CN105548359A - Wood hole defect ultrasonic detection and feature extraction method - Google Patents

Abstract

The invention discloses a wood hole defect ultrasonic detection and feature extraction method. Firstly, according to the wavelet analysis theory, a wavelet function is selected for carrying out five-layer wavelet packet decomposition on ultrasonic signals to obtain 32 frequency-band energies, energy moment analysis is conducted on the basis, and wavelet packet energy moment is subjected to normalization processing and dimension reduction processing through principal component analysis. According to the method, the ultrasonic signals of defects of wood with different hole numbers are processed and analyzed, the wavelet energy moment replaces wavelet packet frequency-band energy to form a wood defect feature vector, the wavelet packet energy moment is obtained after principal component analysis, dimension reduction processing is achieved, and features of defects of wood with different hole numbers can be effectively extracted.

Description

A kind of timber hole defect Ultrasonic Detection feature extracting method

[technical field]

The present invention relates to the technical field of Ultrasonic Detection, the technical field of the particularly Ultrasonic Detection feature extraction of timber hole defect.

[background technology]

Under current timber resources critical shortage situation, utilize the limited forest reserves rationally and effectively, the processing and utilization rate improving timber seems particularly important.Wood nondestructive testing is when not destroying wood quality, accurately detects defects in timber (defect size, position, kind), by processing accordingly the defect of timber in time, improves the utilization factor of timber.Wood nondestructive testing technology is emerging, a comprehensive technology, relates to the multi-door subject such as signal transacting, pattern-recognition.

Feature extraction is the bottleneck problem in current defects in timber identification, and it is directly connected to the accuracy of defect recognition.Carrying out one of signal transacting powerful is wavelet transformation, and it develops from Fourier's change, has good Time-Frequency Localization characteristic, and multiple dimensioned analysis with localization is the elite place of Wavelet transformation.Wavelet transformation wavelet theory solves and cannot describe and positioning signal Mutational part over time and frequency by Fourier transform simultaneously, and this is restricted with regard to making the application of Fourier analysis.Utilize BASED ON WAVELET ANALYSIS signal, the limitation in Fourier analysis application can be overcome.Wavelet analysis is the new branch of science closely grown up during the last ten years, and it is theoretical deep, is widely used.It is the epoch-making development result of Fourier analysis.Wavelet analysis be generally acknowledge in the world at present up-to-date time-frequency analysis tool, particularly be widely used in fields such as signal transacting, seismic prospecting, pattern-recognition, Fault diagnosis and supervise control, image procossing, it is considered to the important breakthrough in recent years on tool and method.The eigenwert of wood internal defect is extracted by the New view of open log detection with wavelet analysis.

[summary of the invention]

Object of the present invention solves the problems of the prior art exactly, a kind of timber hole defect Ultrasonic Detection feature extracting method is proposed, defects in timber proper vector is formed apart from replacing Wavelet Packet Frequency Band Energy with wavelet-packet energy, the wavelet-packet energy distance obtained after principal component analysis (PCA), achieve dimension-reduction treatment, effectively can extract the different numbers of hole defect characteristic of timber.

For achieving the above object, the present invention proposes a kind of timber hole defect Ultrasonic Detection feature extracting method, comprise concrete steps as follows:

A) according to Wavelet Analysis Theory, wavelet function is selected;

B) WAVELET PACKET DECOMPOSITION is carried out to timber hole defect ultrasound wave original signal, obtain 32 dimension frequency band energies, analyze various timber hole defect energy variation rule on each frequency band;

C) energy square is quoted in wavelet analysis, energy span is carried out to 32 dimension frequency band energies and calculates;

D) frequency band energy distance is tieed up in comparative analysis 32, forms defects in timber proper vector with wavelet-packet energy apart from replacing wavelet-packet energy;

E) carry out principal component analysis (PCA) to 32 dimension frequency band energies distances, realize normalized and dimension-reduction treatment defect characteristic information, digital proof extracts the different numbers of hole defect characteristic of timber effectively through the wavelet energy distance of principal component analysis (PCA) by experiment.

As preferably, in described b) step, select the transformation of variable of different decomposition layer to describe the feature of signal, carrying out WAVELET PACKET DECOMPOSITION Decomposition order to timber hole defect ultrasound wave original signal is 5 layers.All there is the best Decomposition order of a denoising effect in unlike signal, Decomposition order is very large for the impact of de-noising effect, and Decomposition order is too much, the useful information of signal will be made to lose serious, after de-noising, the signal to noise ratio (S/N ratio) of signal declines on the contrary, causes operand to increase simultaneously, makes process slack-off.Decomposition order is crossed signal to noise ratio (S/N ratio) at least and is improved seldom but the situation that there will not be signal to noise ratio (S/N ratio) to decline.

