CN106770013A - The method that doping urea milk is differentiated based on two-dimentional near-infrared correlation spectrum invariant moment features - Google Patents

Abstract

The invention discloses a kind of method that doping urea milk is differentiated based on two-dimensional correlation spectra invariant moment features, comprise the following steps：1) prepare plain chocolate and doping milk as training sample, obtain the two-dimentional near-infrared correlation spectrum figure of training sample；2) theoretical according to not bending moment, extraction step 1) two-dimentional near-infrared correlation spectrum figure invariant moment features；3) using PCA to step 2) gained invariant moment features carry out preferably, obtaining preferred invariant moment features；4) supporting vector machine model is set up, Classification and Identification is carried out to the two-dimentional near-infrared correlation spectrum figure of the training sample using SVMs method；5) by the preferred invariant moment features input step 4 of the two-dimentional near-infrared correlation spectrum figure of unknown species milk) in supporting vector machine model after training, you can differentiate the species of the unknown species milk.The method of this law people can be realized being processed while mass data, and identification effect is high.

Description

Differentiate doping urea milk based on two-dimentional near-infrared correlation spectrum invariant moment features Method

Technical field

The invention belongs to milk detection technique field, relate in particular to a kind of constant based on two-dimentional near-infrared correlation spectrum The method that moment characteristics differentiate doping urea milk.

Background technology

The various nutriments such as the protein containing needed by human body, fat, lactose, inorganic salts and calcium in milk, with people Growth in the living standard, the demand to dairy products is also increasing.Illegal retailer under the ordering about of interests to milk in mix The materials such as miscellaneous soybean protein, water, whey, melamine, it is right so as to cause milk safety problem to occur frequently to reduce production cost The healthy of consumer makes a big impact.

Traditional dairy produce quality testing is mainly using some physico-chemical processes, although these method results contrasts are accurate Really, but its process is typically complex, there are certain requirement, and time and effort consuming to experimenter, specific examination is also needed to sometimes Agent, has destructiveness to sample, is not suitable for the timely control of quality in raw material milk purchase.Infrared Spectrum Technology its process Simply, it is less demanding to experimenter, low cost, and without achieving that the Non-Destructive Testing to sample to sample treatment, be adapted to Detection to dopant in dairy products.The application mainly scientific research personnel of current two-dimensional correlation spectra technology is directed to research object With the specific outer synchronous spectrum and asynchronous spectrum disturbed, build contour form, qualitative analysis is carried out with reference to corresponding spectrum rule of reading.With section The continuous progress of technology, two-dimensional correlation spectra technology is just towards recognizing with chemometric model, artificial intelligence is combined, real The road of existing automatic identification.

Chinese invention patent publication number CN104316491A discloses a kind of based on synchronization-asynchronous two-dimentional near-infrared Correlated Spectroscopy The method that urea milk is mixed in detection, the method directly will synchronous-asynchronous two-dimentional near-infrared correlation spectrum matrix and multidimensional stoichiometry Learn the qualitative discrimination that set realizes mixing urea milk.Although its analysis result is preferably, synchronization-asynchronous two-dimentional near-infrared Correlated Spectroscopy Matrix is three-dimensional matrice, and comprising substantial amounts of data, it is necessary to multidimensional Chemical Measurement is modeled, model is complicated, calculates the time Long, efficiency is low.The method of discrimination for mixing urea milk proposed by the invention is built upon being carried out on the basis of extraction invariant moment features , it is not only effectively extracted the characteristic information of two-dimentional near-infrared Correlated Spectroscopy, and have compressed data, improves modeling efficiency.

The content of the invention

In view of the shortcomings of the prior art, it is constant based on two-dimentional near-infrared correlation spectrum it is an object of the invention to provide one kind The method that moment characteristics differentiate doping urea milk, the method sets up classification using the characteristic signal of algorithm of support vector machine and sample Device, by extracting two-dimentional near-infrared correlation spectrum invariant moment features and the algorithm of support vector machine scale-model investigation to sample classification, Whether detection milk adulterates.

The purpose of the present invention is achieved by following technical proposals.

