Citrus storage mode and flavor change detection method
By combining Raman spectroscopy and stoichiometric methods, a classification model is constructed using machine learning algorithms, which solves the sensitivity and complexity of the existing technology in citrus storage and odor detection, and realizes the storage method, storage time and odor distortion of citrus Accurate detection, non-destructive and fast.
Patent Information
- Application Number
- CN202410968153.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-18
- Publication Date
- 2024-11-08
AI Technical Summary
The existing citrus storage and odor detection methods mainly rely on manual identification and sampling and damage measurement. Electronic nose technology and gas chromatography-mass spectrometry combined technology have problems such as limited sensor selection, low sensitivity, high cost, and complex data processing, which are difficult to be effective. Solve the diversity of factors affecting citrus quality.
Raman spectroscopy and stoichiometric methods are combined with machine learning to build a classification model, and baseline correction and normalization of Raman spectral data of citrus are used, and feature band intensity ratio method and machine learning algorithm (support vector machine, minimal Two-machine discriminant analysis) generates a classification model to detect the storage method, storage time and odor of citrus.
It realizes accurate detection of citrus storage methods, storage time and odor changes, and has the advantages of non-destructive, rapid, high sensitivity and online detection, simplifies the analysis process and improves work efficiency.
Smart Images

Figure BDA0004952323030000031 
Figure BDA0004952323030000043 
Figure HDA0004952323040000011
Abstract
Description
Technical Field
[0001] The invention relates to a citrus storage method and a method for detecting taste changes. Background Art
[0002] However, there are still many problems in the post-harvest processing of citrus, especially in the storage and taste spoilage detection of citrus. Currently, most of the storage and taste spoilage detection of citrus are still at the stage of manual identification, sampling and damage measurement.
[0003] The existing detection schemes are mainly electronic nose technology and gas chromatography-mass spectrometry (GC-MS). Since the factors that affect the quality of citrus are not only the color of the citrus skin, but also the maturity stage, preservation environment, and preservation time are all factors that affect the quality of citrus.
[0004] Electronic nose technology is a technology that uses a sensor array to simulate human sense of smell and can be used to detect and identify different odors. The following are the basic steps for electronic nose technology to collect citrus: 1. Prepare the electronic nose device: Make sure the electronic nose device has been calibrated and is in a usable state. Electronic nose devices usually include sensor arrays, data acquisition systems, and analysis software. 2. Prepare samples: Prepare citrus samples, which can be the peel, juice, or volatile component extracts of citrus fruits. 3. Collect sample signals: Place the citrus sample in the sampling chamber of the electronic nose device, and start the data acquisition system to start collecting the volatile component signals of the sample. 4. Analysis and identification: The electronic nose device will convert the collected sample signals into digital signals and analyze and process them through data analysis software. The system will compare the odor characteristics of the sample with the pre-established odor database to match them, thereby identifying the odor characteristics of citrus. 5. Result analysis: According to the results output by the electronic nose device, the quality, freshness, etc. of the citrus sample can be evaluated and analyzed.
[0005] Gas Chromatography-Mass Spectrometry (GC-MS) is a commonly used analytical technique for identifying and quantifying the composition of organic matter. The following are the basic steps for collecting citrus samples using GC-MS technology: 1. Sample preparation: Prepare citrus samples, which can be the peel, juice or volatile component extracts of citrus fruits. The samples need to be properly processed and prepared to ensure that they are suitable for GC-MS analysis. 2. Extract volatile components of samples: Use appropriate extraction methods (such as solid phase microextraction, solvent extraction, etc.) to extract volatile components in citrus samples for GC-MS analysis. 3. GC-MS instrument settings: Adjust the GC-MS instrument to the appropriate working mode, including setting the sample injection method, chromatographic column type and analysis conditions. 4. Sample injection: Inject the extracted volatile component solution of the citrus sample into the GC-MS instrument through an automatic syringe and start sample analysis. 5. GC-MS analysis: The sample is separated in the chromatographic column, and the different components are separated and then enter the mass spectrometer for mass spectrometry analysis. The GC-MS instrument generates mass spectra and chromatograms, which are used to identify the chemical compounds in the citrus samples. 6. Data Analysis: Based on the GC-MS analysis results, the types and amounts of different components present in the citrus samples can be determined, and they can be quantitatively analyzed and identified.
