A method for nondestructive evaluation of ginseng quality based on hyperspectral and x-ray technology

Abstract

The present application provides a ginseng quality nondestructive evaluation method based on hyperspectral and X-ray technology, belongs to the field of spectral analysis technology, and comprises the following steps: collecting information of a complete ginseng sample by using hyperspectral and X-ray technology, and integrating complementary information from the two devices by a modeling method of ensemble learning. The present application provides a fast, nondestructive and accurate ginseng quality evaluation method, and solves the problems of low accuracy of experience identification analysis, complex and time-consuming pretreatment of wet chemical analysis method, and difficulty in adapting to batch rapid detection due to sample crushing treatment of near-infrared spectral analysis technology.

Description

Technical Field

[0001] This invention belongs to the field of spectral analysis technology, and in particular relates to a non-destructive evaluation method for ginseng quality based on hyperspectral and X-ray technology. Background Technology

[0002] Currently, the methods for determining the content of ginsenosides mainly rely on chemical analysis, including visible-ultraviolet spectrophotometry, thin-layer chromatography, capillary electrophoresis, high-performance liquid chromatography (HPLC), and HPLC-MS. While these chemical analysis methods can accurately determine the content of ginsenosides and other chemical components, they require significant time for sample pulverization and component extraction, making real-time online analysis of large numbers of samples difficult. Furthermore, the component extraction process requires large amounts of organic reagents, significantly increasing the cost of the analysis. In addition, near-infrared spectroscopy has been used to predict the content of total ginsenosides, but sample pulverization is also unavoidable during sample processing. As a precious herbal nutritional product and one of the most widely consumed alternative medicines globally, ginseng urgently needs a rapid and non-destructive analytical method to predict the content of its quality-related active ingredients. Summary of the Invention

[0003] In view of this, the present invention proposes a non-destructive evaluation method for ginseng quality based on hyperspectral and X-ray technology, specifically as follows:

[0004] A non-destructive method for evaluating the quality of ginseng based on hyperspectral and X-ray technology includes the following steps:

[0005] Acquire hyperspectral data of the ginseng sample to be tested, and extract information from the hyperspectral data;

[0006] Obtain X-ray data of the ginseng sample to be tested and extract information from the X-ray images;

[0007] Obtain high-performance liquid chromatography data of the ginseng sample to be tested;

[0008] Based on the spectral data, the extracted spectral data is preprocessed;

[0009] Based on the spectral data, characteristic wavelengths are selected from the preprocessed spectral data;

[0010] Based on the image information, feature extraction is performed on the image information;

[0011] Establish a ginseng quality evaluation model based on single data;

[0012] By integrating the prediction results of multiple models using ensemble learning modeling methods, quality information of ginseng can be obtained.

[0013] Furthermore, in step S1,

[0014] The acquisition of the hyperspectral data of the ginseng sample includes the following steps:

[0015] Select ginseng with relatively intact rhizomes, main roots, and lateral roots, and clean the surface soil for later use;

[0016] The acquisition parameters of the hyperspectral imaging spectrometer were set as follows: aperture size f2.8, exposure time 19ms, and distance between the spectrometer lens and the sample 32.5cm, to obtain hyperspectral data;

[0017] Information extraction from the hyperspectral data of the ginseng sample includes the following steps:

[0018] Extracting spectral information from hyperspectral data;

[0019] Extract image information from hyperspectral data, including color information and texture feature information.

[0020] Furthermore, the color information of the ginseng sample includes:

[0021] Nine types of color information are obtained from the first-order color matrix, second-order color matrix, and third-order color matrix calculated based on HSV values;

[0022] The texture feature information of the ginseng sample includes:

[0023] The texture features obtained using the gray-level gradient co-occurrence matrix include: small gradient advantage, large gradient advantage, gray-level distribution non-uniformity, gradient distribution non-uniformity, energy, gray-level average, gradient average, gray-level variance, gradient variance, correlation, gray-level entropy, gradient entropy, mixing entropy, inertia, and inverse difference moment, totaling 15 types of texture features.

[0024] Furthermore, in step S2, the acquisition of X-ray data of the ginseng sample includes the following steps:

[0025] The X-ray emitter has an emission angle of 27.0°;

[0026] The samples are placed on the conveyor belt at certain intervals, and the conveyor belt runs at a speed of 12.0 m / min;

[0027] Simultaneously, a standard sample with a known density is placed, and grayscale images are acquired together with the ginseng sample.

