Method for quickly identifying production place of dried orange peel on site

Abstract

The invention discloses a method for quickly identifying the producing area of dried orange peel on site, which comprises the following steps of: combining an ART-2a model with a random forest model, detecting the content of characteristic elements in dried orange peel samples in known different producing areas, and performing clustering analysis on data in advance through the ART-2a model; and further training a random forest model by using the clustering data to construct a pericarpium citri reticulatae producing area prediction model, wherein the pericarpium citri reticulatae producing area prediction model can realize accurate source tracing of the pericarpium citri reticulatae producing area. Meanwhile, element data of a pericarpium citri reticulatae sample is obtained on site by combining a portable X-ray fluorescence spectrum instrument, pericarpium citri reticulatae producing area information can be rapidly obtained on site by utilizing an ART-2a model and a pericarpium citri reticulatae producing area prediction model, the recognition accuracy of the method for the pericarpium citri reticulatae in the Xinhui producing area reaches 96% or above, the damage to the pericarpium citri reticulatae is small, large precise instruments are not needed, and the method is easy to implement. The method has the advantages of high efficiency, accuracy and the like and is suitable for pericarpium citri reticulatae producing area authentication and quality monitoring.

Description

Technical Field

[0001] The invention relates to the technical field of Chinese medicinal material detection, and in particular to a method for rapid on-site identification of the origin of tangerine peel. Background Art

[0002] Guang Chenpi, derived from the Rutaceae plant citrus ( Citrus reticulata Blanco) and its cultivars. In traditional Chinese medicine, tangerine peel usually refers to citrus peel that has been dried in the sun or dried at low temperature. Guangdong tangerine peel specifically refers to high-quality tangerine peel produced in Guangdong Province, China, especially the top-grade tangerine peel produced in Xinhui, Guangdong, so it is sometimes also called Xinhui tangerine peel. Its medicinal effect and economic value are closely related to the place of origin. There are tangerine peels made from other origins such as Hunan, Jiangxi, Guangxi, etc. and other varieties of tangerine peels on the market, which are called miscellaneous peels. Because Xinhui tangerine peel has the highest price, there is often a phenomenon of tangerine peels from other origins being impersonated as Xinhui tangerine peels. Therefore, being able to quickly identify the origin of tangerine peels on site is of great significance to protecting the rights and interests of consumers.

[0003] Traditional methods for identifying the origin of dried tangerine peel mainly rely on sensory judgment and chemical composition analysis. Sensory judgment is subjective, depends on the experience of the evaluator, and has uncertainty and randomness. Chemical analysis such as gas chromatography-mass spectrometry (GC-MS), high performance liquid chromatography (HPLC), infrared spectroscopy (IR), etc. combined with chemometric methods can accurately distinguish Xinhui dried tangerine peel from other origins, but they all have shortcomings such as long sample testing time, large and expensive analytical instruments, more consumption of harmful organic reagents, and strong destructiveness to dried tangerine peel. In addition, large precision instruments are large in size, expensive, and difficult to maintain and use. Therefore, a method for identifying dried tangerine peel from Xinhui and non-Xinhui production areas that does not require large precision instruments, is less destructive to dried tangerine peel, and has high accuracy is developed. Summary of the invention

[0004] The purpose of the present invention is to overcome the above defects and shortcomings in the prior art and provide a method for identifying the origin of tangerine peel.

[0005] The above-mentioned object of the present invention is achieved through the following technical solutions: The present invention first provides a method for quickly identifying the origin of tangerine peel on site, comprising the following steps: S1. The characteristic element contents of tangerine peel samples from known Xinhui and non-Xinhui production areas were input into the ART-2a model for cluster analysis to obtain the origin cluster labels; the characteristic elements were P, S, Cl, K, Ca, Cr, Mn, Fe, Cu and Zn; S2. Use the sample data and origin cluster labels after cluster analysis in S1 as data sets to train the random forest classification model and build a prediction model for the origin of tangerine peel; S3. The above characteristic element data of the tangerine peel sample of unknown origin is input into the ART-2a model to obtain the origin clustering label, and then the characteristic element data of the tangerine peel sample of unknown origin and the corresponding origin clustering label are input into the tangerine peel origin prediction model to identify whether the tangerine peel sample is tangerine peel produced in Xinhui District.