As preferably, in described c) step, introduce energy span M _ijconcept state energy size on frequency band and the energy characteristic distributions along time shaft, can both reflect energy distribution situation on a timeline by span, also reflects the size of energy on each frequency band, can more effectively extract the energy-distributing feature of signal on each frequency band by span.

As preferably, in described e) step, principal component analysis (PCA) utilizes the thought of dimensionality reduction, under the prerequisite not reducing the information content that raw data comprises, data set is converted into and is become to assign to represent by the validity feature that dimension is less, namely multiple feature is converted into a few comprehensive characteristics.

Beneficial effect of the present invention: the present invention, by carrying out Treatment Analysis to different numbers of hole defects in timber ultrasonic signal, proposes a kind of defect characteristic extracting method of different numbers of holes of wavelet energy distance.The primary bands distribution of wavelet energy distance is wide, and different numbers of hole is also different apart from frequency band number from main energetic, can both reflect energy distribution situation on a timeline by span, also reflects the size of energy on each frequency band, forming defects in timber proper vector with wavelet-packet energy apart from replacing Wavelet Packet Frequency Band Energy.The wavelet-packet energy distance simultaneously obtained after principal component analysis (PCA), achieves dimension-reduction treatment, effectively can extract the different numbers of hole defect characteristic of timber.

Feature of the present invention and advantage will be described in detail by reference to the accompanying drawings by embodiment.

[accompanying drawing explanation]

Fig. 1 is the ultrasonic identification process figure of timber hole defect of a kind of timber of the present invention hole defect Ultrasonic Detection feature extracting method;

Fig. 2 is that the defect characteristic of a kind of timber hole defect of the present invention Ultrasonic Detection feature extracting method extracts process flow diagram.

[embodiment]

Consult Fig. 1, Fig. 2, a kind of timber hole defect of the present invention Ultrasonic Detection feature extracting method, concrete steps:

The first step, according to Wavelet Analysis Theory, selects wavelet function.Wavelet transformation can be divided into continuous wavelet transform, multiresolution wavelet to convert and wavelet transform three major types, signal noise silencing process is exactly the computing of combining the decomposition and reconstruction of signal, select the signal that different wavelet basis process is identical, different results can be obtained, finding one group can the functional form of representation signal feature, does wavelet analysis to signal.When Selection of Wavelet Basis is suitable, characteristic component somewhere in time scale plane will be weakened relatively give prominence to and part dissimilar with wavelet basis, thus realize input and fault diagnosis, in order to effectively disclose the characteristic component of signal, need to select suitable wavelet basis.Wavelet basis is chosen and is mainly considered from rule and concrete object two aspect.

Second step, first carries out 5 layers of WAVELET PACKET DECOMPOSITION to timber hole defect ultrasound wave original signal, obtains 32 dimension frequency band energies, analyzes various timber hole defect energy variation rule on each frequency band.

WAVELET PACKET DECOMPOSITION frequency band energy feature extraction specific algorithm:

Ultrasound wave original signal carries out wavelet packet 5 layers decomposition.Represent original signal with S, represent the jth node of i-th layer that WAVELET PACKET DECOMPOSITION is set, wherein i=0,1,2 with (i, j) ..., N, j=0,1,2 ..., 2 ⁿ-1, N is the number of plies of decomposing, X _ijrepresent WAVELET PACKET DECOMPOSITION coefficient, adopt T=wpdec (s, 5, ' db5') (wherein db5 represents the wavelet type of employing, 5 represent Decomposition order), ultrasonic signal is decomposed, obtains 32 subsignals, extract the signal characteristic of the 5th layer of frequency content from low to high respectively;

WAVELET PACKET DECOMPOSITION coefficient is reconstructed, extracts the signal of each frequency band range.Reconstruction coefficients can adopt function S _5j=wprcoef (T, [5, j]) obtains (wherein T is for being reconstructed signal, and [5, j] represents a 5th layer of jth node), and represent the reconstruction signal of X50 with S50, S51 represents the reconstruction signal of X51, and the rest may be inferred for other, i.e. S _5jrepresent the coefficient of dissociation reconstruct of a 5th layer of jth node.Here, only analyze all nodes of layer 5, then resultant signal S can be expressed as:

S＝S ₅₀+S ₅₁+S ₅₂+...+S ₅₃₁

In test, the sample frequency f of signal _sfor 30kHz, signal is carried out wavelet packet 5 layers and decomposes 32 frequency bands, the frequency range corresponding to each band component sees the following form.

