CN101403689A - Nondestructive detection method for physiological index of plant leaf - Google Patents

Abstract

The invention discloses a nondestructive testing method used for testing the physiological indexes of plant leaves on the basis of invisible-near infrared spectrum, which can carry out the quick and multi-parameter testing on the content of compositions such as chlorophyll, nitrogen, lutein, water and the like simultaneously. The method carries out spectrum collection on calibration samples, subsequently preprocesses the spectrum data, preferably selects the waveband, establishes the calibration model between the spectrum value and the standard value of the content of plant component, and collects the spectrums of the unknown samples; after the spectrum data is pre-processed, the selected waveband data are substituted in the calibration model so as to predict the content of the component to be measured; the technical proposal of the invention adopts full-spectrum information; the measured parameters have strong extensibility and the prediction precision and the model adaptability of the calibration model are improved; the trans-reflective measurement type adopted by the method adopts the spectrum sensitiveness and has stronger adaptability on the leaf type; and the improved wavelet analysis method can simultaneously eliminate the noise of the leaf spectrum data and carries out benchmark line calibration pre-processing on the leaf spectrum and can effectively improve the prediction precision.

Description

A kind of plant leaf blade nondestructive detection method for physiological index

Technical field

The present invention relates to the technical field of nondestructive testing of plant leaf blade physical signs, relate in particular to a kind of method of the plant leaf blade physical signs Non-Destructive Testing based on visible-near-infrared spectrum.

Background technology

The Non-Destructive Testing of plant leaf blade physical signs mainly is meant carries out Non-Destructive Testing to content of material such as chlorophyll, nitrogen, xenthophylls, moisture, it can be used for growing of real-time monitoring plant and health status, analysis moisture and fertility state, by detecting, not only can accurately regulate the supply situation of fertilizer and water, realize accurate fertilizer management, save resource, reduce environmental pollution, and can the scientific guidance cultivation work, so that can ensure cultivated plant healthy growth and good the growth, improve crop yield.

The method of blade Non-Destructive Testing at present has image method and spectroscopic methodology.Image method mainly is some the static or dynamic characteristics of image of plant leaf blade that relies on machine vision method to obtain, and the physiological characteristic of these characteristics of image and plant leaf blade (vegetation index for example, water-intake rate etc.) confidential relation is arranged, by setting up relation between the two, vegetation growth state is detected.This method is mainly used in the measurement aspect of species Classification and Identification and chlorophyll, moisture.This method needs very complicated hardware system, can not accomplish to be convenient for carrying, and most importantly detectable parameter is fewer, and the ratio of precision of model is lower, therefore uses to such an extent that be not very extensive at present.

Spectroscopic methodology obtains the spectral signature of plant by interactions such as the absorption between photon and the material, scatterings, sets up regression relation with the physical signs of plant, thereby detects.Spectrum Non-Destructive Testing commonly used has fluorescence spectrum technology, visible-near-infrared spectrum technology etc.Fluorescent spectrometry is the absorption by the laser excitation material, and present fluorescent spectrometry only is applied to the detection of chlorophyll content, and can't realize for the detection of other materials such as moisture and nitrogen; Visible-near-infrared spectrum application aspect the detection of blade physical signs is very extensive, such as the content of the chlorophyll, moisture, nitrogen and the various nutrients that detect blade.But the visible-near-infrared spectrum lossless detection method that exists all is to adopt a few individual wavelengths of visible and near-infrared band to detect some or the only a few Several Parameters at present, this method principle is simple, be easy to realize, but owing to have only the spectral information of several wavelength, detect when being difficult to realize multiparameter, dirigibility, extendability are relatively poor, in addition when sample is subjected to non-target factor such as measuring condition and sample state and disturbs, this measuring method just can not obtain excellent precision, this adaptation of methods is poor, and antijamming capability is lower.

Summary of the invention

In view of this, fundamental purpose of the present invention is to provide a kind of plant leaf blade nondestructive detection method for physiological index, can solve low, the poor anti jamming capability of model applicability in the existing spectrum lossless detection method, accuracy of detection is low, and technical matters such as detected parameters is less, for achieving the above object, technical scheme of the present invention is achieved in that

A kind of plant leaf blade nondestructive detection method for physiological index comprises the steps:

Steps A, acquisition correction sample set sample, adopt in the reflection mode measurement update sample set sample as seen-continuous spectrum of shortwave near infrared range.