[0006] ART-2a is a neural network clustering algorithm, a variant of the self-organizing feature map network. ART stands for "Adaptive Resonance Theory" and was originally proposed by Stephen Grossberg. ART-2a is specifically designed to process continuous data. Its design allows online learning, that is, as new data arrives, the model can dynamically adjust and generate new classes or adjust existing classes. ART-2a is widely used in image recognition, speech processing, bioinformatics, signal processing and other fields, especially in situations where real-time clustering and recognition of constantly updated data is required. When processing new data that is constantly input, ART-2a does not need to predefine the number of categories, can dynamically adapt to new patterns, and can maintain good clustering effects in the presence of noise interference.

[0007] The present invention combines ART-2a and random forest models for the prediction of tangerine peel origin for the first time. Specifically, the present invention uses several common algorithm models for origin prediction to process tangerine peel sample data in a single process or in combination with each other, and models them to compare their prediction accuracy. The results show that the combination of ART-2a and random forest models can greatly improve the prediction accuracy. The present invention establishes an ART-2a cluster analysis model and a tangerine peel origin prediction model based on a random forest algorithm. The prediction accuracy of the tangerine peel origin in Xinhui District can reach 96.5%, which can achieve rapid and accurate identification of whether the tangerine peel is from the Xinhui District.

[0008] Preferably, the Xinhui production area includes Huicheng Street, Daze Town, Siqian Town, Luokeng Town, Shuangshui Town, Yamen Town, Gujing Town, Sanjiang Town, Muzhou Town, and Da'ao Town.

[0009] Furthermore, the characteristic element content in step S1 is obtained by X-ray fluorescence spectroscopy analysis.

[0010] Preferably, the X-ray fluorescence spectroscopy analysis can be performed using a portable X-ray fluorescence spectrometer, and the characteristic data of the tangerine peel sample can be obtained by using on-site detection technology to achieve on-site rapid identification of the origin of the tangerine peel.

[0011] Furthermore, the cluster analysis in step S1 is to perform cluster analysis on the characteristic elements of tangerine peel samples from known Xinhui and non-Xinhui production areas according to the similarity of element content composition, obtain the relative content data of characteristic elements of tangerine peel samples from different production areas, and obtain the production area clustering label.

[0012] Furthermore, the data set in step S2 is divided into a training set and a test set in a ratio of 8:2.

[0013] Furthermore, the parameters of the training setting in step S2 are: the maximum tree depth "max depth" is: (10, 20, 30) and the number of trees "n estimators" is (10, 50, 100, 200), and the maximum number of features "max features" is (auto).

[0014] Furthermore, the characteristic element data of the tangerine peel sample of unknown origin in step S3 is obtained by detection using a portable x-ray fluorescence spectrometer. The portable x-ray fluorescence spectrometer does not require the use of harmful organic reagents, is less destructive to the tangerine peel, and can also achieve rapid on-site detection to obtain the characteristic element content data of the tangerine peel.

[0015] Furthermore, in step S3, the tangerine peel sample of unknown origin is crushed and then tested.

[0016] Preferably, the pulverization is pulverization to obtain powder with a particle size of 160 to 200 μm.

[0017] The present invention provides a device for identifying the origin of tangerine peel, comprising: An acquisition module is used to obtain the characteristic element contents of tangerine peel samples from known Xinhui and non-Xinhui production areas and tangerine peel samples from unknown production areas; the characteristic elements are P, S, Cl, K, Ca, Cr, Mn, Fe, Cu and Zn; A modeling module is used to perform cluster analysis on the characteristic element contents of tangerine peel samples from known Xinhui and non-Xinhui production areas through the ART-2a model, obtain origin clustering labels, use the clustered sample data and origin clustering labels as data sets to train a random forest classification model, and build a tangerine peel origin prediction model; The prediction module is used to analyze the characteristic elements of the tangerine peel sample of unknown origin through the ART-2a model and the tangerine peel origin prediction model, and identify whether the tangerine peel sample of unknown origin is tangerine peel produced in Xinhui District.

[0018] The present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, any of the above methods is executed.

[0019] The present invention also provides the use of any of the above-mentioned computer-readable storage media in the rapid identification of Xinhui dried tangerine peel on site. The portable X-ray fluorescence spectrometer can be used to quickly detect the characteristic element content of dried tangerine peel in multiple scenarios, and running the above-mentioned computer program can quickly and accurately identify the origin of Xinhui dried tangerine peel at any location.

