CN105784672A - Drug detector standardization method based on dual-tree complex wavelet algorithm - Google Patents

Abstract

The invention discloses a drug detector standardization method based on the dual-tree complex wavelet algorithm. Firstly, multi-scale spectrum decomposition and reconstruction are conducted with the dual-tree complex wavelet, then each layer of reconstructed spectrum is corrected through piecewise direct standardization, predication models based on partial least squares and leave one cross validation are established, and finally all the predication models are fused and evaluated according to a calculated weight. According to the dual-tree complex wavelet and direct standardization combined method, due to the translation invariance and multi-scale property of the dual-tree complex wavelet, the defects of existing model transfer methods are overcome, the capacity of correcting drifting in the X-axis direction and Y-axis direction is excellent, and the method also has the advantages of being exquisite, precise and efficient and can be widely applied to the fields including near-infrared and Raman spectrum.

Description

A kind of drug detecting instrument standardized method based on dual-tree complex wavelet algorithm

Technical field

The present invention relates to chemometric techniques field, particularly relate to a kind of drug detecting instrument standardized method based on dual-tree complex wavelet algorithm.

Background technology

In recent years, along with the continuous of drug issue is paid attention to by China, Raman spectrum is directed initially into drug testing field, becomes one of the modern weapon quick and precisely checking drugs and most preferred technology.Raman spectrum analysis technology is the molecular structure characterization technology set up based on Raman effect, its spectrum peak position, quantity and intensity etc. directly reflect composition and the Constellation information of molecule, there is quick, easy, original position without undermining the features such as agent of being excused from an examination, can directly the sample of different shape be tested.By resolving the Raman dactylogram of complex system, qualitative, the quantitative information of various ingredients can be obtained simultaneously, and can be prevented effectively from sample the interference of moisture.Raman spectrum analysis technology belongs to a layer that technology content in optical detection is the highest, there is stronger database function, as long as the Raman spectrogram of common drugs is built storehouse, can be undertaken looking into spectrum by the automatic search function of spectrogram, determine the one in the sample drugs whether spectrogram storehouse is taken in, can within several seconds the multiple analytical data of one sample of non-destructive determination, possess wide application prospect.

In the real process of illicit drugs inspection, owing to there is certain difference between different Raman spectrometers, cause that the Raman spectrum that same sample detects is existed quite poor by it different, it is unfavorable for follow-up spectral peak identification, therefore, pole is necessary different drug detecting instruments is standardized work, to realize the testing result concordance between different instrument.Equipment Standardization, refers to after Mathematical treatment, makes the model on an instrument can be used in another instrument, again models, thus reducing, the huge workload brought, it is achieved sharing of sample and data resource.The most general and the most successful method of application at present is multivariate calibration algorithm, and wherein better with piecewise direct standardization (PDS) algorithm and innovatory algorithm effect on its basis.

For spectrum, the difference between two instrument spectral can be summarized as two aspects, and one is the difference on x-axis direction, i.e. the displacement of waveform；The difference on y-axis direction on the other hand, i.e. the deformation of waveform.Spectrum itself is multiple dimensioned, and its information comprised also is multiple dimensioned.Difference between the spectrum of different spectrogrphs, it is possible to existing only in a part of frequency domain, the information of other parts is then identical.Piecewise direct standardization algorithm is to be corrected in whole spectrum aspect, it does not have in frequency domain, spectrum is divided, and such Model Transfer is very general, not finely.And the multi-scale wavelet piecewise direct standardization algorithm on piecewise direct standardization algorithm basis, the feature of its multiple dimensioned correction overcomes general, the fine not shortcoming of piecewise direct standardization method, but the variable shortcoming that translates that small echo has causes the poor ability of this method drift on two instrument spectral x-axis directions of correction.

Summary of the invention

Because the drawbacks described above of prior art, the technical problem to be solved is to provide a kind of drug detecting instrument standardized method based on dual-tree complex wavelet algorithm, it introduces dual-tree complex wavelet algorithm, there is advantage fine, accurate, efficient, with the translation transmutability effectively overcoming conventional wavelet to convert, it it is a kind of novel drug detecting instrument standardized instrument.