Choose the plant leaf blade sample, extensively collect the representational sample of some, form the calibration samples collection, be used for training and set up calibration model.Sample selection will have certain representativeness, and according to measurement requirement, test substance content will have a very wide distribution, and has certain gradient, and growth cycle is wanted comprehensively.

The present invention adopts general light spectrometer or self-control spectrometer, and sample is carried out continuous spectroscopic measurement, and spectral range is visible-shortwave near infrared (380nm-1100nm).Spectral measurement process to the concentrated sample of calibration samples is divided into three steps, the collection of the collection of dark noise, the collection of reference signal, measuring-signal.Can calculate reflectivity or absorbance according to above image data, as the spectroscopic data of setting up calibration model.The computing formula of reflectivity and absorbance is as follows:

The reflectivity of sample is:

$R=\frac{S_{\mathrm{sample}}-S_{\mathrm{dark}}}{S_{\mathrm{reference}}-S_{\mathrm{dark}}}$

The absorbance of sample is:

$A=-\mathrm{lgR}=-\lg \left(\frac{S_{\mathrm{sample}}-S_{\mathrm{dark}}}{S_{\mathrm{reference}}-S_{\mathrm{dark}}}\right)$

Wherein, R is the reflectivity of sample, and A is the absorbance of sample, S _SampleThe measuring-signal of representative sample, S _DarkRepresent dark-noise signal, S _ReferenceRepresent reference signal.

This spectral measurement mode can be eliminated the influence that the energy of light source drift causes on the one hand, on the other hand, from measuring-signal and reference signal dark current noise is deducted, and can reduce the noise in the spectrum, thereby improves the signal to noise ratio (S/N ratio) of signal.Dark noise is meant in the output valve that does not have detection system under the illumination condition, mainly is made up of the thermonoise of the electronic circuit of the dark current of detector and instrument internal, and these noises have just constituted dark noise altogether.Reference signal is meant for reflectance spectrum that obtains sample or absorption spectrum, to a measurement of incident intensity.The acquisition method of reference signal and measuring-signal have living space usually two kinds on double light path and monochromatic light road.The Space Double light path is exactly that the light that light source sends is become identical two-beam through light-splitting device, makes it pass through reference sample and sample simultaneously, and reference signal and measuring-signal are measured simultaneously; The monochromatic light road is normally measured its reference signal and measuring-signal respectively in short time.

Among the present invention, the mode of sample being carried out spectrum sample is transflective, detector and light source are placed on the same side, place the diffuse reflection on-gauge plate in the sample bottom, the light that make to see through sample can be reflected once more, and part light will carry out scattered reflection and absorption and once more in the sample surface outgoing and detected in sample inside.Detector detects diffusing of sample simultaneously and penetrates the saturating reflected light of sample twice, so carries a large amount of sample chemical constitution absorption information in the detectable signal.

In the preferred embodiment of the invention, the sampling device adopts optical fiber, owing to light transmitting fiber easy operating and use, can realize long range measurements, and overcome the structural complexity that too much mechanical hook-up brings, and therefore can realize the live body Non-Destructive Testing.

Step B, adopt standard method of analysis to measure the chemical content that calibration samples is concentrated the component to be measured of sample, and with it as standard value;

The chemical content of component to be measured all adopts standard method of analysis to obtain, and described standard can be industry, country or international standard.

Step C, concentrate the spectroscopic data of sample to carry out pre-service, remove noise and background interference calibration samples;

Pre-service mainly be remove in the spectrum measuring data of sample because noise of instrument, measuring condition changes and the light scattering effect of leaf tissue, blade in the high frequency noise and the baseline wander that cause such as other chemical constitution interference.

