CN105758819A

CN105758819A - Method for detecting organic components of soil by utilizing near infrared spectrum

Info

Publication number: CN105758819A
Application number: CN201610113083.4A
Authority: CN
Inventors: 喻文娟; 王瑞斌; 康宏樟; 刘星; 吴节莉; 代傅娜; 吴玉森
Original assignee: Shanghai Jiaotong University
Current assignee: Shanghai Jiaotong University
Priority date: 2016-02-29
Filing date: 2016-02-29
Publication date: 2016-07-13

Abstract

The invention provides a method for detecting organic components of soil by utilizing a near infrared spectrum. The method comprises the following steps: (1) detecting organic matter chemical compositions of a plurality of calibrated soil samples; (2) acquiring diffuse reflection spectrograms of near infrared wavebands of the calibrated soil samples to obtain original spectrograms; (3) carrying out smooth pre-processing on the original spectrograms to obtain processed spectrograms; (4) establishing a quantitative relationship model between the processed spectrograms of the calibrated soil samples and the organic matter chemical compositions by adopting a vector machine method; and (5) acquiring diffuse reflection spectrograms of near infrared wavebands of soil samples to be detected, and detecting the organic matter chemical compositions of the soil samples to be detected according to the quantitative relationship model. The method provided by the invention can be used for rapidly and accurately detecting the organic matter chemical compositions of the soil to be detected by utilizing the near infrared spectrum.

Description

A kind of method of the organic component utilizing near infrared spectrum detection soil

Technical field

The present invention relates to field of ecology, more particularly to a kind of method detecting soil organism chemical composition.

Background technology

The soil organism (Soilorganicmatter, SOM) soil microorganism, animals and plants normal activities are being maintained, preservation of fertility and resiliency, adjusting ambient weather aspect is significant, and studies it is critical only that of SOM and understand its chemical composition and structure in depth.Adopt chemical method, pyrolysis-MS, solid carbon 13 nuclear magnetic resonance method etc. can study SOM structure.Wherein, cross polarization and evil spirit angle of spinning¹³C solid-state nuclear magnetic resonance spectrographic method (Solid-state¹³CNuclearMagneticResonancewithcross-polarizationandmagicanglespinning,CP-MAS¹³CNMR), because it is without extracting organic substance with chemistry or additive method, can more fully provide pedotheque organic composition information, SOM structural research plays more and more important effect.Utilize CP-MAS¹³CNMR studies the basic ideas of SOM chemical composition: 1. first with hydrofluoric acid solution, pedotheque is carried out pretreatment, removes some of which paramagnetism mineral and concentrate organic matter.2. pedotheque good for hydrofluoric acid treatment is carried out¹³C solid-state nuclear magnetic resonance is analyzed.3. the functional group carbon that different on nuclear magnetic spectrogram chemical shifts is corresponding different, such as, spectral peak in 0-50ppm can belong to the alkyl carbon in SOM, Spectra peak recognition in 50-110ppm is in alkoxyl carbon, the Spectra peak recognition in 110-160ppm Spectra peak recognition in aromatic carbon, 160-200ppm is in carboxyl carbon and carbonyl carbon.After being processed by the spectrogram such as phasing, baseline correction, to different-waveband integration, more just can obtain the relative scale of different carbon in SOM by normalization method, thus obtaining the information of SOM chemical composition.Wherein, alkyl carbon/alcoxyl carbon ratio (A/O ratio) has become the generally acknowledged important indicator evaluating SOM stability and degree of decomposition, significant in ecology.Although there being above-mentioned plurality of advantages, utilize nuclear magnetic resonance spectroscopy SOM structure to there is also some shortcomings: as expensive in instrument price, sample treatment is consuming time, test period length, need special messenger to operate, testing cost is high, not easily large-scale promotion etc..Such as, with the magic angle of cross polarization and spin¹³C solid-state nuclear magnetic resonance spectrographic method (CP-MAS¹³CNMR) research soil organism structure just has following deficiency: 1, expensive equipment, and testing cost is high；2, technology requires height, not easily operates；3, testing time length (sample generally requires more than 24h)；4, pre-treatment (processing a sample to need through repeatedly hydrofluoric acid treatment, washing, add last sample lyophilizing, generally require the time of a couple of days) consuming time；5, not easily promote.

Infrared spectrum technology application in soil analysis is risen in the eighties in last century.Currently with near infrared spectrum (NIR) and middle infrared spectrum (MIR) technology, in conjunction with Chemical Measurement means, being widely used in the analysis of the various physicochemical property of soil, result is satisfactory.As: the content of metal, the Microbials etc. such as total carbon content, total nitrogen content, content of tatal phosphorus, moisture, the soil texture, potassium (K), calcium (Ga), ferrum (Fe), manganese (Mn), magnesium (Mg).NIR, MIR spectroscopic analysis methods is a kind of indirect analysis method, it is necessary to first with reference method, the physicochemical property of a large amount of representative pedotheques is measured, and builds calibration model by associating sample spectra and its physicochemical property；Then composition and the character of the unknown pedotheque of calibration model prediction are used.Therefore, tested pedotheque to include the type of predicted pedotheque and the scope of physicochemical property as far as possible, and the physicochemical property of its each component is carried out Accurate Determining.

Near-infrared (NIR) SPECTRAL REGION refers to wavelength electromagnetic wave within the scope of 780～2500nm, its spectral information derives from frequency multiplication and the sum of fundamental frequencies of intramolecule vibration, and mainly reflect that in molecule, hydric group is (such as C-H, N-H, O-H, S-H etc.) frequency multiplication and sum of fundamental frequencies absorption of vibrations.Many Organic substances have characteristic absorption in this SPECTRAL REGION, and the molecular structure of the absorption intensity of different-waveband and this material and concentration exist corresponding relation.Mid-infrared (MIR) SPECTRAL REGION is wavelength electromagnetic wave within the scope of 2500～25000nm, and material is that fundamental frequency, frequency multiplication and sum of fundamental frequencies absorb at the absworption peak of this scope.Different compounds have its special infrared absorption spectroscopy, and the intensity of its bands of a spectrum, position, shape and number are all relevant with compound and state thereof.MIR and NIR spectra are distinctive in that, near infrared spectrum is the frequency multiplication absorption with sum of fundamental frequencies of material molecule internal vibration, and the bands of a spectrum of different component and functional group are easier to overlap and information strength is more weak, cause spectrum elucidation relative difficulty, institute's established model is subject to the impact of extraneous factor, poor stability；And the fundamental frequency that middle infrared spectrum is intramolecule vibration absorbs, its information strength is relatively strong, and information retrieval is relatively easy.