For achieving the above object, the invention provides a kind of drug detecting instrument standardized method based on dual-tree complex wavelet algorithm, it is characterised in that comprise the following steps:

S1, selection also set two illicit drugs inspection instruments respectively as master and slave instrument, the host and slave processors instrument spectral of collected specimens respectively；

Sample spectra is carried out preferably by S2, employing Kennard-Stone algorithm, weeds out abnormal sample spectrum in sample spectra, then selects and set training set and test set；

S3, for sample spectra characteristic, set best Decomposition order, and according to best Decomposition order, sample spectra carried out dual-tree complex wavelet transform, obtain each layer decomposition coefficient；

S4, decomposition coefficient to each layer are reconstructed respectively, obtain the spectrum of each layer reconstruct；

S5, the spectrum that each layer is reconstructed, direct standardization (PDS) algorithm is used to be corrected from machine spectrum, then the main frame spectrum of training set and the chemical true value concentration of sample are used, set up the forecast model based on partial least square method and leave one cross validation, then from machine spectrum, the test set after correction is inputted this model to be predicted the outcome, and obtain the cross validation root-mean-square error RMSECV of each forecast model；

S6, calculated the weights of each forecast model by cross validation root-mean-square error, use weights that all of forecast model carries out Model Fusion, and calculate RMSEP value and correlation coefficient carrys out evaluation model transmission effect.

Further, the master and slave instrument described in step S1 need to be ensured of the instrument of same class model, and gathering spectrum is the spectrum that same sample gathers different instrument under same measuring condition.

Further, the reconstruct described in step S4 is that each layer of wavelet coefficient after the decomposition of the spectrum to different instruments is reconstructed respectively.

Further, direct standardization (PDS) algorithm described in step S5 is a kind of Model Transfer method of multivariate calibration, and its detailed process is as follows:

First, the window of a K+W+1 length is chosen in training set from machine spectrally i-th wavelength points

Zi, order

Z_i=[a_{S, i-k}, a_{S, i-k+1}..., a_{S, i+w-1}, a_{S, i+w}]

Then, the i-th wavelength points of training set main frame is constructed a multiple regression equation with Zi,

Namely

a_{M, i}=Z_ib_i+e_i

Using main constituent (PCA) method to solve this equation, be then placed on the leading diagonal of transition matrix F by solution coefficient b out, other set to 0, and just obtain transition matrix F, as follows:

Finally by transition matrix, test set is converted to, from machine spectrum, the spectrum that main frame spectrum matches, is achieved that Model Transfer, it may be assumed that

X_{S, pds}=X_s·F

In formula: as, i be training set from machine spectrum i-th wavelength points, am, i is training set main frame spectrum i-th wavelength points, bi is conversion coefficient, and ei is error, and F is transition matrix, Xs be test set from machine spectrum, Xs, pds is the correction spectrum obtained after direct standardization (PDS) method.

Further, the cross validation root-mean-square error described in step S5 is RMSECV, and its formula is as follows:

In formula: n is test set sample number, CNIR is a certain actual attribute matrix of sample, is generally concentration array；CREF is the sample properties matrix doped.

Further, described in step S6 use weights forecast model is carried out Model Fusion operation particularly as follows:

$W_i=\frac{{\left(\frac{1}{{\mathrm{RMSECV}}_i}\right)}^2}{{\mathrm{\Σ}}_i^m{\left(\frac{1}{{\mathrm{RMSECV}}_i}\right)}^2}$

Being merged by forecast model according to weights Wi, formula is as follows:

$C=\underset{i}{\overset{m}{\Σ}}{W}_i\·C_{iREF}$

Wherein, CiREF is predicting the outcome of forecast model, and m is the yardstick decomposed, and C is predicting the outcome after forecast model merges, namely final Model Transfer result.

Further, the RMSEP described in step S6 is the predicted root mean square error of forecast model, represents the deviation between model predication value and chemistry true value, and for the most important parameter of quality of evaluation model, formula is as follows:

$RMSEP=\sqrt{\frac{{\mathrm{\Σ}}_1^n{(C_{NIR}-C_{REF})}^2}{n}}$

In formula: n is test set sample number, CNIR is a certain actual attribute matrix of sample, is generally concentration array；CREF is the sample properties matrix doped.

Further, the correlation coefficient described in step S6, represent the degree of correlation between model predication value and chemistry true value, formula is as follows:

$R=\frac{{\Σ}_1^n{C}_{NIR}{C}_{REF}-\frac{{\mathrm{\Σ}}_1^n{C}_{NIR}{\mathrm{\Σ}}_1^n{C}_{REF}}{n}}{\sqrt{({\Σ}_1^n{C_{NIR}}^2-\frac{{\left({\mathrm{\Σ}}_1^n{C}_{NIR}\right)}^2}{n})({\Σ}_1^n{C_{NIR}}^2-\frac{{\left({\mathrm{\Σ}}_1^n{C}_{REF}\right)}^2}{n})}}$

In formula: n is test set sample number, CNIR is a certain actual attribute matrix of sample, is generally concentration array；CREF is the sample properties matrix doped.

