It is a kind of that lysine content is detected based near infrared spectrum characteristic extractive techniqueMethodTechnical field
The invention belongs to feed detection technique fields, and in particular to one kind is based near infrared spectrum characteristic extractive techniqueQuickly detect the method for lysine content in feed addictive L-lysine sulfate.
Background technique
Feed addictive refers to a small amount of or micro substance added in Feed Manufacturing processing, use process, although dosageSeldom but effect is significant.Feed addictive is the raw material that modern feed industry necessarily uses, to strengthen basal feed nutritive value,Improve breeding performonce fo animals, guarantee animal health, save feed cost and improve livestock products quality etc. has apparent effectFruit.
L-lysine is one of eight kinds of essential amino acids required in humans and animals body and that itself cannot be synthesized, people and is movedLysine shortage will cause organism metabolic disorder in object, cause certain physiological function diseases.In animal feeding, it is often necessary toL-lysine is added in feed as exogenous amino acid.L-lysine belongs to one kind of organic matter, in order to maintain the steady of substanceDetermine and be readily transported use, it is often necessary to salt use is made, and the inorganic constituents in lysine salt molecule will not destroy substanceOrganic structure, in the case where retaining L-lysine function, it is easier to storage and transport.L-lysine sulfate is dynamic at presentOne of the amino acids additive being most widely used in object production, not only can be improved the utilization rate of protein in feed,Animal feed conversion rate can also be improved, the nutrition of more general equilibrium can be provided for growth of animal.
L-lysine sulfate is yellowish-brown mobility powder, has specific smell and moisture absorption, passes through biofermentationMethod is made, and by 65%L- lysine sulphate made of spray drying, wherein the content of L-lysine is to L-lysine sulfateThe effect of feed addictive has critically important influence.Currently, always containing about L-lysine in L-lysine sulphate additivesUnified country or professional standard, enterprise mostly use " measurement of amino acid in feed " (GB/T 18246- not yet for the measurement of amount2000) lysine total content in the acid-hydrolysis method measurement product in, acid-hydrolysis method are needed sample in 6mol/L hydrochloric acid, 110 DEG CEnvironment in hydrolyze 22-24h.Although this method precision is relatively high, Parallel testing sample relative deviation < 5% was detectedBut there is long sample hydrolysis time, instrument cost valuableness in journey and the defects of level requirement is high is operated to inspector, especially singlySample detection time-consuming takes around the time of 2d, causes analysis speed slower, is not suitable for the quick detection and analysis of batch samples.In addition, leading to the disunity of target level of product quality due to the difference of each enterprise's production technology, factory product is be easy to cause to formIngredient, L-lysine content, granular size, product colour etc. are irregular, while L-lysine sulfate is in transport, storageIt is easy to deliquesce during hiding and agglomerates, is even rotten, these can all become L-lysine content inspection in limitation L-lysine sulfateThe unfavorable factor of survey, and seriously affect the accuracy of testing result.
Near-infrared spectral analysis technology can quickly be estimated using organic chemicals in the optical characteristics of near infrared spectrumOne or more chemical composition contents in sample, have the advantages that quickly, conveniently, accurately, can analyze simultaneously it is multi-component.Due toNear-infrared analysis is the spectral signal for obtaining sample, can even be measured in former container sometimes, not need other examinationsTherefore agent will not generate any pollution in test process, belong to environmentally protective detection method.In recent years, domestic and international expert answersRelatively broad research, but amino-acid salt have been carried out to the detection of aminoacid ingredient in food, feed with near-infrared spectrum techniqueAnd its correlative study of product has not been reported.The present invention exactly on the basis of using near infrared cheracteristics, has carried out fastThe technique study of L-lysine content in speed measurement L-lysine sulfate feed addictive.
Summary of the invention
For this purpose, a kind of based near infrared spectrum characteristic extraction skill technical problem to be solved by the present invention lies in providingThe method that art quickly detects lysine content in feed addictive, with solution, lysine detection cycle is longer in the prior art and examinesSurvey the unstable problem of accuracy.