Ask the gross energy of each band signal.Because input signal is a random signal, its output is also a random signal.If S _ijcorresponding energy is E _ij, then have:

$E_{ij}=\∫|S_{ij}\left(t\right){|}^{2}dt=\underset{k=1}{\overset{n}{\Σ}}|x_{jk}{|}^2$

Wherein, xij represents the amplitude of the discrete point of reconstruction signal Sij, and trying to achieve each node energy can utility function E _5j=norm (S _5j) * norm (S _5j).

Structural attitude vector.During due to timber existing defects, wood quality there occurs change, can have larger impact to the energy of each inband signal.Therefore, be element with the subsignal energy differences of intact test specimen and defect test specimen, a proper vector can be constructed.Proper vector T is constructed as follows:

$T=\[T_{i0},T_{i1},...,T_{i(2^N-1)}\]$

Wherein $T_j=E_{\mathrm{ij}}^0-E_{\mathrm{ij}},j=0,\mathrm{1,2},...,2^N-1$

3rd step, quotes energy square in wavelet analysis, carries out energy span calculate 32 dimension frequency band energies.

For each inband signal S _ijcan span M _ijfor:

$T^M=\[T_1^M,T_2^M,...,T_{J+1}^M\]=\[M_{Ja},M_{Jd},M_{(J-1)d},...,M_{1d}\]/M$

If S _jacorresponding can span M _ja, S _id(i=J, J-1 ..., 1) corresponding can span be M _id(i=J, J-1 ..., 1) then have $M_{Ja}=\underset{k=1}{\Σ}{k}^{*}|x_{Jak}{|}^2,M_{id}=\underset{k=1}{\Σ}{k}^{*}|x_{idk}{|}^2$

Total can span be:

M＝M _Ja+M _Jd+M _(J-1)d+…+M _1d

4th step, forms defects in timber proper vector with wavelet-packet energy apart from replacing wavelet-packet energy.With normalized energy apart from constitutive characteristic vector:

$T^M=\[T_1^M,T_2^M,...,T_{J+1}^M\]=\[M_{Ja},M_{Jd},M_{(J-1)d},...,M_{1d}\]/M$

In formula, Δ t is sampling time interval; N is total hits; K sampled point.

5th step, principal component analysis (PCA), the characteristic parameter extraction of principal component analysis (PCA).

If characteristic matrix is X, xi _ja jth characteristic ginseng value of the i-th sample, feature samples has n, and Characteristic Number has p.

$X=\left(\begin{array}{cccc}{x}_{11}& x_{12}& \mathrm{...}& x_{1p}\\ x_{21}& x_{22}& \mathrm{...}& x_{2p}\\ .& .& .& .\\ .& .& .& .\\ .& .& .& .\\ x_{n1}& x_{n2}& \mathrm{...}& x_{np}\end{array}\right)$

The dimension impact eliminating different characteristic carries out standardization to initial characteristic data.

$x_{ij}=\frac{x_{ij}-{\stackrel{\‾}{x}}_j}{\sqrt{\mathrm{var}\left(x_j\right)}},(i=1,2,...,n;j=1,2,...,p)$

Wherein ${\stackrel{\‾}{x}}_j=\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{x}_{ij}$

$\mathrm{var}\left(x_j\right)=\frac{1}{n-1}\underset{i=1}{\overset{n}{\Σ}}{(x_{ij}-{\stackrel{\‾}{x}}_j)}^2,(j=\mathrm{1,2},...,p)$

The correlation matrix of eigenmatrix after normalized

$r_{ij}=\frac{\underset{k=1}{\overset{n}{\Σ}}(x_{ki}-{\stackrel{\‾}{x}}_i)(x_{kj}-{\stackrel{\‾}{x}}_j)}{\sqrt{\underset{k=1}{\overset{n}{\Σ}}{(x_{ki}-{\stackrel{\‾}{x}}_i)}^2\underset{k=1}{\overset{n}{\Σ}}{(x_{kj}-{\stackrel{\‾}{x}}_j)}^2}}$

Ask the eigenwert (λ of correlation matrix R ₁, λ ₂..., λ _p) and corresponding proper vector:

a _i＝(a _i1,a _i2,…,a _ip),i＝1,2,…,p

Determine number of principal components

Principal component analysis (PCA) can obtain p major component, in actual operation, how we choose major component, general is not choose p major component, but chooses the large major component of front k each contribution rate of accumulative total, and contribution rate refers to that the information of certain Principle component extraction accounts for the share of total information here.

