Construction method of radix pseudostellariae germplasm identification model and radix pseudostellariae germplasm identification method

Abstract

The invention provides a construction method of a radix pseudostellariae germplasm identification model and a radix pseudostellariae germplasm identification method, and belongs to the technical field of medicinal material identification. The method comprises the following steps: acquiring hyperspectral image data of different radix pseudostellariae; carrying out black and white board correction on the acquired hyperspectral image data of the radix pseudostellariae, and calculating the relative reflectivity of the corrected hyperspectral data; performing threshold segmentation on the relative reflectivity data, extracting a target region of interest, and calculating an average spectral value of the region of interest; removing abnormal data of the average spectral value of the region of interest to obtain original full-band spectral data; and performing characteristic wave band screening on the original full-wave band spectral data, and constructing a radix pseudostellariae germplasm identification model by using the screened characteristic wave band and germplasm information by using a particle optimization algorithm. The model constructed by using the hyperspectral imaging technology can be used for classification and identification of radix pseudostellariae, is high in resolution and sensitivity, good in repeatability, simple, convenient and rapid, and has wide application value.

Description

Technical Field

[0001] The invention belongs to the technical field of medicinal material identification, and in particular relates to a method for constructing a Pseudostellaria heterophylla germplasm identification model and a Pseudostellaria heterophylla germplasm identification method. Background Art

[0002] Pseudostellaria heterophylla is the dried root tuber of the plant Pseudostellaria baicalensis of the Caryophyllaceae family, which has a rich history of medicinal and culinary traditions in various regions of Asia. It contains important bioactive compounds such as cyclic peptides, polysaccharides and amino acids. These components give it significant pharmacological properties, including antifungal, antitumor, immune-enhancing and antioxidant activities. The organic extract of Pseudostellaria heterophylla performed well in inhibiting lipid peroxidation in rat brain homogenate, outperforming the organic extracts of other herbs such as American ginseng, Panax notoginseng and Codonopsis pilosula. With the increasing awareness of health and nutrition, high-quality Pseudostellaria heterophylla germplasm has received increasing attention.

[0003] Pseudostellariae germplasm can meet different market demands, but there is also a risk of germplasm contamination. The use of uncertified germplasm not only infringes on the rights of breeders, but may also cause huge economic losses to Chinese medicine producers and operators. To improve the purity of these germplasms and thus increase their economic value through better selection and processing, accurate means of identifying Pseudostellariae germplasm are essential. Traditional Pseudostellariae germplasm identification methods have been widely used in Chinese medicine research, such as protein fingerprinting or simple sequence repeat detection, with high accuracy and reliability, but these methods require a large amount of organic solvents, which may have adverse effects on the environment. Therefore, there is an urgent need for rapid, non-destructive and environmentally friendly analytical strategies to detect the authenticity of Pseudostellariae germplasm. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a method for constructing a Pseudostellaria heterophylla germplasm identification model. The constructed model can be used for the classification and identification of Pseudostellaria heterophylla, has high resolution and sensitivity, good repeatability, is simple and fast, and has wide application value.

[0005] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions:

[0006] A method for constructing a Pseudostellariae Radix germplasm identification model comprises the following steps: collecting hyperspectral image data of Pseudostellariae Radix with different germplasms; performing black and white plate correction on the collected Pseudostellariae Radix hyperspectral image data to obtain corrected hyperspectral data, and calculating the relative reflectivity of the corrected hyperspectral data; performing threshold segmentation on the relative reflectivity data to extract a target region of interest, and then calculating an average spectral value of the region of interest; eliminating abnormal data of the average spectral value of the region of interest to obtain original full-band spectral data; performing characteristic band screening on the original full-band spectral data, and then using a particle optimization algorithm to construct a Pseudostellariae Radix Germplasm Identification Model by combining the screened characteristic bands and germplasm information.

[0007] Preferably, the collected Pseudostellaria heterophylla hyperspectral image data is divided into a training set and a validation set, the processed training set data is used for modeling, and the processed validation set data is used for debugging the model.

[0008] Preferably, the ratio of the training set to the validation set is 7:3.