[0020] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a method for rapid on-site identification of the origin of dried tangerine peel, which combines an ART-2a model with a random forest model, detects the characteristic element content in dried tangerine peel samples from known Xinhui and non-Xinhui production areas, performs cluster analysis on the data in advance through the ART-2a model, and further trains the random forest model with the cluster data to construct a prediction model for the origin of dried tangerine peel, which can accurately trace the origin of dried tangerine peel. At the same time, the element data of the dried tangerine peel sample is obtained on-site in combination with a portable X-ray fluorescence spectrometer, and the ART-2a model and the prediction model for the origin of dried tangerine peel are used to quickly obtain the information on the origin of dried tangerine peel. The method has an accuracy rate of more than 96% for identifying dried tangerine peel from the Xinhui production area, is less destructive to the dried tangerine peel, does not require large-scale precision instruments, has the advantages of high efficiency and accuracy, and is suitable for the certification of the origin of dried tangerine peel and quality monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is the distribution map of tangerine peel sample collection points in Xinhui area.

[0022] Figure 2 This is the OPLS-DA model analysis diagram.

[0023] Figure 3 This is the PCA analysis result diagram. DETAILED DESCRIPTION

[0024] The present invention is further described below in conjunction with the accompanying drawings and specific examples, but the examples do not limit the present invention in any form. Unless otherwise specified, the reagents, methods and equipment used in the present invention are conventional reagents, methods and equipment in the art.

[0025] Unless otherwise specified, the reagents and materials used in the following examples are commercially available.

[0026] Example 1 1. Sample collection and preparation: Tangerine peel samples from different origins were collected and sampled from the Guangxi Zhuang Autonomous Region and Xinhui District, Jiangmen City, Guangdong Province. The sample collection points in Xinhui District are distributed as follows: Figure 1 110 dried tangerine peel samples from Xinhui and non-Xinhui production areas were selected for drying and grinding. The grinding process was grinding and crushing with a rechargeable portable grinder, and then screening with an 80-mesh standard sieve to prepare samples suitable for portable X-ray fluorescence spectrometry analysis.

[0027] 2. Portable XRF instrument testing and data collation: Take 1.5 g of powder sample and put it into a sample cup with a diameter of 32 mm for XRF detection to obtain the content data of 10 characteristic elements in the sample. The characteristic elements are elements with higher content in the tangerine peel sample. According to the preliminary detection data of the present invention, the selected characteristic elements are P, S, Cl, K, Ca, Cr, Mn, Fe, Cu and Zn.

[0028] 3. Use the artificial neural network algorithm of sampling adaptive resonance theory (ART-2a) on MATLAB software for analysis. Input the content data of characteristic elements of each sample obtained in step 2. ART-2a extracts the content of characteristic elements of tangerine peel samples from each producing area from the data. Use the ART-2a model to cluster the element data to generate several groups of sample clusters. ART-2a will classify similar samples into the same category based on the similarity of elemental composition.

[0029] 4. Random forest model building: The clustering results of ART-2a are used as features, and the clustering results are labeled in combination with the sample origin information to obtain the origin clustering label of ART-2a. The origin clustering label of ART-2a will become one of the input features of the random forest classification model. Use the clustered sample data (clustered characteristic element data) and the origin clustering label to train the random forest model. The random forest model will learn the relationship between each element feature and the origin to enhance classification accuracy. The random forest model provides the contribution of each element feature to distinguishing the origin, helping to identify the key elements that affect the origin classification. Specifically, the process of training the random forest model is as follows: ① Divide the data set into training set and test set, and adjust the ratio according to 8:2, 7:3, and 6:4. The results show that the ratio of training set to test set is 8:2, and the effect is best. The effects of 7:3 and 6:4 are far apart. Therefore, the training set and test set are trained according to the ratio of 8:2. ② Use RandomForestClassifier to train the model and set the hyperparameters, specifically: the maximum tree depth "max depth" is: (10, 20, 30) and the number of trees "n estimators" is (10, 50, 100, 200), and the maximum number of features "max features" is (auto); ③After training the model, predict the training set and test set respectively to generate prediction results.

