Angelica dahurica quality evaluation model method based on artificial intelligence technology
By constructing a fingerprint map and trait feature database of Angelica dahurica and combining it with a deep learning model, the problems of rapid, accurate and non-destructive quality identification of Angelica dahurica were solved, and intelligent and standardized quality detection of Angelica dahurica was achieved.
Patent Information
- Application Number
- CN202510671739.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-23
- Publication Date
- 2025-10-03
AI Technical Summary
Existing technologies make it difficult to achieve rapid, accurate and non-destructive identification of the quality of Angelica dahurica. Traditional methods are cumbersome and highly professional, making it difficult to meet the needs of efficient testing.
An artificial intelligence-based quality evaluation model for Angelica dahurica was adopted. By constructing a fingerprint map and trait feature database of Angelica dahurica, and using deep learning models such as CNN, SVM, MLP, Transformer and k-means for training and evaluation, rapid, non-destructive and accurate quality identification and grade classification of Angelica dahurica were achieved.
It achieves rapid, non-destructive, and accurate identification and grading of Angelica dahurica quality, simplifies manual grading steps, reduces judgment inaccuracy, and improves detection efficiency and accuracy.
Smart Images
Figure SMS_1 
Figure SMS_2 
Figure SMS_3
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of Angelica dahurica quality identification, and specifically to a method for Angelica dahurica quality evaluation model based on artificial intelligence technology. Background Art
[0002] Angelica dahurica (Fisch. ex Hoffm.) Benth. & Hook. f. ex Franch. & Sav.) is a perennial herbaceous plant belonging to the Apiaceae family, native to southeastern Qinghai, southern Gansu (the Tao River basin west of Minxian County), and northern Sichuan. It is insensitive to light and has strong adaptability. It has been designated a National Agricultural Product Geographical Indication by the Ministry of Commerce of the People's Republic of China. Modern pharmacological research has shown that Angelica dahurica is rich in coumarins and their glycosides, volatile oils, polysaccharides, and other active ingredients, exhibiting various pharmacological activities, including analgesic, anti-inflammatory, antioxidant, anti-tumor, and whitening properties. Clinically, traditional Chinese medicines using Angelica dahurica as a primary ingredient include Baizhi Nuan Gong Wan, Fufang Baizhi Tincture, and Baizhi Yangxin Wan. Furthermore, Angelica dahurica is widely used in food and cosmetics, with promising market prospects. Therefore, establishing efficient and accurate quality evaluation methods for Angelica dahurica is crucial to ensure its clinical efficacy and the sustainable development of the Angelica dahurica industry.
[0003] Multispectral imaging technology, as a mainstream rapid nondestructive testing method, can simultaneously reflect a sample's external quality characteristics and internal physical and chemical features, supporting sample location, qualitative, and quantitative analysis. It is currently widely used in the nondestructive identification of traditional Chinese medicines such as alfalfa, castor, and safflower. Deep learning technology, by learning from large amounts of data, can effectively simulate the human brain's neural network for analysis and judgment. It offers advantages such as high recognition rate and high efficiency, and shows promising application prospects in image recognition, authenticity verification, and origin traceability. Research has shown that deep learning models such as convolutional neural networks (CNNs), support vector machines (SVMs), and multilayer perceptrons (MLPs) have been widely used to identify traditional Chinese medicines such as Sichuan peppercorns, ginseng, and dragon bones. Therefore, combining multispectral imaging with machine learning can help achieve rapid and accurate evaluation of Angelica dahurica, improving the accuracy and effectiveness of Angelica dahurica quality assessment.
[0004] The quality identification of Angelica dahurica currently relies mainly on traditional methods of appearance-based identification and chemical analysis. However, these methods are not only highly specialized, cumbersome, and time-consuming, but also difficult to achieve rapid, non-destructive, and accurate identification. Therefore, a rapid, accurate, and non-destructive detection method is urgently needed. Summary of the Invention
[0005] In view of the above problems, the purpose of the present invention is to provide a method for evaluating the quality of Angelica dahurica based on artificial intelligence technology.
