CN102590135B - Herbicide distinguishing method based on least-square support vector machine - Google Patents

Abstract

The invention discloses a herbicide identification method based on a least square support vector machine. It mainly includes the following steps: first, use the terahertz time-domain spectroscopy system to detect the training sample set to obtain the terahertz time-domain spectrum; then extract the model through Fourier transform and terahertz optical parameters, calculate the absorption coefficient spectrum, and use the partial minimum The effective eigenvector is extracted by the square method, and the herbicide identification model database is established based on the effective eigenvector; then the terahertz time-domain spectroscopy system is used to detect and predict the sample set, and the terahertz time-domain spectrum is obtained, and then the Fourier transform and terahertz optical The parameter extraction model is used to calculate the absorption coefficient spectrum, and the partial least square method is used to extract the effective feature vector. Finally, the established herbicide identification model database is called, and the least square support vector machine is used to determine the category of the predicted sample set. The invention can realize rapid and accurate identification of herbicides, and has broad application prospects in the fields of drug analysis and identification.

Description

A kind of herbicide discrimination method based on least square method supporting vector machine

Technical field

The present invention relates to Terahertz Technology Non-Destructive Testing application, relate in particular to a kind of herbicide discrimination method based on least square method supporting vector machine.

Background technology

At present, the detection method of classes of herbicides agricultural chemicals is broadly divided into spectroscopic methodology, enzyme suppresses method and chromatography etc.The complex pretreatment of chromatography sample, requires highly to instrumentation personnel, cannot detect online; It is not high, affected by environment larger that enzyme suppresses method sensitivity, easily occurs undetected, flase drop.The present invention has overcome the shortcoming of classic method, utilizes terahertz time-domain spectroscopy apparatus system, and a kind of easy, workable, herbicide method that sense cycle is short is provided.

In recent years, Terahertz Technology development rapidly, has broad application prospects in fields such as communication, detection, spectrum, imagings.Terahertz (THz) ripple refers to the electromagnetic wave (1THz=10 of frequency within the scope of 0.1～10THz ¹²hz), it in electromagnetic wave spectrum between microwave and far infrared radiation.With respect to the electromagnetic wave of other kind, THz wave has its unique characteristic: 1., with respect to X ray, the energy of THz wave is low, can not damage human body, so its security is higher; 2. tera-hertz spectra can provide the vibration information of acting force, macro-radical etc. between material molecule, can be for the discriminating of material; 3. tera-hertz spectra detects the structure that can not destroy detected material itself, belongs to Non-Destructive Testing category.

In the time calculating the Terahertz absorption coefficient spectrum of herbicide, adopt the terahertz optics parameter extraction model being proposed by Dorney and Duvillaret, refer to list of references (Terahertz (THz) spectral investigation of Imidacloprid, spectroscopy and spectral analysis, Yan Zhigang, Hou Dibo, Cao Binghua, Zhang Guangxin, Zhou Zekui).

Conventional classification discrimination method has at present: Bayesian Method, radial neural network, support vector machine and least square method supporting vector machine.Radial neural network has used sample assumed condition, turns to principle with least risk, but this method is often not being met in actual applications.Support vector machine adopts inequality constrain condition, dimension equals the number of training sample, thereby make the number of matrix element be wherein training sample number square, but in the time that data scale acquires a certain degree, algorithm of support vector machine often cannot be dealt with problems.And least square method supporting vector machine adopts equality constraint, in conjunction with chunking algorithm etc., make up the some shortcomings of support vector machine, reduce to a certain extent solving difficulty, improve the speed of solving.

Summary of the invention

The object of the invention is to have overcome conventional matter detection method, such as the deficiency of liquid phase chromatography, enzyme inhibition method, provide a kind of herbicide discrimination method based on least square method supporting vector machine.

