White spirit identification method based on intelligent sensory technology and machine learning
Through intelligent sensory technology and machine learning methods, using electronic nose and electronic tongue to detect liquor, and construct prediction and classification models, the subjectivity and cost of liquor identification methods in the existing technology are solved, and efficient and accurate identification of liquor is achieved.
Patent Information
- Application Number
- CN202410956462.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-17
- Publication Date
- 2024-11-08
AI Technical Summary
The existing methods for identifying liquors have limitations such as artificial sensory evaluation, high detection cost and long analysis time, and it is difficult to effectively identify the flavor quality of different liquor categories.
Using intelligent sensory technology and machine learning methods, liquor is detected through electronic nose and electronic tongue, data preprocessing and feature extraction, liquor prediction model and classification model are constructed, and liquor is used to use algorithms such as logistic regression, decision trees, and random forests to perform liquor. Prediction and classification.
It realizes efficient, accurate and objective identification of liquor, reduces experimental costs, avoids the influence of subjective factors of artificial sensory evaluation, and improves the accuracy of liquor prediction and classification.
Smart Images

Figure CN118916769A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of liquor analysis, and in particular to a liquor identification method based on intelligent sensory technology and machine learning. Background Art
[0002] Liquor is a traditional Chinese beverage. Influenced by different raw materials, koji, production methods and regional environment, my country now has a very rich variety of liquor. For example, according to the difference in flavor quality, liquor has been divided into twelve types of flavors, including sauce-flavored, strong-flavored, light-flavored and rice-flavored. According to the difference in regional environment, liquor of the same flavor can be further subdivided. For example, the category of strong-flavor liquor includes northern strong-flavor liquor mainly produced in northern my country, Sichuan strong-flavor liquor mainly produced in southwest China and Jianghuai strong-flavor liquor mainly produced in the Huaihe River Basin. The flavor quality of each category of liquor is unique. However, the richness of liquor categories has also increased the difficulty of product identification. At present, common liquor identification methods include identification based on sensory evaluation by professionals, identification based on chromatography or spectroscopy combined with chemometrics and statistics, and identification based on colorimetric sensor arrays. Among them, the method of manual sensory evaluation is greatly affected by subjective factors, and other technical methods also have limitations such as high detection cost and long analysis time.
[0003] Intelligent sensory technology is a bionic technology, in which the principle of electronic nose and electronic tongue is to simulate human sense of smell and taste based on sensor array, quickly identify sample components and generate response signals, and then combine multivariate statistics and other analysis methods to achieve sample differentiation; in addition, for big data samples, machine learning has significant advantages over traditional statistical methods in sample classification and prediction, so there is an urgent need for a liquor identification method based on intelligent sensory technology and machine learning. Summary of the invention
[0004] In view of the shortcomings of the prior art mentioned above, the purpose of the present invention is to provide a liquor identification method based on intelligent sensory technology and machine learning, which is used to solve the problem that the manual sensory evaluation method in the prior art is greatly affected by subjective factors, and other technical methods also have limitations such as high detection cost and long analysis time.
[0005] To achieve the above objectives and other related objectives, the present invention provides the following technical solutions:
[0006] A liquor identification method based on intelligent sensory technology and machine learning comprises the following steps: using an electronic nose and an electronic tongue to detect liquor, and obtaining detection data related to the liquor according to the detection results; preprocessing and feature extraction of the detection data related to the liquor, obtaining characteristic value data related to the liquor according to the preprocessing and feature extraction results, and randomly dividing the characteristic value data into a training set and a test set; constructing a primary liquor prediction model and a primary liquor classification model, and using the characteristic value data in the training set to train the primary liquor prediction model and the primary classification model, and obtaining a liquor prediction model and a liquor classification model according to the training results; inputting the characteristic value data in the test set into the liquor prediction model and the liquor classification model, respectively, and obtaining a liquor prediction result through the liquor prediction model and a liquor classification result through the liquor classification model.