Adopt small echo power spectrumanalysis method to remove high frequency noise simultaneously and carry out baseline correction among the present invention, the specific implementation step is as follows:

C1, select best decomposition scale

Select maximum wavelet to decompose the number of plies as preliminary decomposition scale, after signal decomposition, detail section under each yardstick is carried out power spectrumanalysis, mainly concentrate on the principle of low frequency part according to signal, if the power spectrum of details mainly is to be distributed in low frequency part under a certain yardstick, and the regularity of distribution, this yardstick is exactly final best decomposition scale so.

The selectable maximum decomposition scale of wavelet analysis is [log ₂N], wherein N is a signal length, [] is downward rounding operation.In this method, select the maximum wavelet decomposition number of plies as preliminary decomposition scale.

Then the detail section to reconstruct carries out power spectrumanalysis, instructs determining of final decomposition scale in view of the above.

It is the distribution relation of analytic signal energy with frequency that details under each yardstick is carried out energy spectrum analysis.Make G (Ω)=| F (Ω) | ²Or G (f)=| F (f) | ², wherein, f is the frequency of the circular frequency Ω correspondence of signal, then has:

$W=\frac{1}{2\mathrm{\π}}{\∫}_{-\∞}^{\∞}G\left(\mathrm{\Ω}\right)\mathrm{d\Ω}=\frac{1}{\mathrm{\π}}{\∫}_0^{\∞}G\left(\mathrm{\Ω}\right)\mathrm{d\Ω}=2{\∫}_0^{\∞}G\left(f\right)\mathrm{df}$

As seen, G (Ω) or G (f) have reflected the distribution situation of signal energy on frequency domain, and G (Ω) also claims power spectrum for the energy density spectrum of signal.

Because of noise mainly is present in HFS, and the frequency disunity, so concentration of energy is at high-frequency region in power spectrum, and the irregular details of saying should be regarded noise as and removed.With respect to noise, the energy of signal mainly concentrates on lower frequency range part, and has certain regularity, such detail section be should give reservation, when having begun obviously to show the characteristic of signal on the power spectrum of k layer, then only need carry out the wavelet decomposition of k-1 layer, determine that then k-1 is the best number of plies of decomposing of small echo.

C2, removal high frequency noise

After adopting described best decomposition scale that described spectroscopic data is carried out wavelet decomposition, the high frequency wavelet coefficient that only contains under the noise contribution yardstick is all put 0, and the high frequency wavelet coefficient that contains simultaneously under signal and the noise contribution yardstick is carried out threshold filter, thereby remove high frequency noise.

After spectrum being carried out the wavelet decomposition of k-1 yardstick, decidable yardstick k-2 to the detail section of yardstick 1 be noise, its wavelet coefficient is all put 0.And the HFS of yardstick k-1 comprises signal and noise contribution simultaneously, selects suitable threshold to carry out filtering so tackle under this yardstick the detail section wavelet coefficient, removes noise and keeps useful signal.The method of determining threshold value at present mainly contains four kinds: unified threshold value, SUREShrink threshold value (Stein does not have inclined to one side estimation threshold value), heuristic SURE threshold value and minimax threshold value.Along with the development of instrumental science, institute's employing spectrometer signal to noise ratio (S/N ratio) is generally higher in the experiment at present, and the minimax threshold value determines that with respect to other threshold values method is more conservative, so select this method, its formula is as follows:

${\mathrm{\Δ}}_{\mathrm{Mini}\max i}=\left\{\begin{array}{cc}0& N\≤32\\ {\mathrm{\σ}}_n(0.396+0.1829{\log }_{2}N)& N\≥32\end{array}\right.$

The n-signal length here, σ _nVariance for certain layer of details wavelet coefficient.

C3, baseline correction

The baseline of spectral signal mainly concentrates on low-frequency range, so in low-frequency range, the approximate part that promptly obtains after the wavelet decomposition is carried out baseline correction and replaced full spectrum is proofreaied and correct, to avoid the influence of HFS signal.

Blade is measured, because measuring condition changes and the light scattering effect of leaf tissue, blade in factor such as other chemical constitution interference can cause that recording spectrum produces baseline wander, therefore the approximate part of the spectroscopic data that wavelet decomposition is obtained is carried out the quadratic polynomial match, and match gained curve is removed as baseline.At last the wavelet coefficient under each yardstick is reconstructed, finishes preprocessing process.

