Instant green tea quality intelligent evaluation method based on multi-source data fusion and application

Abstract

The invention relates to an instant green tea quality intelligent evaluation method based on multi-source data fusion and application, belongs to the field of food detection, and relates to system construction of instant green tea quality intelligent identification. The invention mainly provides a rapid and intelligent tea quality evaluation technology, quality indexes of instant green tea are measured through a chemical analysis method, related data are collected through an intelligent evaluation technology, and the relation between the quality indexes and intelligent characteristics is revealed through correlation analysis. Based on an independent data source, an instant green tea variety classification model is constructed. By combining the multi-source data fusion technology and the machine learning algorithm, the efficient, accurate and comprehensive intelligent evaluation system and method for the quality of the instant green tea are constructed, classification of varieties of the instant green tea and prediction of quality indexes are achieved, fusion strategies of different quality index data are compared, and the quality of the instant green tea is improved. The model performance is optimized by introducing a feature selection method, and the practicability and reliability of the system are further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of food detection, and particularly relates to an intelligent evaluation method and application for the quality of instant green tea based on multi-source data fusion. Background Art

[0002] As a convenient and fast beverage, instant green tea is widely popular due to its unique flavor and health benefits. With the accelerating pace of modern life, consumers' requirements for the quality of instant tea are constantly increasing. However, the quality differences among different varieties of instant green tea greatly affect consumers' choices and the market competitiveness of brands. Currently, traditional quality evaluation methods usually rely on manual sensory evaluation and chemical analysis methods, but their subjectivity and inconsistency limit the accuracy and reliability of the evaluation. Therefore, there is an urgent need to seek more objective and efficient intelligent evaluation technologies to improve the accuracy and consistency of instant green tea quality detection. With the rapid development of intelligent evaluation technologies, it has become possible to evaluate the quality of instant green tea using computer vision, electronic nose, and electronic tongue. At the same time, the contents of important compounds such as tea polyphenols, catechins, and caffeine in instant green tea directly affect the quality and nutritional value of instant green tea. Therefore, studying rapid quantitative analysis methods for key components not only has theoretical significance but also has practical application value.

[0003] In recent years, the rapid development of intelligent evaluation technologies has provided new perspectives and methods for tea quality evaluation. These technologies include advanced sensor technologies such as computer vision, electronic nose, and electronic tongue, which can comprehensively analyze the appearance, smell, and taste of tea efficiently and accurately.

[0004] Currently, the application research on intelligent evaluation technologies in the quality detection of instant green tea is still relatively scarce at home and abroad. Existing research mainly focuses on quality detection aspects such as tea variety differentiation, grade differentiation, identification of geographical origin, and monitoring of the processing process.

[0005] Although there is still a lack of systematic data in the application research of instant green tea, by introducing intelligent evaluation technologies combined with machine learning algorithms, it is expected to achieve more accurate and rapid detection and analysis of the quality of instant green tea. Summary of the Invention

[0006] The purpose of the present invention is to overcome the defects of the above-mentioned existing technologies and provide an intelligent evaluation method and application for the quality of instant green tea based on multi-source data fusion. Through advanced technical means, it realizes efficient, accurate, and comprehensive evaluation of the quality of instant green tea to overcome the limitations of traditional quality evaluation methods such as high cost, complex operation, and susceptibility of results to subjective factors.

[0007] The present invention can be realized through the following technical solutions:

[0008] The first object of the present invention is to provide an intelligent evaluation method for the quality of instant green tea based on multi-source data fusion. The intelligent evaluation method includes the following steps:

[0009] Step 1: Select instant green tea as the raw material. Brew a part of the instant green tea with boiling water to obtain tea soup, and use the other part of the instant green tea as tea powder material. Use a computer vision device to collect visual images of the tea soup and tea powder material, use an electronic nose to obtain the aroma information of the tea soup and tea powder material, use an electronic tongue to collect the taste information of the tea soup, and use high performance liquid chromatography to measure the quality component information in the tea soup and tea powder;

