CN108169165A - Maltose mixture quantitative analysis method based on tera-hertz spectra and image information fusion - Google Patents

Abstract

The invention discloses a method for quantitative analysis of maltose mixture based on the fusion of terahertz spectrum and image information. Fusion data modeling of spectra and images for quantitative analysis of maltose mixtures. The beneficial effects of the present invention are: 1. The prediction accuracy of the present invention is significantly better than that of a single spectrum or a single image; 2. The present invention uses the PCA algorithm to extract separate features from spectral data and image data, which can effectively remove noise , can better extract the eigenvectors related to modeling, and then effectively improve the prediction accuracy of the multi-source information fusion model; 3. According to the structural risk minimization theory, a Boosting iteration termination judgment index is proposed, which realizes the minimum Automatic optimization of base model parameters for quadratic support vector machines.

Description

Quantitative analysis of maltose mixture based on fusion of terahertz spectrum and image information method

technical field

The invention relates to a method for quantitative analysis of maltose mixture based on fusion of terahertz spectrum and image information.

Background technique

If food is not stored properly, it is prone to mildew, bud change, hardening, insect erosion, aging and other imaginations, which will cause changes in its smell, color, and composition content. In recent years, electronic noses, machine vision, near-infrared, and Many grain quality detection methods such as X-rays cannot meet the needs of accurate and rapid early detection of grain bud mutation. During the germination process of grain, the main chemical composition change is the conversion of starch into sugars for growth, and maltose molecules have obvious characteristic absorption peaks in the THz band, so the quantitative analysis of maltose molecules in grain particles can To determine the degree of budding of the sample, and then realize the early detection of grain.

As a new and reliable non-destructive detection technology with great potential, THz wave contains rich physical and chemical information in its spectrum, and can obtain information that cannot be obtained by other spectral analysis and imaging techniques. Due to its unique characteristics, THz wave and imaging technology has been widely used in medical imaging, security inspection, quality inspection and quality control and other fields. The existing quantitative analysis methods for maltose mixtures all use a single method, and the prediction accuracy is low due to the insufficient information of the single method.

Contents of the invention

The purpose of the present invention is to provide a quantitative analysis method of maltose mixture based on the fusion of terahertz spectrum and image information, so as to solve the technical problems of insufficient information and low prediction accuracy of a single method.

In view of the mentioned problems, the present invention provides a method for quantitative analysis of maltose mixture based on terahertz spectrum and image information fusion, the steps include:

1) Feature extraction is performed on the spectrum and image sample data of the maltose mixture, and the extracted data are fused;

2) Using the fusion data modeling of spectra and images to quantitatively analyze the maltose mixture.

The preferred solution is: before step 1, it also includes obtaining sample data of spectra and images of the maltose mixture.

The preferred scheme is: obtain the spectral sample data of the maltose mixture, the steps include:

1) The THz-TDS system was used to measure the maltose mixture with various concentrations, and the spectrum was obtained;

2) Obtain the frequency domain spectrum of the sample by Fourier transform;

3) Calculate the absorption spectrum and refractive index of the maltose mixture with different concentrations.

The preferred solution is: place the flakes of maltose mixture with different concentrations on the mobile platform in the THz-TDS system, and perform reflection imaging measurement to obtain partial THz images of the maltose mixture with different concentrations.

The preferred solution is to use the PCA algorithm to perform separate feature extraction on the spectral data and the image data.

The preferred solution is: using the Boosting-LS-SVM multi-source information fusion model to predict the spectrum and image feature data of the maltose mixture, the specific methods include:

1) According to the risk minimization theory, determine the conditions for judging the termination of Boosting iterations;

2) Iterate the Boosting-LS-SVM multi-source information fusion model according to the iteration termination judgment condition;

3) The correlation coefficient and root mean square error are used to evaluate the prediction error of the Boosting-LS-SVM multi-source information fusion model.

The optimal solution is: assuming that there are n samples used to establish the fusion model, for the kth maltose mixture, use the PCA algorithm sample spectrum and image data for feature extraction, if the THz transmission spectrum sample data and reflection image sample data of a maltose mixture sample They are:

_xk1 ={ _xk1,1 , _xk1,2 ,..., _xk1,n1 } and _xk2 ={ _xk2,1 , _xk2,2 ,..., _xk2,n2 } (1)

Among them, x _k1 and x _k2 are the THz transmission spectrum data reflection image data of the k-th maltose mixture respectively, and the feature vector sets after feature extraction are expressed as:

z _k1 = {z _k1,1 , z _k1,2 , . . . z _k1,n1 } and z _k2 ={z _k2,1 , z _k2,2 , ... z _k2,n2 } (2)

Among them, z _k1 and z _k2 are the feature vector sets of the k-th sample spectral data and image data after feature extraction respectively, and the spectral and image feature data are fused to form the k-th input vector of the information fusion model, expressed as :

Among them, z _k is the feature vector set after fusion of the kth sample spectrum and image feature data.

