CN101430276B - Wavelength variable optimization method in spectrum analysis - Google Patents
Wavelength variable optimization method in spectrum analysis Download PDFInfo
- Publication number
- CN101430276B CN101430276B CN2008102395880A CN200810239588A CN101430276B CN 101430276 B CN101430276 B CN 101430276B CN 2008102395880 A CN2008102395880 A CN 2008102395880A CN 200810239588 A CN200810239588 A CN 200810239588A CN 101430276 B CN101430276 B CN 101430276B
- Authority
- CN
- China
- Prior art keywords
- wavelength variable
- wavelength
- variable
- reinheitszahl
- modeling
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Images
Landscapes
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a method for optimizing wavelength variable in spectral analysis. The method comprises the steps as follows: obtained original spectrum is pretreated to obtain a spectral array with useless information eliminated; the purity value of each wavelength variable is calculated in the obtained spectral array to select the wavelength variable with maximum purity value as a first wavelength variable; the relative weighting function of no. j wavelength variable and selected (j-1) wavelength variables is calculated, and the purity value of each wavelength variable after the relative weighting function is added is calculated; the wavelength variable with the maximum purity value is selected as no. j wavelength variable, wherein, j is the integral more than or equal to 2; partial least square regression modeling is carried out by optimized different quantities of the wavelength variables, and predicted root mean square error is calculated; when the predicted root mean square error is minimum, the wavelength variable combination selected for modeling is the optimized wavelength variable combination. The quantity of the wavelength variables selected by the method is small, and the method can minimize redundant information and can improve modeling speed and efficiency obviously.
Description
Technical field
The present invention relates to spectral analysis technique, relate in particular to a kind of method of spectral analysis medium wavelength variable optimization.
Background technology
In conjunction with the spectral analysis technique of polynary calibration model, be one fast, the new technology of component content or character in the Non-Destructive Testing sample.Because sample component content or character can cause the variation of absorption spectrum when changing; So earlier through carrying out related with its concentration or character to sample spectra; Set up polynary calibration model, predict constituent concentration or constitutive property unknown in this sample through the spectral information of polynary calibration model and sample then.But because the existence of various interference, spectral information is complicated and easy overlapping, therefore, need from the spectral information of complicacy, reject redundant information, extract useful information, to set up high-level efficiency, high-precision polynary calibration model.
Yet, when setting up polynary calibration model, sometimes in order not lose the information of spectrum; Can adopt the spectroscopic data in whole wavelength coverages to carry out modeling; But adopt whole spectroscopic data modelings, not only calculated amount is big, and spectral selectivity is relatively poor; The spectral noise of some wavelength useful information big, that comprise is few; Therefore in this case, the precision of prediction of polynary calibration model may not necessarily reach optimum value, selects spectroscopic data modeling under the different wavelength range will directly influence the measuring accuracy of polynary calibration model.It is thus clear that, how to select the wavelength variable, to obtain the most effective spectrogram information in the spectrum, the reduced data computing makes polynary calibration model have best predictive ability, is to set up the major issue that polynary calibration model faces.
In fact; The method for optimizing of wavelength variable is a lot; For example: adopt the method for choice variable forward in the multiple regression analysis, backward delete quantity method or progressively the Return Law carry out the wavelength variable optimization, yet in theory, these wavelength variable optimization methods all are to the data of no correlativity; Under multiple correlation ten minutes serious situation, the reliability that these methods obtain conclusion receives certain influence; And for example: correlation analysis method, show the variation analysis method and carry out the method etc. of wavelength variable optimization according to the ratio of the regression coefficient of the polynary calibration model of tested composition and spectrum residual error; It is not in the complicated especially spectrum of use, for the effect that improves polynary calibration models quality not clearly that but these wavelength variable optimization methods are only applicable to some measurement situation.
At present, have some global optimization methods to be applied on the wavelength variable optimization, like simulated annealing, genetic algorithm etc., wherein, simulated annealing is to receive the inspiration of METAL HEATING PROCESS technology and a kind of stochastic search methods of growing up; Genetic algorithm is a kind of method with the biological natural evolutionary process search optimum solution of computer simulation.Although searching algorithm such as simulated annealing, genetic algorithm has great search capability; But this class methods parameter is provided with complicacy; Influenced the ability that searches globally optimal solution and locally optimal solution; And the setting of these parameters also depends on researcher's experience and the assurance to being studied a question thereof, and therefore has certain subjectivity and randomness.In addition; When adopting genetic algorithm to carry out the wavelength variable optimization; Though the predictive ability of polynary calibration model is very high during single experiment; But, cause the adaptability of polynary calibration model very low, so the wavelength that genetic algorithm is preferably come out does not improve the robustness and the adaptability of polynary calibration model because there is certain randomness in it.