In order to solve the above technical problems, of the present invention a kind of based on the detection of near infrared spectrum characteristic extractive techniqueThe method for establishing model of lysine content, includes the following steps:
(1) sample to be tested for collecting separate sources, carries out pre-treatment respectively, spare；
(2) lysine content of each sample to be tested is measured respectively with art methods；
(3) near infrared spectrum information collection is carried out to sample to be tested obtained in step (1) respectively, obtains calibration set sampleFull spectroscopic data；
(4) the full spectroscopic data of calibration set sample in step (3) is pre-processed, and to pretreated spectroscopic data intoThe extraction of row characteristic information, to establish calibration model；
(5) based on the lysine content measured in step (2), the accuracy of the calibration model of foundation is verified.
In the step (1), the pre-treatment step be sample to be tested is smashed it through into 60 meshes by Cyclone mill, and5h is handled in 105 DEG C of baking oven.
In the step (2), the lysine content detecting step is according to " measurement of amino acid in feed " (GB/TAcid-hydrolysis method measures lysine content in 18246-2000).
In the step (3), the step of the near infrared spectrum information collection for using near-infrared diffusing reflection mode to eachSample to be tested is scanned to obtain spectroscopic data；The scanning mode is continuous wavelength infrared diaphanoscopy, spectra collection wavelengthFor 1000nm-2500nm, resolution ratio 10nm, scanning times 32 times, three times then spectrum is averaged for each sample acquisition, eachThe sweep time of the sample to be tested is 1min.
In the step (4), the Pretreated spectra step is to be handled and gone scattering processing, variable using variable standardizationAt least one of standardization, multiplicative scatter correction, Second Derivative Methods.
In the step (4), the spectroscopic data carries out characteristic information data extraction step using competitive adaptive weightWeighting algorithm (CARS) extracts characteristic information data, specifically comprises the following steps:
(a) sample: model is sampled based on Monte Carlo sampling method, and each CARS sampling in, require fromA certain amount of sample is randomly selected in sample sets as calibration set, to establish PLS model；
(b) remove variable based on decaying exponential function: it is assumed that surveyed sample spectrum battle array is X (m × p), m is sample number, and p isVariable number, the true value matrix of SSC are y (m × 1), then PLS regression model are as follows:
Y=Xb+e；
In formula, b indicates that the coefficient vector of p dimension, e indicate prediction residual；
Wherein, b=Wc=[b1, b2..., bp]TThe linear combination coefficient of score matrix and X (W expression), i-th yuan in bThe absolute value of element | bi| (1≤i≤p) indicates contribution of i-th of variable to SSC value, and variable corresponding to the bigger expression of the value is in SSCPrediction in it is more important；
Utilization index attenuation function removes by force | bi| it is worth relatively small wavelength points, and is sampled using MC, is adopted in i-thAfter sample operation, the storage rate of variable point is calculated by following exponential function:
ri=ae-ki；
In formula, a and k indicate constant respectively at the 1st time and n-th MCS, whole p variables and only 2 variable in sample setParticipate in modeling, i.e. r1=1 and rN=2/p, so that the calculation formula of a and k is as follows:
A=(2/P)1/(N-1)；
In formula, ln indicates natural logrithm, and variable number p is 1499, and setting MC is sampled 50 times；
(c) further variable is screened based on adaptive weight weight sampling technology, by evaluating each variable pointWeight wiVariable Selection is carried out, the calculating of weighted value is as follows:
(d) by the RMSECV value for the new variable subset for calculating and generating more every time, with the smallest change of RMSECV valueQuantum collection is as optimal variable subset；And full spectroscopic data is carried out with spectral signature information data building calibration model with this excellentChoosing.
In the step (4), the step of establishing calibration model used chemometrics method is multiple linearThe Return Law.
In the step (6), the verification step includes: one group of verifying collection sample of acquisition, utilizes established straightening dieType obtains the predicted value of verifying collection lysine content, and the actual value based on art methods measurement compares evaluation, countsThe related coefficient and variance for calculating predicted value and actual value, the accuracy of the calibration model is evaluated with this.
The invention also discloses the method for establishing model in detection feed addictive in lysine content fieldUsing.
The invention also discloses it is a kind of based near infrared spectrum characteristic extractive technique detection lysine content method,Including the near-infrared spectroscopy required according to the method building, and sample containing lysine is contained under this conditionThe step of amount detection.
