Component analysis and quality evaluation method and application of katsumadai seed and katsumadai seed processed product based on HS-GC-MS and machine learning

Abstract

The invention provides an HS-GC-MS and machine learning-based component analysis and quality evaluation method and application of katsumadai seeds and processed products thereof, and belongs to the technical field of quality evaluation. According to the method, HS-GC-MS and electronic nose technologies are adopted, multivariate statistical methods such as principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) are combined, the accuracy and efficiency of processed product classification and sensory evaluation are improved, and odor characteristics and volatile component compositions of alpinia katsumadai from different producing areas and processed products of the alpinia katsumadai are deeply analyzed. The result shows that main volatile components in the katsumadai seed and the processed product thereof are similar in composition and mainly comprise 19 olefins, 8 alcohols, 3 aldehydes, 2 ketones, 2 esters, 1 ether and 1 nitrogen-containing compound, the olefins and the ethers account for 95% or above of the total amount of a sample, and the total amount of the nitrogen-containing compound accounts for 20% or above of the total amount of the sample. Alpha-caryophyllene, p-cymene, phellandrene and other olefin components and ether component eucalyptol are taken as main components.

Description

Technical Field

[0001] This invention relates to the field of quality evaluation technology, and in particular to a method and application for component analysis and quality evaluation of cardamom and its processed products based on HS-GC-MS and machine learning. Background Technology

[0002] Cardamom is the dried, nearly mature seed of the plant *Alpinia katsumadai* Hayata, belonging to the ginger family. It possesses properties that dry dampness, promote qi circulation, warm the stomach, and stop vomiting, and can be used for indigestion, nausea, vomiting, and diarrhea. Due to its unique aroma, cardamom is often used in food processing and cooking with pepper, star anise, and cinnamon to remove the odors of fish, pork, and other foods, and has wide applications in enhancing food flavor. Volatile components are the main source of cardamom's aroma and are also one of its important medicinal components. Modern research shows that cardamom volatile oil has a wide range of pharmacological activities, including protecting the gastric mucosa, anti-gastric ulcer, anti-inflammatory, and antibacterial effects. Studies on the composition of volatile components in cardamom have found that cardamom volatile oil mainly contains terpenes, alcohols, alkenes, and ethers, among which eucalyptol, p-cymene, α-phellandrene, farnesol, caryophyllene, linalool, borneol, 4-terpineol, and α-terpineol are relatively abundant. However, existing studies mostly focus on analyzing single batches of cardamom samples, lacking comprehensive evaluation and analysis of medicinal materials from different origins and batches, resulting in a lack of sufficient representativeness in the research results.

[0003] Furthermore, cardamom needs to be processed into different specifications of medicinal slices for clinical application, including raw cardamom, stir-fried cardamom, and ginger-processed cardamom. Raw cardamom is pungent, warm, and drying, and tends to dispel cold and dampness, and promote qi circulation and relieve stagnation; after stir-frying, its warming properties are moderated, and it tends to warm the spleen and astringe; after ginger processing, it is pungent and warm with the dispersing properties of ginger, and tends to warm the middle jiao and stop vomiting. As the main active ingredient in cardamom, the changes in the composition and content of volatile components before and after processing are inevitably closely related to the changes in efficacy. However, there is currently no systematic research on the effects of processing on volatile components and the correlation between component changes and changes in aroma and efficacy.

[0004] Electronic nose (E-nose) technology, developed in the 1990s, is a novel instrument for rapid food detection. It simulates biological olfaction, capturing the overall odor profile of a sample through a sensor array, thus quickly distinguishing odor differences between various samples. This technology boasts a short response time and fast detection speed, and has been widely applied in the food and traditional Chinese medicine fields. Headspace gas chromatography-mass spectrometry (HS-GC-MS) is an analytical method combining headspace sampling, gas chromatography separation, and mass spectrometry detection. It is commonly used for high-sensitivity analysis of volatile organic compounds (VOCs) such as aldehydes, esters, and heterocyclic compounds, achieving precise qualitative and quantitative analysis of key compounds. It is also widely used in the flavor analysis of traditional Chinese medicine materials and foods. The combined application of electronic nose technology and HS-GC-MS effectively integrates overall odor perception with precise chemical composition analysis, forming a highly efficient research pathway of "rapid initial screening - in-depth analysis." Machine learning (ML), with its capabilities for high-dimensional data processing and complex pattern recognition, can revolutionize the limitations of traditional component analysis through data-driven approaches. Examples include projected importance analysis (VIP) of alternative variables to screen core flavor compounds; deep learning models to capture compound synergies; and the integration of electronic nose / tongue data with sensory evaluations to construct "physical-chemical-sensory" correlation models, quantifying abstract flavor dimensions that are difficult for humans to describe. Combining electronic nose technology with HS-GC-MS and machine learning algorithms will facilitate a complete upgrade in flavor research, from data acquisition to intelligent decision-making. Summary of the Invention

[0005] The purpose of this invention is to provide a method and application for component analysis and quality evaluation of cardamom and its processed products based on HS-GC-MS and machine learning. HS-GC-MS and electronic nose technology are used as the method for analyzing the aroma characteristics and components of cardamom and its processed products for quality evaluation. At the same time, the effectiveness of the multimodal analysis method in robust classification and identification of cardamom processed products is demonstrated, and the effective correlation between component information and electronic nose sensory data is realized, as well as the visualization and quantification of electronic nose data, further improving the reliability of classification and quality evaluation.

