Method for variety identification and quality evaluation of citrus essential oil

Abstract

The invention discloses a method for variety identification and quality evaluation of citrus essential oil, and relates to the field of variety identification and quality evaluation of citrus essential oil. Raman spectrum data of various citrus essential oils in different volatilization stages are collected, and a training data set is constructed; the training data set comprises a sub-data set corresponding to each kind of citrus essential oil; based on the training data set, parameter optimization and performance comparison are carried out on the multiple classification models, and the optimal model is determined as a variety identification model for the multiple kinds of citrus essential oil; constructing quality evaluation models corresponding to different types of the citrus essential oil by utilizing the sub-data set of each type of the citrus essential oil; and performing variety identification and quality evaluation on a to-be-tested sample by using the variety identification model and the plurality of quality evaluation models. A variety identification model and a quality evaluation model constructed on the basis of the model can realize accurate identification of the variety of the citrus essential oil and quality evaluation of the citrus essential oil in different volatilization stages, so that the precision and efficiency of variety identification and quality control are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of citrus essential oil variety identification and quality assessment, and in particular to a method for citrus essential oil variety identification and quality assessment. Background Technology

[0002] Citrus essential oils are natural products extracted from citrus peels using methods such as cold pressing or steam distillation. They possess unique aromas, antioxidant properties, and various pharmacological effects, and are widely used in the food, pharmaceutical, and cosmetic industries. Common citrus essential oil varieties include bergamot, grapefruit, lemon, lime, mandarin orange, and sweet orange. Due to their complex composition, citrus essential oils are prone to volatilization and oxidation when exposed to air, light, or high temperatures, thus affecting their aroma characteristics, biological activity, and pharmacological effects. This instability significantly impacts the quality of essential oils, thus necessitating reliable analytical methods for quality assessment. Currently, gas chromatography-mass spectrometry (GC-MS) has become the gold standard for essential oil component analysis and authenticity identification due to its high sensitivity and accuracy. However, this technology suffers from high cost, complex operation, and long analysis time, limiting its application in rapid on-site testing. Summary of the Invention

[0003] The purpose of this application is to provide a method and equipment for the identification and quality assessment of citrus essential oil varieties. By combining Raman spectroscopy technology with machine learning algorithms, it is possible to achieve rapid and accurate classification of citrus essential oils without damaging the samples, and to dynamically monitor their quality changes, thereby significantly improving the accuracy and efficiency of citrus essential oil quality assessment.

[0004] To achieve the above objectives, this application provides the following solution: This application provides a method for identifying citrus essential oil varieties and assessing their quality, including: Raman spectral data of various citrus essential oils at different volatilization stages were collected to construct a training dataset; the training dataset includes a sub-dataset corresponding to each citrus essential oil. Based on the training dataset, the parameters of various classification models were optimized and their performance compared, and the optimal model was determined to be the variety identification model for various citrus essential oils. Using a subset of data for each type of citrus essential oil, a quality assessment model corresponding to different types of citrus essential oil was constructed. The variety identification model and multiple quality assessment models are used to identify the variety and assess the quality of the test samples.

[0005] According to the specific embodiments provided in this application, the following technical effects are disclosed: ①The method in this application combines Raman spectroscopy and machine learning to establish a variety identification model, which can quickly and accurately identify citrus essential oils with highly similar components; ② By establishing a quality assessment model, we can also dynamically assess the quality changes during the volatilization process and reveal the deterioration pattern of essential oils; ③The method described is fast, easy to operate, and requires no sample pretreatment, making it suitable for field applications. It overcomes the limitations of gas chromatography-mass spectrometry (GC-MS) technology, which is characterized by high cost, complex operation, and long detection cycle. ④ The proposed solution has good scalability and can be extended to the fields of authenticity verification and quality monitoring of other essential oils and natural products. Attached Figure Description

[0006] To more clearly illustrate the technical solutions in the embodiments of this application 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 this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0007] Figure 1 This is a flowchart of a method for identifying citrus essential oil varieties and assessing their quality.