[0009] Preferably, the number of P. heterophylla samples of each germplasm background is ≥200.

[0010] Preferably, a hyperspectral imager is used to perform spectral scanning to collect hyperspectral image data of Pseudostellaria heterophylla, and the spectral scanning conditions are as follows: the scanning mode is focal plane push-scan imaging; the scanning speed is 1.5 m / s; the preheating time is 30 min; the exposure time is 4.5 ms; and the spectral band range is 886 to 1735 nm.

[0011] Preferably, SpecView software is used to perform black and white board correction, and the formula for black and white board correction is as follows: I=(I0-B) / (WB), wherein I is the corrected hyperspectral data, I0 is the original hyperspectral image data; B is the blackboard reference image data, and W is the whiteboard reference image data.

[0012] Preferably, MATLAB R2018b software is used for threshold segmentation, region of interest extraction and average spectral value calculation.

[0013] Preferably, a competitive adaptive reweighted sampling-continuous projection tandem algorithm is used to screen characteristic bands.

[0014] Preferably, the germplasm of Pseudostellaria heterophylla includes Zheshen No. 1, Zheshen No. 2, Zheshen No. 4 and farmer's varieties.

[0015] Another object of the present invention is to provide a method for identifying Pseudostellaria chinensis germplasm, comprising the following steps: constructing a Pseudostellaria chinensis germplasm identification model using the above-mentioned construction method; collecting hyperspectral image data of the sample to be tested, importing the hyperspectral image data of the sample to be tested into the Pseudostellaria chinensis germplasm identification model, and obtaining an identification result.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] The present invention provides a method for constructing a Pseudostellaria heterophylla germplasm identification model. The model constructed by the present invention using hyperspectral imaging technology can be used for the classification and identification of Pseudostellaria heterophylla, has high resolution and sensitivity, good repeatability, is simple and quick, and has wide application value.

[0018] The Pseudostellaria heterophylla germplasm model constructed by the present invention is adopted to identify the Pseudostellaria heterophylla germplasm, and the hyperspectral imaging technology is used to detect the samples to be tested. The samples do not need to undergo complex pretreatment before detection, which saves a lot of labor and effectively improves the detection efficiency. The detection time is about 2 minutes, which shortens the detection time. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a schematic diagram of the process of identifying the germplasm of Radix Pseudostellariae Radix;

[0020] Figure 2 It is the full spectrum band of the short-wave infrared spectrum region of the sample;

[0021] Figure 3 This is the confusion matrix of the CARS-SPA-PSO-BPNN neural network model predicting the finished ginseng germplasm of Pseudostellaria heterophylla, where L is the farm variety, Z1 is Zheshen No. 1, Z2 is Zheshen No. 2, and Z4 is Zheshen No. 4;

[0022] Figure 4 The confusion matrix of the CARS-SPA-PSO-BPNN neural network model predicting the germplasm of the tested samples, where L is the farm variety, Z1 is Zheshen No. 1, Z2 is Zheshen No. 2, and Z4 is Zheshen No. 4;

[0023] Figure 5 Spectral distribution diagram of the characteristic bands screened for the samples. DETAILED DESCRIPTION

[0024] The invention provides a method for constructing a Pseudostellariae Radix Germplasm Identification Model, comprising the following steps: collecting hyperspectral image data of Pseudostellariae Radix with different germplasms; performing black-and-white plate correction on the collected Pseudostellariae Radix Hyperspectral Image Data to obtain corrected hyperspectral data; and calculating the relative reflectivity of the corrected hyperspectral data; performing threshold segmentation on the relative reflectivity data, extracting a target region of interest, and then calculating an average spectral value of the region of interest; eliminating abnormal data of the average spectral value of the region of interest to obtain original full-band spectral data; performing feature band screening on the original full-band spectral data, and then using a particle optimization algorithm to construct a Pseudostellariae Radix Germplasm Identification Model based on the screened feature bands and germplasm information.