Diffuse-reflectance is a kind of conventional near-infrared acquisition method, its ultimate principle is: when light is irradiated to the surface of loose solid sample, except some is by except sample surfaces reflects (be called mirror reflection light) immediately, remaining incident illumination produces unrestrained transmitting at sample surfaces, or toss about in bed reflection between sample microgranule and decay gradually, or the scattering for turning back again after penetrating internal layer.The light being diffusely reflected or scattering out after these contact sample microparticle surfaces has absorption-attenuation characteristic, here it is diffuse-reflectance produces the fundamental cause of spectrum.The effect of diffuse-reflectance device be exactly maximum intensity these diffusions, the luminous energy pinching that scatters out are got up to send into detector to the spectral signal with good signal-to noise ratio.The spectral technology that diffuses is a kind of detection method developed rapidly over nearly 20 years, the method is easy and simple to handle, quick, non-demolition various samples can be analyzed fast, accurately, in addition the development of the digitized of analytical tool and chemometrics method, use chemometrics method can solve the extraction of spectral information and the impact of ambient interferences aspect well, give play to important function in making it in a lot of fields, and achieve good Social and economic benef@.

No matter being NIR or MIR spectrum, in the spectral information collected, comprising some information that From Spectral Signal can be produced interference, thus affecting foundation and the prediction of model, it is therefore desirable to carry out Pretreated spectra.Conventional preprocessing procedures has being used in combination of data smoothing, baseline correction, centralization, multiplicative scatter correction, standardization, derivative, Fourier transform and above several method.Additionally, spectrogram compression and information retrieval can improve the effective information rate analyzed in signal, and its main method has principal component analysis (PCA), wavelet analysis, simulated annealing (SAA), genetic algorithm (GA), moving window (MWPLS) etc..

One of core technology of NIR and MIR spectrum analysis is to set up functional relationship between spectral information and component physicochemical property, namely sets up calibration model.The analysis method that spectrum regression analysis is commonly used has: multiple linear regression (MLR), principal component regression (PCR), partial least square method return (PLSR), artificial neural network (ANN), support vector machine (SVM) etc..MLR, PCR and PLSR are used for solving linear correction problem, and ANN and SVM is used for solving gamma correction problem.

Owing to the impact of spectrum is belonged to non-linear by the factors such as the state of spectrogrph, measurement environment mostly, the relation also having some mass parameters and spectrum is also non-linear.Support vector machine (SVM), as the multivariate calibration methods under nonlinear regression, is avoided that the problem such as over-fitting and local minimum that additive method exists, have also been obtained extensive use in recent years.Support vector machine (SVM) is initial to be proposed the nineties in 20th century by Vapnik, it it is a kind of a kind of novel modeling method grown up by Statistical Learning Theory, it is with structural risk minimization principle for theoretical basis, there is stronger study generalization ability, solve the problems such as dimension non-linear, high, small sample preferably, start to become a kind of approach preferably of the tradition difficulties such as solution " crossing study ", be successfully applied in the field such as pattern recognition, signal processing.The ultimate principle of support vector machine is that the input space is transformed to a higher dimensional space by the nonlinear transformation defined by interior Product function, finds a kind of relation between input variable and output variable in this higher dimensional space.

At present, have some both at home and abroad and utilize infrared spectrum in conjunction with the report of the Chemical Measurement research soil organism, but be concentrated mainly on the prediction to SOM content.Further, since it adopts linear method (such as method of least square) to set up calibration model, it was predicted that precision not high enough (prediction related coefficient of a lot of functional group carbon is below 0.80).

Summary of the invention

Because the drawbacks described above of prior art, the method that the invention provides the organic component of a kind of new detection soil, the organic chemical composition being to utilize near-infrared (NIR) spectrum quick and precisely to detect soil to be solved the technical problem that.

For solving the problems referred to above, the present invention adopts the technical scheme that: a kind of method of organic component utilizing near infrared spectrum detection soil, described method comprises the steps:

1) the organic chemical composition of multiple calibration pedotheque is recorded；

2) gather the diffuse-reflectance spectrogram of the near infrared band of calibration pedotheque, obtain original spectrogram；

3) original spectrogram is carried out smooth pretreatment, spectrogram after being processed；

4) vector machine method is adopted to set up after the process of calibration pedotheque the causes between spectrogram and organic chemical composition；

5) gather the diffuse-reflectance spectrogram of the near infrared band of pedotheque to be measured, calculate the organic chemical composition of pedotheque to be measured according to causes.

Preferably, in described step 1) in, the method for the organic chemical composition recording calibration pedotheque is nuclear magnetic resonance spectrometry.

Preferably, in described step 1) in, the concrete steps of preparation calibration pedotheque include: after soil sample being dewatered, levigate, cross 60 mesh sieves.

Preferably, in described step 1) in, the testing index of the described organic chemical composition recording multiple calibration pedotheque includes content and the A/O ratio of alkyl carbon, alkoxyl carbon, aromatic carbon and carboxylic/base carbon.

Preferably, in described step 3) in, the concrete steps that original spectrogram carries out smooth pretreatment include: the atmospheric background suppresses, and absorbance is changed, and automatic baseline correction and Norris first derivative filtering process.

Preferably, in described step 4) in, the concrete steps setting up causes include: spectral information and organic chemical composition are divided into respectively calibration set and checking collection by SPXY method, adopt RBF kernel function to be supported vector machine to calculate, utilize Web search method and determined kernel functional parameter g and the penalty parameter c of the best by 5 folding cross-validation methods, it is independent variable with the spectral information of calibration set, with stable carbon isotope ratio for dependent variable, set up regression model, and utilize the precision of checking collection inspection calibration model.It is highly preferred that the ratio of sample number that calibration set and checking integrate is as 3:1.It is highly preferred that described utilizes Web search method and determines that the kernel functional parameter g of the best and the step of penalty parameter c include by 5 folding cross-validation methods: allow penalty parameter c and kernel functional parameter g 2^-10To 2¹⁰Between discrete value；For taking fixed kernel functional parameter g and penalty parameter c, as initial data and 5 folding cross validations are utilized to choose the kernel functional parameter g and penalty parameter c making calibration set checking mean square error minimum calibration set；When making the calibration set checking minimum kernel functional parameter g of mean square error and penalty parameter c have many groups, then choose minimum one group of penalty parameter c as optimal parameter；Minimum when choosing penalty parameter c, to there being polykaryon function parameter g, then choose the first group of kernel functional parameter g and penalty parameter c that search as optimal parameter.It is highly preferred that include by the concrete steps verifying the precision integrating inspection calibration model: investigate model performance by prediction related coefficient and predicted root mean square error as index, and use prediction relation analysis error that model is evaluated.