[0010] Step 2: Take the aroma, visual, and taste information as characteristic parameter data respectively for dimensionality reduction processing. After dimensionality reduction, divide the data set, measure the quality indexes of the tea soup and tea powder material, and conduct correlation analysis and significance analysis on the quality indexes and characteristic parameters to form a fusion feature set;

[0011] Step 3: Perform feature selection on the fusion feature set formed in Step 2 to screen out the optimal feature subset. Based on the optimal feature subset, construct SVM, KNN, and RF regression classification models and PLSR, SVR, and RF as prediction models respectively to predict the content of the quality indexes of instant green tea, and use classification accuracy, R values of the training set and prediction set, RMSEC, RMSEP, and RPD and other indexes as evaluation criteria; 2 values, RMSEC, RMSEP, and RPD and other indexes as evaluation criteria;

[0012] Step 4: Based on the optimal feature subset formed in Step 3, use decision-level and feature-level data fusion. In the model architecture at the decision level, use the multiple linear regression (MLR) model and D-S evidence theory. In the feature level, introduce feature selection methods to perform multi-source fusion on the data respectively. According to the evaluation results, obtain a fusion model to improve the accuracy and stability of classification and prediction. For the variety identification of instant green tea, use the model architecture at the decision level, and for the quality identification of instant green tea, use the model architecture at the feature level, and use R 2 values, RMSEC, RMSEP, and RPD as evaluation indexes for evaluation to select the best fusion model.

[0013] Furthermore, the intelligent evaluation method includes the following steps:

[0014] Step 1: Select instant green tea as the raw material. Select a part of it and brew it with boiling water in a certain proportion to obtain tea soup, and the other part is tea powder material. Use an image acquisition device to collect visual images, use an olfactory sensing detection system to obtain the aroma information of the tea soup and tea powder, and use a taste sensing system to collect the taste information of the tea soup.

[0015] Step 2: Take the olfactory, visual, and taste information as feature numbers respectively for dimensionality reduction processing. After dimensionality reduction, divide the dataset, and after performing correlation analysis on the extraction of feature information from the data, conduct classification prediction in the constructed model training set.

[0016] Step 3: Perform PCA dimensionality reduction processing on the computer vision, electronic nose, and electronic tongue feature parameters collected in Step 1 to reduce data redundancy and noise. The 11 components related to the quality of matcha determined by high-performance liquid chromatography are the catechin monomer contents (EGC, C, EC, EGCG, and ECG) of 10 different types of instant green tea; TP, FAA, CAF, and TC. And conduct correlation analysis on the quality indicators and feature parameters to form a fusion feature set.

[0017] Step 4: Perform feature selection on the fusion feature set formed in Step 2 to screen out the optimal feature subset. Based on the optimal feature subset, construct SVM, KNN, and RF regression classification models and PLSR, SVR, and RF as prediction models respectively to predict the quality indicator contents of instant green tea samples. And use classification accuracy, R 2 values of the training set and prediction set, RMSEC, RMSEP, and RPD and other indicators as evaluation criteria.

[0018] Step 5: Apply data fusion at the decision level and feature set respectively to the variety classification and quality classification of instant green tea. In the construction of the decision-level model, use MLR and D-S evidence theory; in the feature level, introduce four feature selection methods of Pearson score, RFE, PSO, and Lasso regression to perform multi-source fusion on the data respectively. According to the evaluation results, optimize the model to improve the accuracy and stability of classification and prediction. For the variety recognition of instant green tea, use the decision-level model construction. For the quality recognition of instant green tea, use the feature-level model architecture and use R 2 values, RMSEC, RMSEP, and RPD as evaluation indicators.

[0019] Furthermore, the computer vision device includes:

[0020] A dark box, a bracket, a light source, a camera, and a container;

[0021] The bracket, light source, camera, and container are all arranged inside the dark box;

[0022] The light source and the camera are respectively connected to the bracket;

[0023] The camera, light source, and container are arranged in sequence from top to bottom;

[0024] The container is used to hold tea soup or tea powder materials.