[0007] In one embodiment of the present invention, the process of using an electronic nose to detect liquor is as follows: dilute the liquor sample 50 to 200 times, take 1.00 to 4.00 mL of each liquor in a 50 to 500 mL headspace bottle, seal and number it, and let it stand at room temperature for 20 to 60 minutes; before the measurement, preheat the electronic nose for 30 to 60 minutes, and clean it for 100 to 200 seconds before each detection; during the measurement, the air flow rate is 1.00 to 2.00 L / min, the sampling time is 50 to 90 seconds, and each liquor sample is tested 20 to 40 times; between every two tests, perform a system clean until the signal peak returns to the baseline.
[0008] In one embodiment of the present invention, the process of using the electronic tongue to detect liquor is as follows: 30 to 50 mL of each liquor is taken into a detection cup, the electronic tongue system is cleaned before each detection, and the sample is tested at room temperature after cleaning. The signal amplification factor is in the range of 100 to 300 times, so that the signal interval is placed in [-2.0, 2.0], and the sensor response signal is recorded. Each liquor sample is tested 20 to 40 times.
[0009] In one embodiment of the present invention, the detection data related to liquor are preprocessed and feature extracted, characteristic value data related to liquor are obtained according to the preprocessing and feature extraction results, and the characteristic value data are randomly divided to form a training set and a test set, including: the preprocessing includes missing value processing and outlier processing, the feature extraction step includes standardization and initialization of polynomial features for feature construction, quantile data conversion, feature selection based on random forest model, principal component analysis dimensionality reduction and random noise injection for data enhancement, and the feature extraction process selects the maximum value of the electronic nose sensor array response value and the smoothing value of the electronic tongue sensor array response value as the characteristic value, and the characteristic value data are randomly divided to form a training set and a test set. This technical solution can improve the accuracy of liquor prediction and classification by preprocessing and feature extraction of the detection data related to liquor.
[0010] In one embodiment of the present invention, the construction of a primary liquor prediction model and a primary liquor classification model includes: the primary liquor prediction model is composed of at least three different prediction sub-models, and the prediction sub-model algorithms include logistic regression, decision tree, random forest, gradient boosting tree, support vector regression and k-nearest neighbor; the primary liquor classification model is composed of at least three classification sub-models, and the classification sub-model algorithms include logistic regression, support vector machine, naive Bayes, k-nearest neighbor and decision tree.
[0011] In one embodiment of the present invention, the primary liquor prediction model and the primary classification model are trained using the eigenvalue data in the training set, and the liquor prediction model and the liquor classification model are obtained according to the training results, including: using the eigenvalue data in the training set to train each prediction sub-model and classification sub-model respectively, and obtaining the liquor prediction model and the liquor classification model according to the training results.
[0012] In one embodiment of the present invention, the establishment of the liquor prediction model and the liquor classification model both use the detection data of the electronic nose and the electronic tongue, and the prediction target is the quantitative descriptive analysis result of the liquor sample, that is, the score of different liquor flavor attributes, wherein the flavor attributes specifically include mellow aroma, aged aroma, fruity aroma, cellar aroma, grain aroma, koji aroma, sweet aroma, dregs aroma, sourness, sweetness, bitterness, mellowness, thickness, freshness, harmony and longevity.
[0013] In one embodiment of the present invention, the characteristic value data in the test set is input into the liquor prediction model, and the prediction result of the liquor is obtained through the liquor prediction model, including: inputting the characteristic value data in the test set into the liquor prediction model, each prediction sub-model in the liquor prediction model operates on the input characteristic value data separately, calculates the R2 and prediction deviation of each prediction sub-model, selects 3 prediction sub-models with better prediction effects and calculates the average prediction result as the prediction result of the liquor prediction model. This technical solution can obtain the prediction result corresponding to the characteristic value data by inputting the characteristic value data in the test set into the liquor prediction model.