[0006] In order to solve the above problems, this application adopts the following technical solutions:
[0007] On the one hand, the quality evaluation and grade classification method of Angelica dahurica based on artificial intelligence technology specifically includes:
[0008] S1. Constructing fingerprints of different grades of Angelica dahurica using the content of imperatorin and isoimperatorin in the Angelica dahurica;
[0009] S2. Spectral images of the samples of different grades of Angelica dahurica were collected. After background removal and region of interest segmentation, a database of properties of different grades of Angelica dahurica was constructed. After processing the data in the database, optimized spectral data of Angelica dahurica was obtained.
[0010] S3. Use the optimized Angelica dahurica spectral data to train an artificial intelligence model to obtain the Angelica dahurica quality evaluation model.
[0011] The method for constructing the fingerprint of Angelica dahurica in step S1 is as follows: using HPLC to detect the effective ingredients of imperatorin and isoimperatorin in Angelica dahurica;
[0012] Specifically, the HPLC column used in this study was an Agilent Zorbax Eclipse XDB-C18 (250 mm x 4.6 mm, 5 μm). The mobile phase consisted of 0.1% formic acid in water (A) and acetonitrile (B); the gradient elution was as follows: 0-10 min, 10%-25% acetonitrile; 10-30 min, 25%-45% acetonitrile; 30-45 min, 45%-65% acetonitrile; 45-47 min, 65%-10% acetonitrile; 47-50 min, 10% acetonitrile; the flow rate was 1.0 mL / min; the injection volume was 10 μL; the column temperature was 30°C; and the detection wavelength was 254 nm.
[0013] Wherein, the Angelica dahurica in step S1 is divided into three grades, the first-grade Angelica dahurica has an imperatorin content of 1.062-1.875 mg / g and an isoimperatorin content of 0.846-1.440 mg / g; the second-grade Angelica dahurica has an imperatorin content of 0.710-0.997 mg / g and an isoimperatorin content of 0.547-1.05 mg / g; the third-grade Angelica dahurica has an imperatorin content of 0.132-0.626 mg / g and an isoimperatorin content of 0.286-0.677 mg / g;
[0014] Specifically, the detection spectrum of the data in step S2 is 365nm-970nm;
[0015] The characteristic feature database in step S2 includes types such as dimensional measurement features, color features, shape features, and texture features; specifically, the features include area, average distance, length, width, CIE Lab model-A value, CIE Lab model-B value, CIE Lab model-brightness, average color difference of standard color scale color values, color band local standard deviation profile, IHS chromaticity average, IHS brightness average, IHS average, average color band ratio, multi-color average, reflectance, reflectance average, reflectance average outside Blob, regional color band absorbance, regional color band component, regional color band average, external area threshold, contour roughness, basic width, compactness, compactness ellipse, perimeter, contour line local deviation, volume, basic image function-maximum pixel, measurement axis length, and regional color band local standard deviation;
[0016] Furthermore, the spectral data processing method described in step S2 is: normalizing, denoising, and enhancing the data to further optimize the quality and accuracy of the data. In this study, the Angelica dahurica sample data was divided into a training set: validation set: test set ratio of 8:1:1 for subsequent model training and evaluation;
[0017] Specifically, the device used in this study runs Windows 10, has an Intel i7-11700 CPU, and an NVIDIA GeForce RTX 3090 GPU. The development environment is Anoconda 3.5 and Python 3.8.
[0018] The Angelica dahurica quality grade classification model in step S3 includes CNN, SVM, MLP, t-SNE, Transformer, and k-means;
[0019] Among them, the Angelica dahurica quality evaluation and grade classification method according to claim 1 is characterized in that the model evaluation indicators in step S3 include recall rate, precision rate, accuracy rate, and F1score.