[0030] 5. Use a portable XRF instrument to test tangerine peel samples of unknown origin to obtain element content data. Input the characteristic element data of tangerine peel samples of unknown origin into the ART-2a model to obtain the origin cluster label, so that the unknown sample can be classified into a certain cluster. Then, the characteristic element data of the unknown sample and the corresponding origin cluster label are input as input features into the trained random forest model. The random forest model will predict the origin of the sample based on the existing classification rules. The random forest outputs the probability distribution of the sample that may belong to each origin, and selects the origin category with the highest probability as the final prediction result to improve the confidence of the classification.

[0031] The ART-2a algorithm can normalize the content of different elements and use the relative concentration ratio of elements to reduce the error caused by individual differences; extract the most important element characteristics of each production area, compare the element composition of tangerine peel samples from unknown production areas with it, preliminarily determine the production area range, and then input the element composition data into the trained random forest model to jointly determine the tangerine peel production area and improve the prediction accuracy. The results show that the prediction accuracy of tangerine peel from Xinhui production area can reach 96.5%, with a Kappa coefficient of 0.896, and the prediction accuracy of tangerine peel from non-Xinhui production areas is 93.9%, with a Kappa coefficient of 0.783.

[0032] In summary, the present invention can detect tangerine peel samples of unknown origin by using a portable XRF instrument, and combine the trained ART-2a and random forest models for analysis and prediction, so as to realize the rapid and accurate identification of tangerine peel from the Xinhui production area on site, with an accuracy rate of up to 96.5%.

[0033] Comparative Example 1: Using Random Forest Model to Analyze and Predict 1. Sample collection and preparation: 100 dried tangerine peel samples from Xinhui and non-Xinhui production areas were selected, dried, ground, and pressed into tablets to prepare samples suitable for X-ray fluorescence spectrometry analysis. A rechargeable portable grinder was used for grinding and pulverization, and then sieved through an 80-mesh standard sieve.

[0034] 2. Portable XRF instrument testing and data collation: Take 1.5 g of powder sample and put it into a sample cup with a diameter of 32 mm for XRF detection. The content data of 10 characteristic elements (P, S, Cl, K, Ca, Cr, Mn, Fe, Cu, Zn) in the sample are obtained, and the element data are normalized.

[0035] 3. Modeling the random forest model in Python: First, log the data to make the features consistent, use the contents of 10 mineral elements as input, and the corresponding origin as the category label. In order to reduce the error caused by the imbalance of the data set, use undersampling, oversampling, SMOTE or ADASYN to balance the data. Divide the data set according to the ratio (8:2, 7:3, 6:4), use RandomForestClassifier to train the model, and optimize the hyperparameters. Calculate the accuracy, classification report, and confusion matrix. Save the prediction results and model for subsequent analysis and use. Finally, calculate the Cohen's Kappa coefficient to evaluate the consistency of the model.

[0036] 4. Input the elemental data of the tangerine peel samples of unknown origin into the trained random forest model. The model will predict the origin of the unknown sample based on the existing classification rules and output its possible origin category. The classification confidence is judged by the prediction probability. If the probability of a certain origin category is much higher than other categories, the prediction result is more reliable; if the probability distribution is relatively even, it means that the model is uncertain about the sample.

[0037] The results of random forest analysis showed that the prediction accuracy of random forest classification for tangerine peel from Xinhui production area was 92.5%, with a Kappa coefficient of 0.852, and the prediction accuracy for tangerine peel from non-Xinhui production areas was 91.7%, with a Kappa coefficient of 0.752.

[0038] Comparative Example 2: Using OPLS-DA model to analyze and predict 1. Sample collection and preparation: 50 dried tangerine peel samples from Xinhui and non-Xinhui production areas were selected, dried, ground, and pressed into tablets to prepare samples suitable for X-ray fluorescence spectrometry analysis. A rechargeable portable grinder was used for grinding and pulverization, and then sieved through an 80-mesh standard sieve.

[0039] 2. Portable XRF instrument testing and data collation: Take 1.5 g of powder sample and put it into a sample cup with a diameter of 32 mm for XRF detection. The content data of 10 characteristic elements (P, S, Cl, K, Ca, Cr, Mn, Fe, Cu, Zn) in the sample are obtained, and the element data are standardized or normalized to eliminate the influence of dimension on the results.