A kind of method that doping urea milk is differentiated based on two-dimentional near-infrared correlation spectrum invariant moment features, including following step Suddenly：

1) prepare plain chocolate and doping milk as training sample, the one-dimensional near red of the training sample is obtained by scanning External spectrum, and Correlated Spectroscopy calculating is carried out by Noda theories, obtain the synchronous-asynchronous two-dimentional near-infrared of each training sample Light spectrum matrix (detailed calculating process is shown in that Yang Ren outstanding persons is based on Two-dimensional spectrum doping milk detection method research [D] University Of Tianjin, 2013.), and by resulting synchronous-asynchronous two-dimentional near infrared light spectrum matrix two-dimentional near-infrared correlation spectrum figure is changed into；

In the step 1) in, the synchronous-asynchronous two-dimentional near infrared light spectrum matrix changes into two-dimentional near-infrared correlation light The method of spectrogram is：By the data normalization of the synchronous-asynchronous two-dimentional near infrared light spectrum matrix between 0-255, pass through The image orders of matlab softwares obtain the two-dimentional near-infrared correlation spectrum figure.

In the above-mentioned technical solutions, using deviation standardization (min-max standardization) to the synchronous-asynchronous near-infrared The data of two-dimension spectrum matrix are normalized, and the standardized conversion formula of deviation is：

New data=(minimums of former data-original data) * 255/ (minimums of the maximum of former data-original data) (1)

Wherein, former data are the data of not normalized synchronous-asynchronous two-dimentional near infrared light spectrum matrix, and new data is to return The data of the synchronous-asynchronous two-dimentional near infrared light spectrum matrix after one change.

2) theoretical according to not bending moment, extraction step 1) the two-dimentional near-infrared correlation spectrum figure of gained invariant moment features；

In the step 2) in, the method for extracting invariant moment features is：

1. two-dimensional matrix f (x, y) of the described two-dimentional near-infrared correlation spectrum figure of m × n is calculated according to formula (2) and (3) P+q rank central moments u_pq：

Wherein, p and q are integer, and p and q=0,1,2 and 3, f (x, y) are the synchronous-asynchronous two-dimentional near-infrared for being obtained Light spectrum matrix is by the two-dimensional matrix after method for normalizing conversion, M₀₀It is the Two-Dimensional Moment of the two-dimentional near-infrared correlation spectrum figure The zeroth order square of battle array f (x, y), M₁₀And M₀₁It is the first moment of two-dimensional matrix f (x, y) of the two-dimentional near-infrared correlation spectrum figure；

2. the M of two-dimensional matrix f (x, y) of the two-dimentional near-infrared correlation spectrum figure is calculated by formula (4)₀₀、M₁₀With M₀₁：

Wherein, m × n is the resolution ratio of the two-dimentional near-infrared correlation spectrum figure.

3. as formula (5) to step it is 1. middle obtained by p+q rank central moments u_pqIt is normalized, obtains η_pq,

η_pq=u_pq/u₀₀ ^γ (5)

Wherein,

4. by step 3. gained η_pqFormula (6) is substituted into successively, obtains invariant moment features φ₁、φ₂、φ₃、φ₄、φ₅、φ₆ And φ₇。

3) using PCA to step 2) gained invariant moment features carry out preferably, obtaining preferred invariant moment features；

In the step 3) in, to step 2) gained invariant moment features carry out preferred method and are：

I is trained the invariant moment features composition characteristic data matrix X '=(x of sample by multiple according to formula (7)_ij')_s×l：

Wherein, x_ij' it is i-th j-th invariant moment features parameter of training sample, l is invariant moment features number, and s is instruction Practice the number of sample；Characteristic matrix X ' described in standardization, obtains X=(x_ij)_s×l, wherein,

Wherein,

II calculates the coefficient correlation of X according to formula (11), obtains correlation matrix R=(r_ij)_s×l：

III calculates the characteristic value (λ of the correlation matrix R₁,λ₂,…,λ_l)；

IV calculates contribution rate：Step III gained characteristic values are substituted into formula (12) successively, l contribution rate is obtained；

V determines principal component：L step IV gained contribution rate is arranged and added up from big to small, contribution rate is chosen and is tired out It is value added be more than or equal to 85% preceding t principal component, determine the t principal component invariant moment features be preferred invariant moment features.