In view of this, fundamental purpose of the present invention is to provide a kind of method of spectral analysis medium wavelength variable optimization, can improve modeling efficiency and precision of prediction.
For achieving the above object, the invention provides a kind of method of spectral analysis medium wavelength variable optimization, comprising:
The near infrared spectrum data of gathering sample through near infrared spectrometer, and the near infrared spectrum data pre-service of current collection obtained to eliminate the near infrared spectrum after the garbage; According to pretreated near infrared spectrum, calculate the Reinheitszahl of each wavelength variable, select the maximum wavelength variable of Reinheitszahl as the 1st wavelength variable, and the Application of MATLAB program is selected preceding j wavelength variable successively automatically; Wherein, Calculate the relevant weight function of j wavelength variable and preceding (j-1) individual wavelength variable of having selected, and calculate the Reinheitszahl that adds each wavelength variable after this relevant weight function, select the maximum wavelength variable of Reinheitszahl as j wavelength variable; Wherein, j is the integer more than or equal to 2; The wavelength variable of the different numbers that usefulness optimizes carries out PLS and sets up polynary calibration model, and calculates predicted root mean square error; A said predicted root mean square error hour wavelength variable combination selected for modeling is optimized wavelength variable combination; Adopt polynary calibration model that the pre-configured sample as forecast set is predicted; Wherein, said pre-service is handled the near infrared spectrum data of being gathered for adopting relevant function method or garbage variable null method or small wave converting method.Wherein, said Reinheitszahl is the standard deviation and the number percent that increases compensating factor average afterwards of said wavelength variable.
Wherein, the wavelength variable of the said different numbers that optimize can be preceding j the wavelength variable of selecting successively.
The method for optimizing of spectral analysis medium wavelength variable provided by the present invention; Concentrating the spectroscopic data of sample to carry out pre-service to calibration samples; Remove noise, background interference and with the incoherent information of analyte after; The Reinheitszahl of each wavelength variable in the spectrum matrix after the calculating pre-service; The wavelength variable of selecting maximum Reinheitszahl, count the relevant weight function of j wavelength variable with preceding (j-1) that selected individual wavelength variable wherein when calculating the Reinheitszahl of the individual wavelength variable of j (j >=2) as the 1st wavelength variable.Said process repeatable fine do not have randomness, is a kind of deterministic algorithm.The inventive method only needs when calculating Reinheitszahl, a compensating factor parameter is set, and can overcome that parameter is provided with a complicated difficult problem in the existing wavelength variable optimization method.It is thus clear that this wavelength variable optimization method of the present invention simply is easy to realize.
In addition, method of the present invention is a kind of wavelength Variables Selection method of self model, promptly analyzes to the data of spectrum itself, is different from more of the prior art and the relevant wavelength Variables Selection method of concentration information work.And, because the wavelength variable of being elected is original wavelength, when analyzing, can be used as reference, with the molecule in the evaluation measured matter or the information of group to predicting the outcome.
In addition; Method of the present invention can select the wavelength variable of different numbers to carry out offset minimum binary (PLS, Partial Least Squares Method) regression modeling when the wavelength variable modeling that application is selected successively; And calculating predicted root mean square error (RMSEP; Root Mean Square Error of Prediction), when predicted root mean square error hour, wavelength variable combination selected for modeling is most preferred wavelength variable combination; So, can obviously improve the precision of prediction of the polynary calibration model of setting up.
This shows that wavelength variable optimization method according to the invention can minimizing redundant information; Owing to introduce relevant weight function; The method of the invention can solve collinearity problem between the wavelength variable; So the wavelength variables of being selected is few; The present invention can set up the spectrum quantitative correction model with higher forecasting precision through the few wavelengths variable of selecting, and can obviously promote modeling speed and efficient.
Description of drawings
Fig. 1 is the schematic flow sheet of optimizing wavelength variable in spectral analysis of the present invention;
Fig. 2 (a) is the original near infrared light spectrogram before the method for the invention embodiment pre-service;
Fig. 2 (b) is that the method for the invention embodiment is through the pretreated near infrared light spectrogram of small wave converting method;
Fig. 3 (a) is the distribution plan of each the wavelength Reinheitszahl curve of the method for the invention embodiment when selecting second wavelength variable;
Fig. 3 (b) is the distribution plan of each the wavelength standard difference curve of the method for the invention embodiment when selecting second wavelength variable;
The distribution plan of the RMSEP value of gained when Fig. 4 selects different number wavelength variable modeling successively for the method for the invention embodiment;
Fig. 5 is the distribution plan of the method for the invention embodiment optimal wavelength variable;
The synoptic diagram that predicts the outcome of the many first calibration models of PLS that Fig. 6 is set up for the method for the invention embodiment most optimum wavelengths variable combination.