[0014] In one embodiment of the present invention, the eigenvalue data in the test set are respectively input into the liquor classification model, and the classification result of the liquor is obtained through the liquor classification model, including: inputting the eigenvalue data in the test set into the liquor classification model, each classification sub-model in the liquor classification model operates on the input eigenvalue data separately, calculates the accuracy, precision and recall rate of each classification sub-model, and selects the results with more output times among the three classification sub-models with better classification effects as the classification results of the liquor classification model. This technical solution can obtain the classification results corresponding to the eigenvalue data by inputting the eigenvalue data in the test set into the liquor classification model.
[0015] As described above, the present invention provides a liquor identification method based on intelligent sensory technology and machine learning, which has the following beneficial effects: the present invention adopts an electronic nose and an electronic tongue to detect liquor samples, the operation process is simple and fast, and the experimental cost is low, and further combines machine learning to establish a liquor prediction model and a liquor classification model based on more data samples and three algorithm integrations, which not only effectively avoids the influence of subjective factors brought by manual sensory evaluation, but also compared with chromatography technology, spectral technology and traditional statistical analysis methods, the present invention can predict and classify different liquor products more efficiently, accurately and objectively, thereby better completing the identification of liquor, and has high promotion value and application potential. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 Shown is an overall flow chart of the liquor identification method based on intelligent sensory technology and machine learning disclosed in an embodiment of the present invention;
[0017] Figure 2 Shown are prediction effect diagrams of different prediction sub-models in the liquor identification method based on intelligent sensory technology and machine learning disclosed in an embodiment of the present invention;
[0018] Figure 3 Shown is a prediction effect diagram of the liquor prediction model in the liquor identification method based on intelligent sensory technology and machine learning disclosed in an embodiment of the present invention. DETAILED DESCRIPTION
[0019] The following describes the embodiments of the present invention through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.
[0020] The present invention relates to a liquor identification method based on intelligent sensory technology and machine learning, the process is as follows Figure 1 As shown, the details are as follows:
[0021] Step 101, using the electronic nose and the electronic tongue to detect the liquor, and obtaining detection data related to the liquor according to the detection results.
[0022] Specifically, electronic noses and electronic tongues are used to detect strong-flavor liquors of different origins and grades in Jiangsu Province. The liquor samples used to establish a liquor prediction model include but are not limited to 26 different strong-flavor liquors from Jiangsu Province. The liquor samples used to establish a liquor classification model include but are not limited to 32 strong-flavor liquors from different origins and 24 different grades of strong-flavor liquors.
[0023] More specifically, the process of using an electronic nose to detect liquor is as follows: dilute the liquor sample 50 to 200 times, take 1.00 to 4.00 mL of each liquor into a 50 to 500 mL headspace bottle, seal it, number it, and let it stand at room temperature for 20 to 60 minutes; before the measurement, preheat the electronic nose for 30 to 60 minutes, and clean it for 100 to 200 seconds before each test; during the measurement, the air flow rate is 1.00 to 2.00 L / min, the sampling time is 50 to 90 seconds, and each liquor sample is tested 20 to 40 times; between each two tests, perform a system clean until the signal peak returns to the baseline;
[0024] The process of using the electronic tongue to test liquor is as follows: take 30 to 50 mL of each liquor into the test cup, clean the electronic tongue system before each test, and start testing at room temperature after cleaning. The signal amplification factor is in the range of 100 to 300 times, so that the signal interval is placed in [-2.0, 2.0], and the sensor response signal is recorded. Each liquor sample is tested 20 to 40 times.
[0025] Step 102, preprocessing and feature extraction are performed on the detection data related to the liquor, characteristic value data related to the liquor is obtained according to the preprocessing and feature extraction results, and the characteristic value data is randomly divided into a training set and a test set.
[0026] Specifically, preprocessing includes missing value processing and outlier processing. The feature extraction steps include standardization and initialization of polynomial features for feature construction, quantile data conversion, feature selection based on random forest model, principal component analysis dimensionality reduction and random noise injection for data enhancement. In addition, the feature extraction process selects the maximum value of the electronic nose sensor array response value and the smoothed value of the electronic tongue sensor array response value as the eigenvalue, and randomly divides the eigenvalue data into training set and test set, among which 80% of the data after division is used as the training set for modeling, and 20% of the data is used as the test set for model verification.