[0008] Figure 2 Raman spectra of pure essential oils from six citrus fruits.

[0009] Figure 3 Characteristic peak analysis diagrams of six pure citrus essential oils.

[0010] Figure 4 Cluster analysis diagram of six pure citrus essential oils using the OPLS-DA algorithm.

[0011] Figure 5 Raman spectra of six citrus essential oils after 28 days of evaporation.

[0012] Figure 6 Characteristic peak analysis of six citrus essential oils after 28 days of volatilization.

[0013] Figure 7 Cluster analysis diagram of six citrus essential oils after 28 days of evaporation using the OPLS-DA algorithm.

[0014] Figure 8 Raman spectra and characteristic peak differences of bergamot essential oil at five evaporation time points.

[0015] Figure 9 Raman spectra and characteristic peak differences of grapefruit essential oil at five evaporation time points.

[0016] Figure 10 Raman spectra and characteristic peak differences of lemon essential oil at five evaporation time points.

[0017] Figure 11 The image shows the Raman spectra and characteristic peak differences of lime essential oil at five evaporation time points.

[0018] Figure 12 Raman spectra and characteristic peak differences of orange essential oil at five evaporation time points are shown in the figure.

[0019] Figure 13 Raman spectra and characteristic peak differences of sweet orange essential oil at five evaporation time points.

[0020] Figure 14 Raman spectra and characteristic peak differences of bergamot essential oil at five evaporation time points.

[0021] Figure 15 Raman spectra and characteristic peak differences of grapefruit essential oil at five evaporation time points.

[0022] Figure 16 Raman spectra and characteristic peak differences of lemon essential oil at five evaporation time points.

[0023] Figure 17 The image shows the Raman spectra and characteristic peak differences of lime essential oil at five evaporation time points.

[0024] Figure 18 Raman spectra and characteristic peak differences of orange essential oil at five evaporation time points are shown in the figure.

[0025] Figure 19 Raman spectra and characteristic peak differences of sweet orange essential oil at five evaporation time points.

[0026] Figure 20 Interpretability analysis diagrams for the citrus essential oil variety identification model on day 0 and day 28. Detailed Implementation

[0027] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0028] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] Raman spectroscopy is a fingerprinting technique based on molecular inelastic scattering. It offers advantages such as being non-contact, non-destructive, and requiring no sample preparation, allowing direct detection of essential oils. This technique can not only specifically identify essential oil varieties but also reflect overall compositional changes. Compared to traditional analytical methods, Raman spectroscopy is rapid, simple, and highly repeatable, making it particularly suitable for large-scale on-site testing and crucial for essential oil quality control. As a non-destructive, rapid, sensitive, and low-cost vibrational spectroscopy technique, Raman spectroscopy is gradually becoming a new tool for essential oil quality testing. It can provide information on molecular structure and chemical bonds, making it suitable for on-site testing. However, traditional Raman spectroscopy has limited distinguishing ability when dealing with citrus essential oils with similar compositions. Therefore, this embodiment combines Raman spectroscopy with machine learning methods to improve the accuracy of essential oil identification and classification.

[0030] In one exemplary embodiment, such as Figure 1 As shown, a method for identifying citrus essential oil varieties and assessing their quality is provided, comprising: Step 101: Collect Raman spectral data of various citrus essential oils at different volatilization stages to construct a training dataset. The training dataset includes a subset for each citrus essential oil. The Raman spectral data were acquired using a handheld Raman spectrometer. The citrus essential oils include: grapefruit, mandarin orange, sweet orange, bergamot, lemon, and lime.

[0031] Step 102: Based on the training dataset, optimize the parameters and compare the performance of various classification models to determine the optimal model as the variety identification model for various citrus essential oils. The classification models include: adaptive augmentation model, decision tree model, random forest model, support vector machine model, and extreme value gradient augmentation model.

[0032] Step 103: Using a subset of data for each type of citrus essential oil, construct a quality assessment model corresponding to different types of citrus essential oil.