[0025] Furthermore, the dark box includes a black matte acrylic plate in a cubic shape (400mm×400mm×600mm), and side sliding grooves and bottom grooves are arranged on the front side of the dark box body to enable the front panel to slide up and down and insert into the bottom groove, so that a closed space is formed inside the dark box to prevent the external light environment from affecting the sample collection;

[0026] The bracket includes a bearing platform (300mm×300mm), a support rod (400mm high) and an adjustment frame. The bearing platform is connected to the support rod, and the adjustment frame is connected to the support rod. A stepped circular groove is designed at the center point of the bearing platform for fixing the container to ensure that the acquisition position is consistent each time, and the adjustment frame is used to adjust the distance between the light source and the camera and the container;

[0027] The light source adopts a D65 annular shadowless light source specified by the International Commission on Illumination (CIE) that is close to the true resolution of daylight, has a color temperature of 6500k, an inner diameter of 75mm, and an outer diameter of 120mm, and can emit uniform light, thereby obtaining higher quality digital image information;

[0028] The camera is Nikon D5100 (CMOS), which is connected to a computer via a data cable and the camera parameters are adjusted by software to achieve image acquisition.

[0029] Furthermore, the olfactory sensing detection system includes an electronic nose.

[0030] Furthermore, the taste sensing system includes an electronic tongue.

[0031] Further, the regression classification model (discriminant model) includes one or more of SVM, KNN and RF;

[0032] The prediction model includes one or more of PLSR, SVR and RF.

[0033] Furthermore, the model evaluation indicators and calculation methods in the evaluation criteria are as follows:

[0034] Classification accuracy

[0035] Coefficient of determination R 2 value

[0036] Cross-validation root mean square error

[0037] Prediction root mean square error

[0038] Standard Deviation

[0039] Relative percentage deviation

[0040] where n is the number of samples in the data set; X i is the actual value of the i-th sample during the establishment of the prediction model; is the average value of the actual values of all samples during the establishment of the prediction model; Y i is the predicted value of the i-th sample during the establishment of the prediction model; is the average value of the predicted values of all samples during the establishment of the prediction model.

[0041] Furthermore, the larger R 2 and RPD are, the smaller RMSEC and RMSEP are, and the more reliable the established model is. When RPD is greater than or equal to 3.0, it indicates that the prediction effect of the model is significant and satisfactory; when RPD is between 2.5 and 3.0, it means that the prediction of the model is acceptable but there is still room for improvement; while when RPD is less than or equal to 2.5, it indicates that the prediction effect of the model fails to reach the acceptable level and needs further optimization and improvement.

[0042] Furthermore, the determination of the quality indexes of the tea soup and tea powder materials includes the following process:

[0043] The components in tea such as TP (tea polyphenols), FAA (total free amino acids), TP / FAA (the ratio of tea polyphenols to total free amino acids), CAF (caffeine), EGC (epigallocatechin), C (catechin), EC (epicatechin), EGCG (epigallocatechin gallate) are determined by liquid chromatography. For the tea powder material: the tea sample is ground into powder (tea powder material), homogenized, and then ultrasonic extraction and centrifugation treatment are carried out with ultrapure water or a specific solvent to obtain the extract. The supernatant is taken and filtered for measurement. A C-18 reverse-phase chromatographic column is selected, and the contents of tea polyphenols, free amino acids, caffeine, catechins, etc. are determined according to the national standard, and their data are recorded.

[0044] Furthermore, in step two, PCA dimensionality reduction processing is adopted for dimensionality reduction processing.

[0045] Furthermore, the feature selection method includes one or more of the four feature selection methods: Pearson score, recursive feature elimination (RFE), particle swarm optimization (PSO), and Lasso regression.

