength; Δ λ _bfor three dB bandwidth; The amplitude of A reflectance spectrum.Index correction function H (λ) expression formula is as shown in formula (3):

$H\left(\mathrm{\&lambda;}\right)=\left\{\begin{array}{c}\frac{1}{\mathrm{\&tau;}}{e}^{\left(\frac{\mathrm{\&lambda;}}{\mathrm{\&tau;}}\right)},S_L<S_R\\ \frac{1}{\mathrm{\&tau;}}{e}^{(-\frac{\mathrm{\&lambda;}}{\mathrm{\&tau;}})},S_L>S_R\end{array}\right.---\left(3\right)$

Wherein, τ is correction.By the definition of function, and in conjunction with the function model of FBG spectrum, the calculating formula of correction τ can be derived as shown in formula (4):

$\mathrm{\&tau;}=\sqrt{7}\mathrm{\&Delta;}{\mathrm{\&lambda;}}_B---\left(4\right)$

By carrying out index correction Gauss curve fitting to peak region spectrum, the accurate peak value at each spectrum peak can be obtained, and the corresponding centre wavelength obtaining FBG reflectance spectrum, meet the demodulation requirement of Distributed FBG sensing network multimodal self-adaptation peak-seeking.

Below in conjunction with accompanying drawing, the preferred embodiments of the present invention are described in detail.

If Fig. 1 is FBG spectrum multimodal self-adaptation peak-seeking algorithm flow chart, as shown in the figure, its concrete implementation step is as follows:

1, smoothing processing.

The noise produced due to external environment, demodulation device etc. is inevitably mixed in spectral signal, in order to eliminate " burr " and " ghost peak " that caused by noise, the present invention adopts 5 slip mean filter methods to the smoothing filtering process of multimodal spectral signal x (n), and filter function is as shown in formula (5):

$\left\{\begin{array}{c}{y}_1=1/5(3x_1+2x_2+x_3-x_4)\\ y_2=1/10(4x_1+3x_2+2x_3+x_4)\\ y_i=1/5(x_{i-2}+x_{i-1}+x_i+x_{i+1}+x_{i+2})\\ y_{n-1}=1/10(x_{n-3}+2x_{n-2}+3x_{n-1}+4x_n)\\ y_n=1/5(-x_{n-3}+x_{n-2}+2x_{n-1}+3x_n)\end{array}\right.---\left(5\right)$

2, peak point is just located.

FBG multimodal reflectance spectrum f (n) after smoothing processing carries out Hilbert conversion.For time-domain signal x (t), its Hilbert transform definition is as shown in formula (6):

$\hat{x}\left(t\right)=H\left[x\left(t\right)\right]=\frac{1}{\mathrm{\&pi;}}{\&Integral;}_{-\&infin;}^{+\&infin;}x\left(\mathrm{\&tau;}\right)\frac{1}{t-\mathrm{\&tau;}}\mathrm{d\&tau;}---\left(6\right)$

As can be seen from formula (6) formula function x (t) is funtcional relationship linearly, and this conversion can be expressed as x (t) and (π t) ^-1the form of convolution is as shown in formula (7):

$\hat{x}\left(t\right)=\frac{1}{\mathrm{\&pi;t}}*x\left(t\right)---\left(7\right)$

Because Hilbert conversion is odd function, then the peak of spectral signal f (n) corresponds to the zero crossing position of rear signal h (n) of conversion.By judging that the zero crossing position of h (n) can obtain each peak λ of original spectrum signal ₀, λ ₁λ _n.

3, spectrum left side band cut-point location.

FBG multimodal reflectance spectrum f (n) after smoothing processing carries out Gabor filtering process.Odd component in one dimension Gabor filter function is acted on FBG spectral signal, may correspond to detect spectrum peak on rise the maximum position of sideband variable quantity, one dimension Gabor odd component filter function is as shown in formula (8):

$s\left(x\right)=\exp [-\frac{1}{2}{\left(\frac{x}{\mathrm{\&delta;}}\right)}^2]\sin \left(w_{x}x\right)---\left(8\right)$

Wherein, δ is filtering extension width parameter; w _xfor the frequency of filtered envelope; δ and w _xthere is reciprocal relation, compose peak bandwidth in conjunction with FBG spectrum 3dB, the parameter of selected Gabor filter is δ=3, w _x=1/3.Gabor filtering process is carried out, as shown in formula (9) to spectral signal f (n) after level and smooth:

g(x)＝f(x)*s(x) (9)

Be may correspond to the mutated site obtaining the spectrum spectrum left side, peak band by the zero crossing position detecting g (x), namely part cutpoint λ ' on the left side ₀, λ ' ₁λ ' _n.

4, peak region segmentation.

By acquired spectral peak position λ ₀, λ ₁λ _nwith part cutpoint λ ' on the left side ₀, λ ' ₁λ ' _nask difference item by item, as shown in formula (10):

Δλ ₀＝λ ₀-λ′ ₀,Δλ ₁＝λ ₁-λ′ ₁,…,Δλ _n＝λ _n-λ′ _n(10)

Then on the right of spectrum with cut-point can be determined by formula (11):

λ″ ₀＝λ ₀+Δλ ₀,λ″ ₁＝λ ₁+Δλ ₁,…,λ″ _n＝λ _n+Δλ _n(11)

Three unique points at each spectrum peak can be obtained, as shown in Figure 2, to compose the left and right cut-point in peak for carrying out region segmentation to FBG multi-peak spectrum in border by aforesaid operations.

5, peak type judges.