[0033] Step 104: Use the variety identification model and multiple quality assessment models to identify the variety and assess the quality of the sample to be tested.

[0034] Step 101 specifically includes: For any given citrus essential oil, M experimental samples were prepared, resulting in N×M experimental samples. N represents the number of different types of citrus essential oils.

[0035] D time points are evenly selected from the preset period as target time points. Each time point represents the number of days of evaporation.

[0036] N×M experimental samples were placed in a dark environment for a pre-set period of volatilization experiment. During the volatilization experiment, when the number of volatilization days reached the target time point, Raman spectral data at X pre-set sites on each sample were acquired using a handheld Raman spectrometer, and Y Raman spectra were collected at each site.

[0037] Using the corresponding citrus essential oil type as the first label for each Raman spectral data, N×M×D×X×Y Raman spectral data were determined as the dataset.

[0038] Identify any citrus essential oil type as the current citrus essential oil type.

[0039] The target time point corresponding to each Raman spectral data point for the current citrus essential oil type is used as the second label, and the M×D×X×Y Raman spectral data points corresponding to the current citrus essential oil type are determined as the subset of data points for the current citrus essential oil type.

[0040] Step 102 specifically includes: training multiple classification models using Raman spectroscopy data of citrus essential oils as input and the first label as output. Based on parameter optimization and performance evaluation indicators, the optimal model is selected as the variety identification model for various citrus essential oils.

[0041] Step 103 specifically includes: determining any citrus essential oil type as the current citrus essential oil type. Using the Raman spectral data from the corresponding subset of the current citrus essential oil as input and the second label as output, train multiple classification models. Based on parameter optimization and performance evaluation metrics, select the optimal model as the quality assessment model corresponding to the current citrus essential oil. Iterate through all citrus essential oil types to obtain the quality assessment model corresponding to each type.

[0042] The parameter optimization method is grid search cross-validation. Performance evaluation metrics include: accuracy, precision, recall, and F1 score. The optimal model is the support vector machine model.

[0043] Step 104 specifically includes: inputting the Raman spectrum of the sample to be tested into the variety identification model to determine the type of citrus essential oil to which the sample belongs; selecting the quality assessment model corresponding to the type of citrus essential oil to which the sample belongs as the target quality assessment model; inputting the Raman spectrum of the sample to be tested into the target quality assessment model to obtain the evaporation days of the sample. The evaporation days are used to characterize the quality of the essential oil.

[0044] When N=6, M=3, D=5, X=3, Y=30, and the preset period is 28 days, the process for building the dataset is as follows: (1) Sample preparation: Six kinds of citrus essential oils, including grapefruit essential oil (CAS 8016-20-4), mandarin orange essential oil (CAS 8008-31-9), sweet orange essential oil (CAS 8008-57-9), bergamot essential oil (CAS 8007-75-8), lemon essential oil (CAS 8008-56-8), were purchased from Shanghai Maclean Biochemical Technology Co., Ltd., China, and lime essential oil (CAS 8008-26-2) was purchased from Shanghai Kee Chemical Technology Co., Ltd., China.

[0045] (2) Volatilization Experiment and Spectral Acquisition: 25 mL of each essential oil was placed in a 30 mL glass bottle (Baikman Biotechnology, China). Three biological replicates were set for each type, for a total of 18 samples. The samples were placed in a dark environment for a 28-day volatileization experiment. Raman spectra were collected at three different sites at five time points: 0, 7, 14, 21, and 28. Each site was sampled 30 times, and Raman spectral data of the six citrus essential oils at the five time points were finally obtained. The use of a handheld Raman spectrometer for rapid detection eliminates the cumbersome sample pretreatment steps in traditional gas chromatography, significantly shortens the detection time, and eliminates the need for large instruments, significantly reducing detection costs. It is particularly suitable for rapid on-site detection scenarios. The effectiveness of this technology mainly comes from the detection technology scheme using a portable Raman spectrometer. The preferred handheld Raman spectrometer is a handheld 785 nm Raman spectrometer with an excitation power of 300 mW, a spectral resolution of 10 cm⁻¹, a spectral acquisition time of 8 s, and a detection spectral range of 400-2300 cm⁻¹.