[0046] Furthermore, Pearson correlation analysis was carried out on 9 color characteristic parameters and quality indexes of instant green tea broth and tea powder materials. At the same time, 14 aroma characteristic parameters of the tea broth and tea powder materials were extracted by electronic nose technology, and the correlations between these aroma characteristics and 10 quality indexes were deeply analyzed. In addition, correlation analysis was also carried out on 18 taste characteristic parameters extracted by electronic tongue sensors and 10 instant green tea quality indexes. Among them, the determination of TP content was carried out with reference to GB / T 8313-2018 "Detection Methods for the Contents of Tea Polyphenols and Catechins in Tea". The determination of catechins and CAF content was carried out with reference to GB / T 8313-2018 "Detection Methods for the Contents of Tea Polyphenols and Catechins in Tea". The determination of FAA content was carried out with reference to GB / T 8314-2013 "Determination of Total Free Amino Acids in Tea".

[0047] Furthermore, in decision-level data fusion, a multiple linear regression (MLR) model was used to classify different varieties of instant green tea, and the D-S evidence theory was combined to evaluate the quality of green tea. At the same time, methods such as Pearson, recursive feature elimination (RFE), particle swarm optimization (PSO), and Lasso regression were used to evaluate the R 2 values, root mean square error of calibration set (RMSEC), root mean square error of prediction set (RMSEP), and relative prediction deviation (RPD) of six models to select the best fusion model.

[0048] The second object of the present invention is to provide an application of an intelligent evaluation method for the quality of instant green tea based on multi-source data fusion, and to use the method for intelligent identification and rapid prediction of the quality of instant green tea.

[0049] Compared with the prior art, the present invention has the following advantages:

[0050] 1. The present invention provides an intelligent evaluation method and application for the quality of instant green tea based on multi-source data fusion. The core lies in the organic combination of three intelligent evaluation technologies: computer vision, electronic nose, and electronic tongue, to construct a multi-source data fusion platform. Through image acquisition and processing, computer vision technology can accurately extract the color characteristics of instant green tea powder and tea broth, which intuitively reflect key information such as the oxidation degree and pigment distribution of instant green tea. The electronic nose technology simulates the olfactory system of mammals and objectively evaluates the aroma quality of tea by detecting the aroma components of instant green tea. The electronic tongue technology further simulates the human taste system and can accurately analyze the taste characteristics of instant green tea, providing more comprehensive data support for quality evaluation.

[0051] 2. The present invention provides an intelligent evaluation method and application for the quality of instant green tea based on multi-source data fusion. On the basis of data collection, the present invention applies a variety of machine learning algorithms, including support vector machine (SVM), random forest (RF), K-nearest neighbor (KNN), etc., to construct an instant green tea variety classification model and a quality index prediction model. Through the training and learning of a large amount of experimental data, these models can accurately identify different varieties of instant green tea and predict the content of its key quality indexes, such as tea polyphenols, catechins, caffeine, etc. The experimental results show that the machine learning-based models have shown excellent performance in both classification and prediction tasks, significantly superior to traditional chemical analysis methods and sensory evaluation methods.

[0052] 3. The present invention provides an intelligent evaluation method and application for the quality of instant green tea based on multi-source data fusion. In order to further improve the accuracy and generalization ability of the model, the present invention also introduces feature-level fusion and decision-level fusion strategies. The feature-level fusion strategy integrates data from different intelligent evaluation technologies to form a more comprehensive quality evaluation data set. The decision-level fusion strategy constructs multiple classification or prediction models based on independent data sources and synthesizes the decision results of these models through certain fusion rules to obtain the final judgment. Experiments prove that the application of the fusion strategy significantly improves the classification accuracy and prediction precision of the model.

[0053] 4. The present invention provides an intelligent evaluation method and application for the quality of instant green tea based on multi-source data fusion, and deeply studies the influence of storage time on the quality of instant green tea. By simulating instant green tea samples under different storage conditions, analyzing the changes in quality indexes and volatile substance content during storage time, an instant green tea storage time classification model is constructed. This model can accurately predict the storage time of instant green tea, providing a scientific basis for the shelf life management and market sales of products.