[0027] Step 103, constructing a primary liquor prediction model and a primary liquor classification model, and using the feature value data in the training set to train the primary liquor prediction model and the primary classification model, and obtaining the liquor prediction model and the liquor classification model according to the training results.
[0028] Specifically, the construction of the primary liquor prediction model and the primary liquor classification model specifically includes: the primary liquor prediction model is composed of at least three different prediction sub-models, and the prediction sub-model algorithms include logistic regression, decision tree, random forest, gradient boosting tree, support vector regression and k-nearest neighbor; the primary liquor classification model is composed of at least three classification sub-models, and the classification sub-model algorithms include logistic regression, support vector machine, naive Bayes, k-nearest neighbor and decision tree;
[0029] The primary liquor prediction model and the primary classification model are trained using the eigenvalue data in the training set, and the liquor prediction model and the liquor classification model are obtained according to the training results. Specifically, the method includes: firstly, each prediction sub-model and classification sub-model are trained respectively using the eigenvalue data in the training set, and then the liquor prediction model and the liquor classification model are obtained according to the training results.
[0030] More specifically, the detection data of electronic nose and electronic tongue are used to establish the liquor prediction model and the liquor classification model. The prediction target is the quantitative descriptive analysis results of the liquor samples, that is, the scores of different liquor flavor attributes, among which the flavor attributes specifically include mellow aroma, aged aroma, fruity aroma, cellar aroma, grain aroma, koji aroma, sweet aroma, dregs aroma, sourness, sweetness, bitterness, mellowness, thickness, freshness, harmony and longevity.
[0031] Step 104, the feature value data in the test set are input into the liquor prediction model and the liquor classification model respectively, and the prediction result of the liquor is obtained by the liquor prediction model and the classification result of the liquor is obtained by the liquor classification model.
[0032] Specifically, the characteristic value data in the test set is input into the liquor prediction model, and the liquor prediction result is obtained by the liquor prediction model, which specifically includes: inputting the characteristic value data in the test set into the liquor prediction model, each prediction sub-model in the liquor prediction model operates on the input characteristic value data separately, calculating the R2 and prediction deviation of each prediction sub-model, selecting 3 prediction sub-models with better prediction effects and calculating the average prediction result as the prediction result of the liquor prediction model;
[0033] The eigenvalue data in the test set are respectively input into the liquor classification model, and the classification results of the liquor are obtained through the liquor classification model. Specifically, the eigenvalue data in the test set are input into the liquor classification model, and each classification sub-model in the liquor classification model operates on the input eigenvalue data separately, and the classification accuracy, precision and recall rate of each classification sub-model are calculated, and the results with more output times among the three classification sub-models with better classification effects are selected as the classification results of the liquor classification model.
[0034] In practical application, Example 1: In this example, prediction models are established for 26 different Luzhou-flavor liquors in Jiangsu Province, and the specific steps are as follows:
[0035] Step 1: Electronic nose detection process: dilute the liquor sample 50 to 200 times, take 1.00 to 4.00 mL of each liquor in a 50 to 500 mL headspace bottle, seal it, number it, and let it stand at room temperature for 20 to 60 minutes. Before the measurement, preheat the electronic nose for 30 to 60 minutes, and clean it for 100 to 200 seconds before each test. During the measurement, the air flow rate is 1.00 to 2.00 L / min, and the sampling time is 50 to 90 seconds. Each liquor sample is tested 20 to 40 times. Between each two tests, the system is cleaned until the signal peak returns to the baseline; Electronic tongue detection process: take 30 to 50 mL of each liquor in the detection cup, clean the electronic tongue system before each test, and start the sample measurement at room temperature after cleaning. The signal amplification factor is 100 to 300 times, so that the signal interval is placed in [-2.0, 2.0], and the sensor response signal is recorded. Each liquor sample is tested 20 to 40 times.