[0046] (3) Data clustering: The Raman spectral data obtained in step (2) are clustered using orthogonal partial least squares discriminant analysis. The clustering performance is evaluated by observing the clustering of Raman spectral sample points in the coordinate quadrants and using three indices: R2X, R2Y and Q2. Figures 2-7 Cluster analysis was performed on the Raman spectra of six citrus essential oils before and 28 days after evaporation. Figure 2 and Figure 5 The image shows the average spectrum of citrus essential oil on day 0 and day 28. The shaded area represents the standard error band of the spectrum. Figure 3 and Figure 6 The deconvolution spectrum represents the spectrum. The deconvolution spectrum can be used to analyze the characteristic peaks of the average Raman spectrum and further refine the differences in the spectral curves. Figure 4 and Figure 7 The graph shows the clustering analysis of citrus essential oils on day 0 and day 28 using the OPLS-DA algorithm. There are still overlapping parts in the graph, indicating that OPLS-DA cannot distinguish the essential oil spectra well. Therefore, a better method needs to be sought. Figures 14-19 Cluster analysis was performed on the Raman spectra of each citrus essential oil at five volatile stages. Figures 14-19In the diagram, (i) is the clustering analysis diagram of the OPLS-DA algorithm, and the gray shaded area represents the standard error band of the spectrum. (ii) is the interpretability analysis of the model.

[0047] (4) Data Identification: Different ensemble learning algorithms were used to automatically analyze all Raman spectral data from step (2). The Raman spectral dataset was divided using the `train_test_split` function, with 80% used as the training set and 20% as the external test set. The training set was further divided into an internal training set and an internal validation set in an 8:2 ratio for model optimization, training, and overfitting assessment. Five machine learning algorithms implemented in the scikit-learn data analysis library (version 1.3.0) were used to train the Raman spectral data: adaptive boosting algorithm, decision tree, random forest, support vector machine, and extreme gradient boosting. The hyperparameters of all models were optimized using GridSearchCV, and the optimal parameters were used for the final model training.

[0048] After building the dataset, the process also includes the classification model building process and the quality assessment model building process.

[0049] Multiple machine learning models were trained using a dataset to obtain classification models. Using Raman spectroscopy data as input and the first label as output, various machine learning models were trained on the dataset to obtain multiple undetermined classification models. Based on model performance evaluation metrics, the optimal undetermined classification model was selected as the classification model. Model performance evaluation metrics included accuracy, precision, recall, F1 score, and five-fold cross-validation. The classification model can identify six different species of citrus essential oils.

[0050] Classification model (essential oil type identification): Objective: To identify the types of essential oil samples (six categories in total: grapefruit, mandarin orange, sweet orange, bergamot, lemon, and lime).

[0051] Input data: Raman spectra from day 0 (before evaporation) were used as the dataset for the classification model, consisting of 6 categories of essential oils × 3 replicates × 90 measurements, totaling 1620 spectra. Raman spectra from day 28 (28 days after evaporation) were used as the dataset for another classification model, with the same spectra.

[0052] Labels: Each spectrum corresponds to the label of the essential oil type it belongs to.

[0053] Model Construction: Five machine learning models (adaptive augmentation, decision tree, random forest, support vector machine, and extreme gradient augmentation) were trained using the above data, and hyperparameters were optimized using GridSearchCV. The best-performing model was selected as the final classification model. By comparing the five machine learning models (including adaptive augmentation, decision tree, random forest, support vector machine, and extreme gradient augmentation), the support vector machine model was selected as the best-performing model for both the classification and quality assessment of citrus essential oils. It demonstrated high classification accuracy on both the training and test sets, outperforming other models and effectively improving the reliability of essential oil classification. This achievement is attributed to the selection of the support vector machine algorithm and the strategy of multi-model performance comparison and optimization.