[0054] 4. The present invention provides an intelligent evaluation method and application for the quality of instant green tea based on multi-source data fusion. In terms of model optimization, the present invention introduces a variety of feature selection methods, such as particle swarm optimization (PSO), recursive feature elimination (RFE), Lasso regression, etc. These methods optimize the feature subset, reduce data redundancy, and improve the prediction accuracy and generalization ability of the model. The experimental results show that the model optimized by the feature selection method shows higher accuracy and stability in both classification and prediction tasks.

[0055] In summary, the present invention constructs an efficient, accurate, and comprehensive intelligent evaluation system and method for the quality of instant green tea by combining multi-source data fusion technology and machine learning algorithms. This system not only realizes the classification of instant green tea varieties and the prediction of quality indicators, but also deeply studies the impact of storage time on quality, providing strong support for the production, quality control, and market sales of instant green tea. At the same time, by introducing a feature selection method to optimize the model performance, the practicality and reliability of the system are further improved, opening up a new path for the intelligent development of the tea industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 In the embodiment of the present invention, the tea powder and tea soup images identified by computer vision are shown, where (A) is the tea powder and (B) is the tea soup.

[0057] Figure 2 In the embodiment of the present invention, the PCA results and (B) LDA results of the electronic nose for different types of instant green tea (A) tea powder, (C) PCA results and (D) LDA results of the tea soup.

[0058] Figure 3 In the embodiment of the present invention, the PCA results and (B) LDA results of the electronic tongue for different types of instant green tea.

[0059] Figure 4 In the embodiment of the present invention, the contents of catechins in different types of instant green tea (A) EGC (B) C (C) EC (D) EGCG (E) ECG (F) TC, note: different lowercase letters indicate significant differences (p<0.05).

[0060] Figure 5 Schematic diagram of the classification fusion strategy of the intelligent evaluation method for the quality of instant green tea based on multi-source data fusion in the present invention, (A) feature-level fusion, (B) decision-level fusion.

[0061] Figure 6 Schematic diagram of the process of the intelligent evaluation method for the quality of instant green tea based on multi-source data fusion in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0062] The present invention will be described in detail below with reference to the drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and gives the detailed implementation manner and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0063] In the present technical solution, the component models, material names, connection structures, applications, algorithms, etc. that are not clearly described are regarded as common technical features disclosed in the prior art.

[0064] In the present invention, unless otherwise clearly specified or limited, terms such as "installation", "connection", "linkage", "fixation" and the like shall be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral body; it may be a mechanical connection or an electrical connection; it may be a direct connection or an indirect connection through an intermediate medium, and may be the communication inside two elements or the interaction relationship between two elements; "upper", "lower", "left", "right", etc. are only used to represent relative position relationships. When the absolute position of the object being described changes, the relative position relationship may change. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0065] It should be noted that in the present invention, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising one..." does not exclude the existence of another identical element in the process, method, article or device comprising the said element.

[0066] In the following embodiments, unless otherwise specified, the raw materials or processing techniques are all conventional commercially available raw materials or conventional processing techniques in the art.

[0067] The present invention relates to an intelligent evaluation method and application for the quality of instant green tea based on multi-source data fusion, belonging to the field of food detection, and is related to the construction of a system for intelligent identification of the quality of instant green tea. The present invention mainly provides a rapid intelligent quality evaluation technology for tea. The quality indexes of instant green tea are determined by chemical analysis methods, and relevant data are collected by intelligent evaluation technologies. Then, correlation analysis is used to reveal the relationship between the quality indexes and intelligent features. Based on independent data sources, a classification model for instant green tea varieties is constructed. By combining multi-source data fusion technology and machine learning algorithms, the present invention constructs a set of efficient, accurate and comprehensive intelligent evaluation systems and methods for the quality of instant green tea, which not only realizes the classification of instant green tea varieties and the prediction of quality indexes, but also deeply studies the influence of storage time on quality, and further improves the practicability and reliability of the system by introducing a feature selection method to optimize the model performance.