The invention has the beneficial effects as follows:

Spectrum is carried out multi-resolution decomposition initially with dual-tree complex wavelet and reconstructs by the present invention, then each layer of spectrum of reconstruct is corrected by application piecewise direct standardization, then set up the forecast model based on partial least square method and leave one cross validation, finally by the weight calculated all of forecast model merged and be evaluated.The method that the dual-tree complex wavelet that the present invention adopts combines with direct standardization, the translation invariance having due to dual-tree complex wavelet and multiple dimensioned characteristic, overcome the deficiency of existing Model Transfer method, the ability of the drift in correction x-axis and y-axis direction is all very outstanding, there is advantage fine, accurate, efficient simultaneously, the field such as near-infrared and Raman spectrum can be widely used in.

Below with reference to accompanying drawing, the technique effect of the design of the present invention, concrete structure and generation is described further, to be fully understood from the purpose of the present invention, feature and effect.

Accompanying drawing explanation

Fig. 1 is the flowchart of the present invention.

Detailed description of the invention

As it is shown in figure 1, a kind of drug detecting instrument standardized method based on dual-tree complex wavelet algorithm, it is characterised in that comprise the following steps:

S1, selection also set two illicit drugs inspection instruments respectively as master and slave instrument, the host and slave processors instrument spectral of collected specimens respectively.

Sample spectra is carried out preferably by S2, employing Kennard-Stone algorithm, weeds out abnormal sample spectrum in sample spectra, then selects and set training set and test set；

S3, for sample spectra characteristic, set best Decomposition order, and according to best Decomposition order, sample spectra carried out dual-tree complex wavelet transform, obtain each layer decomposition coefficient；

S4, decomposition coefficient to each layer are reconstructed respectively, obtain the spectrum of each layer reconstruct；

S5, the spectrum that each layer is reconstructed, direct standardization (PDS) algorithm is used to be corrected from machine spectrum, then the main frame spectrum of training set and the chemical true value concentration of sample are used, set up the forecast model based on partial least square method and leave one cross validation, then from machine spectrum, the test set after correction is inputted this model to be predicted the outcome, and obtain the cross validation root-mean-square error RMSECV of each forecast model；

S6, calculated the weights of each forecast model by cross validation root-mean-square error, use weights that all of forecast model carries out Model Fusion, and calculate RMSEP value and correlation coefficient carrys out evaluation model transmission effect.

Further, the master and slave instrument described in step S1 need to be ensured of the instrument of same class model, and gathering spectrum is the spectrum that same sample gathers different instrument under same measuring condition.Master and slave instrument refers to same model, two different drug detecting instruments.In this example, master and slave instrument is two drug detecting instruments instrument m5 and mp5, m5 is main frame, and mp5 is from machine.

Further, the reconstruct described in step S4 is that each layer of wavelet coefficient after the decomposition of the spectrum to different instruments is reconstructed respectively.Using dual-tree complex wavelet that sample spectra is carried out multi-resolution decomposition, its decomposition scale is relevant with spectrum its length.Assume that spectra length is 2ⁿ, then decomposition scale is not to be exceeded n-1, and decomposition scale should be not too small yet simultaneously, can cause that the useful information in signal is difficult to separate with interference factor, and therefore setting decomposition scale ranges for (n-1)/2 and arrives n-1.In this example, the physical length of flour sample spectrum is 700, then can decompose range scale for [5,9].After determining decomposition scale scope, use RMSEP value and correlation coefficient to determine best decomposition scale, and the wavelet coefficient after decomposing is reconstructed.

Further, direct standardization (PDS) algorithm described in step S5 is a kind of Model Transfer method of multivariate calibration, and its detailed process is as follows:

First, the window of a K+W+1 length is chosen in training set from machine spectrally i-th wavelength points

Zi, order

Z_i=[a_{S, i-k}, a_{S, i-k+1}..., a_{S, i+w-1}, a_{S, i+w}]

Then, the i-th wavelength points of training set main frame is constructed a multiple regression equation with Zi,

Namely

a_{M, i}=Z_ib_i+e_i

Using main constituent (PCA) method to solve this equation, be then placed on the leading diagonal of transition matrix F by solution coefficient b out, other set to 0, and just obtain transition matrix F, as follows:

Finally by transition matrix, test set is converted to, from machine spectrum, the spectrum that main frame spectrum matches, is achieved that Model Transfer, it may be assumed that

X_{S, pds}=X_s·F

In formula: as, i be training set from machine spectrum i-th wavelength points, am, i is training set main frame spectrum i-th wavelength points, bi is conversion coefficient, and ei is error, and F is transition matrix, Xs be test set from machine spectrum, Xs, pds is the correction spectrum obtained after direct standardization (PDS) method.