The sample containing lysine is L-lysine sulfate feed addictive.
The method of quick detection lysine content of the present invention is carried out based near infrared spectrum characteristic extractive techniqueDetection.Near infrared spectrum characteristic extractive technique described herein is to extract sample light by competitive adaptive weighted algorithmData information relevant to lysine, establishes near infrared correction in spectrum, so as to quick, efficient, accurate detection feedThe content of lysine in additive L-lysine sulfate.The method of quick detection lysine content of the present invention, by mentioningTake characteristic relevant to test substance, reduce the data volume for establishing calibration model, improve near-infrared modeling efficiency andThe accuracy of model, detection batch samples that can be faster and better have saved testing cost, have also significantly improved and be obviously improvedIn standard " measurement of amino acid in feed " (GB/T 18246-2000) acid-hydrolysis method measurement amino acid time-consuming, high consumption,The demanding disadvantage of instrumentation.
Detailed description of the invention
In order to make the content of the present invention more clearly understood, it below according to specific embodiments of the present invention and combinesAttached drawing, the present invention is described in further detail, wherein
Fig. 1 is the near-infrared primary light spectrogram of L-lysine sulfate sample after dries pulverizing；
Fig. 2 is the spectral signature information data extracted through competitive adaptive weight weighting algorithm (CARS)；
Fig. 3 is sample to be tested predicted value and actual value correlation scatter plot.
Specific embodiment
Embodiment 1
The present embodiment detects feed addictive L-lysine sulfate sample using near infrared spectrum characteristic extractive techniqueThe content of lysine in this.
The model for quickly detecting lysine content based near infrared spectrum characteristic extractive technique described in the present embodiment is builtCube method specifically comprises the following steps:
(1) sample collection and preparation
The feed addictive L-lysine sulfate sample in source known to 57 parts is taken, which comes from enterprise；By every partSample carries out smashing it through 60 meshes using Cyclone mill；Then it is placed in processing 5h in 105 DEG C of baking ovens again to be dried, as driesCrush state sample to be tested.
(2) measurement of sample lysine content
Every part of sample is placed in processing 5h in 105 DEG C of baking ovens to dry, then by the sample after drying according to " in feedThe measurement of amino acid " acid-hydrolysis method measurement lysine in (GB/T 18246-2000) content.
(3) sample near infrared spectra collection
It is made using SupNIR-2700 type grating near infrared spectrometer (production of optically focused Science and Technology Co., Ltd.) to above-mentionedOven-dried condition under L-lysine sulfate crush sample carry out spectra collection.Near infrared spectrometer preheating is opened at 25 DEG C30min is swept before sample every time using air spectrum as background spectrum.Wherein, the near infrared spectrometer acquisition mode is adopted for diffusing reflectionCollection, spectra collection range are 1000nm-2500nm, and resolution ratio 10nm, scanning times are 32 times.When filling sample, each sampleHeight will maintain an equal level with the edge of sample cell, and to guarantee that sample-loading amount is consistent, each sample multiple scanning 3 times asks its average light to set a song to musicLine.The sample to be tested original spectrum curve collected is as shown in Figure 1.