The preferred solution is: the formula of the iteration termination judgment condition is:

Where C _m is the termination index value obtained after the mth iteration, and C _m >C _m-1 (m>1), z represents the integrated feature vector of the spectrum image of the maltose mixture; y represents the concentration value of the mixture; F _m (z ) represents the basic regression model obtained after the mth Boosting iteration, F _m (z) represents the prediction result of the spectral image fusion model obtained after the m Boosting iteration; β _m represents the LS-SVM basic model during the mth iteration Weight; a _m represents the parameters of the LS-SVM basic model in the mth iteration process.

The preferred solution is: the predicted value of z _k can be expressed as:

where K(z, z _k ) is the kernel function, t is the number of iterations, β _m represents the weight of the LS-SVM basic model in the mth iteration process; a _m represents the weight of the LS-SVM basic model in the mth iteration process parameter, z _k is the feature vector set after fusion of the kth sample spectrum and image feature data.

The beneficial effects of the present invention are as follows:

1. The prediction accuracy of the present invention is significantly better than that of a single spectrum or a single image;

2. The present invention uses the PCA algorithm to perform separate feature extraction on spectral data and image data, which can effectively remove noise and better extract feature vectors related to modeling, thereby effectively improving the prediction accuracy of the multi-source information fusion model ;

3. According to the structural risk minimization theory, a Boosting iteration termination judgment index is proposed, which realizes the automatic optimization of the basic model parameters of the least squares support vector machine.

Description of drawings

Fig. 1 is maltose THz absorption spectrum;

Fig. 2 is the THz absorption spectrum of maltose wheat flour mixture;

Figure 3(a) is the THz image of maltose and polyethylene mixture with different concentrations of maltose mixture;

Figure 3(b) is the THz image of maltose mixture with different concentrations of maltose and wheat flour mixture;

Figure 4 is a scatter diagram of the correlation between the predicted results and the actual results of the mixture of maltose and polyethylene, and the mixture of maltose and wheat flour.

Detailed ways

The present invention will be further described in detail below, so that those skilled in the art can implement it with reference to the description of the polyethylene mixture.

It should be understood that terms such as "having", "comprising" and "including" as used herein do not entail the presence or addition of one or more other elements or combinations thereof.

The invention provides a method for quantitative analysis of maltose mixture based on fusion of terahertz spectrum and image information, the steps comprising:

1) Feature extraction is performed on the spectrum and image sample data of the maltose mixture, and the extracted data are fused;

2) Using the fusion data modeling of spectra and images to quantitatively analyze the maltose mixture.

The preferred solution is: before step 1, it also includes obtaining sample data of spectra and images of the maltose mixture.

The preferred scheme is: obtain the spectral sample data of the maltose mixture, the steps include:

1) The THz-TDS system was used to measure the maltose mixture with various concentrations, and the spectrum was obtained;

2) Obtain the frequency domain spectrum of the sample by Fourier transform;

3) Calculate the absorption spectrum and refractive index of the maltose mixture with different concentrations.

The preferred solution is: place the flakes of maltose mixture with different concentrations on the mobile platform in the THz-TDS system, and perform reflection imaging measurement to obtain partial THz images of the maltose mixture with different concentrations.

The preferred solution is to use the PCA algorithm to perform separate feature extraction on the spectral data and the image data.

The preferred solution is: using the Boosting-LS-SVM multi-source information fusion model to predict the spectrum and image feature data of the maltose mixture, the specific methods include:

1) According to the risk minimization theory, determine the conditions for judging the termination of Boosting iterations;

2) Iterate the Boosting-LS-SVM multi-source information fusion model according to the iteration termination judgment condition;

3) The correlation coefficient and root mean square error are used to evaluate the prediction error of the Boosting-LS-SVM multi-source information fusion model.