[0006] To achieve the above-mentioned objectives, the present invention provides the following technical solution: This invention provides a method for component analysis and quality evaluation of cardamom and its processed products based on HS-GC-MS and machine learning, comprising the following steps: (1) Sample preparation: Obtain samples of cardamom and its processed products, including raw cardamom, roasted cardamom and ginger products; (2) HS-GC-MS analysis: The HS-GC-MS system was used to perform headspace-gas chromatography-mass spectrometry analysis on the samples of cardamom and its processed products to obtain qualitative and quantitative data of volatile components; (3) Machine learning model construction and classification: Using the qualitative and quantitative data of volatile components obtained in step (2) as feature input, a classification model is constructed using a variety of machine learning algorithms, and cardamom samples of different processed product types are classified. (4) Screening of key feature components: Combining the feature selection algorithm with the classification performance of the machine learning model, key feature components that can distinguish between raw cardamom and its different processed products are screened from all volatile components. (5) Quality evaluation: Based on the composition and content of the key characteristic components, establish standards for quality differentiation and evaluation of cardamom and its processed products, and conduct quality evaluation of cardamom and its processed products.

[0007] Preferably, step (1) further includes preprocessing the raw data obtained from HS-GC-MS analysis to remove compounds with missing values ​​higher than a set threshold, thereby forming a dataset for subsequent analysis.

[0008] Preferably, after step (1) and before step (2), the following step is also included: Electronic nose analysis: An electronic nose system was used to obtain the overall odor fingerprint information of cardamom and its processed products. And after step (4), the following step is also included: Data correlation analysis: Correlation analysis was performed on the key characteristic components that can distinguish between raw cardamom and its different processed products and the odor response values ​​obtained from electronic nose analysis to identify the key compounds that characterize the overall odor characteristics of the sample.

[0009] Preferably, the machine learning algorithm described in step (4) includes one or more of neural networks, random forests, support vector machines, k-nearest neighbors, and logistic regression.

[0010] Preferably, in step (4), the feature selection algorithm is combined with the machine learning algorithm in pairs to obtain the model, the model performance is evaluated, the feature subset that maximizes the model performance is obtained, and then the screening model is obtained. The intersection of the volatile compounds screened by the screening model is taken to screen the feature components that can be used for the quality control of cardamom and its processed products.

[0011] Preferably, the feature selection algorithm described in step (5) includes one or more of information gain, analysis of variance, chi-square test and ReliefF.

[0012] Preferably, the key feature components selected in step (5) include: (1) Key components that distinguish between raw and roasted cardamom: Eucalyptol, γ -Pterpinene, Phyllanthrene β-pinene, camphene, 3,6,6-trimethyl-bicyclo(3.1.1)hept-2-ene, (3E)-3-prop-2-enylcyclobutene and (1S,3S)-trans-4-carene; (2) Key components that distinguish between raw cardamom and ginger cardamom: Terpenes, p-cymene, 3,6,6-trimethyl-bicyclo(3.1.1)hept-2-ene, phellandrene, (R)-1-methyl-5-(1-methylvinyl)cyclohexene, γ -Pine, α-pinene, eucalyptol, isovaleraldehyde, camphene, β -pinene, myrcene, (1S,3S)-trans-4-carene, (3E)-3-prop-2-enylcyclobutene and α-terpinene; (3) Key components that distinguish between roasted cardamom and ginger cardamom: isovaleraldehyde, p-cymene, 2-heptanol, β-ocimene, phellandrene, linalool, camphene and terpinene.

[0013] Preferably, in step (5), Pearson correlation analysis is used for correlation analysis, and the key compounds characterizing the overall odor features of the sample include isovaleraldehyde, (3E)-3-prop-2-enylcyclobutene, methyl isovalerate, camphene, phellandrene, p-cymene, eucalyptol, β-ocimene, linalool and 4-phenyl-2-butanone.

[0014] Preferably, the conditions for the HS-GC-MS analysis in step (2) include: the chromatographic column is DB-5; the temperature program is: initial temperature 40℃ held for 2 min, temperature increased to 75℃ at 3℃ / min, temperature increased to 140℃ at 1.5℃ / min, temperature increased to 230℃ at 10℃ / min and held for 2 min, and finally temperature increased to 280℃ at 40℃ / min and held for 2 min.

[0015] The present invention also provides an application of the above-mentioned method for component analysis and quality evaluation of cardamom and its processed products based on HS-GC-MS and machine learning in the quality evaluation of cardamom and its processed products.

[0016] The beneficial effects of this invention compared to the prior art are as follows: (1) This invention employs HS-GC-MS and electronic nose technology, combined with multivariate statistical methods such as principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA), to improve the accuracy and efficiency of processed cardamom classification and sensory evaluation. It deeply analyzes the aroma characteristics and volatile component composition of cardamom and its processed products from different origins. Furthermore, the Orange tool is used to complete and remove data from the HS-GC-MS data, effectively ensuring the scientific validity of the data and providing reliable data support for subsequent chemometric analysis and machine learning applications. This avoids data issues affecting the accuracy of the analytical results and lays the foundation for subsequent research on volatile substances. The results showed that the overall aroma of cardamom differed before and after processing. The main volatile components of cardamom and its processed products were similar, including 19 alkenes, 8 alcohols, 3 aldehydes, 2 ketones, 2 esters, 1 ether, and 1 nitrogen-containing compound. Alkenes and ethers accounted for over 95% of the total sample, with α-caryophyllene, p-cymene, and phellandrene being the main alkenes, and eucalyptol being the main ether. Correlation analysis between HS-GC-MS data and electronic nose response values ​​further clarified the different aroma components of cardamom and its processed products. The results effectively revealed the influence of roasting and ginger processing on the volatile components of cardamom, providing a scientific basis for its quality control and the analysis of the material basis of its medicinal efficacy. It also indicates that the above two methods can be used as methods for analyzing the aroma characteristics and components of cardamom and its processed products, for quality evaluation, and for application in similar studies.