Preferably, in described step 1) in, the concrete steps preparing pedotheque to be measured include: after soil sample being dewatered, levigate, cross 60 mesh sieves.

The invention have the benefit that 1, this method can prediction soil organism chemical composition (being measured by nuclear-magnetism) quickly, accurately, at a low price.2, easy and simple to handle, popularization is strong, applied range.3, suitable in various ecosystems such as forest, farmland, meadows.

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 embodiment of the present invention soil organism¹³The schematic diagram of the chemical shift wave band (for Yunshan Mountain 2-5cm) of C solid state nmr spectrogram and difference in functionality group carbon.

Fig. 2 is the near infrared absorption spectrogram of 4 the different depth soil in embodiment of the present invention sample Pinggu, ground.

Fig. 3 is the schematic diagram of spectral model in the embodiment of the present invention 1 (alkyl carbon) predictive value and actual value dependency.

Fig. 4 is the schematic diagram of spectral model in the embodiment of the present invention 2 (alcoxyl carbon) predictive value and actual value dependency.

Fig. 5 is the schematic diagram of spectral model (aromatic carbon) predictive value and actual value dependency in the embodiment of the present invention 3.

Fig. 6 is the schematic diagram of spectral model in the embodiment of the present invention 4 (carboxyl carbon and carbonyl carbon) predictive value and actual value dependency.

Fig. 7 is the schematic diagram of spectral model in the embodiment of the present invention 5 (A/O ratio) predictive value and actual value dependency.

Detailed description of the invention

As a kind of preferred embodiment, the method for the machine matter chemical composition utilizing near infrared spectrum quickly to detect soil provided by the present invention comprises the steps.

(1) standby soil sample to be checked

By mineral nitrogen layer soil roguing, air-dry, pulverizing, crossing 60 mesh sieves, exsiccator saves backup.Totally 56 samples come from the mineral soil sample of different depth (0-2,2-5,5-10,10-20cm) in the cork oak forest in seven different regions (Pinggu, Hong Yashan, Bai An, Huang Zangyu, Xinyang, Mount Huang, the Yunshan Mountain).

The innovative point of the present invention: sample area, the sample of collection of the present invention comes from 5 provinces on a latitudinal gradient, at a distance of 1500 kilometers from north to south, is gradually transitions the subtropical zone in south from northern warm temperate zone.

(2) CP-MAS is utilized¹³CNMR method records relative amount and the A/O ratio of four class functional group carbon in the soil organism (alkyl carbon, alkoxyl carbon, aromatic carbon and carboxylic (carbonyl) base carbon).

First with hydrofluoric acid solution, pedotheque is carried out pretreatment, remove some of which paramagnetism mineral and concentrate organic matter.Pedotheque good for hydrofluoric acid treatment is carried out¹³C solid-state nuclear magnetic resonance is analyzed.The functional group carbon that chemical shifts different on nuclear magnetic spectrogram is corresponding different, such as, spectral peak in 0-50ppm can belong to the alkyl carbon in SOM, Spectra peak recognition in 50-110ppm is in alkoxyl carbon, the Spectra peak recognition in 110-160ppm Spectra peak recognition in aromatic carbon, 160-200ppm is in carboxyl carbon and carbonyl carbon.After being processed by the spectrogram such as phasing, baseline correction, to different-waveband integration, the relative amount of different carbon just can be obtained in SOM again, by the relative amount of the alkyl carbon relative amount divided by alcoxyl carbon and A/O ratio, the final information obtaining SOM chemical composition by normalization method.

(3) spectra collection

Placing the rustless steel cylinder that cross section is annular of an internal diameter 11mm on low-hydroxy-group squartz glass, its white light that bottom is penetrated up is unobstructed；The 200mg soil sample accurately weighed being placed in it, then have weight by one be that 4g, diameter are also placed in soil sample gently for the bottle of 11mm, it can make, and thickness of sample is homogeneous, have enough dress sample degree of depth and will not press and too tightly produce direct reflection.Utilizing designed, designed of the present invention the sample stage built, gather the diffuse-reflectance spectrogram of near infrared band, instrument configuration is: Fourier transformation infrared spectrometer, and adnexa is near-infrared integrating sphere, white light source, CaF₂Beam splitter, InGaAs detector；Acquisition parameter is: with Jin Jing for background, sweep limits 10000-4000cm^-1, resolution 4cm^-1, scan 64 times.

Innovation about harvester: the general extensive soil sample that gathers has the automatic sampling apparatus of fixed dimension, but it can improve instrument price and testing cost.Based on practical angle, designed, designed of the present invention has also built the irreflexive sample stage of pedotheque, and it is primarily intended to maintenance, and thickness of sample is homogeneous, have enough dress sample degree of depth and will not press and too tightly produce direct reflection.

(4) data prediction

Software Omnic8.2 is carried by whole for original spectrogram wave band (i.e. 10000-4000cm with instrument^-1) carry out the atmospheric background suppression, change into absorption spectrum, then carry out automatic baseline correction.After spectroscopic data imports software matlab7.8, processing with Norris (7,7,1) derivation smoothing techniques, in bracket, first 7 represents that smooth at 7, and second 7 represents 7 differential width, and 1 represents first derivative.Norris derivation smoothing processing can eliminate noise, is eliminated as much as unrelated information variable.

The pretreatment of pedotheque near infrared spectrum, it is possible to efficiently reduce system deviation, noise, granularity and crest cross the impact of point etc..In the preprocessing procedures that some are common, baseline correction mainly eliminates baseline drift；Smoothing processing mainly eliminates noise information；Derivative processing (first derivation or second order derivation) can effectively eliminate needle position misalignment, reduces peak overlapping with peak-to-peak, obtains more effective information；Multiplicative scatter correction is to eliminate the impact on solid diffuse-reflectance spectrum of solid particle size, surface scattering and change in optical path length.