[0020] The beneficial effects of this application are:
[0021] 1. The present application provides a new technical means for rapid, non-destructive and accurate quality identification and grade classification of Angelica dahurica, realizing the standardization and intelligentization of Angelica dahurica quality evaluation.
[0022] 2. The present application provides a method for grading angelica starch, establishes an angelica fingerprint, and determines the content of the active ingredients imperatorin and isoimperatorin in angelica, thereby dividing angelica into 1 to 3 grades according to the range of its active ingredients. This simplifies the tedious steps of manual grading and reduces the inaccurate judgment caused by manual subjective judgment.
[0023] 3. This application establishes a database of Angelica dahurica trait characteristics and divides the Angelica dahurica sample data in this application to facilitate model training and evaluation. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 Figure 1 shows the chemical evaluation results of the Angelica dahurica starch of the present invention. (a) Overlay spectra of 30 batches of medicinal materials; (b) Control spectra; (c) Chromatogram of imperatorin; (d) Chromatogram of isoimperatorin;
[0025] Figure 2 The significance analysis of the contents of different components in 30 batches of samples is shown as follows: (a) significance analysis of imperatorin; (b) significance analysis of isoimperatorin;
[0026] Figure 3 Combination diagram of spectral processing method and average reflectance log value (a) spectral processing method and 19-wavelength fluorescence image; (b) spectral average reflectance log value;
[0027] Figure 4 Model architecture and performance evaluation: confusion matrix and cluster analysis (a) CNN model structure; (b) Transformer model structure; (c) CNN confusion matrix results; (d) MLP confusion matrix results; (e) Transformer confusion matrix results; (f) SVM confusion matrix results; (g) K-means results; (h) t-SNE results. DETAILED DESCRIPTION
[0028] The technical solution of the present application is further described in detail below with reference to the accompanying drawings, but the protection scope of the present application is not limited to the following.
[0029] Example 1
[0030] Angelica dahurica samples were purchased from the Chengdu Lotus Pond Traditional Chinese Medicine Market and identified by Professor Pei Jin of the School of Pharmacy at Chengdu University of Traditional Chinese Medicine as the dried root of Angelica dahurica (Fisch. ex Hoffm.) Benth. et Hook.f. or Angelica dahurica (Fisch. ex Hoffm.) Benth. et Hook.f. var. formosana (Boiss.) Shanet Yuan (Apiaceae). Detailed sample information is provided in Table 1.
[0031] Table 1 Sample information
[0032]
[0033] Take 0.2 g of the dry powder of this product (passed through No. 3 sieve), accurately weigh it, add 10 mL of 50% ethanol solution, weigh the mass, and ultrasonically treat it (power 500 W, frequency 40 kHz) for 1 hour. After cooling, add 50% ethanol solution to make up for the lost mass, shake well, centrifuge at 1500 r / min for 5 minutes, filter the supernatant through a 0.22 μm microporous membrane, and take the filtrate as the test solution.
[0034] Take appropriate amounts of imperatorin (batch number: CHB201201) and isoimperatorin (batch number: CHB210110) reference substances, accurately weigh them, and add methanol to dilute them into reference substance solutions with a mass concentration of 100 μg / ml.
[0035] The HPLC instrument used in this study was an UltiMate 3000 series (Thermo Fisher Scientific, USA). The chromatographic column was an Agilent Zorbax Eclipse XDB-C18 (250 mm x 4.6 mm, 5 μm). The mobile phase consisted of 0.1% formic acid in water (A) and acetonitrile (B); the gradient elution was as follows: 0–10 min, 10%–25% acetonitrile; 10–30 min, 25%–45% acetonitrile; 30–45 min, 45%–65% acetonitrile; 45–47 min, 65%–10% acetonitrile; 47–50 min, 10% acetonitrile; the flow rate was 1.0 mL / min; the injection volume was 10 μL; the column temperature was 30°C; and the detection wavelength was 254 nm.