[0036] The second step is data preprocessing and feature extraction: preprocessing includes missing value processing and outlier processing. The feature extraction steps include standardization and initialization of polynomial features for feature construction, quantile data conversion, feature selection based on random forest model, principal component analysis dimensionality reduction and random noise injection for data enhancement. There is also a feature extraction process that selects the maximum value of the electronic nose sensor array response value and the smoothed value of the electronic tongue sensor array response value as the feature value, and randomly splits the feature value data. After the split, 80% of the data is used as a training set for modeling, and 20% of the data is used as a test set for model verification.
[0037] Step 3: The construction process of the liquor prediction model: The liquor prediction model is composed of at least three different prediction sub-models. The prediction sub-model algorithms include logistic regression, decision tree, random forest, gradient boosting tree, support vector regression and k-nearest neighbor. Input the training set data obtained in the second step, train each prediction sub-model separately, and then obtain the liquor prediction model.
[0038] Step 4: Prediction process of liquor prediction model: Input the test set data obtained in step 2 into the liquor prediction model constructed in step 3. Each prediction sub-model in the liquor prediction model calculates the input feature value separately and calculates the R of each prediction sub-model. 2 and prediction deviation, three prediction sub-models with better prediction effects were selected and the average prediction results were calculated as the prediction results of the liquor prediction model.
[0039] Based on intelligent sensory data and artificial sensory data, logistic regression, decision tree, random forest, gradient boosting tree, support vector regression and k-nearest neighbor were used to establish prediction models for liquors from different Jiangsu regions. The results were verified as follows: Figure 2 As shown, the R 2 All of them are greater than 0.99, and the absolute values of the prediction deviations are all less than 0.001. The average of the prediction results of the three prediction sub-models, namely logistic regression, decision tree, and support vector regression, is selected as the prediction result of the liquor prediction model. The prediction effect is as follows Figure 3 As shown in the figure, the R 2 It is 0.9994, and the absolute value of the prediction deviation is less than 0.001.
[0040] In another practical application, Example 2: In this example, classification models are established for 32 kinds of Luzhou-flavor liquors from different origins in Jiangsu and 24 kinds of Luzhou-flavor liquors of different grades in Jiangsu. The specific steps are as follows:
[0041] Step 1: Electronic nose detection process: dilute the liquor sample 50 to 200 times, take 1.00 to 4.00 mL of each liquor in a 50 to 500 mL headspace bottle, seal it, number it, and let it stand at room temperature for 20 to 60 minutes. Before the measurement, preheat the electronic nose for 30 to 60 minutes, and clean it for 100 to 200 seconds before each test. During the measurement, the air flow rate is 1.00 to 2.00 L / min, and the sampling time is 50 to 90 seconds. Each liquor sample is tested 20 to 40 times. Between each two tests, the system is cleaned until the signal peak returns to the baseline; Electronic tongue detection process: take 30 to 50 mL of each liquor in the detection cup, clean the electronic tongue system before each test, and start the sample measurement at room temperature after cleaning. The signal amplification factor is 100 to 300 times, so that the signal interval is placed in [-2.0, 2.0], and the sensor response signal is recorded. Each liquor sample is tested 20 to 40 times.
[0042] The second step is data preprocessing and feature extraction: preprocessing includes missing value processing and outlier processing. The feature extraction steps include standardization and initialization of polynomial features for feature construction, quantile data conversion, feature selection based on random forest model, principal component analysis dimensionality reduction and random noise injection for data enhancement. There is also a feature extraction process that selects the maximum value of the electronic nose sensor array response value and the smoothed value of the electronic tongue sensor array response value as the feature value, and randomly splits the feature value data. After the split, 80% of the data is used as a training set for modeling, and 20% of the data is used as a test set for model verification.