[0028] The information extraction from the X-ray data of the ginseng sample includes the following steps:

[0029] Calculate the average gray value of the ginseng sample and the standard sample on each X-ray image;

[0030] Through formula Di =D holder *G i / G holder The grayscale value of ginseng is directly converted into a density value, where D i D is the absolute density value of the substance being tested. holder It is the absolute density value of the standard substance, G. i and G holder These are the average gray values ​​of the substance to be tested and the standard substance, respectively.

[0031] Furthermore, in step S4, the method for preprocessing the spectral data includes:

[0032] One or more of the following processes: processing to remove noise from data and processing to remove the influence of the measurement environment on data.

[0033] Furthermore, in step S5, characteristic wavelength selection is performed on the processed spectral data, including:

[0034] Feature wavelengths are selected from all bands in the spectral data based on a feature variable screening algorithm.

[0035] Furthermore, in step S6, feature extraction is performed on the image information, including:

[0036] Feature image information is extracted from image information based on a feature variable filtering algorithm.

[0037] Furthermore, in step S7, a ginseng quality evaluation model is established based on single data, including:

[0038] A random forest regression model is established based on the characteristic wavelength information in hyperspectral data;

[0039] A partial least squares regression model is established based on the feature image information in hyperspectral data.

[0040] A linear regression model is established based on density value information.

[0041] Furthermore, the specific data refers to three types of data: characteristic wavelength information, characteristic image information, and density value information in hyperspectral data.

[0042] Furthermore, in step S8, predictions from multiple models are integrated based on an ensemble learning modeling method to obtain ginseng quality information, including:

[0043] Input the dataset with preset feature variables into the pre-trained basic model;

[0044] The stacking ensemble learning method is used to combine the prediction results of each basic model to obtain the quality information of the ginseng sample to be tested.

[0045] The beneficial effects of this invention are as follows:

[0046] This invention utilizes hyperspectral and X-ray technologies to acquire complete information from ginseng samples, integrating complementary information from both devices through an ensemble learning modeling method. The advantages of this invention lie not only in providing accurate and reliable results, but also in its simplicity and efficiency. Traditional ginseng testing methods typically require complex sample processing and time-consuming steps, while this invention significantly improves testing speed and simplifies the operational process. Therefore, this technology has significant application prospects for ginseng production and trade, providing related industries with a more convenient and effective quality control method. Attached Figure Description

[0047] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0048] Figure 1 This is a flowchart illustrating the implementation of the method for constructing the ginseng quality evaluation model provided in this embodiment of the invention.

[0049] Figure 2 This is a schematic diagram of the device for acquiring hyperspectral data provided in an embodiment of the present invention;

[0050] Figure 3 This is a flowchart of the effective information extraction process for hyperspectral data provided in this embodiment of the invention;

[0051] Figure 4 This is the original spectral image of the hyperspectral data provided in the embodiments of the present invention;

[0052] Figure 5 This is a schematic diagram of the device for acquiring X-ray data provided in an embodiment of the present invention;

[0053] Figure 6 This is a high-performance liquid chromatogram of a ginseng sample provided in an embodiment of the present invention;

[0054] Figure 7 This is a preprocessed spectral image of hyperspectral data provided in the embodiments of the present invention;

[0055] Figure 8 This is a distribution map of characteristic wavelengths selected using UVE, SPA, and CARS algorithms provided in an embodiment of the present invention;

[0056] (A.1) Distribution of characteristic wavelengths related to Rg1+Re selected using the UVE algorithm;

[0057] (A.2) Distribution of characteristic wavelengths related to Rg1+Re selected using the SPA algorithm;

[0058] (A.3) Distribution of characteristic wavelengths related to Rg1+Re selected using the CARS algorithm; (B.1) Distribution of characteristic wavelengths related to Rb1 selected using the UVE algorithm;

[0059] (B.2) Distribution of characteristic wavelengths related to Rb1 selected using the SPA algorithm;

[0060] (B.3) Distribution of characteristic wavelengths related to Rb1 selected using the CARS algorithm;

[0061] Figure 9 This is a distribution map of feature images selected using UVE, SPA, and CARS algorithms, provided in an embodiment of the present invention.