[0054] Evaluation method: The model performance was evaluated using accuracy, precision, recall, F1 score, and five-fold cross-validation, and the optimal model (Support Vector Machine model) was selected. This model achieved the highest accuracy in identifying different types of citrus essential oils and also the highest accuracy in predicting the types of citrus essential oils. A method for identifying citrus essential oil types was constructed by combining Raman spectroscopy with a Support Vector Machine model.

[0055] Multiple machine learning models were trained using subsets of data corresponding to the current citrus essential oil varieties to obtain quality assessment models for each variety. Using Raman spectroscopy data as input and the second label as output, multiple machine learning models were trained using subsets of data corresponding to the current citrus essential oil varieties, resulting in several potential quality assessment models. Based on model performance evaluation metrics, the optimal potential quality assessment model was selected as the quality assessment model for the current citrus essential oil variety. The quality assessment models were then iterated over for all current citrus essential oil varieties to obtain quality assessment models for all varieties. These quality assessment models can identify the Raman spectra of citrus essential oils at different evaporation days. Combining Raman spectroscopy with machine learning techniques, they can be used to identify essential oil quality and detect whether it has deteriorated. Furthermore, the models can accurately identify the Raman spectra of six citrus essential oils at five evaporation days, indicating good generalization ability.

[0056] Quality assessment model (identifying the spectra of essential oils with different evaporation days): Objective: To identify the Raman spectra of essential oil samples at five evaporation time points (day 0, day 7, day 14, day 21, and day 28) to determine whether the essential oils have changed or deteriorated.

[0057] Input data: Raman spectra of six essential oils collected at five evaporation time points, totaling 32,400 spectra (6 categories × 5 time points × 3 repetitions × 4 sites × 90 measurements).

[0058] Labels: Each spectrum corresponds to a label for its evaporation time point.

[0059] Model building: Five machine learning models were used to train the data, and the performance of different models on the volatile days classification task was compared. The best model was selected as the quality evaluation tool.

[0060] Evaluation method: The same indicators as the classification model are used for evaluation, with particular attention to the model's ability to identify subtle Raman spectral changes between different days, thereby reflecting its sensitivity to the essential oil deterioration process.

[0061] (5) Data Evaluation: The test set samples divided in step (4) are selected for predictive ability testing. Different ensemble learning evaluation metrics are used to evaluate the model performance and select the best decision model, specifically including: accuracy, precision, recall, F1 score, and five-fold cross-validation. The generalization ability of each model is also verified through five-fold cross-validation. To ensure the generalization ability of the model, all models are subjected to five-fold cross-validation.

[0062] .

[0063] .

[0064] .

[0065] .

[0066] .

[0067] Where: Accuracy represents accuracy, Precision represents precision, Recall represents recall, and F1 score represents score, which are considered as evaluation metrics; TP, FP, TN, and FN represent true positive, false positive, true negative, and false negative, respectively, and these four cases constitute the model's judgment on the data; CV(5) represents the average performance evaluation value of the model obtained by 5-fold cross-validation; M(f i D i ) represents the model f obtained by training on the training subset in the i-th iteration. i In the i-th verification subset D i The evaluation indicators on the above; f i D represents the model trained in the i-th iteration (obtained by training the other four subsets); i Let represent the i-th subset of data used as the validation set; M represents a performance evaluation function.

[0068] For the classification model, the focus is on examining the model's ability to distinguish between the six types of essential oils; for the quality assessment model, the focus is on examining the model's ability to identify the temporal sequence of spectra at different volatile stages.

[0069] For the quality assessment model, the predictive performance at specific time points was further analyzed, and regression curves were plotted to demonstrate the model's prediction results for the test and training sets. The calibration R-squared and calibration root mean square error were calculated, indicating good model fit and high accuracy. The interpretability analysis of the citrus essential oil variety identification model on day 0 and day 28 is shown in the figure below. Figure 20 Table 1 shows the performance evaluation results of different machine learning algorithms for classifying the Raman spectra of citrus essential oils on day 0 and day 28, with the support vector machine algorithm showing the best performance. Tables 2-7 show the performance evaluation results of the quality assessment models for bergamot essential oil, grapefruit essential oil, lemon essential oil, lime essential oil, mandarin orange essential oil, and sweet orange essential oil.