(4) extraction of near infrared spectrum characteristic and model construction
Handle and go scattering to handle (SNVDT), variable standardization (SNV), multiplicative scatter correction using variable standardization(MSC), second derivative method is pre-processed to the full spectrum of sample correction collection is obtained in step (3).Then using competitive adaptiveWeighting algorithm (CARS) should be weighed and be extracted characteristic information in step (3) after sample preprocessing in spectroscopic data, from every spectrumIt is extracted 31 in 1499 number of wavelengths strong points, extracts result as shown in the following table 1 and Fig. 2, specific extraction step includes:
(a) it samples: model being sampled based on Monte Carlo sampling method (Monte Carlo sampling, MCS), andIn each CARS sampling, require to randomly select a certain amount of sample from sample sets as calibration set, to establish PLS mouldType；
(b) variable is removed based on decaying exponential function (exponentially decreasing function, EDP): falseFixed surveyed sample spectrum battle array is X (m × p), and m is sample number, and p is variable number, and the true value matrix of SSC is y (m × 1), then PLS is returnedReturn model are as follows:
Y=Xb+e；
In formula, b indicates that the coefficient vector of p dimension, e indicate prediction residual；
Wherein, b=Wc=[b1, b2..., bp]TThe linear combination coefficient of score matrix and X (W expression), i-th yuan in bThe absolute value of element | bi| (1≤i≤p) indicates contribution of i-th of variable to SSC value, and variable corresponding to the bigger expression of the value is in SSCPrediction in it is more important；
Utilization index attenuation function removes by force | bi| it is worth relatively small wavelength points, and is sampled using MC, is adopted in i-thAfter sample operation, the storage rate of variable point is calculated by following exponential function:
ri=ae-ki；
In formula, a and k indicate constant respectively at the 1st time and n-th MCS, whole p variables and only 2 variable in sample setParticipate in modeling, i.e. r1=1 and rN=2/p, so that the calculation formula of a and k is as follows:
A=(2/P)1/(N-1)；
In formula, ln indicates natural logrithm, and variable number p is 1499, and setting MC is sampled 50 times；
(c) further right based on adaptive weight weight sampling technology (adaptive reweighted sampling, ARS)Variable is screened, the rule of " survival of the fittest " in the technical modelling Darwinian evolution, by evaluating each variable pointWeight wiVariable Selection is carried out, the calculating of weighted value is as follows:
(d) by the RMSECV value for the new variable subset for calculating and generating more every time, with the smallest change of RMSECV valueQuantum collection is as optimal variable subset；And full spectroscopic data is carried out with spectral signature information data building calibration model with this excellentChoosing.
The Pretreated spectra and characteristic information data are extracted, and are run in Matlab and Pls_toolbox software's.Calibration model is constructed using the spectral signature information data combination multiple linear regression method (MLR) after extraction, it is as a result as followsShown in table 2.
Spectral signature wavelength of the table 1 after CARS is extracted
2 L-lysine sulfate Quantitative Analysis Model verification result of table
(5) verifying and sample measures of calibration model
Other verifying collection samples of acquisition 18, repeat step (1) to (3), and the calibration set mould established using step (4)Type prediction verifying collection sample lysine content, specific verification result is as shown in table 3 below, L-lysine sulfate Quantitative Analysis ModelVerifying collection prediction result see the table below 4, and verifying collection sample to be tested predicted value is with actual value correlation scatter plot as shown in attached drawing 3.
3 L-lysine sulfate Quantitative Analysis Model verification result of table
Table 4 verifies model sample true value and predicted value contrast table
| Sample number into spectrum | True value | Predicted value | Deviation |
| 1 | 51.87 | 52.42 | 0.55 |
| 2 | 52.64 | 52.41 | -0.23 |
| 3 | 53.03 | 53.70 | 0.67 |
| 4 | 54.11 | 53.45 | -0.66 |
| 5 | 54.61 | 56.10 | 1.49 |
| 6 | 54.91 | 56.29 | 1.38 |
| 7 | 55.85 | 56.03 | 0.18 |
| 8 | 56.34 | 56.22 | -0.12 |
| 9 | 56.77 | 58.49 | 1.72 |
| 10 | 57.03 | 57.74 | 0.71 |
| 11 | 57.44 | 56.35 | -1.09 |
| 12 | 57.72 | 57.68 | -0.04 |
| 13 | 57.83 | 58.05 | 0.22 |
| 14 | 58.15 | 57.89 | -0.26 |
| 15 | 58.54 | 58.61 | 0.07 |
| 16 | 58.72 | 59.91 | 1.19 |
| 17 | 59.30 | 59.96 | 0.66 |
| 18 | 60.33 | 60.43 | 0.10 |
From can be seen that in the result of table 3 and 4, lysine content near-infrared in the L-lysine sulfate that the present invention constructs is fixedAnalysis model accuracy with higher is measured, can be used for the content detection of lysine.
Obviously, the above embodiments are merely examples for clarifying the description, and does not limit the embodiments.It is rightFor those of ordinary skill in the art, can also make on the basis of the above description it is other it is various forms of variation orIt changes.There is no necessity and possibility to exhaust all the enbodiments.And it is extended from this it is obvious variation orIt changes still within the protection scope of the invention.