Further, the cross validation root-mean-square error described in step S5 is RMSECV, and its formula is as follows:

In formula: n is test set sample number, CNIR is a certain actual attribute matrix of sample, is generally concentration array；CREF is the sample properties matrix doped.

Further, described in step S6 use weights forecast model is carried out Model Fusion operation particularly as follows:

$W_i=\frac{{\left(\frac{1}{{\mathrm{RMSECV}}_i}\right)}^2}{{\mathrm{\Σ}}_i^m{\left(\frac{1}{{\mathrm{RMSECV}}_i}\right)}^2}$

Being merged by forecast model according to weights Wi, formula is as follows:

$C=\underset{i}{\overset{m}{\Σ}}{W}_i\·C_{iREF}$

Wherein, CiREF is predicting the outcome of forecast model, and m is the yardstick decomposed, and C is predicting the outcome after forecast model merges, namely final Model Transfer result.

Further, the RMSEP described in step S6 is the predicted root mean square error of forecast model, represents the deviation between model predication value and chemistry true value, and for the most important parameter of quality of evaluation model, formula is as follows:

$RMSEP=\sqrt{\frac{{\mathrm{\Σ}}_1^n{(C_{NIR}-C_{REF})}^2}{n}}$

In formula: n is test set sample number, CNIR is a certain actual attribute matrix of sample, generally

For concentration array；CREF is the sample properties matrix doped.

Further, the correlation coefficient described in step S6, represent the degree of correlation between model predication value and chemistry true value, formula is as follows:

$R=\frac{{\Σ}_1^n{C}_{NIR}{C}_{REF}-\frac{{\mathrm{\Σ}}_1^n{C}_{NIR}{\mathrm{\Σ}}_1^n{C}_{REF}}{n}}{\sqrt{({\Σ}_1^n{C_{NIR}}^2-\frac{{\left({\mathrm{\Σ}}_1^n{C}_{NIR}\right)}^2}{n})({\Σ}_1^n{C_{NIR}}^2-\frac{{\left({\mathrm{\Σ}}_1^n{C}_{REF}\right)}^2}{n})}}$

In formula: n is test set sample number, CNIR is a certain actual attribute matrix of sample, is generally concentration array；CREF is the sample properties matrix doped.

Master and slave instrument spectral is carried out multi-resolution decomposition by dual-tree complex wavelet by the present invention, the wavelet coefficient of each layer is reconstructed respectively, then on each layer master and slave instrument spectral is carried out piecewise direct standardization (PDS) respectively, to the establishment of spectrum after correction based on the forecast model of partial least square method and leave one cross validation, finally take effectively to merge means and forecast model is merged.This method more existing Model Transfer method precision is higher, performance is higher, serves important function for Model transfer between different instruments, can be widely applied to multiple spectral range.

The preferred embodiment of the present invention described in detail above.Should be appreciated that those of ordinary skill in the art just can make many modifications and variations according to the design of the present invention without creative work.Therefore, all technical staff in the art, all should in the protection domain being defined in the patent claims under this invention's idea on the basis of existing technology by the available technical scheme of logical analysis, reasoning, or a limited experiment.

Claims (8)

1. the drug detecting instrument standardized method based on dual-tree complex wavelet algorithm, it is characterised in that comprise the following steps:

S1, selection also set two illicit drugs inspection instruments respectively as master and slave instrument, the host and slave processors instrument spectral of collected specimens respectively；

Sample spectra is carried out preferably by S2, employing Kennard-Stone algorithm, weeds out abnormal sample spectrum in sample spectra, then selects and set training set and test set；

S3, for sample spectra characteristic, set best Decomposition order, and according to best Decomposition order, sample spectra carried out dual-tree complex wavelet transform, obtain each layer decomposition coefficient；

S4, decomposition coefficient to each layer are reconstructed respectively, obtain the spectrum of each layer reconstruct；

S5, the spectrum that each layer is reconstructed, direct standardization (PDS) algorithm is used to be corrected from machine spectrum, then the main frame spectrum of training set and the chemical true value concentration of sample are used, set up the forecast model based on partial least square method and leave one cross validation, then from machine spectrum, the test set after correction is inputted this model to be predicted the outcome, and obtain the cross validation root-mean-square error RMSECV of each forecast model；

S6, calculated the weights of each forecast model by cross validation root-mean-square error, use weights that all of forecast model carries out Model Fusion, and calculate RMSEP value and correlation coefficient carrys out evaluation model transmission effect.