[0062] (A) Distribution map of feature images related to Rg1+Re selected using UVE, SPA and CARS algorithms;

[0063] (B) Distribution map of Rb1-related feature images selected using UVE, SPA, and CARS algorithms;

[0064] Figure 10 This is a flowchart of the decision-level data fusion process provided in an embodiment of the present invention;

[0065] Figure 11 This is a scatter plot of the prediction results and actual measurement results of Rg1+Re and Rb1 using decision-level data fusion provided in this embodiment of the invention.

[0066] (A) Scatter plot of the prediction results of the decision-level data fusion model for Rg1+Re and the actual measurement results;

[0067] (B) Scatter plot of the prediction results of the decision-level data fusion model for Rb1 and the actual measurement results.

[0068] Figure label:

[0069] 21: Computer; 22: Hyperspectral Imager; 23: Halogen Lamp

[0070] 24: Sample

[0071] 51: Monitor; 52: Computer; 53: X-ray emitter

[0072] 54: Sample; 55: Conveyor belt; 56: Linear array detector Detailed Implementation

[0073] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0074] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0075] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0076] Example 1

[0077] like Figure 1 As shown, this invention provides a non-destructive method for evaluating the quality of ginseng based on hyperspectral and X-ray technology, comprising the following steps:

[0078] Acquire hyperspectral data of ginseng and extract information from the hyperspectral data of ginseng;

[0079] Hyperspectral data of a clean, whole ginseng root were obtained using a hyperspectral imager.

[0080] Twenty-six batches of five-year-old artificially cultivated ginseng were selected. Four ginseng roots with relatively intact rhizomes, taproots, and lateral roots from each batch were selected and labeled. One ginseng root was randomly selected from each batch as the test set to evaluate the performance of the constructed model (n=26), and the remaining three ginseng roots were used for model training and optimization (n=78). Hyperspectral data were collected after cleaning the ginseng surface.

[0081] The hyperspectral imaging system selected for this experiment consists of one hyperspectral imager (22), two symmetrically placed 75W halogen lamps (23), a computer (21), and image acquisition software (HyperScanner), etc. The schematic diagram of the device is shown below. Figure 2 As shown. The acquisition parameters of the hyperspectral imaging spectrometer (22) were set as follows: aperture size of f2.8, exposure time of 19ms, and distance between the lens of the spectrometer (22) and the sample (24) of 32.5cm. Finally, 128 spectral bands were acquired in the spectral range of 372.84nm to 1049.08nm.

[0082] To fully utilize the information contained in hyperspectral data, spectral information containing the internal quality of the sample is extracted from the hyperspectral data. The extraction process is as follows: Figure 3As shown. The grayscale image of the band with the greatest color difference between the ginseng sample and the background in the hyperspectral data was selected. First, the grayscale image was binarized, and background noise was removed from the binarized image through morphological transformation. Then, threshold segmentation was performed on the binarized image to obtain the contour of the ginseng sample, thereby obtaining the pixel coordinates of the ROI region. Finally, the hyperspectral data of each band was traversed using the pixel coordinates, and the average of the spectral data of all pixels in the ROI region was used as the original spectral data for each sample. Because the acquired hyperspectral images contain a large amount of noise due to the influence of dark current in the camera, uneven intensity distribution of the light source, and other factors, black and white correction was performed on the original spectral data to obtain the final spectral data, as shown in the spectrum. Figure 4 As shown.

[0083] To fully utilize the information contained in hyperspectral data, image information containing external quality, including color and texture features, is extracted from the hyperspectral data. The extraction process is as follows: Figure 3 As shown, since the images acquired by hyperspectral imaging are grayscale images, the grayscale values ​​of the three grayscale images at 438nm, 545nm, and 703nm need to be mapped to the B, G, and R channels respectively and merged into a pseudocolor image. Then, the RGB values ​​of the pseudocolor image are converted to HSV values. Finally, the first-order, second-order, and third-order color matrices are calculated to obtain nine color information. The gray-level gradient co-occurrence matrix is ​​used to extract 15 texture features of the image with the strongest reflectance in the hyperspectral image, including small gradient dominance, large gradient dominance, gray-level distribution non-uniformity, gradient distribution non-uniformity, energy, gray-level average, gradient average, gray-level variance, gradient variance, correlation, gray-level entropy, gradient entropy, mixing entropy, inertia, and inverse difference moment.