[0017] (2) This invention integrates feature selection methods with machine learning algorithms such as neural networks (NN), random forests (RF), support vector machines (SVM), k-nearest neighbors (KNN), and logistic regression (LR) to identify key distinguishing features of cardamom and its processed products. By using various statistical methods to mine the component features of the three processed products and the pairwise relationships between them, and combining the three statistical results, the characteristic components that characterize the differences between raw, roasted, and ginger-processed cardamom can be determined. Key characteristic components such as camphene, benzaldehyde, myrcene, and (1S,3S)-trans-4-carene are screened out, achieving accurate classification of different cardamom samples. The combination of ranking algorithms and machine learning enhances the understanding of subtle differences among the three processed cardamom products and demonstrates the effectiveness of multimodal analysis methods in robust classification and identification of processed cardamom products.

[0018] (3) The present invention further uses kernel density estimation plots and correlation analysis to identify key compounds that can characterize the electronic nose odor characteristics of cardamom samples of different specifications, including isovaleraldehyde, (3E)-3-prop-2-enylcyclobutene, methyl isovalerate, camphene, phellandrene, p-cymene, eucalyptol, β-ocimene, linalool and 4-phenyl-2-butanone, thereby realizing the effective correlation between component information and electronic nose sensory data, as well as the visualization and quantification of electronic nose data. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a peak area packing diagram of raw cardamom products in batches according to an embodiment of the present invention; Figure 2 The following are the analytical results of various compounds in three samples in the embodiments of the present invention, where A is the peak area packing diagram of various compounds in the three processed products; B is the differential compound with a VIP value greater than 1; Figure 3 This is a heatmap of the peak area of ​​the compounds in the three samples in the embodiments of the present invention; Figure 4 The following is a pie chart showing the compound analysis by HS-GC-MS in this embodiment of the invention, where A is the PLS-DA result of all samples, B is the displacement test result of all samples, and C is a pie chart showing the number of each type of compound. Figure 5 The selection of key features in this embodiment of the invention comprises 13 key features from 6 subsets; Figure 6 The following are the compound analysis results based on HS-GC-MS in the embodiments of the present invention, wherein A is the PCA result of raw cardamom vs. roasted cardamom, B is the PLS-DA result of raw cardamom vs. roasted cardamom, C is the displacement experiment result of raw cardamom vs. roasted cardamom, and D is the heat map of volatile compounds with VIP value > 1 in raw cardamom vs. roasted cardamom. Figure 7 The following are the compound analysis results based on HS-GC-MS in the embodiments of the present invention, wherein A is the PCA result of raw cardamom vs. ginger cardamom, B is the PLS-DA result of raw cardamom vs. ginger cardamom, C is the displacement experiment result of raw cardamom vs. ginger cardamom, and D is the heat map of volatile compounds with VIP value > 1 of raw cardamom vs. ginger cardamom. Figure 8The following are the compound analysis results based on HS-GC-MS in the embodiments of the present invention, where A is the PCA result of roasted cardamom vs. ginger cardamom, B is the PLS-DA result of roasted cardamom vs. ginger cardamom, C is the displacement experiment result of roasted cardamom vs. ginger cardamom, and D is the heat map of volatile compounds with VIP value > 1 of roasted cardamom vs. ginger cardamom. Figure 9 This embodiment of the invention uses machine learning-based HS-GC-MS compound analysis, where A represents the model evaluation result of raw cardamom versus roasted cardamom, B represents the feature subset of raw cardamom versus roasted cardamom, C represents the model evaluation result of raw cardamom versus ginger cardamom, D represents the feature subset of raw cardamom versus ginger cardamom, E represents the model evaluation result of roasted cardamom versus ginger cardamom, and F represents the feature subset of roasted cardamom versus ginger cardamom. Figure 10 The above are the electronic nose detection results in this embodiment of the invention, where A is the radar chart, B is the sensor response value of cardamom, C is the sensor response value of the three samples, D is the PCA result of the sensor response value, and E is the PLS-DA result of the three samples. Figure 11 This is the permutation test result based on electronic nose analysis in an embodiment of the present invention; Figure 12 This is a KDE distribution diagram of the training and test sets of the content of eight differentially expressed components in cardamom in an embodiment of the present invention; Figure 13 This is a KDE distribution diagram of the training and test sets of the content of six differentially expressed components in cardamom in an embodiment of the present invention; Figure 14 This is a compatibility analysis of HS-GC-IMS data and E-nose data in an embodiment of the present invention. Detailed Implementation

[0021] Various exemplary embodiments of the present invention will now be described in detail. This detailed description should not be considered as a limitation of the present invention, but rather as a more detailed description of certain aspects, features, and embodiments of the present invention.