Step D, that pretreated spectroscopic data is carried out wave band is preferred, selects best bands at different physical signs;

When adopting whole wave spectrum data to set up calibration model, not only amount of calculation is big, and because the noise that the spectrum of some wave bands comprises is bigger, a little less than the information, perhaps has serious collinearity, and the precision of prediction of calibration model may not necessarily reach optimum value.In order to overcome above-mentioned shortcoming, in the preferred embodiment of the invention, before setting up calibration model, it is preferred at different physical signs the modeling wavelength to be carried out wave band.By the preferred process of wave band, from spectrum, extract the most effective information, make calibration model have best predictive ability, and the reduced data computing.The wave band preferable methods has correlation analysis method, method of analysis of variance, no information variable null method, genetic algorithm etc.

In the preferred embodiment of the invention, it is preferred that the employing relevant function method is carried out wave band.To be the spectroscopic data that will proofread and correct sample in the sample set carry out relatedly with standard value the principle of correlation analysis, obtains the related coefficient at each wavelength place.Set the threshold value of related coefficient in advance, it is then selected that related coefficient surpasses the pairing wavelength of this threshold value.The computing formula of related coefficient is as follows:

$R=\frac{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}(x_i-\stackrel{\‾}{x})(y_i-\stackrel{\‾}{y})}{\sqrt{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{(x_i-\stackrel{\‾}{x})}^2}\sqrt{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{(y_i-\stackrel{\‾}{y})}^2}}$

Wherein, m is the sample number, and x is this wavelength place spectroscopic data value, and x is the mean value of m sample spectra value, and y is the sample standard value, and y is the mean value of m sample standard value, and i is the sample sequence number.

Step e, set up calibration model;

Calibration model is to utilize chemometrics method, regression relation between the standard value of setting up spectral value after the pre-service and in step B, being obtained, if to wave band carry out preferably then be set up the spectral value of preferred bands correspondence after the pre-service and the standard value that in step B, obtained between regression relation.Modeling method commonly used has multiple linear regression (MLR), principal component regression (PCR), partial least squares regression (PLSR), artificial neural network (ANN) model etc.Spectral value comprises reflectivity or absorbance etc.

Step F, measure the spectroscopic data of unknown sample, carry out pre-service after, with corresponding wave band data substitution calibration model, calculate the chemical composition content of the component to be measured of described unknown sample.

In this step, the spectra collection of unknown sample is with step A, and spectroscopic data also will be through the pre-service of step C, then with the calibration model in the pretreated spectroscopic data substitution step e, thereby obtains the chemical composition content of the component to be measured of unknown sample.If pretreated spectroscopic data is preferred through wave band,, thereby obtain the chemical composition content of the component to be measured of unknown sample more accurately then with the calibration model in the spectroscopic data substitution step e of preferred bands correspondence.

Technical scheme of the present invention is to compose data entirely owing to what gather, compares with the discrete type measuring method of a few wavelength, and the precision of prediction of model is improved.Full spectrum data are convenient to adopt effective preprocess method, not only can remove noise, can also reduce measuring condition and change influence as environment temperature, vane thickness, blade surface roughness, non-target factor such as take measurement of an angle, improve spectral signal-noise ratio, thereby improve the precision of prediction and the model adaptability of model.Full spectrum data can realize multi parameter simultaneous measuring, and extensibility is extremely strong, each different biochemical parameter is set up corresponding calibration model get final product, and only need sample is carried out spectral measurement one time.

Adopt the visible light and the shortwave near infrared spectrum of plant sample among the present invention, the a plurality of strong absorption peak that comprises chemical substances such as chlorophyll, nitrogen, xenthophylls in this scope, and absorption peak is not overlapping, spectrum sensitivity and selectivity height, so model prediction precision height.

Adopt saturating reflection measurement mode among the present invention,, therefore carry a large amount of blade biochemical components and absorb information, increased optical length, thereby strengthened the detection sensitivity of spectrum because part light penetrates blade twice.Because collecting simultaneously, detector carried diffusing and contingent reflected light of blade chemical substance optical information, be suitable for measuring the blade sample under all thickness state, avoided conventional transmission and diffuse reflection sample mode requirement for restriction to vane thickness, wider to the applicability of vane type.