The optimal solution is: assuming that there are n samples used to establish the fusion model, for the kth maltose mixture, use the PCA algorithm sample spectrum and image data for feature extraction, if the THz transmission spectrum sample data and reflection image sample data of a maltose mixture sample They are:

_xk1 ={ _xk1,1 , _xk1,2 ,..., _xk1,n1 } and _xk2 ={ _xk2,1 , _xk2,2 ,..., _xk2,n2 } (1)

Among them, x _k1 and x _k2 are the THz transmission spectrum data reflection image data of the k-th maltose mixture respectively, and the feature vector sets after feature extraction are expressed as:

z _k1 = {z _k1,1 , z _k1,2 , . . . z _k1,n1 } and z _k2 ={z _k2,1 , z _k2,2 , ... z _k2,n2 } (2)

Among them, z _k1 and z _k2 are the feature vector sets of the k-th sample spectral data and image data after feature extraction respectively, and the spectral and image feature data are fused to form the k-th input vector of the information fusion model, expressed as :

Among them, z _k is the feature vector set after fusion of the kth sample spectrum and image feature data.

The preferred solution is: the formula of the iteration termination judgment condition is:

Where C _m is the termination index value obtained after the mth iteration, and C _m >C _m-1 (m>1), z represents the integrated feature vector of the spectrum image of the maltose mixture; y represents the concentration value of the mixture; F _m (z ) represents the basic regression model obtained after the mth Boosting iteration, F _m (z) represents the prediction result of the spectral image fusion model obtained after the m Boosting iteration; β _m represents the LS-SVM basic model during the mth iteration Weight; a _m represents the parameters of the LS-SVM basic model in the mth iteration process.

The preferred solution is: the predicted value of z _k can be expressed as:

where K(z, z _k ) is the kernel function, t is the number of iterations, β _m represents the weight of the LS-SVM basic model in the mth iteration process; a _m represents the weight of the LS-SVM basic model in the mth iteration process parameter, z _k is the feature vector set after fusion of the kth sample spectrum and image feature data.

1. THz spectral characteristics of maltose mixture

The THz-TDS system was used to measure the maltose mixtures of various concentrations, each mixture sample was measured three times, the average spectrum was calculated, and a reference information was measured between the measurement of each two concentration sample mixtures, and then Fourier Transform the frequency domain spectrum of the obtained sample, and finally calculate the absorption spectrum and refractive index of the mixture of maltose with different concentrations. Figure 1 shows the absorption spectrum of part of the mixture of maltose and polyethylene, and Figure 2 shows the absorption spectrum of part of the mixture of maltose and wheat flour.

2. THz image characteristics of maltose mixture

Thin slices of maltose mixture with different concentrations were placed on the moving platform in the THz-TDS system, and reflectance imaging measurements were performed. Partial THz images (at 1.0THz) of different concentrations of maltose and polyethylene mixtures are shown in Figure 3(a), and partial THz images (at 1.0THz) of different concentrations of maltose and wheat flour mixtures are shown in Figure 3(b) 0% in the figure represents pure polyethylene or wheat powder. It can be seen from the figure that with the increase of maltose content, the THz images have obvious changes, which shows that it is feasible to use THz imaging technology to realize the quantitative detection of maltose components.

3. Wheat maltose quantitative analysis and modeling based on terahertz spectrum and image information fusion technology

(1) Perform feature extraction on the spectrum and image sample data of the maltose mixture.

Assuming that there are n samples used to establish the fusion model, for the kth maltose mixture, the PCA algorithm is used to extract the sample spectrum and image data. If the THz transmission spectrum sample data and reflection image sample data of a maltose mixture sample are respectively:

_xk1 ={ _xk1,1 , _xk1,2 ,..., _xk1,n1 } and _xk2 ={ _xk2,1 , _xk2,2 ,..., _xk2,n2 } (1)

Among them, x _k1 and x _k2 are the THz transmission spectrum data reflection image data of the k-th maltose mixture respectively, and the feature vector sets after feature extraction are expressed as:

z _k1 = {z _k1,1 , z _k1,2 , . . . z _k1,n1 } and z _k2 ={z _k2,1 , z _k2,2 , ... z _k2,n2 } (2)

Among them, z _k1 and z _k2 are the feature vector sets of the k-th sample spectral data and image data after feature extraction respectively, and the spectral and image feature data are fused to form the k-th input vector of the information fusion model, expressed as :

Among them, z _k is the feature vector set after fusion of the kth sample spectrum and image feature data.

(2) Using the Boosting-LS-SVM multi-source information fusion model to predict the spectral and image feature data of maltose mixture.