[0084] Acquire ginseng X-ray data and extract information from the ginseng X-ray images:

[0085] The online X-ray imaging system used in this research mainly consists of an X-ray emitter (53), a linear array detector (56), a conveyor belt (55), and a computer (52), i.e., a central computer. A schematic diagram of the device is shown below. Figure 5 As shown. To obtain complete and well-defined images, one ginseng sample (54) was placed at a certain interval on a conveyor belt (55) moving at a constant speed (12.0 m / min) each time. In addition, to reduce the impact of equipment instability on the experimental data, a standard sample of known density was placed next to each ginseng sample to collect grayscale images together.

[0086] Calculate the average grayscale value of the ginseng sample and the standard sample on each X-ray image. Finally, use formula D... i =D holder *G i / G holderThe grayscale value of ginseng is directly converted into a density value.

[0087] Obtain ginsenoside content data and perform statistical analysis on the ginseng content data:

[0088] After acquiring hyperspectral and X-ray data, the contents of ginsenosides Rg1, Re, and Rb1 in each batch of ginseng samples were determined according to the methods for determining ginseng sample content in the Chinese Pharmacopoeia. A UltiMate 3000 (Thermo Fisher, USA) high-performance liquid chromatography (HPLC) system was used, with a YMC-Pack ODS-A column (C18, 4.6 × 250 mm, 5 μm). The column temperature was maintained at 30℃. The contents of the three ginsenosides were detected at 203 nm with a flow rate of 1.0 mL / min and an injection volume of 10 μL. Each batch of samples was measured twice.

[0089] The HPLC chromatograms of three ginsenosides in the ginseng sample are shown below. Figure 6 As shown in the table, the content of ginsenoside Rg1+Re ranged from 0.43% to 0.87%, and the content of ginsenoside Rb1 ranged from 0.17% to 0.82%. This indicates that the selected samples are representative and have a wide content range. However, the content of ginsenosides varies greatly among different individual ginseng samples, therefore, quality control of individual samples is necessary. The distribution of ginsenoside content in all samples and the three datasets is shown in Table 1. The distribution of the validation set falls within the range of the calibration set, which is beneficial for model construction.

[0090] Table 1. Statistical analysis results of high performance liquid chromatography data.

[0091]

[0092] Based on the spectral data, the extracted spectral data is preprocessed as follows:

[0093] During the acquisition of spectral data, the spectra are often affected by factors such as noise interference from the instrument itself and the experimental environment, light scattering, baseline drift, sample surface unevenness, and radio scattering. Appropriate preprocessing of the spectral data is generally required. Five-point, first-order Savitzky-Golay (SG), SNV, MSC, 1stDer, and 2ndDer algorithms were used to preprocess the raw spectral data. Then, an RF model was built using the raw spectral data and the preprocessed data to select the optimal preprocessing method for Rg1+Re and Rb1. The default hyperparameters in the sklearn module were used to construct the RF model. The analysis results are shown in Table 2. After SG processing, the performance of the RF model in predicting Rg1+Re and Rb1 was slightly improved. Finally, SG was selected as the optimal data preprocessing method for Rg1+Re and Rb1. Figure 7 The spectral reflectance curves after SG processing are shown, and it is clear that the curves are smoother after preprocessing.

[0094] Table 2 shows the optimal preprocessing method for HSI spectral data selected using the RF model.

[0095]

[0096]

[0097] Based on the spectral data, characteristic wavelength selection is performed on the preprocessed spectral data:

[0098] Because spectral data contains a large amount of irrelevant and redundant information, it is necessary to extract some features for rapid detection to improve the model's prediction accuracy and computational speed. The SPA, CARS, UVE, and LASSO algorithms were selected to filter feature wavelengths on the training set after SG processing. The independent variables in the test set were also retained based on the results obtained from filtering the training set. The distribution of the feature wavelength filtering results is as follows: Figure 8 As shown in Table 3, an RF model was built using the dataset after feature wavelength filtering. Table 3 shows that the number of variables retained by different feature filtering algorithms varies significantly. By considering both model prediction performance and the number of retained variables, CARS and UVE were ultimately selected as the best feature filtering methods for Rg1+Re and Rb1, respectively.