Its corresponding forecast model is had different improvement and impact by the cross-reference of preprocess method not of the same race, so, compare Norris first derivative filtering and the modelling effect of Norris second dervative filtering herein, also compare and add multiplicative scatter correction and be not added with the modelling effect of Norris first derivative filtering during multiplicative scatter correction.

(5) support vector machine method is adopted to set up soil near infrared spectrum and δ¹³Quantitative relationship between C value

Soil organism chemical composition (relative amount of four big functional group carbon and A/O ratio) is also introduced in matlab software.With SPXY algorithm, spectral information and organic chemical composition are divided into calibration set and checking collection respectively by test sample in the ratio of 3:1, are respectively used to model and set up and checking.

The innovative point of the present invention: during modeling, many researchs are all around the preprocessing procedures how choosing the best, the less division methods comparing calibration set and checking collection, but the selection of calibration set and checking collection sample is most important to spectrum Multivariate Calibration.Conventional Method of Sample Selection mainly includes randomized (RS) and K-S (Kennard Stone) method and SPXY (samplesetpartitioningbasedonjointx-ydistance) method at present.Randomized randomness is big, does not ensure that selected sample has enough representativenesses；Sample big for SPECTRAL DIVERSITY is selected into calibration set by K-S method, and all the other samples are included into checking and collect, but low for content or that concentration is low scope, and between sample, spectrum change is only small, and the sample often selected is not representative yet；SPXY algorithm is the sample set system of selection on a kind of Corpus--based Method basis, by spectrum-physics and chemistry value symbiosis distance as according to ensure at utmost to characterize sample distribution, to improve model stability.The spectrum that the present invention processes based on Norris first derivative filtering, compares SPXY method and the effect of K-S method institute established model.

1. data normalization.For preventing dimension between data different, the general spectroscopic data first calibration set and checking concentrated and organic chemical group Chengdu carry out the normalized of [-1,1], after the foundation of calibration model to be done and prediction, renormalization again, the precision of prediction of computation model.

2. optimal parameter c and g is found.For nonlinear problem, support vector machine (Supportingvectormachine, SVM) main thought of homing method is the linear problem that former problem is converted into certain higher dimensional space by nonlinear transformation, and in higher dimensional space, carry out linear solution, the present invention is selected to reduce the Radial basis kernel function (Radialbasisfunction, RBF) of computational complexity in training process

K(||x-x_i| D=exp (-| | x-x_i||²//2g²)

Wherein x_iFor kernel function center, g is the width parameter of function, controls the radial effect scope of function to realize this process.The factor affecting SVM model performance generally has two, i.e. the value of kernel functional parameter g (core width) and penalty parameter c (regularization parameter).C controls sample beyond the punishment degree calculating error, and g is control function regression error then, and directly affects initial eigenvalue and characteristic vector.The too small meeting of g causes over-fitting, and the excessive then model of contrary g is excessively simple, thus impact prediction precision.Therefore, in order to improve study and the generalization ability of SVM, it is necessary to kernel functional parameter g and penalty parameter c are optimized.

The present invention adopts 5 folding cross verifications to choose best penalty parameter c and kernel functional parameter g.Its basic process is, allows c and g all 2^-10To 2¹⁰Between discrete value (its value is changed to 2^-10, 2^-9.5, 2^-9..., 2⁹, 2^9.5, 2¹⁰), utilize 5 folding validation-cross (5fold-CV) methods to obtain organizing calibration set checking mean square error (MSE) under c and g at this as raw data set calibration set for taking fixed c and g, finally take so that calibration set verifies that group c and g minimum for MSE is as optimal parameter.If there being many checking mean square errors corresponding minimum for group c and g, then that minimum for Selecting All Parameters c group is optimal parameter；If corresponding minimum c organizes g more, just choose first group of c and g searched as optimal parameter.Reason for this is that too high c can cause that learning state occurs, namely calibration set MSE is very low and verify collection the MSE significantly high generalization ability of the grader (reduce).

3. regression model, data renormalization, prediction are set up.Utilize grid-search method, after determining best Radial basis kernel function parameter c and g by 5 above-mentioned folding cross verifications, using the infrared spectrum in calibration set as independent variable, with organic chemical composition for dependent variable, both of which is mapped in higher dimensional space, sets up the Support vector regression model between soil near infrared spectrum and organic chemical composition.And the spectral information of individual authentication collection substituted into this model calculate organic chemical composition, data renormalization process after by concentrating the organic chemical composition of actual measurement to compare with checking, testing model precision of prediction.With prediction related coefficient (Correlationcoefficientinprediction, R), predicted root mean square error (Rootmeansquareerrorinprediction, RMSEP) investigating model performance for index, good model should possess the feature that coefficient is high and error is low.Additionally, use prediction relation analysis error (Residualpredictivedeviation, RPD) that model is carried out deep evaluation；In soil spectrum is analyzed, RPD < 1.0, illustrate that model is very poor, it does not have practicality；1.0 < RPD < 1.5, model is poor, and range of application is very limited；1.4 < RPD < 1.8, model is general, can be used for qualitative evaluation；1.8 < RPD < 2.0, model is good, can be used for quantitative forecast；2.0 < RPD < 2.5, model is fine；RPD > 2.5, model is outstanding.Model establishes and can the near-infrared diffuse-reflectance information of soil unknown, that character is similar be substituted in this model afterwards, it was predicted that go out its organic chemical composition.

The present invention gathers the near infrared light spectrogram of cork oak forest soil (from different local and the degree of depth), after initial data is carried out series of preprocessing, in conjunction with the SVM in Chemical Measurement, can accurately, quickly, easy, predict organic chemical composition information in various soil at low cost.

Relative to magnetic nuclear resonance method, the invention have the advantage that

(1) accurate.Prediction related coefficient R is more than 0.85, it was predicted that root-mean-square error less than 0.70, RPD more than 1.80.

(2) quick.One sample collecting only needs 3min, within one day, can gather at least 200 infrared spectrums, and follow-up data process and can complete in 1h.

(3) simple to operate.Sample pre-treatments is simple, and instrumentation is simple.

(4) testing cost is low.Compare nuclear magnetic resonance chemical analyser, infrared spectrometer low price.

(5) popularization is strong, it is easy to promote.Instrumentation is simple, and price is relatively cheap.

(6) applied widely, cannot be only used for the detection of forest soil, it is also possible to other ecosystem in farmland, grassland etc..