Propose a kind of improved wavelet analysis method among the present invention the blade spectroscopic data is carried out the pre-service of noise remove and baseline correction simultaneously, can effectively improve precision of prediction.This method instructs determining of best decomposition scale with high frequency wavelet capacity factor analysis of spectrum result under each yardstick, more traditional pass through to attempt or experience determines that the method for decomposition scale has stronger general applicability, also make the definite more objective, accurate of decomposition scale simultaneously.High frequency wavelet coefficients by using relatively conservative minimax threshold mode carried out filtering, be more suitable for the present situation that spectral instrument nowadays generally has degree of precision thereafter.Simultaneously, to the baseline wander that produces in the low frequency wavelet coefficients by using quadratic polynomial match spectral measurement, and removed, be different from traditional direct method low frequency signal zero setting, be more suitable for the baseline that biosome produces because of scattering, and reduced the loss of useful information to a greater extent.

Description of drawings

Fig. 1 is the needed hardware system figure of the method for the invention;

Fig. 2 is the process flow diagram of lossless detection method of the present invention;

Fig. 3 is the spectrogram of calibration samples collection original spectrum data;

Fig. 4 is that wavelet pretreatment is decomposed number of plies synoptic diagram in the method for the invention;

Fig. 5 is for to carry out pretreated spectrogram to calibration samples collection original spectrum data;

Fig. 6 is for adopting the predict the outcome figure of calibration model to the chlorophyll content of unknown sample.

Embodiment

For making the purpose, technical solutions and advantages of the present invention clearer, by the following examples and with reference to accompanying drawing, the present invention is described in more detail.

One embodiment of the present invention measures chlorophyll content in leaf blades with saturating reflection sample mode and moisture is that example is done detailed introduction to the present invention.

Be illustrated in figure 1 as the needed hardware system of this method of realization, this system is by halogen tungsten lamp light source 1, y-type optical fiber 2, and structure of fiber_optic 3, standard reflecting plate 4, spectral range is the portable spectrometer 5 of visible-shortwave near infrared range, computing machine 6 is formed.Specifically be connected to: y-type optical fiber 2 is fixed on above the support 3, assurance optical length unanimity, the light that halogen tungsten lamp light source 1 sends is radiated at blade surface through the incident optical bundle of y-type optical fiber 2, under the effect of standard reflecting plate 4, reflect and penetrate the blade outgoing behind the light penetration blade, the outgoing fibre bundle of y-type optical fiber 2 is collected in these light together, beam splitting system through portable spectrometer 5, project the photoelectric detector of portable spectrometer 5, obtain the spectral signal of blade, portable spectrometer 5 is exported to computing machine 6 with these signals, spectral measurement of the present invention is simple to operate, only need to use the special software of spectrometer outfit, just can obtain the spectroscopic data of blade.

The used portable spectrometer of present embodiment, effective wavelength range are 500nm-1100nm, and light source is the 5V halogen tungsten lamp, and the diameter of diffuse reflection optical fiber is 400um, ceramic standard reflecting plate, structure of fiber_optic, and computing machine.The parameter of spectrometer is provided with as follows during measurement: average time is 10, and be 20ms integral time, and smoothly counting is 0.Adopt reflection way, sample is placed on above the standard reflecting plate, collect and to penetrate the transmitted light that blade returns and diffuse.Optical fiber becomes miter angle to place with blade surface, avoid blade direct reflection effect.The chlorophyll content in leaf blades value can be recorded by the chlorophyll instrument of market public offering.

With reference to Fig. 2, this embodiment is as follows to the implementation procedure of plant leaf blade Non-Destructive Testing:

Steps A, adopt sample in the saturating reflection mode measurement update sample set as seen-continuous spectrum of shortwave near infrared range.

At first need to select the calibration samples collection, the sample that calibration samples is concentrated should have certain representativeness, and present embodiment is chosen 68 blades of same kind of plant kind, the thickness that as far as possible guarantees each blade is identical, the green intensity difference, from withered and yellow to dark green, and moisture is evenly distributed, scope is wide.Above-mentioned 68 samples are divided into calibration samples collection and forecast set, and calibration samples collection number of samples is 56, and the forecast set number of samples is 12.Calibration samples collection sample is used to set up calibration model, and the forecast set sample is used to estimate the measuring accuracy of chlorophyll of the present invention and moisture detection method.