[0043] Step 3: The construction process of the liquor classification model: The liquor classification model is composed of at least three classification sub-models. The classification sub-model algorithms include logistic regression, support vector machine, naive Bayes, k-nearest neighbor and decision tree. Input the training set data obtained in the second step, train each classification sub-model separately, and then obtain the liquor classification model.
[0044] Step 4. Classification process of the liquor classification model: Input the test set data obtained in the second step into the liquor classification model constructed in the third step. Each classification sub-model in the liquor classification model operates on the input feature values separately, calculates the accuracy, precision and recall rate of each sub-model classification, and selects the results with the largest number of outputs from the three classification sub-models with better classification effects as the classification results of the liquor classification model.
[0045] In this embodiment, five algorithms, namely, logistic regression, support vector machine, naive Bayes, k-nearest neighbor and decision tree, were used to establish classification models for 32 kinds of Luzhou-flavor liquors from different origins in Jiangsu Province, and the verification results are shown in Table 1. Classification models were established for 24 kinds of Luzhou-flavor liquors of different grades in Jiangsu Province, and the verification results are shown in Table 2. Except for naive Bayes, the accuracy, precision and recall rate of the classification sub-models established by other algorithms all reached the highest values, and the results with more occurrences in the classification results of logistic regression, support vector machine and decision tree were selected as the classification results of the liquor classification model;
[0046] Table 1 Verification results of the classification model for 32 kinds of Luzhou-flavor liquors from different origins in Jiangsu Province
[0047]
[0048] Table 2 Verification results of the classification model for 24 different grades of Luzhou-flavor liquor in Jiangsu Province
[0049]
[0050]
[0051] In summary, the present invention adopts electronic nose and electronic tongue to detect different Luzhou-flavor liquors in Jiangsu area, and extracts characteristic values respectively, and then establishes a machine learning model based on the fused data by using logistic regression, decision tree, random forest, gradient boosting tree, support vector regression and k-nearest neighbor algorithm. Both the liquor prediction model and the liquor classification model are composed of three well-trained sub-models, which can efficiently predict and classify different liquor samples, improve the accuracy and objectivity of product classification and prediction, can better complete the identification of liquor, and has high application value and promotion potential.
[0052] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. All equivalent modifications or changes made by those skilled in the art without departing from the spirit and technical ideas disclosed by the present invention shall still be covered by the claims of the present invention.
Claims
1. A liquor identification method based on intelligent sensory technology and machine learning, characterized in that: The following steps are involved: Using electronic nose and electronic tongue to test liquor, and obtaining test data about liquor according to the test results; Preprocessing and feature extraction are performed on the detection data of the liquor, characteristic value data of the liquor is obtained according to the preprocessing and feature extraction results, and the characteristic value data is randomly divided into a training set and a test set; Constructing a primary liquor prediction model and a primary liquor classification model, and using the feature value data in the training set to train the primary liquor prediction model and the primary classification model, and obtaining the liquor prediction model and the liquor classification model according to the training results; The characteristic value data in the test set are respectively input into the liquor prediction model and the liquor classification model, and the prediction result of the liquor is obtained by the liquor prediction model and the classification result of the liquor is obtained by the liquor classification model.
2. The liquor identification method based on intelligent sensory technology and machine learning according to claim 1, characterized in that: The process of using the electronic nose to detect liquor is as follows: dilute the liquor sample 50 to 200 times, take 1.00 to 4.00 mL of each liquor into a 50 to 500 mL headspace bottle, seal and number it, and then let it stand at room temperature for 20 to 60 minutes; before the measurement, preheat the electronic nose for 30 to 60 minutes, and clean it for 100 to 200 seconds before each detection; during the measurement, the air flow rate is 1.00 to 2.00 L / min, the sampling time is 50 to 90 seconds, and each liquor sample is tested 20 to 40 times; between each two tests, perform a system clean until the signal peak returns to the baseline.