[0022] It should be understood that the terminology used in this invention is merely for describing particular embodiments and is not intended to limit the invention. Furthermore, with respect to numerical ranges in this invention, it should be understood that each intermediate value between the upper and lower limits of the range is also specifically disclosed. Every smaller range between any stated value or intermediate value within a stated range, and any other stated value or intermediate value within said range, is also included in this invention. The upper and lower limits of these smaller ranges may be independently included or excluded from the range.

[0023] Unless otherwise stated, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. While only preferred methods and materials have been described herein, any methods and materials similar or equivalent to those described herein may be used in the implementation or testing of this invention. All references to this specification are incorporated by way of citation to disclose and describe methods and / or materials associated with those references. In the event of any conflict with any incorporated reference, the content of this specification shall prevail.

[0024] Various modifications and variations can be made to the specific embodiments described in this specification without departing from the scope or spirit of the invention, as will be apparent to those skilled in the art. Other embodiments derived from this specification will also be apparent to those skilled in the art. This specification and embodiments are merely exemplary.

[0025] The terms “include,” “including,” “have,” “contain,” etc., used in this article are all open-ended terms, meaning that they include but are not limited to.

[0026] Example 1 (1) Grouping The experiment was divided into three groups: raw cardamom group (S1-S6), stir-fried cardamom group (C1-C6), and ginger cardamom group (J1-J6), with 6 batches in each group. Detailed information is shown in Table 1.

[0027] Table 1. Source information of 6 batches of cardamom samples

[0028] (2) HS-GC-MS analysis and data analysis of volatile compounds (2.1) HS-GC-MS analysis The flavor profile of cardamom relies on its rich and complex volatile aromatic compounds rather than a single substance. Aroma analysis was performed using a gas chromatography-mass spectrometry (HS-GC-MS) system (Thermo Scientific™ TSQ 9610, USA) with a DB-5 column (30 m × 0.25 mm × 0.25 μm).

[0029] The three groups of samples were powdered separately, and 0.5 g of each batch of each group of samples was placed in a 20 mL headspace vial and incubated at 120°C for 30 minutes.

[0030] HS-GC-MS system parameter settings: Sample loop temperature set to 140℃, transfer line temperature set to 280℃. Helium (99.999%) was used as the carrier gas at a flow rate of 1 ml / min. The initial temperature was held at 40℃ for 2 min, then increased to 75℃ at 3℃ / min, to 140℃ at 1.5℃ / min, to 230℃ at 10℃ / min, held for 2 min, and finally increased to 280℃ at 40℃ / min, held for 2 min.

[0031] The headspace vial was injected with 1.0 mL of air at the top, the EI ion source temperature was 230 °C, the MS ionization potential was 70 eV, and full scan data acquisition mode was used. All analyses were performed in triplicate. The volatile components in each cardamom sample were analyzed by HS-GC-MS identification of mass spectra and RIs against the NIST 23 (National Institute of Standards and Technology, Gaithersburg, MD, USA) database. After removing compounds with missing values ​​higher than 50%, the results are shown in Table 2.

[0032] Table 2 HS-GC-MS Analysis of Volatile Compounds

[0033]

[0034] HS-GC-MS analysis showed that cardamom contains 36 common volatile components (Table 2), including 19 alkenes, 8 alcohols, 3 aldehydes, 2 ketones, 2 esters, 1 ether, and 1 nitrogen-containing compound. Among them, the alkenes mainly include α-caryophyllene, (R)-1-methyl-5-(1-methylvinyl)cyclohexene, p-cymene, phellandrene, trans-caryophyllene, myrcene, β-pinene, camphene, 3,6,6-trimethyl-bicyclo(3.1.1)hept-2-ene, eucalyptol, 4-phenyl-2-butanone, L-camphor, (-)-terpine-4-ol, linalool, etc. The ether is mainly eucalyptol.

[0035] (2.2) Multivariate data analysis and machine learning Multivariate data analysis and machine learning were performed using SIMCA version 14.1 (Umetrics, Sweden), Origin Pro 2025 (Origin Lab Corporation, MA, USA), Matlab R2023b (MathWorks, Natick, MA, USA), and Orange software, respectively.

[0036] (2.2.1) Analysis of differences in volatile components of raw cardamom Principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) were used for dimensionality reduction to screen for significantly different components (VIP > 1 and P < 0.05) in raw cardamom products from different origins. The results are as follows: Figure 1 As shown.

[0037] Figure 1 The results showed that samples from Wenchang, Hainan (S5) and Guangxi (S6) had higher total volatile compound content, followed by samples from Yunnan (S1). Samples from Zhanjiang, Guangdong (S3) and a certain location in Hainan (S4) had basically the same volatile compound content, while the sample from Sanya, Hainan (S2) had the lowest volatile compound content. PCA and PLS-DA results for raw cardamom did not clearly separate the six batches, indicating that the differences in volatile compounds among cardamom samples from different origins were not significant.

[0038] (2.2.2) Composition analysis of cardamom and its processed products A differential analysis of the component composition of cardamom and its processed products was conducted, and the results are as follows: Figure 2 , 3 As shown.