Relative to utilizing near infrared spectrum to predict soil organism composition in conjunction with partial least square method, the invention have the advantage that and significantly improve predictablity rate.

The prediction of alkyl carbon relative amount in embodiment 1 soil organism

(1) soil sample to be checked is prepared.By mineral nitrogen layer soil roguing, air-dry, pulverizing, crossing 60 mesh sieves, exsiccator saves backup.Totally 56 samples come from the pedotheque of different depth (0-2,2-5,5-10,10-20cm) in the cork oak forest in seven different regions (Pinggu, Hong Yashan, Bai An, Huang Zangyu, Xinyang, Mount Huang, the Yunshan Mountain).

(2) CP-MAS is utilized¹³CNMR method records the relative amount of alkyl carbon in the soil organism.First with hydrofluoric acid solution, pedotheque is carried out pretreatment, remove some of which paramagnetism mineral and concentrate organic matter.Pedotheque good for hydrofluoric acid treatment is carried out¹³C solid-state nuclear magnetic resonance is analyzed.On nuclear magnetic spectrogram, the spectral peak in 0-50ppm can belong to the alkyl carbon in SOM.After being processed by the spectrogram such as phasing, baseline correction, to this waveband integral, more just can obtain the relative amount (Fig. 1) of alkyl carbon in SOM by normalization method.

(3) spectra collection.Weighing 200mg soil sample, be placed in the rustless steel cylinder that cross section is annular of 11mm, bottom is low-hydroxy-group squartz glass, and sample top flattens.Gathering the diffuse-reflectance spectrogram of near infrared band, instrument configuration is: Fourier transformation infrared spectrometer, and adnexa is near-infrared integrating sphere, white light source, CaF₂Beam splitter, InGaAs detector；Acquisition parameter is: with Jin Jing for background, sweep limits 10000-4000cm^-1, resolution 4cm^-1, scan 64 times.For Pinggu, sample ground, accompanying drawing 2 shows the near-infrared spectrogram of 4 depth of soil.

(4) data prediction.Software Omnic8.2 is carried by whole for original spectrogram wave band (i.e. 10000-4000cm with instrument^-1) carry out the atmospheric background suppression, change into absorption spectrum, then carry out automatic baseline correction.After the spectroscopic data of 56 samples is imported software matlab7.8, first full spectrum is carried out multiplicative scatter correction, then carry out the process such as 7 smooth, first derivatives by Norris method.

The near infrared spectrum of pedotheque has been attempted pretreatment 4 kinds common by this embodiment: 1. Norris first derivative filtering+SPXY method divides calibration set and checking collection；2. multiplicative scatter correction+Norris first derivative filtering+SPXY method divides calibration set and checking collection；3. Norris second dervative filtering+SPXY method divides calibration set and checking collection；4. Norris first derivative filtering+K-S method divides calibration set and checking collection, reapplies support vector machine (SVM) method and sets up the quantitative estimation model of soil alkyl carbon.Result (table 1) shows, the precision of prediction of built soil alkyl carbon content estimation models is had certain impact by different pretreatments method, the spectrum modeling accuracy of preprocess method that this patent adopts (multiplicative scatter correction+Norris first derivative filtering+SPXY method divides calibration set and checking collection) is significantly high, is just slightly below the spectrum modeling accuracy 1. planting preprocess method.

1 model preprocessing procedures of table predicts the precision comparison of soil organism alkyl carbon content with common preprocess method

(5) support vector machine method is adopted to set up the model of alkyl carbon relative amount in soil near infrared spectrum quantitative correction organic matter.56 corresponding soil alkyl carbon relative amounts are also introduced in matlab software.With SPXY algorithm, spectral information and alkyl carbon relative amount being divided in the ratio of 3:1 by test sample calibration set and checking collection respectively, calibration set is containing 42 samples, and checking collection is containing 14 samples.After data carry out [-1,1] normalized, based on calibration set, utilizing grid-search method to choose Radial basis kernel function the best c in 5 folding interactive verification process is 0.5, and best g is 9.77 × 10^-4.In all-wave spectral limit, the regression model between soil near infrared spectrum and alkyl carbon relative amount is set up with calibration set again with best c, g of choosing, and the spectral information of individual authentication collection is substituted into this model calculating alkyl carbon relative amount, by concentrating actual measurement alkyl carbon relative amount to compare with checking after the process of data renormalization, testing model precision of prediction.Prediction related coefficient (R) is 0.8552 (Fig. 3), it was predicted that root-mean-square error (RMSEP) is 0.6467, it was predicted that relation analysis error (RPD) is 1.86.And the calibration model R set up with method of least square be 0.8963, RMSEP be 1.055, RPD is 1.75 (tables 2).

The precision comparison of table 2 support vector machine (SVM) and offset minimum binary (PLS) model prediction soil organism chemical composition

This result shows, the method is suitable in forest soil different depth (0-2cm, 2-5cm, 5-10cm, 10-20cm) detection of alkyl carbon relative amount, can quickly detect alkyl carbon relative amount in the soil organism within a short period of time, and satisfied accuracy of detection can be reached.When setting up calibration model in conjunction with Chemical Measurement, being different from traditional linear correction method (such as PLS), the present invention adopts a kind of gamma correction model SVM, achieves good precision of prediction equally.Meanwhile, multiplicative scatter correction, Norris first derivative filtering and SPXY method divide the preprocess methods such as data set to utilizing near-infrared diffuse-reflectance technology to play an important role in conjunction with SVM method harmless quantitative detection soil alkyl carbon relative amount.56 samples in test model are from the cork oak forest soil crossing over two climate zones and five provinces, and spectrum all can be produced impact by different weather conditions and soil property.But the model set up at such complex condition just just has the wider array of scope of application, therefore, the near-infrared spectrum technique based on support vector machine method is to be suitable for the efficient detection technology that in the soil organism, alkyl carbon content is predicted.

The prediction of alcoxyl carbon relative amount in embodiment 2 soil organism

(2) CP-MAS is utilized¹³CNMR method records the relative amount of alcoxyl carbon in the soil organism.First with hydrofluoric acid solution, pedotheque is carried out pretreatment, remove some of which paramagnetism mineral and concentrate organic matter.Pedotheque good for hydrofluoric acid treatment is carried out¹³C solid-state nuclear magnetic resonance is analyzed.On nuclear magnetic spectrogram, the spectral peak in 50-110ppm can belong to the alcoxyl carbon in SOM.After being processed by the spectrogram such as phasing, baseline correction, to this waveband integral, more just can obtain the relative amount (Fig. 1) of alcoxyl carbon in SOM by normalization method.