[0006] The disadvantages of electronic nose technology include: 1. Limited choice of sensors: Currently, the types of sensors used in electronic noses are limited, and it is difficult to cover all odor components that may need to be detected. 2. Low sensitivity: Some electronic nose devices are not sensitive to low-concentration odors, and are prone to misjudgment or omissions. 3. Poor robustness: It is greatly affected by environmental conditions, humidity, temperature and other factors, which can easily cause instability in detection results. 4. High complexity of data processing and analysis: The amount of data collected by the electronic nose is large and complex, and it takes a lot of time and resources to process and analyze the data. 5. High cost: The current price of electronic nose technology equipment is relatively high, which limits its popularity and promotion in large-scale applications.
[0007] The disadvantages of gas chromatography-mass spectrometry (GC-MS) include: 1. Expensive equipment cost: The purchase, maintenance and operation costs of GC-MS equipment are relatively high, making it difficult to popularize and promote its application. 2. Requirement of professional operators: GC-MS technology requires operators to have certain professional knowledge and skills, and the operation is complicated, requiring special training to be proficient in use. 3. High sample preparation requirements: GC-MS requires complex sample preparation and pretreatment, such as extraction, enrichment, derivatization and other steps, which affects the efficiency and speed of analysis. 3. Limited detection sensitivity: Although GC-MS has high sensitivity, there are still certain challenges for the detection of some low-concentration substances or trace components. 4. Complex data processing: The amount of data generated by GC-MS is huge, and complex data processing and analysis are required, which requires high computer skills and data interpretation capabilities of the operator. 5. Some compounds are difficult to separate and detect: Because some compounds may have overlapping peaks or are difficult to separate on the chromatographic column, the accuracy and reliability of the GC-MS analysis results may be affected.
[0008] Partial Least Squares Discriminant Analysis
[0009] Partial Least Squares Discriminant Analysis (PLS-DA) is a multivariate statistical analysis method used to process data sets with multiple independent variables and one or more dependent variables. In PLS-DA, the discriminant function is a function used to classify samples. The discriminant function predicts and classifies the dependent variable through a linear combination of latent variables. Specifically, assuming there are m latent variables, the formula of the discriminant function is shown in (2):
[0010] Y = b0 + b1*Z1 + b2*Z2 +··· + bm*Zm (2)
[0011] Where Y represents the dependent variable (usually the classification label), b0, b1, b2, ..., bm represent the coefficients of the discriminant function, and Z1, Z2, ..., Zm represent m latent variables. The coefficients of the discriminant function are solved by minimizing the residual sum of squares between the independent variable and the dependent variable, so that the discriminant function can best distinguish different categories. In statistics, the sum of the squares of the difference between the actual value and the predicted value is called the residual sum of squares, which represents the effect of random errors. The smaller the residual sum of squares of a set of data, the better its fit. The specific calculation formula is shown in (3):
[0012]
[0013] The steps are as follows: Step 1, data preprocessing. First, preprocess the data, including centering and standardization, to eliminate the dimensional differences between variables, making the data easier to compare and analyze. Step 2, build a model. PLS-DA finds a set of new variables (called latent variables) that are linear combinations of the original independent variables and dependent variables, and can explain the covariance between the independent variables and the dependent variables to the greatest extent. These latent variables are used to construct discriminant functions to achieve prediction and classification of dependent variables. Step 3, solve the model. PLS-DA uses the least squares method to solve the model parameters and fits the model by continuously minimizing the residual sum of squares between the independent variables and the dependent variables. In this process, PLS-DA will perform multiple rounds of iterations, and by continuously adjusting the latent variables and the discriminant function, the residual sum of squares will gradually decrease, and finally achieve the optimal model fitting effect. Step 4, prediction and classification. Through the constructed discriminant function, new samples can be predicted and classified, thereby achieving prediction and classification of dependent variables.
[0014] Support Vector Machine
[0015] The mathematical expression of Support Vector Machine (SVM) can be described by the following formula: Assume that the training data set is {(x 1 ,y 1 ),(x 2 ,y 2 ),...,(x n ,y n )}, where x i is the feature vector, y i is the category label, if y i ∈{-1, 1}.
[0016] The optimization problem of SVM can be expressed as:
[0017] Minimize:
[0018] Constraints:
[0019] Here, is the normal vector of the hyperplane, and b is the bias term. By using the Lagrangian multiplier method, the Lagrangian function can be obtained as shown in formula (4).
[0020]
[0021] Among them, α i is the Lagrange multiplier. Finally, by taking the derivative of L(ω, b, α) with respect to ω and b and setting the derivative equal to 0, the optimal ω and b can be obtained, and then the decision function of the linear support vector machine is obtained as shown in formula (5).