[0036] Methodological investigations (precision, linearity, repeatability, and stability tests) were conducted on the Angelica dahurica sample solution. Correlation coefficients (r) and RSDs were calculated, with the RSD of the relative peak area being less than 3% as the standard. The final RSDs were all less than 3%, indicating good instrument performance, reliable experimental methods, and stable sample properties, fully meeting the standard requirements for quality testing.
[0037] Thirty batches of Angelica dahurica were prepared with test solutions according to the method described in "2.3.1" and injected according to the conditions described in "2.3.2" to obtain fingerprints for each batch of samples. The HPLC fingerprints of the Angelica dahurica samples were imported into the "Traditional Chinese Medicine Chromatographic Fingerprint Similarity Evaluation Software (2012 Edition)" using the S1 sample chromatogram as a reference. The averaging method was used with a time window width of 0.1 min. After multi-point calibration and Mark peak matching, the 30 HPLC fingerprints of Angelica dahurica and the reference were generated.
[0038] HPLC results of 30 batches of Radix Angelicae Dahuricae of the present invention are as follows Figure 1 shown. Figure 1 (a) Overlapping chromatograms of 30 batches of samples. Figure 1(b) After matching, eight common peaks were identified. By comparison with the reference sample chromatogram 1 (cd), two components were identified: imperatorin 6 and isoimperatorin 8. Using imperatorin 6, which has a larger peak area and better resolution, as the reference peak, the relative peak areas of the 30 batches of Angelica dahurica samples were calculated. The similarity between the chromatograms of the 30 batches of Angelica dahurica samples and the reference chromatogram was calculated using the 2012 version of the "Similarity Evaluation System for Chromatographic Fingerprints of Traditional Chinese Medicine" software. The relative peak areas and similarity results for the 30 batches of Angelica dahurica samples are shown in Table 2. Table 2 shows that the similarity of the 30 batches of Angelica dahurica samples ranged from 0.951 to 0.999, all greater than 0.9, indicating that the main chemical components were relatively consistent among the batches, while the content of the active ingredients varied significantly.
[0039] Table 2 Relative peak area and similarity analysis of 30 batches of Angelica dahurica samples
[0040]
[0041]
[0042] Legend: A represents first class, B represents second class, C represents third class; 1 to 8 are Figure 3 The eight peaks marked in
[0043] The results of the determination of imperatorin and isoimperatorin in 30 batches of Angelica dahurica are shown in Table 3. It can be seen from Table 3 that the content of first-class imperatorin is between 1.062-1.875, second-class is between 0.710-0.997, and third-class is between 0.132-0.626 mg / g. The content of first-class, second-class and third-class isoimperatorin is 0.846-1.440, 0.547-1.050 and 0.286-0.677 mg / g respectively. According to the provisions of the Pharmacopoeia, the content of imperatorin (C 0.004) in Angelica dahurica is 0.004-0.006 mg / g, calculated on a dry basis. 16 H 14 O4) shall not be less than 0.080%. According to the results in the table, all 10 batches of the first grade meet the requirements, three of the 10 batches of the second grade do not meet the requirements, and only one batch of the third grade meets the requirements of the pharmacopoeia. The results of the significance analysis of the content data of imperatorin and isoimperatorin are as follows: Figure 2 As shown in the figure, the average values of first-, second-, and third-grade imperatorin were 1.365 mg / g, 0.884 mg / g, and 0.422 mg / g, respectively, while the average values of first-, second-, and third-grade isoimperatorin were 1.117 mg / g, 0.764 mg / g, and 0.549 mg / g, respectively. The figure shows that the content of both components in Angelica dahurica is in the order of first-grade > second-grade > third-grade, with significant differences in imperatorin between different grades, indicating a positive correlation between grade and composition.