Contribution rate is larger, and the information representing the original variable that this major component comprises is stronger, generally chooses the eigenwert that contribution rate of accumulative total reaches 85-95%.

Calculate major component

According to standardized raw data, substitute into major component expression formula respectively, just can obtain new feature under each major component.

The course of work of the present invention:

A kind of timber hole defect of the present invention Ultrasonic Detection feature extracting method in the course of the work, when the wood quality that ultrasound wave passes through changes, the effects such as signal reflects, reflect, scattering, ultrasonic signal energy is decayed, the form showed is that the amplitude of ultrasonic signal can change, and its frequency and phase place also change simultaneously.Wavelet decomposition launches by signal decomposition on different frequency bands by wavelet basis, and the approximate good and bad degree of this signal places one's entire reliance upon the selection of wavelet basis.Because wavelet transformation possesses following two functions: various interference noise can be filtered, thus the impact of stress release treatment, can effectively detected defect characteristic and defect point, so utilize wavelet transformation can extract the information of defect characteristic.

The present invention, by carrying out Treatment Analysis to different numbers of hole defects in timber ultrasonic signal, proposes a kind of defect characteristic extracting method of different numbers of holes of wavelet energy distance.First 5 layers of WAVELET PACKET DECOMPOSITION are carried out to ultrasonic signal, obtain 32 frequency band energies, carry out the analysis of energy span on this basis, and by pivot constituent analysis, normalized and dimension-reduction treatment are carried out to wavelet-packet energy distance.Experimental result shows: wavelet energy is apart from finding out compared with frequency band energy, the primary bands distribution of wavelet energy distance is wide, and different numbers of hole is also different apart from frequency band number from main energetic, can both reflect energy distribution situation on a timeline by span, also reflects the size of energy on each frequency band, therefore form defects in timber proper vector with wavelet-packet energy apart from replacing Wavelet Packet Frequency Band Energy.The wavelet-packet energy distance of 16 major components simultaneously obtained after principal component analysis (PCA), achieves dimension-reduction treatment, effectively can extract the different numbers of hole defect characteristic of timber.

Above-described embodiment is to explanation of the present invention, is not limitation of the invention, anyly all belongs to protection scope of the present invention to the scheme after simple transformation of the present invention.

Claims (4)

1. a timber hole defect Ultrasonic Detection feature extracting method, is characterized in that: comprise concrete steps as follows:

A) according to Wavelet Analysis Theory, wavelet function is selected;

B) WAVELET PACKET DECOMPOSITION is carried out to timber hole defect ultrasound wave original signal, obtain 32 dimension frequency band energies, analyze various timber hole defect energy variation rule on each frequency band;

C) energy square is quoted in wavelet analysis, energy span is carried out to 32 dimension frequency band energies and calculates;

D) frequency band energy distance is tieed up in comparative analysis 32, forms defects in timber proper vector with wavelet-packet energy apart from replacing wavelet-packet energy;

E) carry out principal component analysis (PCA) to 32 dimension frequency band energies distances, realize normalized and dimension-reduction treatment defect characteristic information, digital proof extracts the different numbers of hole defect characteristic of timber effectively through the wavelet energy distance of principal component analysis (PCA) by experiment.

2. a kind of timber hole defect Ultrasonic Detection feature extracting method as claimed in claim 1, it is characterized in that: in described b) step, select the transformation of variable of different decomposition layer to describe the feature of signal, carrying out WAVELET PACKET DECOMPOSITION Decomposition order to timber hole defect ultrasound wave original signal is 5 layers.

3. a kind of timber hole defect Ultrasonic Detection feature extracting method as claimed in claim 1, is characterized in that: in described c) step, and introducing can span M _ijconcept state energy size on frequency band and the energy characteristic distributions along time shaft, can both reflect energy distribution situation on a timeline by span, also reflects the size of energy on each frequency band, can more effectively extract the energy-distributing feature of signal on each frequency band by span.

4. a kind of timber hole defect Ultrasonic Detection feature extracting method as claimed in claim 1, it is characterized in that: in described e) step, principal component analysis (PCA) utilizes the thought of dimensionality reduction, under the prerequisite not reducing the information content that raw data comprises, data set is converted into and is become to assign to represent by the validity feature that dimension is less, namely multiple feature is converted into a few comprehensive characteristics.