Earlier blade surface is carried out simple dust cleaning, select the zone line of blade, and avoid vein, as the measuring point.Next sample set is carried out spectral measurement, open portable spectrometer 5, gather dark current signals, open halogen tungsten lamp light source 1, y-type optical fiber 2 is fixed on above the support 3 with 45 degree, support 3 is placed on gathers reference signal above the standard reflecting plate 4, then blade is placed on above the standard reflecting plate, the invariant position that keeps y-type optical fiber 2 and support 3, the measuring-signal of measurement sample.Calculate the absorbance of sample according to following formula, spectrogram as shown in Figure 3.

$A=-\lg \left(\frac{S_{\mathrm{sample}}-S_{\mathrm{dark}}}{S_{\mathrm{reference}}-S_{\mathrm{dark}}}\right)$

Wherein, the absorbance of A representative sample, S _SampleThe representative sample signal, S _DarkRepresent dark current signals, S _ReferenceRepresent reference signal.

Step B, detection calibration samples are concentrated the chemical content of the component to be measured of sample.

Chlorophyll content is represented (unit is mg/g) with the quality of contained chlorophyll content in every gram cured leaf sheet, and its national standard measuring method is to adopt spectrophotometric method.This method need utilize organism extract extract and separate to go out chlorophyll in the blade, measures absorbance by colorimetric method on double beam spectrophotometer, calculates chlorophyll content with the Arnon formula then.Computing formula is:

X ₁＝7.12A ₆₆₀+16.8A ₆₄₂

${\mathrm{<figure-callout\; id="2"\; label="X"\; filenames="A20081022693000161.png,A20081022693000182.png"\; state="\{\{state\}\}">X</figure-callout>}}_2=\frac{{\mathrm{<figure-callout\; id="1"\; label="X"\; filenames="A20081022693000161.png"\; state="\{\{state\}\}">X</figure-callout>}}_1\&times;V\&times;F}{m\&times;1000}$

Wherein, X ₁Be total chlorophyllous content in the sample liquid, unit is every liter (mg/L) of milligram; A be the respective wavelength place (660,642nm) record absorbance; X ₂Be total chlorophyllous massfraction in the sample, the every gram of unit milligram (mg/g); V is the cumulative volume of extract, and unit is a milliliter (mL); F sample extension rate; M is a sample mass, and unit is a gram (g).

Oven drying method is adopted in the standard method of measurement of moisture, and blade is placed in 103 degree ± 2 degree drying boxes, heats 4 hours, add a cover taking-up, in exsiccator, be cooled to room temperature, weighing, place drying box heating 1 hour again, add a cover taking-up, and cool off weighing in the exsiccator, repeat to heat 1 hour operation, be no more than 0.005g until double weighing difference, be constant weight, be as the criterion with minimum.

Moisture is represented with massfraction, is calculated as follows.

Wherein, M ₁Be the quality before sample and the baking of aluminium matter baking ware, the g of unit; M ₂Be the quality after sample and the baking of aluminium matter baking ware, the g of unit; M ₀Be the quality of sample, the g of unit.

In order to reduce the complexity of experiment, the widely used in the industry chlorophyll meter of available rows replaces the analytical chemistry method of GB to measure the chlorophyll content of blade, and this does not influence the correctness of method disclosed by the invention.What we adopted in this embodiment is SPAD chlorophyll instrument.In selected blade spectral measurement position, use this chlorophyll instrument to measure the chlorophyll content value (SPAD of unit) of 3 these positions, calculating mean value is as the chlorophyll content in leaf blades standard value.The chlorophyll content distribution range of 68 samples that this method is measured is: 10.8SPAD-53.5SPAD.

And above-mentioned standard method is just adopted in the measurement of moisture, when blade is dried to mass conservation in the vacuum air dry oven, calculates moisture.The moisture distribution range of 68 samples that this method is measured is: 62.5%-90.5%.