[0099] Table 3. Optimal Feature Selection Method for HSI Spectral Data Using RF Model

[0100]

[0101]

[0102] Based on the image information, feature extraction is performed on the image information:

[0103] Since image features inevitably contain some redundant information, the SPA, CARS, UVE, and LASSO algorithms were used to filter out image features that are significantly correlated with Rg1+Re and Rb1. The distribution of the filtered feature variables is as follows: Figure 9As shown in the figure. Then, a PLSR model is built based on the selected feature variables. An appropriate number of latent variables can significantly improve the prediction performance of the PLSR model. The optimal number of latent variables is selected from 1 to 20 by performing five-fold cross-validation on the training set. For each number of latent variables, the PLSR model is cross-validated 100 times. The RMSE values ​​obtained from the cross-validation are averaged to obtain the root mean square error (RMSECV) of the model corresponding to this number of latent variables. Finally, the number of latent variables with the smallest RMSECV value is selected as the optimal number of latent variables. The number of feature variables selected by each feature selection algorithm and the prediction results of the optimized PLSR model are shown in Table 4. It can be seen that the performance of the PLSR model is improved to varying degrees after different feature selection algorithms. By comparing the prediction results of the PLSR model on the test set and the number of variables retained by the feature selection method, the LASSO algorithm is finally selected as the best selection method for hyperspectral image features. This indicates to some extent that the image feature information contains information related to the content of organic components. Although the prediction results were not satisfactory, the image information may provide some additional information for predicting the content of Rg1+Re and Rb1 in the sample.

[0104] Table 4. Optimal Feature Selection Method for HSI Image Features Using the PLSR Model

[0105]

[0106]

[0107] Use the prediction set to validate the performance of the optimal model:

[0108] Since the sample density information is only one-dimensional data, the simplest univariate linear regression algorithm was chosen to analyze the correlation between sample density and the contents of Rg1+Re and Rb1. The specific analysis results are shown in Table 5. The Rp2 values ​​of the LR model for Rg1+Re and Rb1 are 0.9059 and 0.7656, respectively, and the RPD values ​​are 3.3244 and 2.1065, respectively. This indicates a significant correlation between sample density and the content of its organic components.

[0109] The contents of three ginsenosides in the prediction set samples were predicted using optimized PLSR and RF models. Table 5 shows that the RF model built based on HSI spectral information achieved the best prediction results (Rg1+Re:Rp2=0.8236, Rb1:Rp2=0.7774), indicating that the spectral information from HSI contains richer information related to ginsenoside content. However, the RPD values ​​for all three models were less than 2.5. It can be seen that regression models built using single data are difficult to achieve more accurate predictions. The idea of ​​data fusion will subsequently be used to improve the predictive performance of the three ginsenosides.

[0110] Table 5. Prediction results of the optimized regression models constructed using different data.

[0111]

[0112]

[0113] Construction of data fusion model:

[0114] To verify whether ensemble data of spectral data, image feature data, and density data from ginseng's HSI images could optimize the prediction model and improve its accuracy, ensemble learning models Rg1+Re and Rb1 were developed based on selected feature spectral data, feature image data, and density values. A stacking ensemble learning method was used to combine each basic model, as illustrated in the diagram below. Figure 10 As shown in the diagram. First, the RF, PLSR, and LR models all use 10-fold cross-validation to predict on the training set and simultaneously predict on the test set. Then, the average of the results of each model's 10 predictions on the test set is calculated. Finally, the RF model is rebuilt using a new training set and used to predict on a new test set.

[0115] Table 6 shows the prediction performance of the constructed ensemble learning model on the training and test sets. (Scatter plot shown below) Figure 11 As shown in Table 4, the predicted Rp2 for Rg1+Re is 0.9712, RMSEP is 2.06%, and RPD is 6.0073, representing a significant performance improvement compared to the previous model built using individual data. The predicted Rp2 for Rb1 is 0.8433, RMSEP is 7.69%, and RPD is 2.5765. By comparing the prediction results obtained using a single learner in Table 4, it can be seen that the Stacking algorithm can effectively amplify the strengths of each basic learner and compensate for its weaknesses, thereby improving prediction accuracy and generalization performance. Compared to classical HPLC analysis methods, the method of this invention has the advantages of non-destructive testing and rapid analysis.