Embodiment
For make above-mentioned purpose of the present invention, feature and advantage can be more obviously understandable, below in conjunction with accompanying drawing and specific embodiment the present invention done further detailed explanation.
Basic thought of the present invention is: at first, concentrate the spectroscopic data of sample to carry out pre-service to calibration samples, more pretreated spectroscopic data is carried out self model wavelength variable optimization, the wavelength variable of selecting maximum Reinheitszahl is as the 1st wavelength variable; When calculating the Reinheitszahl of the individual wavelength variable of j (j >=2), the Reinheitszahl of j wavelength variable is counted the relevant weight function of j wavelength variable and preceding (j-1) individual wavelength variable of having selected; Select the wavelength variable of different numbers to carry out the PLS regression modeling then successively, and calculate predicted root mean square error; When predicted root mean square error hour, selected modeling wavelength variable combination is most preferred wavelength variable combination.
Need to prove; Before carrying out the wavelength variable optimization; Can the spectrum samples of testing acquisition in advance be divided into training set and forecast set; Wherein, said training set sample is used for carrying out the training of the preferred and polynary calibration model of wavelength variable, and said forecast set sample is used for estimating the precision of prediction of the preferred and polynary calibration model of wavelength variable.
Usually; Possibly comprise the high frequency noise that causes owing to noise of instrument, measuring condition variation etc. in the spectrum measuring data of said sample; And the garbages such as background interference that produce of other chemical constitution light absorption, therefore, said pre-service mainly is exactly in order to remove in the spectrum constituent concentration or the incoherent information of character of these and sample; Constituent concentration or the nature parameters with sample is relevant as much as possible for the variable of selecting with assurance, and then improves spectral quality.
In fact, original spectrum is carried out pretreated method has several kinds: relevant function method, garbage variable null method and small wave converting method etc. specify the realization of preprocess method below for example.
First kind of preprocessing procedures---relevant function method, this method may further comprise the steps:
Steps A 101, with the spectroscopic data X of the constituent concentration of measuring or constitutive property data Y (n * 1) and sample (it is related that n * m) carries out, and by formula the related coefficient C of each wavelength is obtained in (1):
Wherein, n is a number of samples, and m is the variable number, x
iBe the element among the spectroscopic data X, y
iBe the element in the data Y, x, y are respectively x
iAnd y
iThe average of place row.
The threshold value C of steps A 102, setting related coefficient
0, it is C that related coefficient surpasses this threshold value
k＞C
0Pairing k wavelength is selected.
Steps A 103, form new matrix X by the wavelength variable of selecting
NEW, to be used for further spectral manipulation.
Second kind of preprocessing procedures---garbage variable removing method, this method may further comprise the steps:
Steps A 201, (n * m) and concentration matrix Y (n * 1) carry out the PLS regretional analysis, and choose major component number f, and wherein f is a positive integer with sample set spectrum matrix X;
Wherein, n representes the number of sample, and m representes the number of wavelength variable.
Steps A 202, (n * m) is combined into matrix XR (n * 2m) with X and R to produce random noise matrix R;
Here, m classifies X as before the matrix XR that is combined into, and back m classifies R as.
Steps A 203, matrix XR and Y are carried out PLS return, reject the validation-cross of a sample at every turn, obtain a regression coefficient vector b, obtain n PLS regression coefficient composition matrix B (n * 2m) altogether.
Steps A 204, (standard deviation std (b) and the mean value mean (b) of n * 2m) calculate C then by the column count matrix B
i=mean (bi)/std (bi), i=1 wherein, 2 ..., 2m.
Steps A 205, get the maximum value C of C in [m+1,2m] interval
Max=max (abs (C)).
Steps A 206, at [1, m] the interval corresponding C of matrix X that removes
i＜C
MaxVariable, and surplus variable formed the new matrix X that chooses through garbage variable removing method
NEW, for follow-up spectral manipulation is got ready.
The third preprocessing procedures---small wave converting method.Wavelet transformation has the multiresolution analysis characteristic; Because in spectral signal; The background that noise and other composition light absorption produce finds expression in low yardstick detail coefficients and high yardstick proximate component more, therefore utilizes wavelet transformation can remove garbages such as various noises and background interference simultaneously.Adopting wavelet transformation that original spectrum is carried out pretreated process may further comprise the steps:
Steps A 301, produce a random noise and background matrix R (n * m), (n * m) and R are combined into matrix XR (n * 2m) with sample sets spectrum matrix X;
Wherein, n is that number of samples, m are the variable number, and m classifies X as before the combinatorial matrix XR, and back m classifies R as.
Steps A 302, with the wavelet decomposition of carrying out of every bars of matrix XR, select wavelet basis and wavelet decomposition number of plies k, obtain small echo detail coefficients matrix D (k * 2m/ (i * 2)) and proximate component matrix A (k * 2m), i=1 wherein, 2 ..., k.