The near infrared spectrum of pedotheque has been attempted pretreatment 4 kinds common by this embodiment: 1. Norris first derivative filtering+SPXY method divides calibration set and checking collection；2. multiplicative scatter correction+Norris first derivative filtering+SPXY method divides calibration set and checking collection；3. Norris second dervative filtering+SPXY method divides calibration set and checking collection；4. Norris first derivative filtering+K-S method divides calibration set and checking collection, reapplies support vector machine (SVM) method and sets up the quantitative estimation model of soil alcoxyl carbon.Result (table 3) shows, the precision of prediction of built soil alcoxyl carbon content estimation models is had certain impact by different pretreatments method, and the spectrum modeling accuracy of the preprocess method (multiplicative scatter correction+Norris first derivative filtering+SPXY method divides calibration set and checking collection) that this patent adopts is significantly higher than other three kinds of methods.

3 model preprocessing procedures of table predict the precision comparison of soil organism alcoxyl carbon content with common preprocess method

(5) support vector machine method is adopted to set up the model of alcoxyl carbon relative amount in soil near infrared spectrum quantitative correction organic matter.56 corresponding soil alcoxyl carbon relative amounts are also introduced in matlab software.With SPXY algorithm, spectral information and alcoxyl carbon relative amount being divided in the ratio of 3:1 by test sample calibration set and checking collection respectively, calibration set is containing 42 samples, and checking collection is containing 14 samples.After data carry out [-1,1] normalized, based on calibration set, utilizing grid-search method to choose Radial basis kernel function the best c in 5 folding interactive verification process is 0.4, and best g is 9.77 × 10^-4.In all-wave spectral limit, the regression model between soil near infrared spectrum and alcoxyl carbon relative amount is set up with calibration set again with best c, g of choosing, and the spectral information of individual authentication collection is substituted into this model calculating alcoxyl carbon relative amount, by concentrating actual measurement alcoxyl carbon relative amount to compare with checking after the process of data renormalization, testing model precision of prediction.Prediction related coefficient (R) is 0.9256 (Fig. 4), it was predicted that root-mean-square error (RMSEP) is 0.6874, it was predicted that relation analysis error (RPD) is 2.00.And the calibration model R set up with method of least square be 0.7505, RMSEP be 0.9414, RPD is 1.07 (tables 2).

This result shows, the method is suitable in forest soil different depth (0-2cm, 2-5cm, 5-10cm, 10-20cm) detection of alcoxyl carbon relative amount, can quickly detect alcoxyl carbon relative amount in the soil organism within a short period of time, and satisfied accuracy of detection can be reached.When setting up calibration model in conjunction with Chemical Measurement, it is different from traditional linear correction method (such as PLS), the present invention adopts a kind of gamma correction model SVM, significantly improves the precision of prediction of model, makes infrared method prediction soil alcoxyl carbon content be possibly realized.Meanwhile, multiplicative scatter correction, Norris first derivative filtering and SPXY method divide the preprocess methods such as data set to utilizing near-infrared diffuse-reflectance technology to play an important role in conjunction with SVM method harmless quantitative detection soil alcoxyl carbon relative amount.56 samples in test model are from the cork oak forest soil crossing over two climate zones and five provinces, and spectrum all can be produced impact by different weather conditions and soil property.But the model set up at such complex condition just just has the wider array of scope of application, therefore, the near-infrared spectrum technique based on support vector machine method is to be suitable for the efficient detection technology that in the soil organism, alcoxyl carbon content is predicted.

The prediction of aromatic carbon relative amount in embodiment 3 soil organism

(2) CP-MAS is utilized¹³CNMR method records the relative amount of aromatic carbon in the soil organism.First with hydrofluoric acid solution, pedotheque is carried out pretreatment, remove some of which paramagnetism mineral and concentrate organic matter.Pedotheque good for hydrofluoric acid treatment is carried out¹³C solid-state nuclear magnetic resonance is analyzed.On nuclear magnetic spectrogram, the spectral peak in 110-160ppm can belong to the aromatic carbon in SOM.After being processed by the spectrogram such as phasing, baseline correction, to this waveband integral, more just can obtain the relative amount (Fig. 1) of aromatic carbon in SOM by normalization method.

The near infrared spectrum of pedotheque has been attempted pretreatment 4 kinds common by the present invention: 1. Norris first derivative filtering+SPXY method divides calibration set and checking collection；2. multiplicative scatter correction+Norris first derivative filtering+SPXY method divides calibration set and checking collection；3. Norris second dervative filtering+SPXY method divides calibration set and checking collection；4. Norris first derivative filtering+K-S method divides calibration set and checking collection, reapplies support vector machine (SVM) method and sets up the quantitative estimation model of soil aromatic carbon.Result (table 4) shows, the precision of prediction of built soil aromatic carbon content estimation models is had certain impact by different pretreatments method, the spectrum modeling accuracy of preprocess method that this patent adopts (multiplicative scatter correction+Norris first derivative filtering+SPXY method divides calibration set and checking collection) is significantly high, is just slightly below the spectrum modeling accuracy 1. planting preprocess method.

4 model preprocessing procedures of table predict the precision comparison of soil organism aromatic carbon content with common preprocess method

(5) support vector machine method is adopted to set up the model of aromatic carbon relative amount in soil near infrared spectrum quantitative correction organic matter.56 corresponding soil aromatic carbon relative amounts are also introduced in matlab software.With SPXY algorithm, spectral information and aromatic carbon relative amount being divided in the ratio of 3:1 by test sample calibration set and checking collection respectively, calibration set is containing 42 samples, and checking collection is containing 14 samples.After data carry out [-1,1] normalized, based on calibration set, utilizing grid-search method to choose Radial basis kernel function the best c in 5 folding interactive verification process is 2.0, and best g is 9.77 × 10^-4.In all-wave spectral limit, the regression model between soil near infrared spectrum and aromatic carbon relative amount is set up with calibration set again with best c, g of choosing, and the spectral information of individual authentication collection is substituted into this model calculating aromatic carbon relative amount, by concentrating actual measurement aromatic carbon relative amount to compare with checking after the process of data renormalization, testing model precision of prediction.Prediction related coefficient (R) is 0.8786 (Fig. 5), it was predicted that root-mean-square error (RMSEP) is 0.5194, it was predicted that relation analysis error (RPD) is 2.02.And the calibration model R set up with method of least square be 0.9311, RMSEP be 0.4543, RPD is 2.51 (tables 2).