The step of the herbicide discrimination method based on least square method supporting vector machine is as follows:

1) select maleic acid hydrazide, 2 kinds of herbicides of 2-first-4-chloropropionic acid to prepare training sample sets and Prediction, wherein train in sample sets and contain 7 maleic acid hydrazide samples and 7 2-first-4-chloropropionic acid samples, in Prediction, contain 7 maleic acid hydrazide samples and 7 2-first-4-chloropropionic acid samples;

2) utilize terahertz time-domain spectroscopy system to detect training sample sets, obtain terahertz time-domain spectroscopy, and through Fourier transform and terahertz optics parameter extraction model, calculate the absorption coefficient spectrum of training sample sets, utilize partial least square method to extract validity feature vector to absorption coefficient spectrum, and to train the validity feature vector of sample sets as basis, set up herbicide and differentiate model database X;

3) set output vector Y, differentiate that taking herbicide model database X and output vector Y, as basis, utilize least square method supporting vector machine to set up herbicide and differentiate model;

4) utilize terahertz time-domain spectroscopy system to detect Prediction, obtain terahertz time-domain spectroscopy, and through Fourier transform and terahertz optics parameter extraction model, calculate the absorption coefficient spectrum of Prediction, utilize partial least square method to extract validity feature vector, using the validity feature vector of Prediction as forecast set Z;

5) finally by forecast set Z input, the herbicide based on least square method supporting vector machine is differentiated model, for verifying the discriminating accuracy of herbicide discriminating model.

Described training sample, prediction sample preparation method be: select polyethylene powders as experiment compressing tablet material respectively with maleic acid hydrazide, the former medicine of two kinds of herbicides of 2-first-4-chloropropionic acid mixes, by maleic acid hydrazide, two kinds of former medicines of herbicide of 2 first-4-chloropropionic acid and the polyethylene powders temperature with 80 DEG C in vacuum drying chamber is dried two hours, and mix with 1: 1 part by weight respectively, putting into clean agate mortar grinds evenly, finally by the maleic acid hydrazide of 160mg, two kinds of former medicines of herbicide of 2-first-4-chloropropionic acid and poly potpourri are pressed into the thin rounded flakes that diameter is 13mm under 20MPa pressure, as maleic acid hydrazide, 2-first-4-chloropropionic acid sample.

Described utilizes terahertz time-domain spectroscopy system to training sample sets, Prediction carries out detection method: to training sample sets, before Prediction detects, be filled with nitrogen toward terahertz time-domain spectroscopy system, make relative humidity in system be less than 4.0%, and indoor relative ambient humidity is controlled at below 50%, when terahertz time-domain spectroscopy system works, stepper motor stroke range is made as 0-2cm, sampling step length is made as 0.01cm, to train sample sets, Prediction is put into terahertz time-domain spectroscopy system and is detected, each sample detection three times, be averaged, eliminate stochastic error.

Described step 2) be: utilize partial least square method to extract validity feature vector to the absorption coefficient spectrum of training sample sets, set up herbicide and differentiate model database X and output vector Y, X, Y expression formula are as follows:

$X=\left[\begin{array}{cccccc}{x}_{11}& x_{12}& \·\·\·& x_{1k}& \·\·\·& x_{1n}\\ x_{21}& x_{22}& \·\·\·& x_{2k}& \·\·\·& x_{2n}\\ \·& \·& & \·& & \·\\ \·& \·& & \·& & \·\\ \·& \·& & \·& & \·\\ x_{m1}& x_{m2}& \·\·\·& x_{m3}& \·\·\·& x_{\mathrm{mn}}\end{array}\right],$

Y＝[y ₁?y ₂?…?y _k?…?y _n] ^T

In formula, m represents to train sample size in sample sets, and n represents to train the validity feature vector dimension of sample sets, wherein, and m=14, n=2.

Described step 3) be:

According to the discriminating model database X of input, by kernel function, herbicide is differentiated to model database X is mapped to higher dimensional space S, in S space, construct optimal classification face, the kernel function of employing is radial basis kernel function, formula is as follows:

$K(x_p,x_q)=\exp (-\frac{{\left|\right|x_p-x_q\left|\right|}^2}{{2\mathrm{\δ}}^2})$

In above formula, δ is kernel functional parameter, x _p, x _qthe validity feature vector of training sample sets, p, q ∈ [1, n].Optimal classification problem is converted into the minimum value of asking class interval φ (w, ε):

$\mathrm{\φ}(w,\mathrm{\ϵ})=\frac{1}{2}{\left|\right|w\left|\right|}^2+\frac{1}{2}\mathrm{\γ}\underset{1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ϵ}}_k^2$