Steps A 303, by the standard deviation std (di) and the mean value mean (di) of column count matrix D (k * 2m/ (i * 2)), calculate C then
Di=mean (di)/std (di).
Steps A 304, (standard deviation std (bi) and the mean value mean (bi) of k * 2m) calculate C then by the column count matrix A
Ai=mean (ai)/std (ai).
Steps A 305, if [1, m/ (i * 2)] interval | C
Di| value and [m+1,2m] interval | C
Di| be worth similarly, then the composition of this detail coefficients representative is rejected as noise, i=1 wherein, and 2 ..., k; [if 1, m] interval | C
Ai| value is less than [m+1,2m] interval | C
Ai| value, then its corresponding composition is removed as a setting.
Steps A 306, with denoising, go k layer low frequency and high frequency coefficient after the background to carry out signal reconstruction, utilize the reconstruct spectral signal to set up calibration model, can select the optimal wavelet base according to predicted root mean square error; Each bar spectral signal of reconstruct is formed new spectrum matrix X
NEW, so that carry out the spectral manipulation work such as preferred of wavelength variable.
The preprocess method that it is pointed out that the preferred embodiment of the present invention is not limited only to said method, and the preprocess method of garbages such as other any removal noises, background all should belong to protection scope of the present invention.
Based on preprocess method recited above, as shown in Figure 1, the schematic flow sheet of optimizing wavelength variable in spectral analysis of the present invention comprises the steps:
Step S101, the original spectrum data of experiment all samples are carried out pre-service, to obtain the spectrum matrix of eliminating after the garbage;
Eliminate garbage through pre-service, can improve spectral quality, make the relation between spectrum and analyte constituent concentration or the character tightr.In above-mentioned preprocess method, said relevant function method and garbage variable null method are applicable to the uncomplicated situation of spectrum, generally only are used to remove noise; And small wave converting method relies on its multiresolution analysis characteristic, can remove garbages such as noise and background simultaneously.Can select the pretreated method of spectrum as the case may be.
Step S102, at pretreated spectrum matrix X
NEWIn, calculate the Reinheitszahl of each wavelength variable, and select the maximum wavelength variable of Reinheitszahl as the 1st the wavelength variable of selecting;
Said Reinheitszahl can be expressed as the dispersion degree and the number percent that has increased compensating factor central tendency afterwards of said wavelength variable in order to characterize the contribution of each variable to polynary calibration model; Said dispersion degree is the standard deviation of wavelength variable, and said central tendency is the average of said wavelength variable.In addition, under, the noise fainter situation suitable, can regulate through compensating factor with signal at signal.Usually, said compensating factor can be set to 1% to 5% of said average.
Spectrum matrix X
NEWIn, shown in the following formula of the computing method of the Reinheitszahl of each wavelength variable i (2):
p
i，1＝σ
i/(μ
i+α) (2)
Wherein, σ
iBe standard deviation, μ
iBe average, α is a compensating factor.The Reinheitszahl p of each wavelength variable i of trying to achieve by formula (2)
I, 1, judge p then
I, 1The size of value has maximum p
I, 1I wavelength variable of value is the 1st the wavelength variable of selecting.
Step S103, calculate the relevant weight function of j wavelength variable, and calculate each wavelength variable and count the Reinheitszahl after the relevant weight function, j wavelength variable selecting the maximum wavelength variable conduct of Reinheitszahl to select;
Wherein, j is the integer more than or equal to 2; Said relevant weight function is in order to the significance level that concerns between preceding (j-1) the individual wavelength variable that characterizes j wavelength variable and selected.
Select the general process of the individual wavelength variable of j (j >=2) specific as follows:
Calculate spectrum matrix X
NEWIn the length l of each wavelength variable i
i, shown in formula (3):
Wherein, d
I, jBe spectrum matrix X
NEWIn the capable j column element of i, have:
Obtain relational matrix C=D (l) D (l)
T/ n, wherein, D (l) is by element d (l)
I, jThe matrix of forming; And the relevant weight function ρ of calculating
I, j, shown in formula (4).
Wherein, j representes the sequence number of j wavelength variable to be determined, P
J-1The sequence number of (j-1) individual wavelength variable in relational matrix C that expression has at present been selected, p
1Represent the sequence number of selected the 1st wavelength variable in relational matrix C, so, j wavelength variable Reinheitszahl p
I, jFor:
p
i，j＝ρ
i，j(σ
i/(μ
i+α)) (5)
Has maximum p
I, jThe wavelength variable of value is j the wavelength variable of selecting, and corresponding standard difference s
I, jExpression formula is suc as formula shown in (6).
s
i，j＝ρ
i，jσ
i (6)
Usually, has maximum p
I, jThe standard difference s of the wavelength variable of value
I, jAlso can be higher relatively, so standard difference s
I, jCan be used as reference value, to supervise the wavelength variable of being selected.