[0040] 3. Use SIMCA software to build an OPLS-DA model. The model parameters R2X and R2Y represent the explanation rate of the model for the X and Y matrices, respectively. 2 Indicates the predictive ability of the model, Q 2>0.5 is considered a valid model. The permutation test is used to analyze and verify whether the model has an overfitting intercept. The intercept of R2Y of a good model does not exceed 0.3-0.4, and the intercept of Q2Y does not exceed 0.05 (usually a negative value). The horizontal axis of the OPLS-DA score graph represents the difference between groups, and the vertical axis represents the difference within the group.

[0041] 4. Input the data of tangerine peel samples of unknown origin into the trained OPLS-DA model to obtain the prediction score of the sample. Observe the position of the unknown sample in the predictive principal component diagram to determine which cluster of known origin it is closer to. Use classification threshold or discriminant analysis to further confirm the origin category to which the unknown sample belongs. Based on the prediction confidence score of the model, evaluate the reliability of the classification result of the unknown sample.

[0042] OPLS-DA analysis results ( Figure 2 ) shows that R2X=0.817, R2Y=0.711, Q 2 =0.694, indicating that the explanatory rate and predictive power of the established model are good and reliable; in addition, the permutation test results also show that this model is relatively ideal and there is no overfitting phenomenon. The prediction accuracy of this model for tangerine peel in Xinhui production area is 86.7%, and the prediction accuracy in non-Xinhui production area is 82.5%.

[0043] Comparative Example 3: PCA analysis and prediction 1. Sample collection and preparation: 62 dried tangerine peel samples from Xinhui and non-Xinhui production areas were selected, dried, ground, and pressed into tablets to prepare samples suitable for X-ray fluorescence spectrometry analysis. A rechargeable portable grinder was used for grinding and pulverization, and then sieved through an 80-mesh standard sieve.

[0044] 2. Portable XRF instrument testing and data collation: Take 1.5 g of powder sample and put it into a sample cup with a diameter of 32 mm for XRF detection to obtain the content data of 10 characteristic elements (P, S, Cl, K, Ca, Cr, Mn, Fe, Cu, Zn) in the sample.

[0045] 3. SPSS software was used for PCA analysis, and principal component analysis was performed on the 10 mineral element indicators in Xinhui and non-Xinhui dried tangerine peel. First, the data was tested for applicability, KMO=0.833>0.5, indicating that it is more suitable for factor analysis; Bartlett's sphericity test was used to measure the degree of correlation between the indicator variables, where the P value was <0.01, indicating that there was a collinearity problem between the above variable indicators.

[0046] 4. Use the trained PCA model to transform the elemental composition data of the unknown origin sample to obtain the principal component score of the sample, and project it into the principal component space of the known sample. Observe the position of the unknown sample in the principal component space and determine the origin cluster center it is close to. Determine the distance between the unknown sample and each origin cluster center, and select the closest cluster center as the possible origin of the sample. Use the K-means clustering method based on the principal component score to further determine the category to which the unknown sample belongs.

[0047] PCA results ( Figure 3 ) showed that the cumulative contribution rate of the first principal component (PC1) and the second principal component (PC2) was 67.6%, of which PC1 could explain 42.4% of the variation of the overall data, and PC2 explained 25.2% of the variation of the overall data. Among them, the mineral element content of Xinhui and non-Xinhui dried tangerine peel samples was relatively stable, and the sample points were relatively concentrated. Except for individual samples, Xinhui and non-Xinhui dried tangerine peel samples could be basically separated. Xinhui dried tangerine peel was mainly located on the positive semi-axis of the first principal component, and non-Xinhui dried tangerine peel was located on the negative semi-axis of the first principal component. The two dried tangerine peel producing areas could be initially distinguished. According to the distance from the sample to the cluster center of each producing area and K-means clustering, the results showed that the prediction accuracy of the Xinhui producing area dried tangerine peel using PCA analysis was 87.5%, the prediction accuracy of the non-Xinhui producing area was 78.6%, and the KMO value was 0.833.

[0048] Comparative Example 4: Analysis and prediction using OPLS-DA + random forest model 1. Sample collection and preparation: 55 dried tangerine peel samples from Xinhui and non-Xinhui production areas were selected, dried, ground, and pressed into tablets to prepare samples suitable for X-ray fluorescence spectrometry analysis. A rechargeable portable grinder was used for grinding and pulverization, and then sieved through an 80-mesh standard sieve.