According to the iteration termination judgment condition formula:

Where C _m is the termination index value obtained after the mth iteration, and C _m >C _m-1 (m>1), z represents the integrated feature vector of the spectrum image of the maltose mixture; y represents the concentration value of the mixture; F _m (z ) represents the basic regression model obtained after the mth Boosting iteration, F _m (z) represents the prediction result of the spectral image fusion model obtained after the m Boosting iteration; β _m represents the LS-SVM basic model during the mth iteration Weight; a _m represents the parameters of the LS-SVM basic model in the mth iteration process. The optimal number of iterations of the calculated Boosting-LS-SVM fusion model is t times. The weights of the LS-SVM basic model in the first t iterations are: β ₁ , β ₂ , ... β _n , and the parameters of the basic model are respectively is: a ₁ , a ₂ ,...a _t and a ₁ , b ₂ ,...b _t , then the predicted value of the combined model for z _k can be expressed as:

Among them, K(z, z _k ) is the kernel function.

In order to prevent the impact on the prediction accuracy of the information fusion model due to the unbalanced amount of spectral and image data information, this method first uses the PCA algorithm to extract the features of the THz spectral data samples and image data samples respectively, and controls the number of characteristic signals of each data sample , select the first 4 principal components in the THz spectral data and the first 5 principal components in the image data as the input of the LS-SVM model, and then use the Boosting-LS-SVM algorithm to carry out information fusion modeling on the feature data. At the same time, LS-SVM algorithm uses radial basis function (RBF) and grid search algorithm to evaluate the generalization performance of spectral image combination model. In the Boosting-LS-SVM algorithm, it is first necessary to set the two parameters C and γ of the basic model LS-SVM according to the empirical values, and then perform Boosting iterations on the LS-SVM basic model according to the above-mentioned iteration termination judgment conditions, and finally use the correlation coefficient R and the root mean square error RMSE are used to evaluate the prediction error of the model, and the calculation formula is as follows:

in, Indicates the standard measurement of maltose content, Indicates the predicted value of the Boosting-LS-SVM fusion model. The smaller the RMSE value, the higher the prediction accuracy of the fusion model.

Then the iteration termination judgment index value _Cm after m Boosting iterations in the Boosting-LS-SVM algorithm can be expressed as:

Among them, F _m (z) represents the prediction result of the spectral image fusion model obtained after m iterations of Boosting; β _m represents the weight of the LS-SVM basic model in the m iteration process; a _m represents the LS - Parameters of the underlying model of the SVM.

(3) Establishment of fusion model for quantitative analysis of wheat maltose based on Boosting integration method

Under the corresponding optimal feature data combination, the Boosting-LS-SVM modeling was carried out by using single THz spectral feature data, single THz image feature data, and the combination of THz spectral feature data and image feature data. Table 1 lists the The best prediction results and corresponding parameters of three Boosting-LS-SVM fusion models for ethylene mixture and maltose wheat flour mixture.

It can be seen from Table 1 that the prediction accuracy of the Boosting-LS-SVM model constructed with THz spectral feature data and image feature data is significantly better than that of the Boosting-LS-SVM model constructed with single spectral data and single image data feature extraction. best prediction result. The results prove that the PCA+Boosting-LS-SVM information fusion model proposed in this paper can effectively remove noise, and can better extract the feature vectors related to modeling, thereby effectively improving the prediction accuracy of the multi-source information fusion model.

Table 1 The prediction results and corresponding parameters of the Boosting-LS-SVM model for the maltose mixture

As shown in Figure 4, the mixture of maltose and polyethylene, and the mixture of maltose and wheat flour, after separate feature extraction of spectral data and image data by PCA method, the prediction results of the Boosting-LS-SVM information fusion model constructed by fusion feature data Scatter plot of correlation with actual results. It can be seen from the figure that the PCA+Boosting-LS-SVM multi-source information fusion model adopted in this paper can better approach the actual measured value for the predicted value of the maltose mixture, and the linear correlation between the two is high, which shows that the method proposed in this paper It can significantly improve the prediction accuracy of maltose content in the maltose mixture, is an effective quantitative detection method of maltose, lays an important theoretical and technical basis for the early detection of grain germination, and has great application and promotion value.

Although the embodiment of the present invention has been disclosed as above, it is not limited to the use listed in the specification and implementation, it can be applied to various fields suitable for the present invention, and it can be easily understood by those skilled in the art Therefore, the invention is not limited to the specific details without departing from the general concept defined by the claims and their equivalents.