[0039] Figure 2 , 3 The results show that the three grades of cardamom samples have basically the same types of compounds, and the differences are mainly reflected in the content of the components. Figure 2 According to data from the analysis, olefins and ethers constitute the largest proportion of volatile components in cardamom, accounting for approximately 95% of all volatile compounds and covering the main aroma information of cardamom. Roasting and ginger processing significantly increase the content of volatile ethers in cardamom. Figure 2 (B) Heatmap analysis results of peak areas of 36 volatile substances ( Figure 3 The results showed that camphene, γ-terpinene, α-terpinene, 4-terpinene acetate, p-cymene, (R)-1-methyl-5-(1-methylvinyl)cyclohexene, 3,6,6-trimethyl-bicyclo(3.1.1)hept-2-ene, terpinene, phellandrene, β-pinene, eucalyptol, (1S,3S)-trans-4-carene, myrcene, β-ocimene, thujone, α-pinene, and (3E)-4,8-dimethyl-1,3,7-nonatriene were significantly higher in processed cardamom than in raw cardamom, and higher in ginger cardamom than in roasted cardamom.

[0040] Using the contents of the above 36 volatile components as independent variables and the contents of 3 types of cardamom samples as dependent variables, PLS-DA analysis can effectively distinguish between raw cardamom, roasted cardamom, and ginger products. Figure 4 China A, R 2 X = 0.632,R 2 Y = 0.835, Q 2 =0.687). 200 permutation tests demonstrated the reliability of the PLS-DA model. Figure 4 B, R 2 = 0.221, Q 2 = -0.429). Using VIP (Variable Importance in the Projection) > 1 as an indicator, 36 compounds were screened, and 14 compounds were identified as key components for distinguishing between raw cardamom, roasted cardamom, and ginger cardamom. These compounds include isovaleraldehyde, isopentyl alcohol, 2-methylbutanol, (3E)-3-prop-2-enylcyclobutene, methyl isovalerate, 2-heptanol, thujone, camphene, phellandrene, p-cymene, eucalyptol, β-ocimene, linalool, and 4-phenyl-2-butanone. Figure 4 As shown in C.

[0041] (2.2.3) Machine learning based on three processed cardamom products By combining feature selection algorithms (IG, ANOVA) with machine learning methods (SVM, RF), and utilizing machine learning algorithms NN (hidden layer size = 25, activation = ReLU, solver = Adam), RF (number of trees = 4, maximum depth = 5), SVM (kernel = RBF, C = 1.2), and KNN (number of neighbors = 4, metric = Euclidean, weight = uniform) to classify different types of cardamom samples, the results are as follows: Figure 5 As shown. The validation process using random sampling was performed 10 times, with a training set to test set ratio of 4:1. The model performance was evaluated based on the area under the curve, accuracy, F1 score, precision, and recall metrics (AUC > 0.5; CA ≥ 0.7; F1 ≥ 0.7; Prec ≥ 0.7; Recall ≥ 0.7; MCC ≥ 0.4). Information gain (IG), feature weights (Relief F), and X were used. 2 Chi-square test and analysis of variance (ANOVA) were used to rank the importance of eigenvalues. Machine learning and feature selection were combined to distinguish the subset of features that maximized model performance. This process aimed to evaluate the model's predictive accuracy on different subsets of features to determine the most effective subset, as shown in Table 3.

[0042] Thirteen key features were identified in the HS-GC-MS dataset. Figure 5The following components were selected: 2-methyl-3-buten-2-ol, isovaleraldehyde, 2-methylbutanal, isopentanol, 2-methylbutanol, (3E)-3-prop-2-enylcyclobutene, methyl isovalerate, styrene, 2-heptanol, camphene, benzaldehyde, myrcene, and (1S,3S)-trans-4-carene. These components can be used as characteristic components to distinguish the three processed cardamom products. Furthermore, the validation results show that the characteristic components selected according to this strategy can characterize the quality differences between different processed cardamom products and can be used for their quality control.

[0043] Table 3. Model evaluation results based on feature subsets

[0044] (2.2.4) Analysis of the differences in volatile components among the three processed cardamom products in pairwise comparisons. To further clarify the differences in volatile components among raw, roasted, and ginger-processed cardamom products, principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA) were performed on their volatile components after grouping. The results are as follows: Figures 6-8 As shown.

[0045] Figure 6 The results showed significant differences between raw and roasted cardamom, according to PCA results ( Figure 6 (A) shows that the principal components explain 84.3% of the total variance. R 2 X(cum) = 0.843, Q 2 (cum) = 0.474). The OPLS-DA model exhibits clearer clustering characteristics. Figure 6 (B) R 2 X(cum) = 0.688, R 2 Y(cum) = 0.937, Q 2 (cum) = 0.851, and the results of 200 permutation tests indicate that the model does not exhibit overfitting, thus demonstrating the reliability of the OPLS-DA model. Figure 6 C in the middle R 2 = 0.449, Q 2 = -0.869). For example... Figure 6 As shown in D, 17 compounds with a VIP>1 can serve as differentiating components between raw and roasted cardamom, including camphene, eucalyptol, and... β -pinene, phellandrene, γ-Pine, (3E)-3-prop-2-enylcyclobutene, 3,6,6-trimethyl-bicyclo(3.1.1)hept-2-ene, (1S,3S)-trans-4-carene, 4-phenyl-2-butanone, α -Pine resinene, p-cymene, terpinene α -pinene, (R)-1-methyl-5-(1-methylvinyl)cyclohexene, myrcene, 6,8a-dimethyl-3-propyl-2-yl-2,4,5,8-tetrahydro-1H-azaene, 4-terpinene acetate. The thermal map results in this group show that, except for 4-phenyl-2-butanone, trans-caryophyllene and... α Apart from caryophyllene, the contents of the other 14 compounds in roasted cardamom were higher than those in raw cardamom samples.