[0044] Table 330 Determination results of imperatorin and isoimperatorin in batches of Angelica dahurica
[0045]
[0046]
[0047] Example 2
[0048] After background removal and region of interest segmentation using typical discriminant analysis and threshold setting, the original image with background subtraction and the fluorescence image of the original image at 19 wavelengths are obtained. Figure 3 (a) As shown. It can be seen that the larger the wavelength, the greater the proportion of blue and the darker the color. The spectral average reflectance log value of the three grades of samples is shown in Figure 3 (b) It can be seen that the overall average reflectivity has a trend of first grade > second grade > third grade. Based on the characteristics of Angelica dahurica in Table 4, a database of shape characteristics of different grades of Angelica dahurica was constructed.
[0049] Table 4 Classification of characteristics of Angelica dahurica
[0050]
[0051]
[0052] Example 3
[0053] This application demonstrates two models, CNN and Transformer. For the specific model structure, see Figure 4 (ab). The results of the deep learning model evaluation indicators are shown in the table. It can be seen that the accuracy of the CNN model is 85.4%, the accuracy of the MLP is 80.6%, the accuracy of the SVM is 85.4%, and the accuracy of the Transformer is 88.7%. The Transformer model performs best in comprehensive comparison. The confusion matrix of the four deep learning models and the results of the two unsupervised learning models are shown in the figure below. Figure 4 ,from Figure 4 It can be seen that the confusion matrix of the Transformer model is the best, and the separation between the first and second classes of the two unsupervised models is not ideal. In general, the classification effect of the Transformer model is the best.
[0054] Table 5 Results of various evaluation indicators of deep learning models
[0055]
[0056] The above description is merely a preferred embodiment of the present application. It should be understood that the present application is not limited to the form disclosed herein and should not be construed as excluding other embodiments. Instead, the present application can be used in various other combinations, modifications, and environments and can be modified within the scope of the concept described herein through the above teachings or techniques or knowledge in related fields. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present application should be protected by the claims appended hereto.
Claims
1. A method for constructing a quality evaluation model of Angelica dahurica based on artificial intelligence technology, characterized in that: include: S1. Constructing fingerprints of different grades of Angelica dahurica using the content of imperatorin and isoimperatorin in the Angelica dahurica; S2. Spectral images of the samples of different grades of Angelica dahurica were collected. After background removal and region of interest segmentation, a database of properties of different grades of Angelica dahurica was constructed. After processing the data in the database, optimized spectral data of Angelica dahurica was obtained. S3. Use the optimized Angelica dahurica spectral data to train an artificial intelligence model to obtain the Angelica dahurica quality evaluation model.
2. The method according to claim 1, characterized in that The method for constructing the fingerprints of different grades of Angelica dahurica in step S1 is: using HPLC to detect the effective ingredients of imperatorin and isoimperatorin in Angelica dahurica.
3. The method according to claim 1, characterized in that The angelica dahurica in step S1 is divided into three grades, the imperatorin content of the first-grade angelica dahurica is 1.062-1.875 mg / g, and the isoimperatorin content is 0.846-1.440 mg / g; the imperatorin content of the second-grade angelica dahurica is 0.710-0.997 mg / g, and the isoimperatorin content is 0.547-1.05 mg / g; the imperatorin content of the third-grade angelica dahurica is 0.132-0.626 mg / g, and the isoimperatorin content is 0.286-0.677 mg / g.
4. The method according to claim 1, wherein The detection spectrum of the data in step S2 is 365nm-970nm.
5. The method according to claim 1, wherein The property feature database in step S2 includes size measurement features, color features, shape features, and texture features.
6. The method according to claim 1, characterized in that The Angelica dahurica spectral data processing method described in step S2 is: normalizing, denoising, and enhancing the data to further optimize the quality and accuracy of the data. In this study, the Angelica dahurica sample data was divided into a training set: validation set: test set ratio of 8:1:1 for subsequent model training and evaluation.
7. The method according to claim 1, characterized in that The intelligent model described in step S3 includes CNN, SVM, MLP, t-SNE, Transformer, and k-means.
8. The method according to claim 1, characterized in that The intelligent model evaluation indicators in step S3 include recall rate, precision rate, accuracy rate, and F1 score.