[0022] h(x)=sign(ω·x+b) (5)
[0023] The above is the mathematical formula description of the linear support vector machine calculation. Summary of the invention
[0024] The technical problem to be solved by the present invention is to provide a citrus storage method and taste change detection method, which is based on Raman spectroscopy and chemometrics methods, and cooperates with machine learning to obtain a classification model, and uses this model to analyze the Raman spectrum of unknown citrus to obtain accurate results of the storage method, storage time, and whether the corresponding citrus has changed taste.
[0025] In order to solve the above technical problems, the present invention provides a citrus storage method and a method for detecting taste changes, comprising the following steps:
[0026] (1) Model building:
[0027] The same type of citrus is stored in cold storage and at room temperature. Before storage and on the 30th day, several citrus samples are taken every day to collect Raman spectrum data, with a total of no less than 450;
[0028] Take equal amounts of normal citrus and citrus with spoiled taste to collect Raman spectra, and the Raman spectrum data of citrus samples in both states shall not be less than 510;
[0029] The collected spectral data were baseline corrected and normalized, and characteristic band analysis was performed. The characteristic band intensity ratio method was used to analyze the wave number 757 cm -1 、865cm -1 、1158cm -1 、1438cm -1 、1529cm -1 and 1606cm -1 The peak intensity and displacement of I 1438 / I 1529 ,I 1606 / I 1529 ,I 865 / I 1158 and I 1610 / I 1529 The ratio results are recorded, and the recorded information is matched one by one with the current taste change of the corresponding citrus, storage method and storage time; then the analysis results are subjected to machine learning, and the support vector machine, decision tree and partial least squares discriminant analysis classification models are used for model optimization to obtain the optimal classification model, wherein the support vector machine is used to generate the classification model of the taste change, and the partial least squares discriminant analysis is used to generate the classification model of the storage method and storage time;
[0030] (2) Detection operation:
[0031] By collecting the Raman spectrum of the citrus to be tested and inputting the obtained spectrum into the model obtained in step (1), accurate results of the storage method, storage time, and whether the taste of the corresponding citrus can be obtained.
[0032] Principle analysis:
[0033] Storage time:
[0034] 1438cm -1 and 1606cm -1 The peak intensity and displacement changes have a strong correlation with the storage time. -1 and 1606cm -1 The characteristic peaks correspond to limonene and terpineol one by one. The changing trends of limonene and terpineol with storage time are consistent with the changing trends of the corresponding wavenumbers. Therefore, limonene and terpineol can be used as a kind of prediction information for the storage time of citrus.
[0035] The characteristic band intensity ratio method was used to analyze the 1438 cm -1 and 1606cm -1 After comparing the characteristic band ratios, it is found that the overall data is indeed consistent with the above peak change trend. 1438 and I 1606 Both are related to I 1529 The ratio is because at 1529cm -1 The peak of is the same size in all samples, and the dimensional differences between samples can be eliminated by comparing it.
[0036] Regarding taste change:
[0037] Wave number 757cm -1 、1438cm -1 and 1606cm -1 The characteristic peak is related to essential oils, with a wave number of 865 cm -1 The characteristic peaks are related to carbohydrates. It is found that the changes in chemical substances inside essential oils and carbohydrates are consistent with the changing trends of the corresponding wavenumbers.
[0038] The characteristic band intensity ratio method was used to analyze the 865cm -1 and 1606cm -1 After calculating the characteristic band ratio, it is found that the overall data does have the same peak change trend as mentioned above.
[0039] About storage method:
[0040] Wave number 757cm -1 、865cm -1 and 1606cm -1The peak intensity has a clear correlation with the displacement;
[0041] Wave number 757cm -1 and 1606cm -1 The characteristic peak is related to essential oils, with a wave number of 865 cm -1 The characteristic peaks are related to carbohydrates. It is found that the changes in chemical substances inside essential oils and carbohydrates are consistent with the changing trends of the corresponding wavenumbers.
[0042] For PLS-DA models, the size of n_component can usually be adjusted to optimize the model. MSE (Mean Squared Error) and Score parameters can be used to evaluate model performance.