[0046] Figure 7 The results showed that both PCA and OPLS-DA could effectively distinguish between raw cardamom and ginger-processed cardamom. Figure 7 (A, B, C). Based on OPLS-DA analysis, a total of 20 differentially volatile compounds were screened, including those related to cymene, etc. γ -Pterpinene, Phyllanthrene α -pinene, 3,6,6-trimethyl-bicyclo(3.1.1)hept-2-ene, terpinene, eucalyptol, (1S,3S)-trans-4-carene, β -pinene, (R)-1-methyl-5-(1-methylvinyl)cyclohexene, camphene, isovaleraldehyde, (3E)-3-prop-2-enylcyclobutene, β -Osimerne, Myrcene α -Caryophyllene, 2-Heptaneol, α -terpinene, 4-terpinene acetate, trans-caryophyllene ( Figure 7 (D). Among them, cardamom contains a high content of 2-heptanol, α -Caryophyllene and trans-caryophyllene, and γ -Pterpinene, Phyllanthrene α The content of 17 compounds, including pinene, was significantly higher in ginger cardamom than in raw cardamom.

[0047] Figure 8 PCA results for roasted and ginger-processed cardamom were poor, but OPLS-DA results showed that they could be separated. R 2 X(cum) = 0.624, R 2 Y(cum)=0.93, Q 2 (cum) = 0.721). The results of 200 permutation tests demonstrate the reliability of the OPLS-DA model. Figure 8 C in the middle R2 = 0.444, Q 2 = -0.912). Based on OPLS-DA, 18 compounds with VIP > 1 were screened as key differentiators, including those for cymene, 2-heptanol, linalool, and β -Ocimethylene, phellandrene, camphene, terpinene, (R)-1-methyl-5-(1-methylvinyl)cyclohexene, isovaleraldehyde, 4-phenyl-2-butanone, 2-methyl-3-buten-2-ol, α -pinene, myrcene, trans-caryophyllene, 2-methylbutanol, thujone, 2-methylbutanal, isoamyl alcohol. (Examples: ...) Figure 8 As shown in Figure D, isoamyl alcohol, 2-methylbutanol, 2-heptanol, linalool, 4-phenyl-2-butanone, and trans-caryophyllene are concentrated in some roasted cardamom samples, while ginger cardamom is characterized by higher contents of isopentaldehyde, 2-methylbutanol, thujone, myrcene, terpinene, phellandrene, α-pinene, p-cymene, (R)-1-methyl-5-(1-methylvinyl)cyclohexene, and β-ocimene compounds.

[0048] (2.2.5) Based on machine learning methods, the characteristic components of the three processed cardamom products were determined in pairwise comparisons. Based on the data obtained from HS-GC-MS, sorting algorithms (IG, ANOVA, X) were used. 2 The ReliefF method, combined with machine learning methods (SVM, NN, kNN, RF, LR), was used to determine the feature components of three processed cardamom products in pairwise comparisons. The results are as follows: Figure 9 As shown in Tables 4-6.

[0049] Figure 9 Figures A and B show that, by taking the intersection of the four feature subsets with the highest overall scores from raw and roasted cardamom, eight feature components were identified, including eucalyptol. γ -Pterpinene, Phyllanthrene β -Pinene, camphene, 3,6,6-trimethyl-bicyclo(3.1.1)hept-2-ene, (3E)-3-prop-2-enylcyclobutene, and (1S,3S)-trans-4-carene can be used to distinguish between raw and roasted cardamom; all of the above compounds have a VIP>1 as determined by OPLS-DA analysis.

[0050] Figure 9 Figures C and D show that four feature subsets with high overall scores were selected from *Amomum villosum* and *Amomum tsao-ko*, identifying 16 characteristic components, including terpinene, p-cymene, 3,6,6-trimethyl-bicyclo(3.1.1)hept-2-ene, phellandrene, (R)-1-methyl-5-(1-methylvinyl)cyclohexene, γ-Pine resin, α -Pinene, eucalyptol, isovaleraldehyde, camphene, β -pinene, myrcene, (1S,3S)-trans-4-carene, (3E)-3-prop-2-enylcyclobutene, 2-methylbutanal, α -Pine resinene can be used to distinguish between raw cardamom and ginger cardamom; except for 2-methylbutanal, the rest are compounds with VIP>1 obtained by OPLS-DA analysis.

[0051] Figure 9 E and F show that three feature subsets with high overall scores were selected from roasted cardamom and ginger cardamom, identifying isovaleraldehyde, p-cymene, 2-heptanol, and β Eight characteristic components, including ocimene, phellandrene, linalool, camphene, and terpinene, were used to distinguish between roasted cardamom and ginger cardamom; all eight components were compounds with a VIP>1 as determined by OPLS-DA analysis.

[0052] In conclusion, the high consistency between the machine learning and OPLS-DA results also demonstrates the reliability of the machine learning model.

[0053] Table 4. Model evaluation results based on feature subsets: raw cardamom vs. roasted cardamom

[0054] Table 5. Model evaluation results based on feature subsets: Cardamom vs. Ginger Cardamom

[0055] Table 6. Model evaluation results based on feature subsets: roasted cardamom vs. ginger cardamom

[0056] (3) Intelligent sensory evaluation The aromatic characteristics of the three samples were evaluated using an electronic nose system (PEN 3 Airsence, Schwerin, Germany). 0.2 g of powder was placed in a 20 mL headspace vial and equilibrated at room temperature for 10 min.