4) set up supporting vector machine model, using SVMs method to it is described training sample two-dimentional near-infrared correlation light Spectrogram carries out Classification and Identification, is input into as step 3) obtained by preferred invariant moment features, be output as train sample species：Plain chocolate Or doping milk, wherein, call the training function Svmtrain of supporting vector machine model to be trained the training sample；

5) calculate according to the method described above unknown species milk two-dimentional near-infrared correlation spectrum figure preferably bending moment is special, will The preferred invariant moment features input step 4 of the two-dimentional near-infrared correlation spectrum figure of unknown species milk) in training after support to In amount machine model, you can differentiate the species of the unknown species milk.

Compared to prior art, the method for the present invention is characterized pure using the invariant moment features of two-dimentional near-infrared correlation spectrum figure The information that milk is included with doping milk, image processing techniques is combined with milk doping differentiation, is different from conventional The synchronous spectrum and asynchronous spectrum of contour form are built, qualitative analysis is carried out with reference to corresponding spectrum rule of reading.The method can be realized largely Processed while data, identification effect is high.In addition, the present invention is by two-dimentional near-infrared correlation spectrum figure invariant moment features and props up Vector machine is held to be combined, by extract the invariant moment features of two-dimentional near-infrared correlation spectrum figure by two-dimentional near-infrared spectrum technique with Pattern-recognition is combined, and realizes the qualitative discrimination of doping milk and plain chocolate.Spectrogram human eye is the method overcome directly to compare To haveing the shortcomings that subjective erroneous judgement and largely spectrogram cannot compare simultaneously, identification effect and accuracy are high.

Brief description of the drawings

Fig. 1 is the flow chart of the inventive method；

Fig. 2 is the schematic diagram of SVMs.

Specific embodiment

Below in conjunction with the accompanying drawings technical scheme is further illustrated with specific embodiment.

As shown in Fig. 1~2, a kind of method that doping urea milk is differentiated based on two-dimensional correlation spectra invariant moment features, bag Include following steps：

1) prepare plain chocolate and doping milk as training sample, the one-dimensional near red of the training sample is obtained by scanning External spectrum, and Correlated Spectroscopy calculating is carried out by Noda theories, obtain the synchronous-asynchronous two-dimentional near-infrared of each training sample Light spectrum matrix, two-dimentional near-infrared correlation spectrum figure is changed into by the synchronous-asynchronous two-dimentional near infrared light spectrum matrix, sets up sample Product picture library.

Wherein, the method for obtaining synchronous-asynchronous two-dimentional near infrared light spectrum matrix is recorded in Publication No.： (also refer to document Yang Ren outstanding persons to be ground based on Two-dimensional spectrum doping milk detection method in the patent of invention of CN104316491A Study carefully [D] University Of Tianjin, 2013).Synchronous-asynchronous near-infrared two-dimension spectrum matrix changes into two-dimentional near-infrared correlation spectrum figure Method is：By the data normalization of the synchronous-asynchronous two-dimentional near infrared light spectrum matrix between 0-255, normalized side Method has many kinds, using deviation standardization (min-max standardization) to synchronous-asynchronous in specific embodiment of the invention The data of two-dimentional near infrared light spectrum matrix are normalized, and end value is mapped between [0-255].The standardized conversion of deviation Formula is：

New data=(minimums of former data-original data) * 255/ (minimums of the maximum of former data-original data) (1)

Wherein, former data are the data of not normalized synchronous-asynchronous two-dimentional near infrared light spectrum matrix, and new data is to return The data of the synchronous-asynchronous two-dimentional near infrared light spectrum matrix after one change.

Data to normalized synchronous-asynchronous two-dimentional near infrared light spectrum matrix are ordered using the image () of matlab softwares Order can obtain two-dimentional near-infrared correlation spectrum figure.

2) theoretical according to not bending moment, extraction step 1) the two-dimentional near-infrared correlation spectrum figure of gained invariant moment features；

In the step 2) in, the method for extracting invariant moment features is：

1. two-dimensional matrix f (x, y) of the described two-dimentional near-infrared correlation spectrum figure of m × n is calculated according to formula (2) and (3) P+q rank central moments u_pq：

Wherein, p and q are integer, and p and q=0,1,2 and 3, f (x, y) are synchronous-asynchronous two-dimentional near infrared light spectrum matrix Two-dimensional matrix after being converted by foregoing method for normalizing, M₀₀It is the two-dimensional matrix of the two-dimensional correlation spectra figure

The zeroth order square of f (x, y), M₁₀And M₀₁It is two-dimensional matrix f (x, y) of the two-dimentional near-infrared correlation spectrum figure First moment；

2. the M of two-dimensional matrix f (x, y) of two-dimentional near-infrared correlation spectrum figure is calculated by formula (4)₀₀、M₁₀And M₀₁：

Wherein, m × n is the resolution ratio of the two-dimentional near-infrared correlation spectrum figure.