[0049] 2. Portable XRF instrument testing and data collation: Take 1.5 g of powder sample and put it into a sample cup with a diameter of 32 mm for XRF detection to obtain the content data of 10 characteristic elements (P, S, Cl, K, Ca, Cr, Mn, Fe, Cu, Zn) in the sample.

[0050] 3. OPLS-DA + Random Forest Model Training: First use SIMCA software to establish an OPLS-DA model. First, reduce the dimension of the data and identify the key variables that distinguish different origins. This model can maximize the differences between origins and minimize the variation between samples from the same origin. Through the S-plot method, find the elements that contribute most to distinguishing the origins. The important elements screened out by OPLS-DA are input into the random forest model for classification training. As an integrated learning method, random forest can automatically process high-dimensional data and capture nonlinear relationships. Finally, the random forest provides the contribution of each element feature to distinguishing the origin, helping to identify the key elements that affect the origin classification.

[0051] 4. Perform k-fold cross validation on the OPLS-DA + random forest model to evaluate the stability and generalization ability of the model. Calculate indicators such as accuracy, precision, and recall to evaluate the performance of the model in distinguishing different origins. Analyze the contribution of each feature in the random forest to further identify the most important elements.

[0052] The results show that the prediction accuracy of OPLS-DA+random forest model for tangerine peel from Xinhui production area is 90.3%, with a Kappa coefficient of 0.751. The prediction accuracy for tangerine peel from non-Xinhui production areas is 89.2%, with a Kappa coefficient of 0.802. The elements with higher importance in production area prediction are K, Cl, Cu, and Ca.

[0053] Comparative Example 5: Using PCA+Random Forest Model to Analyze and Predict 1. Sample collection and preparation: 65 dried tangerine peel samples from Xinhui and non-Xinhui production areas were selected, dried, ground, and pressed into tablets to prepare samples suitable for X-ray fluorescence spectrometry analysis. A rechargeable portable grinder was used for grinding and pulverization, and then sieved through an 80-mesh standard sieve.

[0054] 2. Portable XRF instrument testing and data collation: Take 1.5 g of powder sample and put it into a sample cup with a diameter of 32 mm for XRF detection to obtain the content data of 10 characteristic elements (P, S, Cl, K, Ca, Cr, Mn, Fe, Cu, Zn) in the sample.

[0055] 3. PCA+Random Forest Model Training: PCA is used to convert the original high-dimensional data into several principal components. The principal components retain most of the variance information in the data, which helps to retain important information while reducing the data dimension. Select the first few principal components that explain more than 90% of the cumulative variance to reduce the dimension and retain information that is helpful for classification. The principal components after PCA dimensionality reduction are used as input features, and the random forest model is used to classify the origin of the samples. Random forest performs well in handling nonlinear relationships by integrating multiple decision trees. The hyperparameters of random forest are optimized, and grid search is used to further improve model performance.

[0056] 4. Perform k-fold cross validation on the PCA+random forest model to evaluate the robustness and generalization ability of the model. Use indicators such as confusion matrix and classification report to evaluate the accuracy, precision, recall rate, and F1 score of the model to ensure the classification effect of the model.

[0057] The results showed that the PCA+random forest model had a prediction accuracy of 89.6% for tangerine peel from Xinhui production area, with a Kappa coefficient of 0.906, and a prediction accuracy of 88.5% for tangerine peel from non-Xinhui production area, with a Kappa coefficient of 0.775. Comparative Example 6: PCA+ATR-2a analysis and prediction 1. Sample collection and preparation: 80 dried tangerine peel samples from Xinhui and non-Xinhui production areas were selected, dried, ground, and pressed into tablets to prepare samples suitable for X-ray fluorescence spectrometry analysis. A rechargeable portable grinder was used for grinding and pulverization, and then sieved through an 80-mesh standard sieve.

[0058] 2. XRF instrument testing and data collation: Take 1.5 g of powder sample and put it into a sample cup with a diameter of 32 mm for XRF testing to obtain the content data of 10 characteristic elements (P, S, Cl, K, Ca, Cr, Mn, Fe, Cu, Zn) in the sample. The magnitude of the concentration values ​​of different elements may vary greatly, and the influence of each variable is balanced after standardization. Outliers are eliminated to improve the quality of data and the stability of the model.