Step S104, select the wavelength variable of different numbers to carry out the PLS regression modeling successively, and calculate predicted root mean square error RMSEP;
Usually, the computing formula of RMSEP is:
Wherein,
Be predicted value, y
iBe reference value.
Predicted root mean square error RMSEP has reflected that measurement data departs from the degree of actual value, and generally speaking, the value of RMSEP is more little, and the expression measuring accuracy is high more, and therefore available RMSEP is as the standard of this measuring process precision of evaluation.When the value of RMSEP hour, selected modeling wavelength variable combination is optimized wavelength variable combination.And step S104 does the PLS regretional analysis through the wavelength variable of iteration selection successively, and then selected modeling wavelength variables is judged.
Step S105, judge whether the value of RMSEP reaches minimum value, if, execution in step S106 then, otherwise return step S103;
Usually, the value of the RMSEP that obtains when this modeling obtained during greater than its preceding modeling as a result the time, think that promptly said previous RMSEP is a minimum value; If the value of RMSEP does not reach minimum value, repeating step S103 then, iteration successively is until selecting optimized wavelength variable combination.
In addition, also can, algorithm preestablish the RMSEP value of a minimum, when beginning in order to condition as end loop.
Step S106, when RMSEP reaches minimum value, can finish the preferred of this wavelength variable, accomplish modeling.
Wavelength variable optimization method of the present invention can the Application of MATLAB program design, can realize the automatic selection of wavelength variable.Wavelength variable according to iteration is successively selected is set up polynary calibration model, utilizes the RMSEP of validation-cross to judge again, and the variable combination that hour is used for modeling when the RMSEP value is the most optimum wavelengths variable combination of selection.
It is emphasized that; Wavelength variable optimization method of the present invention also can be selected the wavelength variable of some earlier; Carry out the PLS regression modeling with these wavelength variablees of having selected again; Wherein, more preferably, the wavelength variable of used different numbers can be the wavelength variable of selecting successively during modeling; Otherwise, can further combine some Variables Selection methods (like the method for exhaustion, genetic algorithm and DSMC etc.) to select part wavelength variable and carry out modeling.Calculate the value of RMSEP then, and judge whether the value of RMSEP is minimum, if occurred minimum RMSEP value, then can stop the selection of wavelength variable; Otherwise, proceed the preferred of wavelength variable.
Below in conjunction with a specific embodiment, the method for optimizing of spectral analysis medium wavelength variable of the present invention is described in detail.
With the experiment of human plasma near infrared spectrum blood sugar test is example, and concentration of glucose in the sample is carried out forecast analysis.Fourier transformation infrared spectrometer is adopted in the experiment of human plasma near infrared spectrum in the present embodiment; The scope of gathering spectrum is 900～3600nm; The detecting device that adopts is the InSb detecting device of LIN cooling, and instruments such as 1mm quartz sample pool, peristaltic pump automatic sample handling system and automatic clinical chemistry analyzer have also been selected in this experiment for use in addition.
The configuration step of blood plasma laboratory sample is: in whole blood, add the anticoagulant heparin agent and in hydro-extractor, isolate behind 1500 commentaries on classics of process per minute, the 10min; In the blood plasma of separating, add glucose then, demarcate blood glucose value with automatic clinical chemistry analyzer and glucose oxidase method.This blood plasma experiment obtains 33 in sample altogether, and wherein 22 samples are used for carrying out the training of wavelength variable optimization and polynary calibration model as training set; 11 samples are used for estimating the precision of prediction of wavelength variable optimization and polynary calibration model as forecast set.In addition, the concentration of glucose scope is 10.4～44.4mg/dL, and stochastic distribution, and its concentration standard difference is 8.5mg/dL.
The implementation procedure of the wavelength variable optimization method of present embodiment may further comprise the steps:
Step S201, the spectroscopic data of experiment all samples is carried out pre-service, remove garbages such as noise and background.
The selection spectral range of analysis is 1000～1890.36nm, and every spectrum has 4711 wavelength variablees.To every spectrum all the Application of wavelet method remove garbage, select wavelet basis db3, decompose through Mallat, its decomposition scale is 12, removes yardstick and is respectively 1,2,3,10 pairing compositions, reconstruct spectral information then.
Shown in Fig. 2 (a), be the original near infrared light spectrogram before the present embodiment pre-service, and Fig. 2 (b) be present embodiment through the pretreated near infrared light spectrogram of small wave converting method, spectrum is new spectrum matrix X after the pre-service
NEWBecause spectrum matrix X
NEWIn spectroscopic data amount big (33 * 4711 matrixes), can be with reference to shown in Fig. 2 (b).