2. a kind of drug detecting instrument standardized method based on dual-tree complex wavelet algorithm as claimed in claim 1, it is characterized in that, master and slave instrument described in step S1 need to be ensured of the instrument of same class model, and gathering spectrum is the spectrum that same sample gathers different instrument under same measuring condition.

3. a kind of drug detecting instrument standardized method based on dual-tree complex wavelet algorithm as claimed in claim 1, it is characterised in that each layer of wavelet coefficient that the reconstruct described in step S4 is after the decomposition of the spectrum to different instruments is reconstructed respectively.

4. a kind of drug detecting instrument standardized method based on dual-tree complex wavelet algorithm as claimed in claim 1, it is characterized in that, direct standardization (PDS) algorithm described in step S5 is a kind of Model Transfer method of multivariate calibration, and its detailed process is as follows:

First, choose the window Zi of a K+W+1 length in training set from machine spectrally i-th wavelength points, order

Z_i=[a_{S, i-k}, a_{S, i-k+1}..., a_{S, i+w-1}, a_{S, i+w}]

Then, the i-th wavelength points of training set main frame is constructed a multiple regression equation with Zi, namely

a_{M, i}=Z_ib_i+e_i

Using main constituent (PCA) method to solve this equation, be then placed on the leading diagonal of transition matrix F by solution coefficient b out, other set to 0, and just obtain transition matrix F, as follows:

Finally by transition matrix, test set is converted to, from machine spectrum, the spectrum that main frame spectrum matches, is achieved that Model Transfer, it may be assumed that

X_{S, pds}=X_s·F

In formula: as, i be training set from machine spectrum i-th wavelength points, am, i is training set main frame spectrum i-th wavelength points, bi is conversion coefficient, and ei is error, and F is transition matrix, Xs be test set from machine spectrum, Xs, pds is the correction spectrum obtained after direct standardization (PDS) method.

5. a kind of drug detecting instrument standardized method based on dual-tree complex wavelet algorithm as claimed in claim 1, it is characterised in that the cross validation root-mean-square error described in step S5 is RMSECV, and its formula is as follows:

In formula: n is test set sample number, CNIR is a certain actual attribute matrix of sample, is generally concentration array；CREF is the sample properties matrix doped.

6. a kind of drug detecting instrument standardized method based on dual-tree complex wavelet algorithm as claimed in claim 1, it is characterised in that described in step S6 use weights forecast model is carried out Model Fusion operation particularly as follows:

$W_i=\frac{{\left(\frac{1}{{\mathrm{RMSECV}}_i}\right)}^2}{{\Σ}_i^m{\left(\frac{1}{{\mathrm{RMSECV}}_i}\right)}^2}$

Being merged by forecast model according to weights Wi, formula is as follows:

$C=\underset{i}{\overset{m}{\Σ}}{W}_i\·C_{iREF}$

Wherein, CiREF is predicting the outcome of forecast model, and m is the yardstick decomposed, and C is predicting the outcome after forecast model merges, namely final Model Transfer result.

7. a kind of drug detecting instrument standardized method based on dual-tree complex wavelet algorithm as claimed in claim 1, it is characterized in that, RMSEP described in step S6 is the predicted root mean square error of forecast model, represent the deviation between model predication value and chemistry true value, for the most important parameter of quality of evaluation model, formula is as follows:

$RMSEP=\sqrt{\frac{{\Σ}_1^n{(C_{NIR}-C_{REF})}^2}{n}}$

In formula: n is test set sample number, CNIR is a certain actual attribute matrix of sample, is generally concentration array；CREF is the sample properties matrix doped.

8. a kind of drug detecting instrument standardized method based on dual-tree complex wavelet algorithm as claimed in claim 1, it is characterised in that the correlation coefficient described in step S6, represents the degree of correlation between model predication value and chemistry true value, and formula is as follows:

$R=\frac{{\Σ}_1^n{C}_{NIR}{C}_{REF}-\frac{{\Σ}_1^n{C}_{NIR}{\Σ}_1^n{C}_{REF}}{n}}{\sqrt{({\Σ}_1^n{C_{NIR}}^2-\frac{{\left({\Σ}_1^n{C}_{NIR}\right)}^2}{n})({\Σ}_1^n{C_{NIR}}^2-\frac{{\left({\Σ}_1^n{C}_{REF}\right)}^2}{n})}}$

In formula: n is test set sample number, CNIR is a certain actual attribute matrix of sample, is generally concentration array；CREF is the sample properties matrix doped.