[0059] 3. PCA+ATR-2a: PCA is used to convert high-dimensional elemental composition data into several principal components, which can explain most of the variance in the data and retain the main information. The principal components that can explain more than 90% of the cumulative variance are retained to reduce the data dimension and remove redundant information. The artificial neural network algorithm (ART-2a) based on sampling adaptive resonance theory in MATLAB software extracts the content of characteristic elements in each production area from the spectral data, and the training set data is clustered using the ART-2a model to generate several groups of sample clusters. ART-2a will classify similar samples into the same category based on the similarity of elemental composition.

[0060] 4. Input the elemental data of tangerine peel samples of unknown origin into the trained PCA+ATR-2a model, and use indicators such as silhouette coefficient and cluster purity to evaluate the clustering effect of the ART-2a model to ensure the differentiation between origins. According to the evaluation results, optimize the parameters of the ART-2a network to further improve the classification effect and stability of the model.

[0061] The results showed that the prediction accuracy of PCA+ATR-2a model for tangerine peel in Xinhui production area was 88.4%, and the prediction accuracy in non-Xinhui production area was 79.6%, and the KMO value was 0.866.

[0062] In summary, by comparing the common analysis models used for tangerine peel origin analysis, the results show that the combination of ART-2a and the random forest model has the highest accuracy in predicting tangerine peel from Xinhui and non-Xinhui production areas. At the same time, combined with a portable X-ray fluorescence spectrometer, it can realize on-site rapid identification of tangerine peel origin.

Claims

1. A method for identifying the origin of tangerine peel, characterized in that: The following steps are involved: S1. The characteristic element contents of tangerine peel samples from known Xinhui and non-Xinhui production areas were input into the ART-2a model for cluster analysis to obtain the origin cluster labels; the characteristic elements were P, S, Cl, K, Ca, Cr, Mn, Fe, Cu and Zn; S2. Use the sample data and origin cluster labels after cluster analysis in S1 as data sets to train the random forest classification model and build a prediction model for the origin of tangerine peel; S3. The above characteristic element data of the tangerine peel sample of unknown origin is input into the ART-2a model to obtain the origin clustering label, and then the characteristic element data of the tangerine peel sample of unknown origin and the corresponding origin clustering label are input into the tangerine peel origin prediction model to identify whether the tangerine peel sample is tangerine peel produced in Xinhui District.

2. The method according to claim 1, characterized in that: The characteristic element content in step S1 is obtained by X-ray fluorescence spectroscopy analysis.

3. The method according to claim 1, characterized in that: The cluster analysis described in step S1 is to perform cluster analysis on the characteristic elements of tangerine peel samples from known Xinhui and non-Xinhui production areas according to the similarity of element content composition, obtain the relative content data of characteristic elements of tangerine peel samples from different production areas, and obtain the production area cluster label.

4. The method according to claim 1, characterized in that: The data set in step S2 is divided into a training set and a test set in a ratio of 8:

2.

5. The method according to claim 1, characterized in that: The characteristic element data of the tangerine peel sample of unknown origin described in step S3 is obtained by using a portable X-ray fluorescence spectrometer.

6. The method according to claim 1, characterized in that: In step S3, the tangerine peel sample of unknown origin is crushed and then tested.

7. The method according to claim 6, characterized in that: The pulverization is to obtain powder with a particle size of 160 to 200 μm.

8. A device for identifying the origin of tangerine peel, characterized in that: include: An acquisition module is used to obtain the characteristic element contents of tangerine peel samples from known Xinhui and non-Xinhui production areas and tangerine peel samples from unknown production areas; the characteristic elements are P, S, Cl, K, Ca, Cr, Mn, Fe, Cu and Zn; A modeling module is used to perform cluster analysis on the characteristic element contents of tangerine peel samples from known Xinhui and non-Xinhui production areas through the ART-2a model, obtain origin clustering labels, use the clustered sample data and origin clustering labels as data sets to train a random forest classification model, and build a tangerine peel origin prediction model; The prediction module is used to analyze the characteristic elements of the tangerine peel sample of unknown origin through the ART-2a model and the tangerine peel origin prediction model, and identify whether the tangerine peel sample of unknown origin is tangerine peel produced in Xinhui District.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is executed.

10. Use of the computer-readable storage medium of claim 9 in rapid on-site identification of Xinhui dried tangerine peel.

Images