[0070] Table 1. Comparison of performance evaluation results of different machine learning algorithms for classifying Raman spectra of citrus essential oil on day 0 and day 28. Table 2 Comparison of Performance Evaluation Results of Quality Assessment Models for Bergamot Essential Oil Table 3 Comparison of Performance Evaluation Results of Grapefruit Essential Oil Quality Assessment Model Table 4. Comparison of Performance Evaluation Results of Quality Assessment Models for Lemon Essential Oil Table 5. Comparison of Performance Evaluation Results of Quality Assessment Model for Lime Essential Oil Table 6. Comparison of Performance Evaluation Results of Quality Assessment Models for Orange Essential Oil Table 7 Comparison of Performance Evaluation Results of Quality Assessment Models for Sweet Orange Essential Oil (6) Model Interpretation: The best-performing ensemble learning model from step (5) is subjected to model interpretability analysis. The analysis is based on the class-specific weight vector (coef_). By averaging the weight vectors of all classes, the average contribution value of each feature is calculated, thus reflecting its overall importance in the classification process. Finally, these feature contributions are visualized in the form of a scatter plot, revealing the role of different Raman shifts in model decision-making.

[0071] Feature importance analysis was performed on both the best-performing classification model and the quality assessment model.

[0072] By extracting the category-specific weight vector (coef_) for each model, we analyze the key Raman shifts that the model relies on when distinguishing different essential oil types or evaporation time points. We visualize the feature contribution values ​​using scatter plots to aid in understanding the model's decision-making logic.

[0073] like Figures 14-20 By employing an interpretability algorithm (category-specific weight vector coef), key characteristic peaks in the Raman spectra of citrus essential oils were identified, clarifying the components that play a major role in model decision-making and enabling interpretable analysis of the model. This achievement stems from the design that combines Raman characteristic spectral lines with an interpretability algorithm. Therefore, this license not only maintains the accuracy of traditional gas chromatography detection but also achieves significant improvements in detection efficiency, cost control, and field adaptability, making it particularly suitable for various scenarios in the essential oil industry, such as raw material acceptance, process control, and market supervision.

[0074] Compared with the results obtained by gas chromatography, Raman spectroscopy can detect the component differences among six different citrus essential oils, which are reflected in spectral differences. Even after 28 days of evaporation, the six essential oils can still be distinguished. It can also detect component differences of the same essential oil at days 0, 7, 14, 21, and 28 of evaporation. Gas chromatography requires sample preparation, expensive instruments, and a long detection time. This method requires no sample preparation, and can be performed anytime and anywhere using a portable Raman spectrometer. The detection time is between 1 and 2 minutes, and the accuracy obtained by combining Raman spectroscopy with the SVM model is high. The Raman spectra of the six citrus essential oils at five evaporation time points (the gray shaded area represents the standard error band of the spectrum) and the difference analysis diagram of characteristic peaks are shown below. Figures 8-13 As shown. Figures 8-13 The three images in the upper left half show the main characteristic peaks of the spectrum; the one image in the lower left half shows the average spectrum at five evaporation time points; and the three images on the right... Figure 3 Spectral difference analysis of the main characteristic peaks, with letters representing differences between groups.

[0075] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0076] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for identifying citrus essential oil varieties and evaluating their quality, characterized in that, include: Raman spectral data of various citrus essential oils at different volatilization stages were collected to construct a training dataset; The training dataset includes a sub-dataset corresponding to each citrus essential oil; Based on the training dataset, the parameters of various classification models were optimized and their performance compared, and the optimal model was determined to be the variety identification model for various citrus essential oils. Using a subset of data for each type of citrus essential oil, a quality assessment model corresponding to different types of citrus essential oil was constructed. The variety identification model and multiple quality assessment models are used to identify the variety and assess the quality of the test samples.