Constraint condition is:

y _k[(ψ(x _k)·w+b)]≥1-ε _k

Finally obtain herbicide by method of Lagrange multipliers and differentiate model:

Y in formula _kthe element in output vector Y, y _k=+1, and-1}, k ∈ [1, n], xk is the validity feature vector of training sample sets, and ε is error, and γ is error penalty factor, Ψ (x _k) be validity feature vector x _kat the mapping of feature space S, α _kbe Lagrange multiplier, b is that herbicide is differentiated the intercept of model optimal classification face at coordinate plane.

Terahertz time-domain spectroscopy system involved in the present invention, works at normal temperatures and pressures, and this system has very high signal to noise ratio (S/N ratio), can carry out online Non-Destructive Testing to herbicide sample, there will not be flase drop and undetected phenomenon.In addition, utilize least square method supporting vector machine succinct to two kinds of herbicide discrimination processes, fast, accurately, aspect the discriminating of medicine and analysis, having higher using value.

Brief description of the drawings

Fig. 1 is the identification result figure that two kinds of herbicides are differentiated model.

Embodiment

The step of the herbicide discrimination method based on least square method supporting vector machine is as follows:

1) select maleic acid hydrazide, 2 kinds of herbicides of 2-first-4-chloropropionic acid to prepare training sample sets and Prediction, wherein train in sample sets and contain 7 maleic acid hydrazide samples and 7 2-first-4-chloropropionic acid samples, in Prediction, contain 7 maleic acid hydrazide samples and 7 2-first-4-chloropropionic acid samples;

2) utilize terahertz time-domain spectroscopy system to detect training sample sets, obtain terahertz time-domain spectroscopy, and through Fourier transform and terahertz optics parameter extraction model, calculate the absorption coefficient spectrum of training sample sets, utilize partial least square method to extract validity feature vector to absorption coefficient spectrum, and to train the validity feature vector of sample sets as basis, set up herbicide and differentiate model database X;

3) set output vector Y, differentiate that taking herbicide model database X and output vector Y, as basis, utilize least square method supporting vector machine to set up herbicide and differentiate model;

4) utilize terahertz time-domain spectroscopy system to detect Prediction, obtain terahertz time-domain spectroscopy, and through Fourier transform and terahertz optics parameter extraction model, calculate the absorption coefficient spectrum of Prediction, utilize partial least square method to extract validity feature vector, using the validity feature vector of Prediction as forecast set Z;

5) finally by forecast set Z input, the herbicide based on least square method supporting vector machine is differentiated model, and for verifying the discriminating accuracy of herbicide discriminating model, identification result is good, and accuracy has reached 100%, sees Fig. 1.

Described training sample, prediction sample preparation method be: select polyethylene powders as experiment compressing tablet material respectively with maleic acid hydrazide, the former medicine of two kinds of herbicides of 2-first-4-chloropropionic acid mixes, by maleic acid hydrazide, two kinds of former medicines of herbicide of 2 first-4-chloropropionic acid and the polyethylene powders temperature with 80 DEG C in vacuum drying chamber is dried two hours, and mix with 1: 1 part by weight respectively, putting into clean agate mortar grinds evenly, finally by the maleic acid hydrazide of 160mg, two kinds of former medicines of herbicide of 2-first-4-chloropropionic acid and poly potpourri are pressed into the thin rounded flakes that diameter is 13mm under 20MPa pressure, as maleic acid hydrazide, 2-first-4-chloropropionic acid sample.

Described utilizes terahertz time-domain spectroscopy system to training sample sets, Prediction carries out detection method: to training sample sets, before Prediction detects, be filled with nitrogen toward terahertz time-domain spectroscopy system, make relative humidity in system be less than 4.0%, and indoor relative ambient humidity is controlled at below 50%, when terahertz time-domain spectroscopy system works, stepper motor stroke range is made as 0-2cm, sampling step length is made as 0.01cm, to train sample sets, Prediction is put into terahertz time-domain spectroscopy system and is detected, each sample detection three times, be averaged, eliminate stochastic error.