Claims (9)

1. The method for quantitatively analyzing the maltose mixture based on the fusion of the terahertz spectrum and the image information is characterized by comprising the following steps of:

1) performing characteristic extraction on spectrum and image sample data of the maltose mixture, and fusing the extracted data;

2) the maltose mixture was quantitatively analyzed by using fused data modeling of spectra and images.

2. The method for quantitatively analyzing the maltose mixture based on the fusion of the terahertz spectrum and the image information according to claim 1, wherein step 1 is preceded by obtaining sample data of the spectrum and the image of the maltose mixture.

3. The method for quantitatively analyzing the maltose mixture based on the fusion of the terahertz spectrum and the image information according to claim 2, wherein the steps of acquiring the spectrum sample data of the maltose mixture comprise:

1) measuring the maltose mixture with each concentration by adopting a THz-TDS system to obtain a spectrum;

2) obtaining a frequency domain spectrum of the sample by adopting Fourier transform;

3) the absorption spectrum and refractive index of the maltose mixture at different concentrations were calculated.

4. The quantitative analysis method for the maltose mixture based on the terahertz spectrum and image information fusion as claimed in claim 2, characterized in that, different concentrations of maltose mixture slices are placed on a moving platform in the THz-TDS system, and reflection imaging measurement is carried out to obtain partial THz images of the maltose mixtures with different concentrations.

5. The quantitative analysis method for the maltose mixture based on the fusion of the terahertz spectrum and the image information as claimed in claim 1, characterized in that the PCA algorithm is adopted to perform the individual feature extraction on the spectrum data and the image data.

6. The quantitative analysis method for the maltose mixture based on the terahertz spectrum and image information fusion as claimed in claim 1, wherein the spectrum and image feature data of the maltose mixture are quantitatively analyzed by using a Boosting-LS-SVM multi-source information fusion model, and the specific steps comprise:

1) determining Boosting iteration termination judgment conditions according to a risk minimization theory;

2) iterating the Boosting-LS-SVM multi-source information fusion model according to iteration termination judgment conditions;

3) and evaluating the prediction error of the Boosting-LS-SVM multi-source information fusion model by adopting the correlation coefficient and the root mean square error.

7. The method for quantitatively analyzing the maltose mixture based on the fusion of the terahertz spectrum and the image information as claimed in claim 6, wherein assuming that n samples are used for establishing the fusion model, for the kth maltose mixture, the PCA algorithm is used to extract the features of the sample spectrum and the image data, if the THz transmission spectrum sample data and the reflection image sample data of one maltose mixture sample are respectively:

x_k1＝{x_k1，1，x_k1，2，…，x_k1，n1and x_k2＝{x_k2，1，x_k2，2，…，x_k2，n2} (1)

Wherein x is_k1And x_k2The THz transmission spectrum data reflection image data of the kth maltose mixture respectively represent that a feature vector set after feature extraction is respectively as follows:

z_k1＝{z_k1，1，z_k1，2，…z_k1，n1and z_k2＝{z_k2，1，z_k2，2，…z_k2，n2} (2)

Wherein z is_k1And z_k2Respectively fusing the spectral and image feature data to form a kth input vector of the information fusion model, wherein the kth input vector is expressed as:

wherein z is_kAnd fusing a feature vector set of the kth sample spectrum and the image feature data.

8. The method for quantitatively analyzing the maltose mixture based on the fusion of the terahertz spectrum and the image information as claimed in claim 7, wherein the iteration termination judgment condition is formulated as:

wherein C is_mIs the termination index value obtained after the mth iteration, and C_m＞C_m-1(m > 1), and z represents a comprehensive characteristic vector of a maltose mixture spectral image; y represents a concentration value of the mixture; f_m(z) represents the basic regression model obtained after the mth Boosting iteration, F_m(z) represents the spectral image fusion model prediction result obtained after m Boosting iterations,. beta_mRepresenting the weight of the LS-SVM basic model in the mth iteration process; a is_mAnd representing parameters of the LS-SVM basic model in the mth iteration process.

9. The method for quantitatively analyzing maltose mixture based on terahertz spectrum and image information fusion as claimed in claim 8, wherein z is_kThe predicted value of (c) can be expressed as:

wherein K (z, z)_k) is a kernel function, t is the number of iterations, β_mRepresenting the weight of the LS-SVM basic model in the mth iteration process; a is_mRepresents the parameters of the LS-SVM base model in the mth iteration process, z_kAnd fusing a feature vector set of the kth sample spectrum and the image feature data.