3. as formula (5) to step it is 1. middle obtained by p+q rank central moments u_pqIt is normalized, obtains η_pq,

η_pq=u_pq/u₀₀ ^γ (5)

Wherein,

4. by step 3. gained η_pqFormula (6) is substituted into successively, obtains invariant moment features φ₁、φ₂、φ₃、φ₄、φ₅、φ₆ And φ₇。

3) using PCA to step 2) gained invariant moment features carry out preferably, obtaining preferred invariant moment features；

Wherein, to step 2) gained invariant moment features carry out preferred method and are：

I is trained the invariant moment features composition characteristic data matrix X '=(x of sample by multiple according to formula (7)_ij')_s×l：

Wherein, x_ij' it is i-th j-th invariant moment features parameter of training sample, l is invariant moment features number, and s is instruction Practice the number of sample；Original invariant moment features data dimension is different, certain influence can be produced to result, in order to eliminate difference The dimension impact of invariant moment features, it is necessary to characteristic matrix X ' described in standardization, obtains X=(x_ij)_s×l, wherein,

Wherein,

II calculates the coefficient correlation of X according to formula (11), obtains correlation matrix R=(r_ij)_s×l：

III calculates the characteristic value (λ of the correlation matrix R₁,λ₂,…,λ_l)；

IV calculates contribution rate：Contribution rate refers to that the variance of certain principal component accounts for the proportion of whole variances, it is actual namely certain Individual characteristic value accounts for the total proportion of All Eigenvalues.Step III gained characteristic values are substituted into formula (12) successively, l tribute is obtained Offer rate；

V determines principal component：L step IV gained contribution rate is arranged and added up from big to small, contribution rate is chosen and is tired out It is value added be more than or equal to 85% preceding t principal component, determine the t principal component invariant moment features be preferred invariant moment features；

4) SVMs (SVM) model, SVMs principle are set up as shown in Fig. 2 input is for X1, X2 ..., Xn, Y is output as, inner product kernel function is K1 (X), K2 (X) ... ..., Kn (X).Using SVMs method to the two of the training sample Dimension near-infrared correlation spectrum figure carry out Classification and Identification, be input into as step 3) obtained by preferred invariant moment features, be output as train sample The species of product：Plain chocolate or doping milk, wherein, the training function Svmtrain of supporting vector machine model is called to the training Sample is trained；

5) by the preferred invariant moment features input step 4 of the two-dimentional near-infrared correlation spectrum figure of unknown species milk) middle instruction In supporting vector machine model after white silk, you can differentiate the species of the unknown species milk.

Below by taking plain chocolate and doping urea milk as an example, technical method of the invention is illustrated.

1) prepare 27 plain chocolates and 27 doping urea milk as training sample, 54 training samples are obtained by scanning The synchronization of product-asynchronous two-dimentional near infrared light spectrum matrix, two-dimentional near-infrared is changed into by synchronization-asynchronous two-dimentional near infrared light spectrum matrix Correlation spectrum figure；

2) theoretical according to not bending moment, extraction step 1) the two-dimentional near-infrared correlation spectrum figure of gained invariant moment features, the He of table 1 Table 2 is respectively 5 plain chocolates and 5 invariant moment features data of doping urea milk in 54 training samples；

The invariant moment features data of the synchronization of the plain chocolate of table 1-asynchronous two-dimentional atlas of near infrared spectra

The invariant moment features data of the synchronization-asynchronous two-dimentional atlas of near infrared spectra of the doping urea milk of table 2

3) using PCA to step 2) gained invariant moment features carry out preferably, obtain 4 preferably bending moment spy Levy, as shown in table 3.

The principal component analysis result of table 3

As shown in Table 3, by principal component analysis, plain chocolate with doping urea synchronous-asynchronous two-dimentional atlas of near infrared spectra Preceding 4 principal component contribution rate of accumulative total be 92.94%, reach more than 85%, can very well extract the distribution of former characteristic statistic Characteristic, i.e. not bending moment φ₁, not bending moment φ₃, not bending moment φ₂Not bending moment φ₆It is preferred invariant moment features.