[0025] In the present invention, the germplasm of Pseudostellariae Radix includes Zheshen No. 1, Zheshen No. 2, Zheshen No. 4 and farm variety, and the number of samples of Pseudostellariae Radix from each germplasm background is ≥200. The present invention adopts a double-sided image acquisition method to obtain the overall characteristics of the sample. When collecting the hyperspectral image data of Pseudostellariae Radix, the data of one side of Pseudostellariae Radix is ​​first collected, and then the Pseudostellariae Radix is ​​turned over, and the data of the back of Pseudostellariae Radix is ​​collected again. The collected hyperspectral image data of Pseudostellariae Radix is ​​divided into a training set and a validation set, and the processed training set data is used for modeling, and the processed validation set data is used for debugging the model; the ratio of the training set to the validation set is 7:3. The farm variety described in the specific embodiment of the present invention is the Pseudostellariae Radix from Shanghuangbai Village, Zherong County, Ningde City, Fujian Province.

[0026] In the present invention, a hyperspectral imager is used to perform spectral scanning to collect hyperspectral image data of Pseudostellaria heterophylla, and the spectral scanning conditions are as follows: the scanning mode is focal plane push scanning imaging; the scanning speed is 1.5m / s; the preheating time is 30min; the exposure time is 4.5ms; the spectral band range is 886-1735nm. When performing spectral scanning, the hyperspectral imager is located 25cm above Pseudostellaria heterophylla, and the distance between each Pseudostellaria heterophylla is 1.5cm. The spectrometer model in the specific embodiment of the present invention is GaiaFied-N17E-HR.

[0027] In the present invention, SpecView software is used to perform black and white board correction, and the formula for black and white board correction is as follows: I=(I0-B) / (WB), wherein I is the corrected hyperspectral data, I0 is the original hyperspectral image data; B is the blackboard reference image data, and W is the whiteboard reference image data.

[0028] In the present invention, MATLAB R2018b software is used to perform threshold segmentation, region of interest extraction and average spectral value calculation; the threshold segmentation is to select the band with the largest difference in spectral reflectance between the sample and the background, and the background and seed samples are clearly separated by setting a suitable threshold; the region of interest is obtained by using a method of morphological filtering combined with mask processing.

[0029] The present invention has found that the extracted region of interest (ROI) is relatively subject to relatively large noise interference in the first and last band regions of the hyperspectral system. After obtaining the average spectral reflectance file of each sample ROI, it is necessary to screen out abnormal data and uniformly eliminate the band information that is greatly affected by random noise. In the present invention, the abnormal data refers to the band information with a low noise ratio, and the signal-to-noise ratio of the band information that is greatly affected by random noise is low. The signal-to-noise ratio refers to the ratio of signal power to noise power. When the random noise interferes with the signal greatly, the signal power is smaller than the noise power, and the signal-to-noise ratio will decrease. In the present invention, spectral data in the range of 886 to 1735nm is collected, and the spectral data in the range of 960.79nm to 1600.71nm is retained after eliminating the data with a low signal-to-noise ratio.

[0030] In the present invention, the competitive adaptive reweighted sampling-continuous projection tandem algorithm is used for feature band screening. Among them, the successive projections algorithm (SPA) is a widely used feature variable selection algorithm, which can extract low collinearity and low redundancy variables through the projection algorithm to avoid information overlap and collinearity. Competitive adaptive reweighted sampling (CARS) is a method for variable selection in multi-component spectral analysis, combining Monte Carlo sampling and partial least squares (PLS) model regression coefficients, retaining important variables through adaptive weighted sampling, removing unimportant variables, thereby improving model performance, so that the accuracy of the discriminant model constructed due to the reduction of data volume has less loss.

[0031] The present invention screens characteristic bands and Pseudo-P ...

[0032] The present invention also provides a method for identifying Pseudostellaria heterophylla germplasm, comprising the following steps: constructing a Pseudostellaria heterophylla germplasm identification model using the above-mentioned construction method; collecting hyperspectral image data of the sample to be tested, and importing the hyperspectral image data of the sample to be tested into the Pseudostellaria heterophylla germplasm identification model to obtain an identification result. Figure 1 shown.

[0033] The technical solutions provided by the present invention are described in detail below in conjunction with the embodiments, but they should not be construed as limiting the protection scope of the present invention.