3. The liquor identification method based on intelligent sensory technology and machine learning according to claim 1 is characterized in that: The process of using the electronic tongue to detect liquor is as follows: 30 to 50 mL of each liquor is taken into a detection cup, the electronic tongue system is cleaned before each detection, and the sample is tested at room temperature after cleaning. The signal amplification factor is in the range of 100 to 300 times, so that the signal interval is placed in [-2.0, 2.0], and the sensor response signal is recorded. Each liquor sample is tested 20 to 40 times.
4. The liquor identification method based on intelligent sensory technology and machine learning according to claim 1, characterized in that: The method of preprocessing and feature extracting the detection data of the liquor, obtaining feature value data of the liquor according to the preprocessing and feature extraction results, and randomly dividing the feature value data into a training set and a test set includes: The preprocessing includes missing value processing and outlier processing, the feature extraction step includes standardizing and initializing polynomial features for feature construction, quantile data conversion, feature selection based on random forest model, principal component analysis dimensionality reduction and random noise injection for data enhancement, and the feature extraction process selects the maximum value of the electronic nose sensor array response value and the smoothing value of the electronic tongue sensor array response value as feature values, and randomly divides the feature value data into training set and test set.
5. The liquor identification method based on intelligent sensory technology and machine learning according to claim 1, characterized in that: The construction of the primary liquor prediction model and the primary liquor classification model includes: The primary liquor prediction model is composed of at least three different prediction sub-models, and the prediction sub-model algorithms include logistic regression, decision tree, random forest, gradient boosting tree, support vector regression and k-nearest neighbor; the primary liquor classification model is composed of at least three classification sub-models, and the classification sub-model algorithms include logistic regression, support vector machine, naive Bayes, k-nearest neighbor and decision tree.
6. The liquor identification method based on intelligent sensory technology and machine learning according to claim 5 is characterized in that: The method uses the characteristic value data in the training set to train the primary liquor prediction model and the primary classification model, and obtains the liquor prediction model and the liquor classification model according to the training results, including: The eigenvalue data in the training set are used to train each prediction sub-model and classification sub-model respectively, and the liquor prediction model and the liquor classification model are obtained according to the training results.
7. The liquor identification method based on intelligent sensory technology and machine learning according to claim 6 is characterized in that: The liquor prediction model and the liquor classification model are both established using the detection data of the electronic nose and the electronic tongue. The prediction target is the quantitative descriptive analysis result of the liquor samples, that is, the scores of different liquor flavor attributes, wherein the flavor attributes specifically include mellow aroma, aged aroma, fruity aroma, cellar aroma, grain aroma, koji aroma, sweet aroma, dregs aroma, sourness, sweetness, bitterness, mellowness, richness, freshness, harmony and longevity.
8. The liquor identification method based on intelligent sensory technology and machine learning according to claim 1, characterized in that: The step of inputting the characteristic value data in the test set into the liquor prediction model and obtaining the liquor prediction result through the liquor prediction model comprises: The eigenvalue data in the test set is input into the liquor prediction model. Each prediction sub-model in the liquor prediction model operates on the input eigenvalue data separately, and calculates the R of each prediction sub-model. 2 and prediction deviation, three prediction sub-models with better prediction effects were selected and the average prediction results were calculated as the prediction results of the liquor prediction model.
9. The liquor identification method based on intelligent sensory technology and machine learning according to claim 1, characterized in that: The step of inputting the characteristic value data in the test set into the liquor classification model respectively and obtaining the classification result of the liquor through the liquor classification model comprises: The eigenvalue data in the test set is input into the liquor classification model. Each classification sub-model in the liquor classification model operates on the input eigenvalue data separately, calculates the classification accuracy, precision and recall rate of each classification sub-model, and selects the results with the largest number of outputs among the three classification sub-models with better classification effects as the classification results of the liquor classification model.
Citation Information
Cited By
Method for identifying different-quality-grade alcohol-sweet base liquor of Maotai-flavor liquor based on GC-MS combined with multiple machine learning models
CN119291104A
Baijiu flavor perception evaluation method based on electroencephalogram and surface myoelectricity
CN120240981A
Apple identification method, device and equipment, medium and product
CN121119824A