This result shows, the method is suitable in forest soil different depth (0-2cm, 2-5cm, 5-10cm, 10-20cm) detection of aromatic carbon relative amount, can quickly detect aromatic carbon relative amount in the soil organism within a short period of time, and satisfied accuracy of detection can be reached.When setting up calibration model in conjunction with Chemical Measurement, being different from traditional linear correction method (such as PLS), the present invention adopts a kind of gamma correction model SVM, achieves good precision of prediction equally.Meanwhile, multiplicative scatter correction, Norris first derivative filtering and SPXY method divide the preprocess methods such as data set to utilizing near-infrared diffuse-reflectance technology to play an important role in conjunction with SVM method harmless quantitative detection soil aromatic carbon relative amount.56 samples in test model are from the cork oak forest soil crossing over two climate zones and five provinces, and spectrum all can be produced impact by different weather conditions and soil property.But the model set up at such complex condition just just has the wider array of scope of application, therefore, the near-infrared spectrum technique based on support vector machine method is to be suitable for the efficient detection technology that in the soil organism, aromatic carbon content is predicted.

The prediction of carbonyl carbon and carboxyl carbon relative amount in embodiment 4 soil organism

(2) CP-MAS is utilized¹³CNMR method records the relative amount of carboxylic in the soil organism (carbonyl) base carbon.First with hydrofluoric acid solution, pedotheque is carried out pretreatment, remove some of which paramagnetism mineral and concentrate organic matter.Pedotheque good for hydrofluoric acid treatment is carried out¹³C solid-state nuclear magnetic resonance is analyzed.On nuclear magnetic spectrogram, the spectral peak in 160-200ppm can belong to carboxylic (carbonyl) the base carbon in SOM.After being processed by the spectrogram such as phasing, baseline correction, to this waveband integral, more just can obtain the relative amount (Fig. 1) of carboxylic in SOM (carbonyl) base carbon by normalization method.

The near infrared spectrum of pedotheque has been attempted pretreatment 4 kinds common by the present invention: 1. Norris first derivative filtering+SPXY method divides calibration set and checking collection；2. multiplicative scatter correction+Norris first derivative filtering+SPXY method divides calibration set and checking collection；3. Norris second dervative filtering+SPXY method divides calibration set and checking collection；4. Norris first derivative filtering+K-S method divides calibration set and checking collection, reapplies support vector machine (SVM) method and sets up the quantitative estimation model of soil carboxylic (carbonyl) base carbon.Result (table 5) shows, the precision of prediction of built soil carboxylic (carbonyl) base carbon content estimation models is had certain impact by different pretreatments method, and the spectrum modeling accuracy of the preprocess method (multiplicative scatter correction+Norris first derivative filtering+SPXY method divides calibration set and checking collection) that this patent adopts is significantly higher than other three kinds of methods.

5 model preprocessing procedures of table predict the precision comparison of soil organism carbonyl (carboxylic) base carbon content with common preprocess method

(5) support vector machine method is adopted to set up the model of carboxylic (carbonyl) base carbon relative amount in soil near infrared spectrum quantitative correction organic matter.56 corresponding soil carboxylic (carbonyl) base carbon relative amounts are also introduced in matlab software.With SPXY algorithm, spectral information and carboxylic (carbonyl) base carbon relative amount being divided in the ratio of 3:1 by test sample calibration set and checking collection respectively, calibration set is containing 42 samples, and checking collection is containing 14 samples.After data carry out [-1,1] normalized, based on calibration set, utilizing grid-search method to choose Radial basis kernel function the best c in 5 folding interactive verification process is 2.8, and best g is 9.77 × 10^-4.In all-wave spectral limit, the regression model between soil near infrared spectrum and carboxylic (carbonyl) base carbon relative amount is set up with calibration set again with best c, g of choosing, and the spectral information of individual authentication collection is substituted into this model calculating carboxylic (carbonyl) base carbon relative amount, by concentrating actual measurement carboxylic (carbonyl) base carbon relative amount to compare with checking after the process of data renormalization, testing model precision of prediction.Prediction related coefficient (R) is 0.9173 (Fig. 6), it was predicted that root-mean-square error (RMSEP) is 0.3628, it was predicted that relation analysis error (RPD) is 2.08.And the calibration model R set up with method of least square be 0.5877, RMSEP be 0.741, RPD is 0.97 (table 2).

This result shows, the detection that the method is suitable in forest soil different depth (0-2cm, 2-5cm, 5-10cm, 10-20cm) carboxylic (carbonyl) base carbon relative amount, can quickly detect carboxylic in the soil organism (carbonyl) base carbon relative amount within a short period of time, and satisfied accuracy of detection can be reached.When setting up calibration model in conjunction with Chemical Measurement, it is different from traditional linear correction method (such as PLS), the present invention adopts a kind of gamma correction model SVM, significantly improves the precision of prediction of model, makes infrared method prediction soil carboxylic (carbonyl) base carbon content be possibly realized.Meanwhile, multiplicative scatter correction, Norris first derivative filtering and SPXY method divide the preprocess methods such as data set to utilizing near-infrared diffuse-reflectance technology to play an important role in conjunction with SVM method harmless quantitative detection soil carboxylic (carbonyl) base carbon relative amount.56 samples in test model are from the cork oak forest soil crossing over two climate zones and five provinces, and spectrum all can be produced impact by different weather conditions and soil property.But the model set up at such complex condition just just has the wider array of scope of application, therefore, the near-infrared spectrum technique based on support vector machine method is to be suitable for the efficient detection technology that in the soil organism, carboxylic (carbonyl) base carbon content is predicted.

The prediction of alkyl carbon/alcoxyl carbon ratio (A/O ratio) in embodiment 5 soil organism

(2) CP-MAS is utilized¹³CNMR method records the A/O ratio in the soil organism.First with hydrofluoric acid solution, pedotheque is carried out pretreatment, remove some of which paramagnetism mineral and concentrate organic matter.Pedotheque good for hydrofluoric acid treatment is carried out¹³C solid-state nuclear magnetic resonance is analyzed.On nuclear magnetic spectrogram, the spectral peak in 0-50ppm can belong to the alkyl carbon (Fig. 1) in SOM, the spectral peak in 50-110ppm can belong to the alcoxyl carbon (Fig. 1) in SOM.After being processed by the spectrogram such as phasing, baseline correction, to the two waveband integral, more just can obtaining the relative amount of two kinds of functional group carbon in SOM by normalization method, the ratio of the former with the latter is A/O ratio.A/O is than the important indicator evaluating SOM stability and degree of decomposition being well recognized as, significant in ecology.