Step S202, to through pretreated spectroscopic data X
NEWCarry out the wavelength variable optimization, calculation training collection sample light spectrum matrix X
NEWIn the Reinheitszahl of each wavelength variable i, to select the 1st wavelength variable;
The value of present embodiment compensating factor α is set to 5% of average.Through each wavelength variable place's Reinheitszahl and standard difference of relatively being calculated by step B1, can draw maximum Reinheitszahl is p
1,1=0.0431, therefore can confirm that the 1st the wavelength variable of being selected is 1 wavelength variable (wavelength is 1000nm) for the variable label.
Step S203, second wavelength variable of selection;
According to Reinheitszahl and the standard difference after formula (5) and the relevant weight function of formula (6) calculating adding; The Reinheitszahl that is obtained and corresponding result between each wavelength variable are the distribution plan of each the wavelength Reinheitszahl curve of present embodiment when selecting second wavelength variable shown in Fig. 3 (a); The standard difference that is obtained and corresponding result between each wavelength variable are the distribution plan of each the wavelength standard difference curve of present embodiment when selecting second wavelength variable shown in Fig. 3 (b).As can be seen from Figure 3, selected second wavelength variable is for having peaked the 4711st wavelength variable (wavelength is 1890.36nm).
Repeating step S203, and then obtain selected the 3rd～16 wavelength variable and be respectively: the 4223rd variable (wavelength is 1730.7nm), the 1944th variable (wavelength is 1241.16nm), the 2655th variable (wavelength is 1361.29nm), the 4700th variable (wavelength is 1886.44nm), the 3281st variable (wavelength is 1488.1nm), the 4684th variable (wavelength is 1880.76nm), the 2973rd variable (wavelength is 1422.88nm), the 3857th variable (wavelength is 1627.6nm), the 2814th variable (wavelength is 1391.4nm), the 1232nd variable (wavelength is 1140.38nm), the 2558th variable (wavelength is 1343.54nm), the 4078th variable (wavelength is 1688.33nm).
The wavelength variables that step S204, judgement are selected.
Set up PLS with the 1st to the 16th the wavelength variable of selecting successively and return polynary calibration model, the RMSEP in the time of can obtaining adopting the wavelength variable of different numbers to carry out modeling by cross verification.
As shown in Figure 4, the distribution plan of the RMSEP value of gained when selecting different number wavelength variable modeling successively for present embodiment, inverted triangle is represented the RMSEP value among Fig. 4, the trend that curve representation RMSEP value changes along with the variation of wavelength variables.It is thus clear that when preceding 14 the wavelength variable modelings of selection, the RMSEP value is minimum; So the optimized wavelength variable number is 14, this moment, these selected preceding 14 wavelength variable combination were optimized wavelength variable combination, and are as shown in Figure 5; Distribution plan for present embodiment optimal wavelength variable; Can reflect the wavelength range of variables that is optimized, circle is represented the wavelength variable selected, the curve representation curve of spectrum among Fig. 5.
Through wavelength variable optimization method of the present invention; Set up PLS and return polynary calibration model; The forecast set sample is predicted the value that can obtain RMSEP is 1.9mg/dL, the related coefficient that predicts the outcome (Correlation) of this polynary calibration model is 0.94; Its correlationship is as shown in Figure 6, the synoptic diagram that predicts the outcome of the many first calibration models of being set up for present embodiment most optimum wavelengths variable combination of PLS.In Fig. 6, stain is represented the correlativity of reference value and predicted value, and straight line is then represented a benchmark of this correlativity.When said stain during the closer to straight line, the correlativity of expression predicted value and reference value is big more.
As shown in table 1; For selecting the Prediction Parameters of different wave length variable range modeling for use; Present embodiment application self model wavelength variable optimization method is selected 14 wavelength variablees and is carried out modeling; Compose the comparison of modeling entirely with selecting 4711 wavelength variablees for use, self model wavelength variable optimization method of the present invention not only is simple and easy to realize, modeling efficiency is high, and the precision of prediction of its polynary calibration model of building also obviously improves.
Table 1
In the above-described embodiments, the description of each embodiment is all emphasized particularly on different fields, do not have the part that details among certain embodiment, can get final product referring to the associated description of other embodiment.
The above is merely preferred embodiment of the present invention, just is used for the present invention that explains, and is not to be used to limit protection scope of the present invention.For each above-mentioned embodiment, for simple description, so it all is expressed as a series of combination of actions; But those skilled in the art should know; The present invention does not receive the restriction of described sequence of movement, because according to the present invention, some step can adopt other orders or carry out simultaneously.In addition, within spirit of the present invention and claim protection domain, any modification and change to effect of the present invention all fall into protection scope of the present invention.