[0116] Table 6. Prediction results of the ensemble learning model

[0117]

[0118] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for non-destructive evaluation of ginseng quality based on hyperspectral and X-ray techniques, characterized in that, The method comprises the following steps: S1, obtaining hyperspectral data of a ginseng sample to be detected, and performing information extraction on the hyperspectral data; S2, obtaining X-ray data of the ginseng sample to be detected, and performing information extraction on the X-ray data; S3, obtaining high-performance liquid chromatography data of the ginseng sample to be detected; S4, based on the spectral data, pre-processing the extracted spectral data; S5, based on the spectral data, selecting characteristic wavelengths from the pre-processed spectral data; S6, based on image information, extracting features from the image information; the image information is image information in the hyperspectral data; S7, based on single data, establishing a ginseng quality evaluation model; S8, based on an integrated learning modeling method, integrating prediction results of multiple models to obtain ginseng quality information; In the step S7, the ginseng quality evaluation model is established based on single data, comprising: Based on the characteristic wavelength information in the hyperspectral data, a random forest regression model is established; Based on the characteristic image information in the hyperspectral data, a partial least squares regression model is established; Based on the density value information of the X-ray data, a linear regression model is established; In the step S8, based on the integrated learning modeling method, the prediction of multiple models is integrated to obtain the quality information of the ginseng, comprising: Inputting the data set of the preset characteristic variable into the pre-trained basic model; Using the Stacking integrated learning method to combine the prediction results of each basic model, and taking the quality information of the ginseng sample to be detected.

2. The method for non-destructive evaluation of ginseng quality based on hyperspectral and X-ray techniques according to claim 1, characterized in that, In the step S1, The acquisition of the hyperspectral data of the ginseng sample comprises the following steps: Selecting ginseng with relatively complete root head, main root and lateral root, and cleaning the surface soil for standby; The acquisition parameter setting of the hyperspectral imaging spectrometer is: the aperture size is f2.8, the exposure time is 19 ms, the distance between the lens of the spectrometer and the sample is 32.5 cm, and the hyperspectral data is obtained; The information extraction of the hyperspectral data of the ginseng sample comprises the following steps: Extracting spectral information from the hyperspectral data; Extracting image information from the hyperspectral data, including color information and texture feature information.

3. The ginseng quality non-destructive evaluation method based on hyperspectral and X-ray technology according to claim 2, wherein The color information of the ginseng sample comprises: 9 kinds of color information obtained from the first-order color matrix, the second-order color matrix and the third-order color matrix calculated based on the HSV value; The texture feature information of the ginseng sample comprises: 15 kinds of texture feature information obtained by calculating the small gradient advantage, the large gradient advantage, the gray distribution non-uniformity, the gradient distribution non-uniformity, the energy, the gray average, the gradient average, the gray variance, the gradient variance, the correlation, the gray entropy, the gradient entropy, the mixed entropy, the inertia and the inverse difference matrix of the gray level gradient co-occurrence matrix.

4. The method for ginseng quality nondestructive evaluation based on hyperspectral and X-ray technology according to claim 1, characterized in that, In the step S2, the acquisition of the X-ray data of the ginseng sample comprises the following steps: The emission angle of the X-ray emitter is 27.0°; The sample is placed on the conveyor belt at a certain interval, and the running speed of the conveyor belt is 12.0 m / min; Meanwhile, a standard sample with a known density is placed together with the ginseng sample to collect a gray-scale image; The information extraction of the X-ray data of the ginseng sample comprises the following steps: Calculating the average gray-scale values of the ginseng sample and the standard sample on each X-ray image; By formula D i = D holder * G i / G holder The gray value of ginseng is directly converted into density value, wherein D i is the absolute density value of the measured substance, D holder is the absolute density value of the standard substance, G i and G holder are the average gray values of the measured substance and the standard substance respectively.

5. The method for ginseng quality nondestructive evaluation based on hyperspectral and X-ray technology according to claim 1, characterized in that, In the step S4, the method for pre-processing the spectral data comprises: One or more of the following processes: removing noise data and removing data affected by the measurement environment.

6. The method for ginseng quality nondestructive evaluation based on hyperspectral and X-ray technology according to claim 1, characterized in that, In the step S5, the processed spectral data are subjected to feature wavelength selection, comprising: Selecting feature wavelengths from all wavelength bands in the spectral data based on a feature variable screening algorithm.

7. The method for ginseng quality nondestructive evaluation based on hyperspectral and X-ray technology according to claim 1, characterized in that, In the step S6, the image information is subjected to feature extraction, comprising: Screening feature image information from the image information based on a feature variable screening algorithm.

8. The method for ginseng quality nondestructive evaluation based on hyperspectral and X-ray technology according to claim 1, characterized in that, The single data specifically are three kinds of data: feature wavelength information in the hyperspectral data, feature image information in the hyperspectral data and density value information of the X-ray data.

Images