4) set up supporting vector machine model, using SVMs method to it is described training sample two-dimentional near-infrared correlation light Spectrogram carries out Classification and Identification, is input into as step 3) obtained by 54 training sample preferred invariant moment features, be output as train sample The species of product：Plain chocolate or doping milk, wherein, the training function Svmtrain of supporting vector machine model is called to the training Sample is trained.External prediction is carried out to calibration set using supporting vector machine model, its result shows, in 54 training samples There are 3 differentiation mistakes, wherein there is a plain chocolate to be mistaken for the milk of doping urea, two doping urea milk are not mistaken for Plain chocolate.Built supporting vector machine model is 94.4444% to the differentiation accuracy rate of calibration set.

5) using 13 plain chocolates and 13 doping urea milk as unknown species milk, by 26 unknown species milk The preferred invariant moment features input step 4 of two-dimentional near-infrared correlation spectrum figure) in training after supporting vector machine model in, you can Differentiate the species of the unknown species milk.External prediction is carried out to calibration set using model, its result shows, 26 sample numbers There are 4 differentiation mistakes in, wherein plain chocolate differentiates that result is all correct, and the milk of 4 doping urea is mistaken for plain chocolate.Institute Established model is 84.6154% to the differentiation accuracy rate of calibration set.

Exemplary description is done to the present invention above, it should explanation, do not departed from the situation of core of the invention Under, any simple deformation, modification or other skilled in the art can not spend the equivalent of creative work equal Fall into protection scope of the present invention.

Claims (5)

1. it is a kind of based on two-dimensional correlation spectra invariant moment features differentiate doping urea milk method, it is characterised in that including with Lower step：

1) prepare plain chocolate and doping milk as training sample, the one-dimensional near infrared light of the training sample is obtained by scanning Spectrum, and Correlated Spectroscopy calculating is carried out by Noda theories, obtain the synchronous-asynchronous two-dimentional near infrared spectrum of each training sample Matrix, and resulting synchronous-asynchronous two-dimentional near infrared light spectrum matrix is changed into two-dimentional near-infrared correlation spectrum figure；

2) theoretical according to not bending moment, extraction step 1) the two-dimentional near-infrared correlation spectrum figure of gained invariant moment features；

3) using PCA to step 2) gained invariant moment features carry out preferably, obtaining preferred invariant moment features；

4) set up supporting vector machine model, using SVMs method to it is described training sample two-dimentional near-infrared correlation spectrum figure Carry out Classification and Identification, be input into as step 3) obtained by preferred invariant moment features, be output as train sample species：Plain chocolate is mixed Miscellaneous milk, wherein, call the training function Svmtrain of supporting vector machine model to be trained the training sample；

5) by the preferred invariant moment features input step 4 of the two-dimentional near-infrared correlation spectrum figure of unknown species milk) in training after Supporting vector machine model in, you can differentiate the species of the unknown species milk.

2. method according to claim 1, it is characterised in that in the step 1) in, the synchronous-asynchronous two dimension is near The method that infrared spectrum matrix changes into two-dimentional near-infrared correlation spectrum figure is：By the synchronous-asynchronous two-dimentional near infrared spectrum The data normalization of matrix obtains the two-dimentional near-infrared correlation light between 0-255 by the image orders of matlab softwares Spectrogram.

3. method according to claim 2, it is characterised in that near to the synchronous-asynchronous two dimension using deviation standardization The data of infrared spectrum matrix are normalized, and the standardized conversion formula of deviation is：

New data=(minimums of former data-original data) * 255/ (minimums of the maximum of former data-original data) (1) its In, former data are the data of not normalized synchronous-asynchronous two-dimentional near infrared light spectrum matrix, and new data is same after normalizing The data of step-asynchronous two-dimentional near infrared light spectrum matrix.