[0034] Example 1

[0035] A method for constructing a Pseudostellaria heterophylla germplasm identification model comprises the following steps:

[0036] The double-sided image acquisition method of the sample was adopted, and the hyperspectral imaging technology was used to perform spectral scanning on different germplasms of Pseudostellaria heterophylla. The hyperspectral image data of both sides of Pseudostellaria heterophylla were collected. The germplasms of Pseudostellaria heterophylla were Zheshen No. 1, Zheshen No. 2, Zheshen No. 4 and farmer's species. The number of samples of each germplasm was ≥ 200. Pseudostellaria heterophylla was placed on the black cloth plane, with a distance of 1.5 cm between each Pseudostellaria heterophylla. The hyperspectral imager was used for spectral scanning. The hyperspectral imager was located 25 cm above the Pseudostellaria heterophylla. The spectral scanning conditions were as follows: the scanning mode was focal plane push scanning imaging; the scanning speed was 1.5 m / s; the preheating time was 30 min; the exposure time was 4.5 ms; the spectral band range was 886-1735 nm. The collected hyperspectral image data of Pseudostellaria heterophylla were divided into training set and validation set in a ratio of 7:3. The processed training set data was used for modeling, and the processed validation set data was used for debugging the model.

[0037] The collected Pseudostellaria heterophylla hyperspectral image data was corrected by black-and-white board to obtain the corrected hyperspectral data, and the relative reflectance of the corrected hyperspectral data was calculated; the black-and-white board correction formula was I=(I0-B) / (WB), where I was the corrected hyperspectral data, I0 was the original hyperspectral image data; B was the blackboard reference image data, and W was the whiteboard reference image data.

[0038] MATLAB R2018b software is used to perform threshold segmentation on the relative reflectance data, extract the target region of interest, and then calculate the average spectral value of the region of interest;

[0039] Eliminate the band information with low noise ratio in the average spectral value of the region of interest (keep the spectral data in the range of 960.79nm to 1600.71nm) to obtain the original full-band spectral data;

[0040] The original full-band spectral data was subjected to characteristic band screening using a competitive adaptive reweighted sampling-continuous projection tandem algorithm, and then a Pseudo-P ...

[0041] Example 2

[0042] A method for identifying Pseudostellaria heterophylla germplasm comprises the following steps:

[0043] A total of 967 Pseudostellariae tuberculosis were collected, including Zheshen No. 1, Zheshen No. 2, Zheshen No. 4 and farm-grown Pseudostellariae tuberculosis. The specific information of the samples is shown in Table 1, among which the farm-grown Pseudostellariae tuberculosis is from Shanghuangbai Village, Zherong County, Ningde City, Fujian Province.

[0044] 1934 sets of data were obtained by scanning the front and back sides of Pseudostellaria heterophylla, and the training set and validation set were divided into a ratio of 7:3. Pseudostellaria heterophylla was placed on a black cloth plane, with a distance of 1.5 cm between each Pseudostellaria heterophylla, and the hyperspectral imager was located 25 cm above the Pseudostellaria heterophylla for scanning. The spectral scanning conditions were as follows: the scanning mode was focal plane push scanning imaging; the scanning speed was 1.5 m / s; the preheating time was 30 min; the exposure time was 4.5 ms; and the spectral band range was 886-1735 nm.

[0045] Table 1 Sample information

[0046]

[0047] The SpecView software configured with the hyperspectral sorter is used to calibrate the scanned original spectral image with a black and white plate to obtain the corrected hyperspectral data. The black and white plate calibration formula is I = (I0-B) / (WB), where I is the corrected hyperspectral data and I0 is the original hyperspectral image data; B is the blackboard reference image data and W is the whiteboard reference image data. The relative reflectivity of the corrected hyperspectral data is then calculated.