The near infrared spectrum of pedotheque has been attempted pretreatment 4 kinds common by the present invention: 1. Norris first derivative filtering+SPXY method divides calibration set and checking collection；2. multiplicative scatter correction+Norris first derivative filtering+SPXY method divides calibration set and checking collection；3. Norris second dervative filtering+SPXY method divides calibration set and checking collection；4. Norris first derivative filtering+K-S method divides calibration set and checking collection, reapplies support vector machine (SVM) method and sets up the quantitative estimation model of soil organism A/O ratio.Result (table 6) shows, built soil organism A/O is had the spectrum modeling accuracy of preprocess method (multiplicative scatter correction+Norris first derivative filtering+SPXY method divides calibration set and checking collection) that certain impact, this patent adopt the highest than the precision of prediction of estimation models by different pretreatments method.

6 model preprocessing procedures of table predict the precision comparison of soil organism A/O ratio with common preprocess method

(5) support vector machine method is adopted to set up the model of A/O ratio in soil near infrared spectrum quantitative correction organic matter.56 corresponding soil organism A/O ratios are also introduced in matlab software.With SPXY algorithm, spectral information and organic A/O ratio being divided in the ratio of 3:1 by test sample calibration set and checking collection respectively, calibration set is containing 42 samples, and checking collection is containing 14 samples.After data carry out [-1,1] normalized, based on calibration set, utilizing grid-search method to choose Radial basis kernel function the best c in 5 folding interactive verification process is 0.5, and best g is 9.77 × 10^-4.In all-wave spectral limit, set up the regression model of soil near infrared spectrum and organic A/O ratio again with calibration set with best c, g of choosing, and the spectral information of individual authentication collection is substituted into the organic A/O ratio of this model calculating, by concentrating the organic A/O of actual measurement to compare relatively with checking after the process of data renormalization, testing model precision of prediction.Prediction related coefficient (R) is 0.9366 (accompanying drawing 7), it was predicted that root-mean-square error (RMSEP) is 0.0174, it was predicted that relation analysis error (RPD) is 2.69.And the calibration model R set up with method of least square be 0.6788, RMSEP be 0.0293, RPD is 1.22 (tables 2).

This result shows, the method is suitable for the detection of the middle organic A/O ratio of forest soil different depth (0-2cm, 2-5cm, 5-10cm, 10-20cm), can quickly detect the soil organism A/O ratio within a short period of time, and satisfied accuracy of detection can be reached.When setting up calibration model in conjunction with Chemical Measurement, it is different from traditional linear correction method (such as PLS), the present invention adopts a kind of gamma correction model SVM, significantly improves the precision of prediction of model, makes infrared method prediction soil organism A/O ratio be possibly realized.Meanwhile, multiplicative scatter correction, Norris first derivative filtering and SPXY method divide the preprocess methods such as data set to utilizing near-infrared diffuse-reflectance technology to detect soil organism A/O compared with important function in conjunction with SVM method harmless quantitative.56 samples in test model are from the cork oak forest soil crossing over two climate zones and five provinces, and spectrum all can be produced impact by different weather conditions and soil property.But the model set up at such complex condition just just has the wider array of scope of application, therefore, the near-infrared spectrum technique based on support vector machine method is to be suitable for the soil organism A/O efficient detection technology than prediction.

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

1. the method for the organic component utilizing near infrared spectrum detection soil, it is characterised in that described method comprises the steps:

2. the method for claim 1, it is characterised in that in described step 1) in, the method for the organic chemical composition recording calibration pedotheque is nuclear magnetic resonance spectrometry.

3. the method for claim 1, it is characterised in that in described step 1) in, the concrete steps of preparation calibration pedotheque include: after soil sample being dewatered, levigate, cross 60 mesh sieves.

4. the method for claim 1, it is characterised in that in described step 1) in, the testing index of the described organic chemical composition recording multiple calibration pedotheque includes content and the A/O ratio of alkyl carbon, alkoxyl carbon, aromatic carbon and carboxylic/base carbon.

5. the method for claim 1, it is characterised in that in described step 3) in, the concrete steps that original spectrogram carries out smooth pretreatment include: the atmospheric background suppresses, and absorbance is changed, and automatic baseline correction and Norris first derivative filtering process.

6. the method for claim 1, it is characterized in that, in described step 4) in, the concrete steps setting up causes include: spectral information and organic chemical composition are divided into respectively calibration set and checking collection by SPXY method, adopt RBF kernel function to be supported vector machine to calculate, utilize Web search method and determined kernel functional parameter g and the penalty parameter c of the best by 5 folding cross-validation methods, it is independent variable with the spectral information of calibration set, with stable carbon isotope ratio for dependent variable, set up regression model, and utilize the precision of checking collection inspection calibration model.

7. method as claimed in claim 6, it is characterised in that the ratio of the sample number that calibration set and checking integrate is as 3:1.

8. method as claimed in claim 6, it is characterised in that described utilizes Web search method and determine that the kernel functional parameter g of the best and the step of penalty parameter c include by 5 folding cross-validation methods: allow penalty parameter c and kernel functional parameter g 2^-10To 2¹⁰Between discrete value；For taking fixed kernel functional parameter g and penalty parameter c, as initial data and 5 folding cross validations are utilized to choose the kernel functional parameter g and penalty parameter c making calibration set checking mean square error minimum calibration set；When making the calibration set checking minimum kernel functional parameter g of mean square error and penalty parameter c have many groups, then choose minimum one group of penalty parameter c as optimal parameter；Minimum when choosing penalty parameter c, to there being polykaryon function parameter g, then choose the first group of kernel functional parameter g and penalty parameter c that search as optimal parameter.

9. method as claimed in claim 6, it is characterized in that, the concrete steps integrating the precision checking calibration model with checking include: investigate model performance by prediction related coefficient and predicted root mean square error as index, and use prediction relation analysis error that model is evaluated.

10. the method for claim 1, it is characterised in that in described step 1) in, the concrete steps preparing pedotheque to be measured include: after soil sample being dewatered, levigate, cross 60 mesh sieves.