Described step 2) be: utilize partial least square method to extract validity feature vector to the absorption coefficient spectrum of training sample sets, set up herbicide and differentiate model database X and output vector Y, X, Y expression formula are as follows:

$X=\left[\begin{array}{cccccc}{x}_{11}& x_{12}& \·\·\·& x_{1k}& \·\·\·& x_{1n}\\ x_{21}& x_{22}& \·\·\·& x_{2k}& \·\·\·& x_{2n}\\ \·& \·& & \·& & \·\\ \·& \·& & \·& & \·\\ \·& \·& & \·& & \·\\ x_{m1}& x_{m2}& \·\·\·& x_{m3}& \·\·\·& x_{\mathrm{mn}}\end{array}\right],$

Y＝[y ₁?y ₂?…?y _k?…?y _n] ^T

In formula, m represents to train sample size in sample sets, and n represents to train the validity feature vector dimension of sample sets, wherein, and m=14, n=2.

Described step 3) be:

According to the discriminating model database X of input, by kernel function, herbicide is differentiated to model database X is mapped to higher dimensional space S, in S space, construct optimal classification face, the kernel function of employing is radial basis kernel function, formula is as follows:

$K(x_p,x_q)=\exp (-\frac{{\left|\right|x_p-x_q\left|\right|}^2}{{2\mathrm{\δ}}^2})$

In above formula, δ is kernel functional parameter, x _p, x _qthe validity feature vector of training sample sets, p, q ∈ [1, n].Optimal classification problem is converted into the minimum value of asking class interval φ (w, ε):

$\mathrm{\φ}(w,\mathrm{\ϵ})=\frac{1}{2}{\left|\right|w\left|\right|}^2+\frac{1}{2}\mathrm{\γ}\underset{1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ϵ}}_k^2$

Constraint condition is:

y _k[(ψ(x _k)·w+b)]≥1-ε _k

Finally obtain herbicide by method of Lagrange multipliers and differentiate model:

Y in formula _kthe element in output vector Y, y _k=+1 ,-1}, k ∈ [1, n], x _kbe the validity feature vector of training sample sets, ε is error, and γ is error penalty factor, Ψ (x _k) be validity feature vector x _kat the mapping of feature space S, α _kbe Lagrange multiplier, b is that herbicide is differentiated the intercept of model optimal classification face at coordinate plane.

Claims (3)