[0048] After the black and white plate calibration, each root of Radix Pseudostellariae Radix is ​​segmented from the background of the hyperspectral image as an independent region of interest (ROI): by ENVI 5.1 version, first open the calibrated spectral image, use Basic Tools→Region of Interest→ROI Tool to segment the entire area into a single sample and background, open the segmented single sample and background spectral image, use Basic Tools→Region of Interest→Band Threshold to ROI to extract the region of interest, set the maximum and minimum thresholds (the threshold needs to be debugged to find the best threshold. When the region of interest is extracted using ENVI 5.1 software, when the pixel points of the region of interest are most consistent with the true spectral image of the sample, it is the best threshold. In this embodiment, the maximum and minimum thresholds are 1 and 0.2, respectively). Then use Basic Tools→Statistics→Compute Statistics to calculate the sample and background spectral reflectances in different bands, and select the band with the largest difference between the sample and background reflectances. By setting appropriate thresholds (maximum and minimum thresholds are 1 and 0.2, respectively) in the band with the largest difference in reflectance between the sample and the background, the background and seed samples are clearly separated, and then the morphological filtering combined with mask processing is used to obtain the target ROI. The spectral image of the target ROI is analyzed and extracted using MATLAB R2018b software to obtain HSI data (average spectral reflectance). The sorted sample short-wave infrared spectral data is shown in Figure 1. Figure 2 shown.

[0049] After obtaining the average spectral reflectance file of each sample ROI, the band information with low noise ratio was eliminated (a total of 126 band information was eliminated), and only 386 spectral data in the range of 960.79nm to 1600.71nm of each ROI file were retained to obtain the original full-band spectral data;

[0050] The original full-band spectral data was screened for characteristic bands using the competitive adaptive reweighted sampling-continuous projection tandem algorithm, and then the characteristic bands obtained by screening and the germplasm information were used to construct a Pseudo-P ...

[0051] Table 2 Results of Pseudostellaria heterophylla germplasm identification model

[0052]

[0053] As shown in Table 2, the constructed PSO-BPNN four-classification model has a prediction accuracy of 98.2% and 96.4% for the training set and validation set, respectively. Figure 3 The confusion matrix is ​​shown.

[0054] Collecting the hyperspectral image data of the samples to be tested: 25 roots of Pseudostellaria heterophylla samples were selected for each germplasm (Zheshen No. 1, Zheshen No. 2, Zheshen No. 4 and farm species), and the overall characteristics of the samples were obtained by using the double-sided image acquisition method. 50 sets of data (25 roots × double sides) were obtained for each germplasm. The hyperspectral image data of the samples to be tested were imported into the Pseudostellaria heterophylla germplasm identification model. The test results showed that the accuracy rate was 96.50%. The prediction results of the samples to be tested are as follows: Figure 4 The confusion matrix shown in the figure shows that the method provided by the present invention can accurately identify Pseudo-ginseng germplasm.

[0055] Example 3

[0056] This example compares the effects of different preprocessing of the original full-band spectral data on the identification of Pseudostellaria heterophylla germplasm.

[0057] RAW: adopt the method of Example 2 (without pretreatment);

[0058] SNV: The difference from Example 2 is that the original full-band spectral data uses the standard normal variate (SNV) algorithm to correct the information using the mean and variance of the sample data to eliminate scattered noise;

[0059] MSC: The difference from Example 2 is that the original full-band spectral data uses a multiplicative scatter correction (MSC) algorithm to eliminate the baseline shift and offset problems caused by light scattering and enhance the effective spectral information;

[0060] MC: The difference from Example 2 is that the original full-band spectral data uses mean centering (MC), and the mean of each variable in the data set is subtracted from the mean of the variable, so that the mean of the processed data is zero, reducing the collinearity between the dependent variable and the interaction term and improving the convergence speed and performance of the algorithm;

[0061] FD: The difference from Example 2 is that the original full-band spectral data uses the first-order derivative (FD) in the direct derivation method to eliminate the baseline drift and spectral band overlap in the spectral data, thereby improving the spectral band characteristics and spectral resolution.

[0062] DT: The difference from Example 2 is that the original full-band spectral data uses detrend correction (Detrend, DT) to eliminate the long-term trend or systematic deviation in the data, and uses polynomial regression to fit more complex data trends.

[0063] After different preprocessing of the original full-band spectral data, the identification effect of the constructed identification model on P. heterophylla germplasm is shown in Table 3.