Step C, adopt small echo power spectrumanalysis method to carry out pre-service, remove noise and background interference the spectroscopic data that obtains in the steps A;

At first adopt the db7 wavelet basis to carry out wavelet decomposition to original spectrum, because wavelength has more than 2700, so small echo is maximum, and to decompose the number of plies be 11, and it is defined as preliminary decomposition scale, calculates its power spectrum.Only provided ground floor among Fig. 4 to the signal of the energy spectrum of layer 6 details as following analysis.Owing to there is not its frequency of signal of noise to concentrate in the 20HZ basically, and its frequency spectrum of signal that is mixed with noise can broaden, as shown in Figure 4, the power spectrum that can find the 5th layer of details has concentrated within the 20HZ basically, therefore according to the method for wavelet decomposition disclosed by the invention, best wavelet decomposition yardstick is defined as 4 layers.Then, original signal is carried out 4 layers of wavelet decomposition, the wavelet coefficient of the 4th layer details is carried out threshold filter, threshold value is determined according to following formula

${\mathrm{\&Delta;}}_{\mathrm{Mini}\max i}=\left\{\begin{array}{cc}0& N\&le;32\\ {\mathrm{\&sigma;}}_n(0.396+0.1829{\log }_{2}N)& N\&GreaterEqual;32\end{array}\right.$

To be made as 0 less than the wavelet coefficient of this threshold value, greater than the former wavelet coefficient of the reservation of this threshold value, and the wavelet coefficient on other layer all is made as 0, obtains removing the purpose of noise.And the approximate part of wavelet decomposition is carried out conic fitting, and as baseline, and it is deducted, realize to the denoising of original spectrum and remove baseline that the result is as shown in Figure 5.

Step D, pretreated spectroscopic data is carried out band selection, select its best modeled wave band at different physical signs.

Here the parameter of Ce Lianging has two, and one is chlorophyll content, and one is moisture, respectively these two parameters is carried out correlation analysis and carries out band selection.To be the spectroscopic data value of will proofread and correct the sample set sample carry out relatedly with standard value the principle of correlation analysis, obtains the related coefficient at each wavelength place.The threshold value of setting related coefficient is 0.7, and it is then selected that related coefficient surpasses the pairing wavelength of this threshold value.The computing formula of related coefficient is as follows:

$R=\frac{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}(x_i-\stackrel{\&OverBar;}{x})(y_i-\stackrel{\&OverBar;}{y})}{\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_i-\stackrel{\&OverBar;}{x})}^2}\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{(y_i-\stackrel{\&OverBar;}{y})}^2}}$

Wherein m is the sample number, and y is the sample standard value, and x is this wavelength place spectroscopic data value.Through correlation analysis, the wave band that discovery and chlorophyll content correlativity are big is 660-740nm, and the wave band big with the moisture correlativity is 950-980nm.

Step e, in the wave band of determining, set up calibration model, set up the regression relation between spectral value and the standard value.

56 samples that above-mentioned calibration samples is concentrated are set up the PLS regression model to moisture and chlorophyll concentration value respectively, obtain two calibration models.

Step F, measure the spectroscopic data of unknown sample, spectroscopic data is carried out pre-service after, calculate the chemical composition content of the component to be measured of unknown sample in the spectrum numerical value substitution calibration model with corresponding wave band place.

The final purpose of spectral measurement method is the physical signs that the spectral information of input unknown sample can be predicted this sample, so just realized harmless, detect in real time, therefore the final step of this method is carried out spectral measurement and pre-service according to foregoing method to unknown sample exactly, will obtain the predicted value of chlorophyll content in leaf blades and moisture in the calibration model among the spectroscopic data substitution E of selected wave band.

Provided predicting the outcome of chlorophyll content of plant and moisture below.