[0057] Parameter settings: detection time 300 s, internal flow rate 400 mL / min, injection time 5 s, washing time 180 s. The sensors used by Enose are shown in Table 7.

[0058] Table 7 Electronic nose sensor performance

[0059] Electronic nose detection results as follows Figure 10 As shown.

[0060] Figure 10 The data from the A-test shows that sensors W5S (nitrogen oxides) and W1W (sulfides) have higher response values ​​for six batches of raw cardamom compared to other sensors, indicating that cardamom contains relatively high levels of nitrogen oxides and sulfides. Among these, sensor W5S showed the highest response value for the cardamom sample (S3) from Guangdong. Figure 10 (B) , the remaining differences are not significant, indicating that Guangdong has a higher content of nitrogen oxides than other producing areas. There is no significant difference in the composition of raw and processed products in terms of the types of components. Figure 10 (C)

[0061] PCA and PLS-DA analyses were performed using the response values ​​of each sensor of the electronic nose to cardamom and its processed products as variables. The results showed that raw cardamom and its different processed products all clustered together, and there was significant differentiation between different types of samples. Figure 10 (DE). Three principal components were extracted from both the raw and processed products, with a cumulative variance contribution rate of 93.5%. R 2 X=42.2%, R 2 X = 36.9% R 2 X=14.4%, R 2 X(cum) = 99.2%, Q2(cum) = 90.2%. Supervised PLS-DA analysis was performed based on PCA, and its R... 2 X(cum) = 0.982, R²Y(cum) = 0.859, Q²(cum) = 0.757, indicating that the prediction model is stable and has good predictive accuracy. The results of 200 permutation tests show that the model does not exhibit overfitting (R² = 0.982, R²Y(cum) = 0.859, Q²(cum) = 0.757). 2 =0.179, Q 2 =-0.506), which can be used to discriminantly analyze the inter-group differences in cardamom and its processed products. The results also indicate that roasting and ginger processing affect the aroma characteristics of cardamom.

[0062] (4) Correlation analysis between HS-GC-MS data and electronic nose response values Using the peak areas of 10 electronic nose sensor data points and 14 differential components (VIP>1, as mentioned in (2.2.2)) as sample features, data normalization was performed using Orange's "Preprocess" function. Then, the "Data Sampler" command was used to randomly divide the dataset into training and test sets at an 8:2 ratio. Finally, kernel density estimation (KDE) plots were generated using Matlab. Figure 12 , 13 As shown.

[0063] Figure 12 , 13 The results show that the distribution of each component in the training set and the test set is basically consistent, indicating that the model can be well applied to the test set.

[0064] Data from 10 electronic nose sensors and the peak areas of 14 differentially expressed components (VIP>1) were input into Origin Pro 2025 software for Pearson correlation analysis and visualization heatmaps were generated. Figure 13 As shown in the figure. An absolute value of the correlation coefficient r > 0.8 is considered a strong correlation, 0.5 to 0.8 is considered a moderate correlation, 0.3 to 0.5 is considered a low correlation, and < 0.3 is considered a negligible correlation.

[0065] Figure 14 The results showed a low positive correlation between W1C and 2-heptanol ( P <0.05), with phellandrene, β -Ocimene showed a significant negative correlation ( P <0.05, showing a moderate negative correlation with p-cymene and eucalyptol. P <0.05. W5S showed a low positive correlation with linalool ( P <0.05), with methyl isovalerate, phellandrene, p-cymene and β -Ociemene showed a moderate negative correlation ( P <0.05). W3C and phellandrene, β -Ocimene showed a significant negative correlation ( P <0.05, showing a moderate negative correlation with p-cymene and eucalyptol. P <0.05. W6S showed a moderate positive correlation with 4-phenyl-2-butanone (P2H2O). P <0.05, and showed a moderate negative correlation with camphene, eucalyptol and linalool. P <0.05). W5C and phellandrene, β -Ocimene showed a significant negative correlation ( P <0.05, showing a moderate negative correlation with p-cymene and eucalyptol. P <0.05. W1S showed a low positive correlation with 4-phenyl-2-butanone ( P <0.05, and showed a moderate negative correlation with (3E)-3-prop-2-enylcyclobutene and p-cymene ( P <0.05). W1W with methyl isovalerate, phellandrene, p-cymene and β -Ociemene showed a moderate negative correlation ( P <0.05%. W2S showed a low positive correlation with 4-phenyl-2-butanone ( P<0.05, and showed a moderate negative correlation with (3E)-3-prop-2-enylcyclobutene, camphene, and p-cymene. P <0.05. W2W showed a low positive correlation with 4-phenyl-2-butanone ( P <0.05, showing a low negative correlation with (3E)-3-prop-2-enylcyclobutene and camphene. P <0.05. W3S showed a moderate positive correlation with 4-phenyl-2-butanone (P2H2O). P <0.05, showing a significant negative correlation with camphene ( P <0.05, and showed a moderate negative correlation with phellandrene, p-cymene, eucalyptol and linalool. P <0.05).

[0066] In summary, isovaleraldehyde, (3E)-3-prop-2-enylcyclobutene, methyl isovalerate, camphene, phellandrene, p-cymene, eucalyptol, β-ocimene, linalool, and 4-phenyl-2-butanone can be used as key compounds for characterizing the electronic nose odor characteristics of cardamom samples of different specifications.