2. The method for identifying citrus essential oil varieties and evaluating their quality according to claim 1, characterized in that, The Raman spectral data were acquired using a handheld Raman spectrometer.

3. The method for identifying citrus essential oil varieties and evaluating their quality according to claim 1, characterized in that, The types of citrus essential oils include: grapefruit essential oil, mandarin orange essential oil, sweet orange essential oil, bergamot essential oil, lemon essential oil, and lime essential oil.

4. The method for identifying citrus essential oil varieties and evaluating their quality according to claim 1, characterized in that, The classification models include: adaptive augmentation model, decision tree model, random forest model, support vector machine model, and extreme value gradient augmentation model.

5. The method for identifying citrus essential oil varieties and evaluating their quality according to claim 1, characterized in that, Raman spectral data of various citrus essential oils at different volatilization stages were collected to construct a training dataset, including: For any type of citrus essential oil, M experimental samples were prepared, resulting in N×M experimental samples; N is the number of types of citrus essential oils. D time points are evenly selected from a preset period as target time points; the time points are the number of days of evaporation. N×M experimental samples were placed in a dark environment for a pre-set period of volatilization experiment. During the volatilization experiment, when the number of volatilization days reached the target time point, Raman spectral data at X pre-set sites on each sample were acquired using a handheld Raman spectrometer, and Y Raman spectra were collected at each site. Using the corresponding citrus essential oil type as the first label for each Raman spectral data, N×M×D×X×Y Raman spectral data were determined as the dataset; Select any citrus essential oil type as the current citrus essential oil type; The target time point corresponding to each Raman spectral data point for the current citrus essential oil type is used as the second label, and the M×D×X×Y Raman spectral data points corresponding to the current citrus essential oil type are determined as the subset of data points for the current citrus essential oil type.

6. The method for identifying citrus essential oil varieties and evaluating their quality according to claim 5, characterized in that, N=6; M=3; D=5; X=3; Y=30; The preset period is 28 days.

7. The method for identifying citrus essential oil varieties and evaluating their quality according to claim 1, characterized in that, The variety identification model and multiple quality assessment models are used to identify the variety and assess the quality of the test samples, specifically including: The Raman spectrum of the sample to be tested is input into the variety identification model to determine the type of citrus essential oil to which the sample belongs; The quality assessment model corresponding to the type of citrus essential oil to which the sample to be tested belongs is selected as the target quality assessment model; The Raman spectrum of the sample to be tested is input into the target quality assessment model to obtain the number of days the sample has been volatilized; the number of days the sample has been volatilized is used to characterize the quality of the essential oil.

8. The method for identifying citrus essential oil varieties and evaluating their quality according to claim 5, characterized in that, Based on the training dataset, parameter optimization and performance comparison were performed on various classification models, and the optimal model was determined to be a variety identification model for various citrus essential oils, specifically including: Using Raman data of citrus essential oil as input and the first label as output, multiple classification models were trained. Based on parameter optimization and performance evaluation indicators, the optimal model was selected as the variety identification model for various citrus essential oils.

9. The method for identifying citrus essential oil varieties and evaluating their quality according to claim 8, characterized in that, Using a subset of data for each type of citrus essential oil, a quality assessment model corresponding to different citrus essential oil varieties was constructed, specifically including: Select any citrus essential oil type as the current citrus essential oil type; Multiple classification models are trained using the Raman spectral data of the current citrus essential oil in the corresponding subset as input and the second label as output. Based on parameter optimization and performance evaluation indicators, the optimal model is selected as the quality assessment model for the current citrus essential oil. By iterating through all types of citrus essential oils, a quality assessment model corresponding to each type of citrus essential oil is obtained.

10. The method for identifying citrus essential oil varieties and evaluating their quality according to claim 9, characterized in that, The parameter optimization method is grid search cross-validation. The performance evaluation metrics include: accuracy, precision, recall, F1 score, and five-fold cross-validation; The optimal model is the support vector machine model.

Images