With the left and right cut-point at each spectrum peak for border, reconstruct original spectrum signal, and just to locate centered by peak point, to the left and right half-peak closed region integration respectively that original spectrum and horizontal ordinate are formed, try to achieve integral area S _lwith S _r, and the size of foundation left and right half-peak judges the drift condition of peak type.

6, the accurate peak-seeking of peak correction

Based on index correction Gauss curve fitting method, in conjunction with the peak type judged, accurate peak-seeking is carried out to each peak.The function expression of index correction Gauss curve fitting method is as shown in formula (12):

H _EMG(λ)＝G(λ)*H(λ) (12)

Wherein, H _eMG(λ) be revised Gauss curve fitting function; The FBG that G (λ) is standard composes peak Gauss model; H (λ) is index correction function.FBG composes peak model can by formula (13) approximate representation:

$G\left(\mathrm{\&lambda;}\right)=\mathrm{Aexp}[-4\ln 2{\left(\frac{\mathrm{\&lambda;}-{\mathrm{\&lambda;}}_B}{\mathrm{\&Delta;}{\mathrm{\&lambda;}}_B}\right)}^2]---\left(13\right)$

Wherein, λ is FBG wavelength; λ _bfor FBG centre wavelength; Δ λ _bfor three dB bandwidth; The amplitude of A reflectance spectrum.As shown in Figure 3, different index correction functions is adopted to revise peak value to different peak types.

If 1. S _l> S _r, then FBG peak value is to the left, carries out index correction to it, and its index correction function is peak center wavelength X is obtained after correction _p;

If 2. S _l=S _r, carry out Gauss curve fitting, then FBG peak center wavelength X _pbe λ _b;

3. S _l< S _r, then FBG peak value is to the right, carries out index correction to it, and its index correction function is peak center wavelength X is obtained after correction _p.

By carrying out the accurate peak-seeking of index correction Gauss curve fitting to each spectrum peak after segmentation, the final accurate peak obtaining spectrum all spectrums peak.

What finally illustrate is, above preferred embodiment is only in order to illustrate technical scheme of the present invention and unrestricted, although by above preferred embodiment to invention has been detailed description, but those skilled in the art are to be understood that, various change can be made to it in the form and details, and not depart from claims of the present invention limited range.

Claims (4)

1. the accurate Peak Search Method of Distributed FBG sensing network multimodal self-adaptation, is characterized in that: in the method, comprises the following steps:

Step one: adopt 5 level and smooth spectral signals of slip mean filter method;

Step 2: peak region segmentation is carried out to multimodal spectrum in conjunction with Hilbert conversion and Gabor filtering algorithm;

Step 3: just to locate centered by peak point the left and right half-peak integration in spectrum peak after segmentation;

Step 4: judge peak type by the area of more left and right half-peak, adopts index correction Gauss curve fitting algorithm to each accurate peak-seeking in spectrum peak, with the accurate peak-seeking of the self-adaptation realizing multimodal on this basis.

2. the accurate Peak Search Method of a kind of Distributed FBG sensing network multimodal self-adaptation according to claim 1, it is characterized in that: in step 2, adopt Hilbert to convert the method combined with Gabor filtering algorithm and realize spectrum multimodal adaptive peak region segmentation, for follow-up high precision matching peak-seeking algorithm provides possibility, specifically comprise the following steps:

1) spectral signal after level and smooth converts through Hilbert, according to the odd function character of Hilbert conversion, obtains the first anchor point of original spectrum peak value by the zero crossing correspondence of signal after detecting conversion;

2) spectral signal after level and smooth, through Gabor filtering process, is extracted the character of signal edge unique point, is obtained the variable quantity maximum point of original spectrum left side band, namely part cutpoint on the left side by the zero crossing of signal after detection filter according to Gabor filtering;

3) centered by spectrum peak just anchor point, be spaced apart step-length just to locate between peak point and left side band cut-point, get a little on the right side of first anchor point, part cutpoint on the right side as spectrum;

4) with left and right cut-point for border, region segmentation is carried out to each peak in multimodal spectrum.

3. the accurate Peak Search Method of a kind of Distributed FBG sensing network multimodal self-adaptation according to claim 1, it is characterized in that: in step 3, the asymmetrical characteristic at peak is composed in conjunction with FBG, after spectrum peak region segmentation, take cut-point as border reconstruct original spectrum signal, and just to locate peak point as center, left and right half-peak is quadratured respectively, by more left and right half peak area S _lwith S _rsize judge the drift condition of peak type, for correction of peak value provides reference.

4. the accurate Peak Search Method of a kind of Distributed FBG sensing network multimodal self-adaptation according to claim 1, it is characterized in that: in step 4, the accurate Peak Search Method of described index correction Gauss curve fitting is the correction of peak value method put forward for spectrum peak type asymmetrical characteristic; According to the spectrum peak drift condition judged, convolution index correction function on FBG standard gaussian fitting function basis, effectively improve peak-seeking precision, index correction function expression formula is as follows:

$H\left(\mathrm{\&lambda;}\right)=\left\{\begin{array}{c}\frac{1}{\mathrm{\&tau;}}{e}^{\left(\frac{\mathrm{\&lambda;}}{\mathrm{\&tau;}}\right)},S_L=S_R\\ \frac{1}{\mathrm{\&tau;}}{e}^{(-\frac{\mathrm{\&lambda;}}{\mathrm{\&tau;}})},S_L>S_R\end{array}\right.$

Wherein, τ is correction; By the definition of function, and in conjunction with the function model of FBG spectrum, correction can be derived Δ λ _bfor spectral bandwidth.