Claims (4)
1. the method for a spectral analysis medium wavelength variable optimization is characterized in that, this method comprises:
The near infrared spectrum data of gathering sample through near infrared spectrometer, and the near infrared spectrum data pre-service of current collection obtained to eliminate the near infrared spectrum after the garbage;
According to pretreated near infrared spectrum, calculate the Reinheitszahl of each wavelength variable, select the maximum wavelength variable of Reinheitszahl as the 1st wavelength variable, and the Application of MATLAB program is selected preceding j wavelength variable successively automatically;
Wherein, Calculate the relevant weight function of j wavelength variable and preceding (j-1) individual wavelength variable of having selected, and calculate the Reinheitszahl that adds each wavelength variable after this relevant weight function, select the maximum wavelength variable of Reinheitszahl as j wavelength variable; Wherein, j is the integer more than or equal to 2;
The wavelength variable of the different numbers that usefulness optimizes carries out PLS and sets up polynary calibration model, and calculates predicted root mean square error; A said predicted root mean square error hour wavelength variable combination selected for modeling is optimized wavelength variable combination; Adopt polynary calibration model that the pre-configured sample as forecast set is predicted;
Wherein, said pre-service is handled the near infrared spectrum data of being gathered for adopting relevant function method or garbage variable null method or small wave converting method;
Wherein, said Reinheitszahl is the standard deviation and the number percent that increases compensating factor average afterwards of said wavelength variable.
2. method according to claim 1 is characterized in that, said compensating factor is 1% to 5% of a said average.
3. method according to claim 1 is characterized in that, this method further comprises: preestablishing a predicted root mean square error value is its minimum value.
4. according to claim 1 or 3 described methods, it is characterized in that during the predicted root mean square error value of the predicted root mean square error value of this modeling gained during greater than a preceding modeling, previous predicted root mean square error value is a minimum value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2008102395880A CN101430276B (en) | 2008-12-15 | 2008-12-15 | Wavelength variable optimization method in spectrum analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2008102395880A CN101430276B (en) | 2008-12-15 | 2008-12-15 | Wavelength variable optimization method in spectrum analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101430276A CN101430276A (en) | 2009-05-13 |
| CN101430276B true CN101430276B (en) | 2012-01-04 |
Family
ID=40645798
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2008102395880A Expired - Fee Related CN101430276B (en) | 2008-12-15 | 2008-12-15 | Wavelength variable optimization method in spectrum analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101430276B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104502306A (en) * | 2014-12-09 | 2015-04-08 | 西北师范大学 | Near infrared spectrum wavelength selecting method based on variable significance |
Families Citing this family (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102313712B (en) * | 2011-05-30 | 2013-04-24 | 中国农业大学 | Correction method of difference between near-infrared spectrums with different light-splitting modes based on fiber material |
| CN103955711B (en) * | 2014-05-20 | 2017-07-07 | 北京航空航天大学 | A kind of mode identification method in imaging spectral target identification analysis |
| CN104865228B (en) * | 2015-06-02 | 2017-08-15 | 中国科学院上海技术物理研究所 | The quantitative LIBS detection method solved based on fusion entropy optimization |
| CN105630743B (en) * | 2015-12-24 | 2018-05-01 | 浙江大学 | A kind of system of selection of spectrum wave number |
| CN105891141A (en) * | 2016-03-30 | 2016-08-24 | 南京富岛信息工程有限公司 | Method for rapidly measuring gasoline property data |
| CN106950193B (en) * | 2017-05-24 | 2019-04-26 | 长春理工大学 | Based on the near infrared spectrum Variable Selection from weight variable combination cluster analysis |
| CN107728602A (en) * | 2017-09-28 | 2018-02-23 | 合肥工业大学 | A kind of Personalized service method of hydroforming equipment failure |
| CN108918446B (en) * | 2018-04-18 | 2021-05-04 | 天津大学 | Ultra-low concentration sulfur dioxide ultraviolet difference feature extraction algorithm |
| CN109409350B (en) * | 2018-10-23 | 2022-05-31 | 桂林理工大学 | PCA modeling feedback type load weighting-based wavelength selection method |
| CN109596558B (en) * | 2018-12-17 | 2020-05-19 | 华中科技大学 | Spectrogram basic dimension correction and differential analysis method based on moving least square method |
| CN110069895B (en) * | 2019-05-20 | 2021-06-01 | 中国水利水电科学研究院 | Method for establishing winter wheat nitrogen content full-growth period spectrum monitoring model |