[0064] Table 3 Effect of Pseudostellaria heterophylla germplasm identification model after original full-band pretreatment

[0065]

[0066] It can be seen that the method provided in this application is more effective in identifying Pseudoradix heterophylla germplasm.

[0067] Example 4

[0068] This example compares the effects of different characteristic band screening methods on the identification of Pseudostellaria heterophylla germplasm.

[0069] CARS-SPA: adopt the method of Example 2 (using competitive adaptive reweighted sampling-continuous projection tandem algorithm to screen feature bands);

[0070] SPA: The difference from Example 2 is that: a continuous projection method is adopted to screen the characteristic bands;

[0071] CARS: The difference from Example 2 is that competitive adaptive reweighted sampling is adopted to screen characteristic bands.

[0072] After adopting different methods to screen characteristic bands, the identification effect of the constructed identification model on P. heterophylla germplasm is shown in Table 4. The distribution of characteristic bands screened by CARS-SPA tandem algorithm is shown in Table 4. Figure 5 shown.

[0073] Table 4 Effects of constructing models using different characteristic band screening methods

[0074]

[0075] From Table 4 and Figure 5 The results show that the model constructed by the present invention uses only 16 characteristic bands, which only account for 4.2% of the number of full spectrum bands. The constructed model is simple and has high accuracy.

[0076] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for constructing a Pseudostellaria heterophylla germplasm identification model, characterized in that: The following steps are involved: Collect hyperspectral image data of Pseudostellaria heterophylla with different germplasm backgrounds; Perform black and white plate correction on the collected Pseudostellaria heterophylla hyperspectral image data to obtain corrected hyperspectral data, and calculate the relative reflectance of the corrected hyperspectral data; Performing threshold segmentation on the relative reflectance data, extracting the target region of interest, and then calculating the average spectral value of the region of interest; Eliminate the abnormal data of the average spectrum value of the region of interest to obtain the original full-band spectrum data; The original full-band spectral data is subjected to characteristic band screening, and then a Pseudo-Pseudo-Pseudo-Ginseng germplasm identification model is constructed by combining the characteristic bands obtained through screening with germplasm information using a particle optimization algorithm.

2. The construction method according to claim 1, characterized in that: The collected Pseudostellaria heterophylla hyperspectral image data were divided into training set and validation set. The processed training set data were used for modeling, and the processed validation set data were used for debugging the model.

3. The construction method according to claim 2, characterized in that: The ratio of the training set to the validation set is 7:

3.

4. The construction method according to claim 1, characterized in that: The number of samples of P. heterophylla from each germplasm background was ≥200.

5. The construction method according to claim 1, characterized in that: A hyperspectral imager was used to perform spectral scanning to collect hyperspectral image data of Pseudostellaria heterophylla. The spectral scanning conditions were as follows: the scanning mode was focal plane push-scan imaging; the scanning speed was 1.5 m / s; the preheating time was 30 min; the exposure time was 4.5 ms; and the spectral band range was 886-1735 nm.

6. The construction method according to claim 1, characterized in that: SpecView software was used to perform black and white plate calibration, and the formula for black and white plate calibration was as follows: I = (I0-B) / (WB), where I is the calibrated hyperspectral data, I0 is the original hyperspectral image data; B is the blackboard reference image data, and W is the whiteboard reference image data.

7. The construction method according to claim 1, characterized in that: MATLAB R2018b software was used for threshold segmentation, region of interest extraction and average spectral value calculation.

8. The construction method according to claim 1, characterized in that: The competitive adaptive reweighted sampling-continuous projection cascade algorithm is used to screen characteristic bands.

9. The construction method according to claim 1, characterized in that: The germplasm of Pseudoradix heterophylla includes Zheshen No. 1, Zheshen No. 2, Zheshen No. 4 and the farmer's variety.

10. A method for identifying Pseudostellaria heterophylla germplasm, characterized in that: The method comprises the following steps: constructing a Pseudostellaria heterophylla germplasm identification model by using the construction method described in any one of claims 1 to 9; The hyperspectral image data of the sample to be tested is collected, and the hyperspectral image data of the sample to be tested is imported into the Pseudostellaria heterophylla germplasm identification model to obtain an identification result.

Images