[0068] In the following embodiments:

[0069] The computer vision device includes: a dark box, a bracket, a light source, a camera, and a holding container; the bracket, the light source, the camera, and the holding container are all arranged inside the dark box; the light source and the camera are respectively connected to the bracket; the camera, the light source, and the holding container are arranged in sequence from top to bottom; the holding container is used to hold tea soup or tea powder materials. The dark box includes a black matte acrylic board, which is cube-shaped. Side chutes and bottom grooves are provided on the front side of the dark box body to enable the front panel to slide up and down and insert into the bottom groove, so as to form a closed space inside the dark box and avoid the influence of the external light environment on sample collection; the bracket includes a bearing platform, a support rod, and an adjustment frame. The bearing platform is connected to the support rod, and the adjustment frame is connected to the support rod. A stepped circular groove is designed at the center point of the bearing platform for fixing the holding container to ensure the consistency of the collection position each time, and the adjustment frame is used to adjust the distances between the light source, the camera, and the holding container; the light source uses an annular shadowless light source; the camera is connected to a computer through a data cable and the camera parameters are adjusted with software to achieve image collection.

[0070] The dark box includes a black matte acrylic board, which is cube-shaped (400mm×400mm×600mm). Side chutes and bottom grooves are provided on the front side of the dark box body to enable the front panel to slide up and down and insert into the bottom groove, so as to form a closed space inside the dark box and avoid the influence of the external light environment on sample collection;

[0071] The bracket includes a bearing platform (300mm×300mm), a support rod (400mm high), and an adjustment frame. The bearing platform is connected to the support rod, and the adjustment frame is connected to the support rod. A stepped circular groove is designed at the center point of the bearing platform for fixing the holding container to ensure the consistency of the collection position each time, and the adjustment frame is used to adjust the distances between the light source, the camera, and the holding container;

[0072] The light source uses a D65 annular shadowless light source that approximates the true resolution of daylight as specified by the International Commission on Illumination (CIE). Its color temperature is 6500k, the inner diameter is 75mm, and the outer diameter is 120mm. It can emit uniform light, thereby obtaining higher-quality digital image information;

[0073] The camera selected is Nikon D5100 (CMOS). It is connected to a computer through a data cable and the camera parameters are adjusted with software to achieve image collection.

[0074] The electronic nose used is the CNose-14 electronic nose produced by Shanghai Baosheng Industrial Development Co., Ltd., which consists of a sampling system, a sensor array, a signal acquisition system and a gas cleaning channel. Among them, the sampling system is responsible for the collection of sample gas and the control of gas flow rate; the sensor array is the core component of the electronic nose system for detecting sample gas, consisting of 14 imported metal oxide semiconductor (MOS) sensors with different properties, and having different sensitivities to different volatile compounds; the signal acquisition system can control the electronic nose system to collect response signals.

[0075] For the flavor collection of the electronic tongue, the Smart Tongue electronic tongue produced by Shanghai Baosheng Industrial Development Co., Ltd. is used, which consists of a sampling system, a sensor array, a data acquisition and analysis system. Its sensor array consists of 6 chemically sensitive sensors with cross-sensitivity (platinum S1, gold S2, palladium S3, tungsten S4, titanium S5, silver S6). Before each test, the electronic tongue system needs to be preheated for more than 30 minutes, and the sensors are cleaned with deionized water, and it is confirmed that the protective liquid in the Ag / AgCl reference electrode is not dry; during sample detection, the sample to be tested in the beaker is immersed in the bottom of the electrode, so that the taste signal causes a change in the potential of the lipid-sensitive membrane through the sensor array and converts it into an electrical signal; finally, the software receives, records, processes the electrical signal response value and saves it in the computer.

[0076] Embodiment

[0077] As Figures 5 - 6 shown, this embodiment provides an intelligent evaluation method and application for the quality of instant green tea based on multi-source data fusion. The intelligent evaluation method includes the following steps:

[0078] Step 1: Select instant green tea as the raw material. Brew a part of the instant green tea with boiling water into tea soup, and use the other part of the instant green tea as tea powder material. Use a computer vision device to collect visual images of the tea soup and tea powder material, use an electronic nose to obtain the aroma information of the tea soup and tea powder material, use an electronic tongue to collect the taste information of the tea soup, and use high-performance liquid chromatography to measure the quality component information in the tea soup and tea powder.