4. method according to claim 3, it is characterised in that in the step 2) in, the method for extracting invariant moment features For：

1. the p+q of two-dimensional matrix f (x, y) of the described two-dimentional near-infrared correlation spectrum figure of m × n is calculated according to formula (2) and (3) Rank central moment u_pq：

$u_{pq}=\underset{x=1}{\overset{m}{\Σ}}\underset{y=1}{\overset{n}{\Σ}}{(x-\stackrel{\‾}{x})}^p{(y-\stackrel{\‾}{y})}^{q}f(x,y)---\left(2\right)$

$\stackrel{\‾}{x}=\frac{M_{10}}{M_{00}},\stackrel{\‾}{y}=\frac{M_{01}}{M_{00}}---\left(3\right)$

Wherein, p and q are integer, and p, q=0,1,2 and 3, f (x, y) are the synchronous-asynchronous two-dimentional near infrared light spectrum matrix Two-dimensional matrix after being converted according to method for normalizing described in claim 3, M₀₀It is the two of the two-dimentional near-infrared correlation spectrum figure The zeroth order square of dimension matrix f (x, y), M₁₀And M₀₁It is the one of two-dimensional matrix f (x, y) of the two-dimentional near-infrared correlation spectrum figure Rank square；

2. the M of two-dimensional matrix f (x, y) of the two-dimentional near-infrared correlation spectrum figure is calculated by formula (4)₀₀、M₁₀And M₀₁：

$M_{pq}=\underset{x=1}{\overset{m}{\Σ}}\underset{y=1}{\overset{n}{\Σ}}{x}^p{y}^{q}f(x,y)---\left(4\right)$

Wherein, m × n is the resolution ratio of the two-dimentional near-infrared correlation spectrum figure.

3. as formula (5) to step it is 1. middle obtained by p+q rank central moments u_pqIt is normalized, obtains η_pq,

η_pq=u_pq/u₀₀ ^γ (5)

Wherein,

4. by step 3. gained η_pqFormula (6) is substituted into successively, obtains invariant moment features φ₁、φ₂、φ₃、φ₄、φ₅、φ₆With φ₇。

5. method according to claim 1, it is characterised in that in the step 3) in, to step 2) bending moment is not special for gained Levying carries out preferred method and is：

I is trained the invariant moment features composition characteristic data matrix X '=(x of sample by multiple according to formula (7)_ij')_s×l：

$X^{\′}=\left(\begin{array}{cccc}{x_{11}}^{\′}& {x_{12}}^{\′}& \mathrm{...}& {x_{1l}}^{\′}\\ {x_{21}}^{\′}& {x_{22}}^{\′}& \mathrm{...}& {x_{2l}}^{\′}\\ .& .& .& .\\ .& .& .& .\\ .& .& .& .\\ {x_{s1}}^{\′}& {x_{s2}}^{\′}& \mathrm{...}& {x_{sl}}^{\′}\end{array}\right)---\left(7\right)$

Wherein, x_ij' it is i-th j-th invariant moment features parameter of training sample, l is invariant moment features number, and s is training sample The number of product；Characteristic matrix X ' described in standardization, obtains X=(x_ij)_s×l, wherein,

$x_{ij}=\frac{{x_{ij}}^{\′}-{{\stackrel{\‾}{x}}_j}^{\′}}{\sqrt{\mathrm{var}\left({x_j}^{\′}\right)}},(i=1,2,...,s;j=1,2,...,l)---\left(8\right)$

Wherein,

$\mathrm{var}\left({x_j}^{\′}\right)=\frac{1}{s-1}\underset{i=1}{\overset{s}{\Σ}}{({x_{ij}}^{\′}-{x_j}^{\′})}^2,(j=1,2,...,l)---\left(10\right)$

II calculates the coefficient correlation of X according to formula (11), obtains correlation matrix R=(r_ij)_s×l：

$r_{ij}=\frac{\underset{k=1}{\overset{s}{\Σ}}(x_{ki}-{\stackrel{\‾}{x}}_i)(x_{kj}-{\stackrel{\‾}{x}}_j)}{\sqrt{\underset{k=1}{\overset{s}{\Σ}}{(x_{ki}-{\stackrel{\‾}{x}}_i)}^2\underset{k=1}{\overset{s}{\Σ}}{(x_{kj}-{\stackrel{\‾}{x}}_j)}^2}}---\left(11\right)$

III calculates the characteristic value (λ of the correlation matrix R₁,λ₂,…,λ_l)；

IV calculates contribution rate：Step III gained characteristic values are substituted into formula (12) successively, l contribution rate is obtained；

V determines principal component：L step IV gained contribution rate is arranged and added up from big to small, contribution rate accumulated value is chosen Preceding t principal component more than or equal to 85%, the invariant moment features for determining the t principal component are preferred invariant moment features.