| CN110338813B (en) * | 2019-06-04 | 2022-04-12 | 西安理工大学 | Noninvasive blood glucose detection method based on spectrum analysis |
| CN110726694A (en) * | 2019-10-22 | 2020-01-24 | 常州大学 | Characteristic wavelength selection method and system of spectral variable gradient integrated genetic algorithm |
| CN111982949B (en) * | 2020-08-19 | 2022-06-07 | 东华理工大学 | Method for separating EDXRF spectrum overlapping peak by combining fourth derivative with three-spline wavelet transform |
| CN113406058A (en) * | 2021-05-28 | 2021-09-17 | 中国科学院沈阳自动化研究所 | LIBS iron ore pulp quantitative analysis method for screening PLS based on mutual information characteristics |
| CN113592743B (en) * | 2021-08-11 | 2024-01-23 | 北华航天工业学院 | Spectral high-frequency information and low-frequency information separation and coupling method based on complex wavelet transformation |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101299022A (en) * | 2008-06-20 | 2008-11-05 | 河南中医学院 | Method for evaluating Chinese medicine comprehensive quality using near infrared spectra technique |
-
2008
- 2008-12-15 CN CN2008102395880A patent/CN101430276B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101299022A (en) * | 2008-06-20 | 2008-11-05 | 河南中医学院 | Method for evaluating Chinese medicine comprehensive quality using near infrared spectra technique |
Non-Patent Citations (3)
| Title |
|---|
| 李丽娜等.基于交互式自模型混合物分析的近红外光谱波长变量优选方法.《分析化学》.2009,第37卷(第6期),823-827. * |
| 柳艳云等.近红外分析中光谱波长选择方法进展与应用.《药物分析杂志》.2010,第30卷(第5期),968-975. * |
| 褚小立等.近红外分析中光谱预处理及波长选择方法进展与应用.《化学进展》.2004,第16卷(第4期),528-542. * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104502306A (en) * | 2014-12-09 | 2015-04-08 | 西北师范大学 | Near infrared spectrum wavelength selecting method based on variable significance |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101430276A (en) | 2009-05-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101430276B (en) | Wavelength variable optimization method in spectrum analysis | |
| CN101915744B (en) | Near infrared spectrum nondestructive testing method and device for material component content | |
| JP3245157B2 (en) | Measurement and correction of spectral data | |
| CN103913432B (en) | Based on the near-infrared spectrum wavelength system of selection of particle cluster algorithm | |
| CN104062256B (en) | A kind of flexible measurement method based near infrared spectrum | |
| CN101413885A (en) | Near-infrared spectrum method for rapidly quantifying honey quality | |
| CN107014769A (en) | A kind of fresh meat K value Fast nondestructive evaluation models based on THz spectrum analyses | |
| CN110726694A (en) | Characteristic wavelength selection method and system of spectral variable gradient integrated genetic algorithm | |
| CN106644978A (en) | Detection method capable of judging precision and used for analyzing content of soil nutrients based on spectral characteristic wavelength | |
| CN108830253B (en) | Screening model establishing method, spectrum screening device and method | |
| CN107632010B (en) | Method for quantifying steel sample by combining laser-induced breakdown spectroscopy | |
| CN106770058A (en) | The quick special purpose device and its application method of the soil nitrate-N based on infrared spectrum | |
| CN106198447A (en) | Chemical Mixed Fertilizer main component harmless quantitative detection method based on near-infrared spectrum technique | |
| CN114136900B (en) | Water quality detection method combining ultraviolet and visible light absorption spectrum technology | |
| Jiang et al. | Qualitative and quantitative analysis in solid-state fermentation of protein feed by FT-NIR spectroscopy integrated with multivariate data analysis | |
| Yu et al. | Prediction of soil properties based on characteristic wavelengths with optimal spectral resolution by using Vis-NIR spectroscopy | |
| CN113049499A (en) | Indirect remote sensing inversion method for water total nitrogen concentration, storage medium and terminal equipment | |
| CN102128805A (en) | Method and device for near infrared spectrum wavelength selection and quick quantitative analysis of fruit | |
| CN103134770A (en) | Method for eliminating influence on infrared spectrum detection of soil total nitrogen content from moisture | |
| CN107247033A (en) | Differentiate the method for Huanghua Pear maturity based on rapid decay formula life cycle algorithm and PLSDA | |
| CN103134763A (en) | Method for predicting crude oil density by infrared spectroscopy | |
| CN112630180B (en) | Ultraviolet/visible light absorption spectrum model for detecting concentration of organophosphorus pesticide in water body | |
| CN106960091A (en) | A kind of THz optical spectrum rapid nondestructive detection methods of fresh meat freshness K values | |
| CN107271389A (en) | A kind of spectral signature variable fast matching method based on index extreme value | |
| CN107356556A (en) | A kind of double integrated modelling approach of Near-Infrared Spectra for Quantitative Analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120104 Termination date: 20141215 |
|
| EXPY | Termination of patent right or utility model |