[0043] Advantages of the present invention: The present invention is based on Raman spectroscopy and chemometric methods, and cooperates with machine learning to obtain a classification model, and uses this model to analyze the Raman spectrum of unknown citrus, so as to obtain the storage method, storage time, and whether the corresponding citrus has changed its taste. Accurate results. The use of Raman spectroscopy has the following advantages: 1. Non-destructive: Raman spectroscopy technology is a non-destructive analysis method. No special treatment or destruction is required for citrus samples, and analysis can be performed without causing damage. 2. Rapidity: Raman spectroscopy technology has a fast detection speed and can quickly obtain the spectral information of citrus samples, which is suitable for high-throughput analysis needs. 3. High sensitivity: Raman spectroscopy technology has high sensitivity and can detect and distinguish trace components in citrus samples. 4. Online detection: Raman spectroscopy technology can monitor the chemical composition and structural information of citrus samples in real time, which is suitable for online process monitoring. 5. No sample preparation is required: Compared with some traditional analysis methods, Raman spectroscopy technology does not require a complex sample preparation process, simplifies the analysis process, and improves work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1 It is a flow chart of the classification model optimization of the present invention. DETAILED DESCRIPTION
[0045] A citrus storage method and a method for detecting taste change, comprising the following steps:
[0046] (1) Model building:
[0047] The same type of citrus is stored in cold storage and at room temperature. Before storage and on the 30th day, several citrus samples are taken every day to collect Raman spectrum data, with a total of no less than 450;
[0048] Take equal amounts of normal citrus and citrus with spoiled taste to collect Raman spectra, and the Raman spectrum data of citrus samples in both states shall not be less than 510;
[0049] See also Figure 1 The collected spectral data were baseline corrected and normalized, and characteristic band analysis was performed. The characteristic band intensity ratio method was used to analyze the wave number 757cm -1 、865cm -1 、1158cm -1 、1438cm -1 、1529cm -1 and 1606cm -1 The peak intensity and displacement of I 1438 / I 1529 ,I 1606 / I 1529 ,I 865 / I 1158 and I 1610 / I 1529 The ratio results are recorded, and the recorded information is matched one by one with the current taste change of the corresponding citrus, storage method and storage time; then the analysis results are subjected to machine learning, and the support vector machine, decision tree and partial least squares discriminant analysis classification models are used for model optimization to obtain the optimal classification model, wherein the support vector machine is used to generate the classification model of the taste change, and the partial least squares discriminant analysis is used to generate the classification model of the storage method and storage time;
[0050] (2) Detection operation:
[0051] By collecting the Raman spectrum of the citrus to be tested and inputting the obtained spectrum into the model obtained in step (1), accurate results of the storage method, storage time, and whether the taste of the corresponding citrus can be obtained.
[0052] Verification experiment:
[0053] A relevant model is established for a certain type of citrus grown in a certain place by adopting the method of step (1) in the above embodiment. After the model is established, 100 citrus samples with known storage time, storage method and whether they have spoiled are prepared. After the relevant information on the citrus is erased, they are handed over to testers for testing according to the method of step (2) in the above embodiment. The test results are recorded and compared with the data before the experiment. According to statistics, among the 100 samples, the storage method and taste spoilage test results are completely consistent with the known information, and there is an error of ±3 days in the storage time.
Claims
1. A citrus storage method and a method for detecting taste changes, characterized in that: The following steps are involved: (1) Model building: The same kind of citrus is stored in cold storage and at room temperature. Before storage and on the 30th day, several citrus samples are taken every day to collect Raman spectrum data, with a total of no less than 450; Take equal amounts of normal citrus and citrus with spoiled taste to collect Raman spectra, and the Raman spectrum data of citrus samples in both states shall not be less than 510; The collected spectral data were baseline corrected and normalized, and characteristic band analysis was performed. The characteristic band intensity ratio method was used to analyze the wave number 757 cm -1 、865cm -1 、1158cm -1 、1438cm -1 、1529cm -1 and 1606cm -1 The peak intensity and displacement of I 1438 / I 1529 ,I 1606 / I 1529 ,I 865 / I 1158 and I 1610 / I 1529 The ratio results are recorded, and the recorded information is matched one by one with the current taste change of the corresponding citrus, storage method and storage time; then the analysis results are subjected to machine learning, and the support vector machine, decision tree and partial least squares discriminant analysis classification models are used for model optimization to obtain the optimal classification model, wherein the support vector machine is used to generate the classification model of the taste change, and the partial least squares discriminant analysis is used to generate the classification model of the storage method and storage time; (2) Detection operation: By collecting the Raman spectrum of the citrus to be tested and inputting the obtained spectrum into the model obtained in step (1), accurate results of the storage method, storage time, and whether the taste of the corresponding citrus can be obtained.