1. a herbicide identification method based on least squares support vector machine, is characterized in that its steps are as follows: 1) Two herbicides, maleic hydrazide and 2-methyl-4-chloropropionic acid, were selected to prepare the training sample set and prediction sample set, in which the training sample set contained 7 maleic hydrazide samples and 7 2-methyl-4-chloropropionic acid samples. For 4-chloropropionic acid samples, it is predicted that the sample set contains 7 maleic hydrazide samples and 7 2-methyl-4-chloropropionic acid samples; 2) Use the terahertz time-domain spectroscopy system to detect the training sample set, obtain the terahertz time-domain spectrum, and calculate the absorption coefficient spectrum of the training sample set through Fourier transform and terahertz optical parameter extraction model. The square method extracts the effective eigenvectors from the absorption coefficient spectrum, and based on the effective eigenvectors of the training sample set, the herbicide identification model database X is established; the output vector Y is set, based on the herbicide identification model database X and the output vector Y , using the least squares support vector machine to establish a herbicide identification model; 3) Use the terahertz time-domain spectroscopy system to detect the predicted sample set, obtain the terahertz time-domain spectrum, and calculate the absorption coefficient spectrum of the predicted sample set through Fourier transform and terahertz optical parameter extraction model. The effective eigenvector is extracted by the square method, and the effective eigenvector of the predicted sample set is used as the predicted set Z; 4) Finally, the prediction set Z is input into the herbicide identification model based on the least squares support vector machine, which is used to verify the identification accuracy of the herbicide identification model; The described step 2) is: Use the partial least squares method to extract effective feature vectors from the absorption coefficient spectrum of the training sample set, and establish the herbicide identification model database X and output vector Y. The expressions of X and Y are as follows: $\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\left[\begin{array}{cccccc}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 11</span>}}& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 12</span>}}& <span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&CenterDot;</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}\mathrm{<span\; class="notranslate">\; k</span>}}& <span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}\mathrm{<span\; class="notranslate">\; no</span>}}\\ {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; one</span>}}& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; two</span>}}& <span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; k</span>}}& <span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&Center\; Dot;</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; no</span>}}\\ <span\; class="notranslate">\; \&CenterDot;</span>& <span\; class="notranslate">\; \&CenterDot;</span>& & <span\; class="notranslate">\; \&CenterDot;</span>& & <span\; class="notranslate">\; \&CenterDot;</span>\\ <span\; class="notranslate">\; \&CenterDot;</span>& <span\; class="notranslate">\; \&CenterDot;</span>& & <span\; class="notranslate">\; \&CenterDot;</span>& & <span\; class="notranslate">\; \&CenterDot;</span>\\ <span\; class="notranslate">\; \&CenterDot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& & <span\; class="notranslate">\; \&CenterDot;</span>& & <span\; class="notranslate">\; \&CenterDot;</span>\\ {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; 2</span>}}& <span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; 3</span>}}& <span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; mn</span>}}\end{array}\right]<span\; class="notranslate">\; ,</span>$ Y＝[y ₁ y ₂ ... y _k ... y _n ] ^T In the formula, m represents the number of samples in the training sample set, and n represents the effective feature vector dimension of the training sample set, where m=14, n=2; According to the input identification model database X, the herbicide identification model database X is mapped to the high-dimensional space S through the kernel function, and the optimal classification surface is constructed in the S space. The kernel function used is the radial basis kernel function, and the formula is as follows: $\mathrm{<span\; class="notranslate">\; K</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\frac{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; q</span>}}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; \&delta;</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span>$ In the above formula, δ is the kernel function parameter, x _p and x _q are the effective feature vectors of the training sample set, p,q∈[1,n], The optimal classification problem is transformed into finding the minimum value of the classification interval φ(w,ε): $\mathrm{<span\; class="notranslate">\; \&phi;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; w</span>}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; \&gamma;</span>}\underset{\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}$ The constraints are: y _k [(ψ(x _k ).w+b)]≥1-ε _k Finally, the herbicide identification model is obtained by Lagrangian multiplier method:

where y _k is the element in the output vector Y, y _k ={+1,-1}, k∈[1,n], x _k is the effective feature vector of the training sample set, ε is the error, γ is the error penalty factor, Ψ(x _k ) is the mapping of the effective feature vector x _k in the feature space S, α _k is the Lagrangian multiplier, b is the intercept of the optimal classification surface of the herbicide identification model on the coordinate plane, and n represents the training Valid eigenvector dimensions for the sample set. 2. a kind of herbicide identification method based on least squares support vector machine according to claim 1, it is characterized in that the preparation method of described training sample, prediction sample is: select polyethylene powder as the material of experimental tabletting Mixed with maleic hydrazide and 2-methyl-4-chloropropionic acid two herbicides respectively, maleic hydrazide, 2-methyl-4-chloropropionic acid two herbicides and polyethylene powder Dry in a vacuum oven at a temperature of 80°C for two hours, mix them at a weight ratio of 1:1, put them into a clean agate mortar and grind them evenly, and finally add 160 mg of maleic hydrazide, 2-methyl-4- The mixture of the two herbicides of chloropropionic acid and polyethylene was pressed into a circular sheet with a diameter of 13 mm under a pressure of 20 MPa, which was used as a sample of maleic hydrazide and 2-methyl-4-chloropropionic acid. 3. a kind of herbicide discrimination method based on least squares support vector machine according to claim 1, is characterized in that: described utilization terahertz time-domain spectroscopy system carries out detection method to training sample set, prediction sample set as : Before testing the training sample set and predicted sample set, fill the terahertz time-domain spectroscopy system with nitrogen to make the relative humidity in the system less than 4.0%, and the indoor relative humidity is controlled below 50%, so that the terahertz time-domain spectroscopy system works , the travel range of the stepping motor is set to 0-2cm, and the sampling step is set to 0.01cm. The training sample set and the predicted sample set are put into the terahertz time-domain spectroscopy system for detection. Each sample is tested three times and averaged. Eliminate random errors.