Distribution of chlorophyll concentration value and the statistical parameter of proofreading and correct sample set and forecast set sample in the present embodiment are as shown in the table: unit: SPAD

Table one sample chlorophyll concentration value distribution situation

	Number (individual)	Maximal value (SPAD)	Minimum value (SPAD)	Mean value (SPAD)	Variance (SPAD)
						Total sample	68	53.5	10.8	34.7	10.46
The calibration samples collection	56	53.5	10.8	34.5	11.11
						Forecast set	12	43.5	22.2	35.4	6.96

56 samples that above-mentioned calibration samples is concentrated are set up the PLSR calibration model of chlorophyll content, and the coefficient R between chlorophyll predicted value and the standard value is 0.99, and predicted root mean square error (RMSEP) is 1.4SPAD.In the spectrum substitution calibration model with the pretreated corresponding wave band of 12 unknown sample in the forecast set, the prediction case of chlorophyll content as shown in Figure 6, the predicted value of sample and the related coefficient between the standard value are 0.96 in the forecast set, and prediction standard deviation (RMSEP) is 1.8SPAD.

56 samples that calibration samples is concentrated are set up the PLSR calibration model of moisture, and the coefficient R between moisture predicted value and the standard value is 0.98, and predicted root mean square error (RMSEP) is 2.1%.In the spectrum substitution calibration model with the pretreated corresponding wave band of 12 unknown sample in the forecast set, the predicted value of the moisture of sample and the related coefficient between the standard value are 0.96 in the forecast set, and prediction standard deviation (RMSEP) is 3.0%.

Above-mentioned embodiment is used for the present invention that explains, rather than limits the invention, and 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 (8)

1, a kind of plant leaf blade nondestructive detection method for physiological index is characterized in that, may further comprise the steps:

Gather representational sample as the calibration samples collection, adopt in the reflection mode measurement update sample set sample as seen-continuous spectrum of shortwave near infrared range;

Adopt standard method of analysis to measure the component concentration to be measured that calibration samples is concentrated sample, and with it as standard value;

Adopt small echo power spectrumanalysis method the spectroscopic data of calibration samples collection to be carried out the pre-service of noise remove and baseline correction;

Set up calibration model;

Utilize described calibration model to measure the chemical content of the component to be measured of unknown sample.

2, method according to claim 1 is characterized in that, carries out simultaneously a plurality of biochemical parameters being carried out modeling under the situation of a spectra collection at the sample that calibration samples is concentrated, and each different biochemical parameter is set up different calibration models.

3, method according to claim 1 is characterized in that, comprises in the step of described small echo power spectrumanalysis method:

Determine preliminary decomposition scale, the signal detail under each decomposition scale of tentatively determining is partly carried out power spectrumanalysis, thereby determine best decomposition scale;

After adopting described best decomposition scale that described spectroscopic data is carried out wavelet decomposition, the high frequency wavelet coefficient that only contains under the noise contribution yardstick is all put 0, and the high frequency wavelet coefficient that contains simultaneously under signal and the noise contribution yardstick is carried out threshold filter, remove high frequency noise;

With quadratic polynomial baseline being carried out least square fitting in the low frequency estimating part of maximum decomposition scale is also removed.

4, method according to claim 1 is characterized in that, described calibration model is to utilize chemometrics method, and the regression relation of setting up between described standard value and the described pretreated spectral value forms.

5, method according to claim 1 is characterized in that, the step of chemical content of utilizing described calibration model to measure the component to be measured of unknown sample comprises:

Adopt the reflection mode measure unknown sample as seen-continuous spectrum of shortwave near infrared range;

Adopt small echo power spectrumanalysis method the spectroscopic data of unknown sample to be carried out the pre-service of noise remove and baseline correction;

Thereby the described calibration model of pretreated spectroscopic data substitution of described unknown sample is obtained the chemical content of the component to be measured of described unknown sample.

6, method according to claim 1, it is characterized in that, after the spectroscopic data to described calibration samples collection carries out described pre-service, also comprise at different physical signs and carry out the preferred step of wave band, correspondingly, described calibration model then is the regression relation of setting up between the spectral value of preferred bands correspondence after described standard value and the described pre-service.

7, method according to claim 1, it is characterized in that, the spectral measurement process of the sample that described calibration samples is concentrated comprises collection to dark noise, to the collection of reference signal and to the collection of measuring-signal, dark noise and reference signal are used for eliminating influence that the energy of light source drift causes and the noise that reduces spectrum.

8, method according to claim 1 is characterized in that, the sampling device adopts optical fiber.

Images