[0067] All statistical analyses were performed in triplicate. Data analysis and visualization were performed using Excel (Microsoft Corporation, USA) and Orange 3.38.1 (University of Ljubljana, Republic of Slovenia). Radar chart analysis, stacked histogram analysis, heatmap analysis, and Venn diagram analysis were performed using Origin Pro 2025 software (Origin Lab Corporation, USA) and the cloud-based biomedical data visualization system Hiplot (https: / / hiplot.com.cn / home / index.html).

[0068] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for the analysis and quality evaluation of ingredients of Amomum villosum Lour. and its processed products based on HS-GC-MS and machine learning, characterized in that, The method comprises the following steps: (1) sample preparation: obtaining samples of Amomum villosum and its processed products, including samples of Amomum villosum crude, fried and ginger processed products; (2) HS-GC-MS analysis: using a HS-GC-MS system to perform headspace-gas chromatography-mass spectrometry analysis on the samples of Amomum villosum and its processed products respectively, to obtain qualitative and quantitative data of volatile components; (3) machine learning model construction and classification: taking the qualitative and quantitative data of volatile components obtained in step (2) as feature input, using multiple machine learning algorithms to construct classification models, and classifying Amomum villosum samples of different processed product types; (4) key feature component screening: combining feature selection algorithms with the classification performance of the machine learning model to screen key feature components from all volatile components that can distinguish Amomum villosum crude and its different processed products; (5) quality evaluation: based on the composition and content of the key feature components, establishing a standard for quality differentiation and evaluation of Amomum villosum and its processed products, and performing quality evaluation of Amomum villosum and its processed products.

2. The method for analyzing and evaluating the components of Amomum tsjaubai and its processed products based on HS-GC-MS and machine learning according to claim 1, characterized in that, In step (1), the raw data obtained by HS-GC-MS analysis is also preprocessed to remove compounds with missing values higher than a set threshold, forming a data set for subsequent analysis.

3. The method for analyzing and evaluating the components of Amomum tsjaubai and its processed products based on HS-GC-MS and machine learning according to claim 1, characterized in that, After step (1) and before step (2), the following step is included: Electronic nose analysis: using an electronic nose system to obtain the overall odor fingerprint information of Amomum villosum and its processed product samples; And after step (4), the following step is included: Data correlation analysis: correlation analysis of key feature components that can distinguish Amomum villosum crude and its different processed products and odor response values obtained by electronic nose analysis to determine key compounds representing the overall odor characteristics of the samples.

4. The method according to claim 1 or 2, characterized in that, The machine learning algorithm in step (4) includes one or more of neural networks, random forests, support vector machines, k-nearest neighbors, and logistic regression.

5. The method according to claim 1 or 2, characterized in that, In step (4), the feature selection algorithm is combined with the machine learning algorithm to obtain a model, the performance of the model is evaluated, the feature subset that maximizes the performance of the model is obtained, and then the screening model is obtained. The intersection of the volatile compounds screened by the screening model is taken to screen the feature components that can be used for quality control of Amomum villosum and its processed products.

6. The method of claim 1 or 2, wherein, The feature selection algorithm in step (5) includes one or more of information gain, analysis of variance, chi-square test, and ReliefF.

7. The method according to claim 1 or 2, characterized in that, The key feature components screened in step (5) include: (1) key components for distinguishing Amomum villosum crude and fried Amomum villosum: cineol, γ - pinene, phellandrene, β - pinene, phellandrene, 3,6,6-trimethyl-bicyclo(3.1.1)hept-2-ene, (3E)-3- prop-2-enylcyclobutene and (1S,3S)-trans-4-carene; (2) key components for distinguishing Amomum villosum crude and ginger Amomum villosum: terpinolene, p-cymene, 3,6,6-trimethyl-bicyclo(3.1.1)hept-2-ene, phellandrene, (R)-1-methyl-5-(1-methylethenyl)cyclohexene, γ - pinolene, a-pinene, eucalyptol, isovaleraldehyde, camphene, β - pinene, myrcene, (1 S,3S)-trans-4-carene, (3E)-3-prop-2-enylcyclobutene and a-pinolene; (3) key components for distinguishing fried Amomum villosum and ginger Amomum villosum: isoamyl aldehyde, p-cymene, 2-heptanol, beta-ocimene, phellandrene, linalool, camphene, and terpinolene.

8. The method of claim 2, wherein, In step (5), Pearson correlation analysis is used for correlation analysis, and the key compounds representing the overall odor characteristics of the samples include isoamyl aldehyde, (3E)-3-prop-2-enyl cyclobutene, isoamyl acetate, camphene, phellandrene, p-cymene, eucalyptol, beta-ocimene, linalool, and 4-phenyl-2-butanone.

9. The method of claim 1, wherein, The conditions of the HS-GC-MS analysis in step (2) include: the chromatographic column is DB-5; the temperature rising program is: the initial temperature is 40℃, keeping for 2 min, rising to 75℃ at a rate of 3℃ / min, rising to 140℃ at a rate of 1.5℃ / min, rising to 230℃ at a rate of 10℃ / min and keeping for 2 min, and finally rising to 280℃ at a rate of 40℃ / min and keeping for 2 min.

10. Application of the method for analyzing and evaluating the quality of Amomum villosum and its processed products based on HS-GC-MS and machine learning according to any one of claims 1-9 in the quality evaluation of Amomum villosum and its processed products.

Images