[0079] Step 2: Take the aroma, visual, and taste information as feature parameter data respectively for dimensionality reduction processing. After dimensionality reduction, divide the data set, measure the quality indicators of the tea soup and tea powder material, and perform a correlation analysis on the quality indicators and feature parameters to form a fusion feature set.

[0080] Step 3: Perform feature selection on the fusion feature set formed in Step 2 to screen out the optimal feature subset. Based on the optimal feature subset, construct SVM, KNN, and RF regression classification models and PLSR, SVR, and RF as prediction models respectively to predict the content of the quality indicators of instant green tea, and use the classification accuracy rate, R of the training set and prediction set2 Indexes such as value, RMSEC, RMSEP, and RPD are used as evaluation criteria;

[0081] Step 4: Based on the optimal feature subset formed in Step 3, data fusion at the decision level and feature level is used. In the model architecture at the decision level, a multiple linear regression (MLR) model and D-S evidence theory are used. In the feature level, a feature selection method is introduced to perform multi-source fusion on the data respectively. According to the evaluation results, a fusion model is obtained to improve the accuracy and stability of classification and prediction. For the variety identification of instant green tea, the model architecture at the decision level is used, and for the quality identification of instant green tea, the model architecture at the feature level is used, and R 2 Values, RMSEC, RMSEP, and RPD are used as evaluation indicators for evaluation to select the best fusion model. The models used in Step 4 are mainly based on statistics for feature fusion, and are divided into different fusion strategies such as the decision level and the feature level. The model used in Step 3 is a regression classification prediction model, following a progressive relationship of first predicting and then performing feature correlation analysis.

[0082] Specifically, it includes the following steps:

[0083] Six kinds of instant green tea (purchased from Zhejiang Mingbao Company): G302, G305, G306, G307, J753, ZTHG505 were selected, and were used as samples in the form of tea powder and tea soup respectively, and were detected by computer vision, electronic nose, and electronic tongue.

[0084] In the computer vision detection, a computer vision device was built independently to collect images of tea powder and tea soup samples. 3 g of tea powder samples were weighed and evenly spread on a glass dish, and for tea soup samples, 0.5 g of tea powder was brewed with 150 mL of ultrapure water (boiling water), and an equal amount was taken and placed in a porcelain evaluation cup. All images were collected under the same conditions and saved in JPG format. Subsequently, Python software was used to preprocess the images, automatically select the region of interest of 600×600 pixels near the center point, and extract multiple color feature parameters such as the red, green, and blue channels, as well as hue, saturation, and brightness. Each sample was prepared 6 times, and each time 5 detections were repeated, a total of 360 groups of data were obtained. The specific images of tea powder and tea soup recognized by computer vision are as Figure 1 shown, where Figure 1 (A) is tea powder, Figure 1 (B) is tea soup. Before the electronic nose detection, a pre-experiment was carried out to determine the best detection parameters. The amount of tea powder sample was 5 g, the headspace time was 60 min, and the ambient temperature was 25°C; the amount of tea soup sample was 15 mL, and other conditions were the same. After the samples were placed in a headspace bottle and sealed and left standing, the sample gas was aspirated at a specific rate for detection. Each sample was also repeated 6 times, and each time 5 detections were repeated, a total of 360 groups of data sets were obtained. As Figure 2As shown, the schematic diagram of the results of the electronic nose for different types of instant green tea is given. Figure 2 (A) is the PCA result of tea powder and Figure 2 (B) is the LDA result of tea powder, Figure 2 (C) is the PCA result of tea soup and Figure 2 (D) is the LDA result of tea soup.

[0085] For the electronic tongue detection, the optimal detection parameters of the tea soup samples were determined according to the pre-experiment results. 0.5 g of tea powder was dissolved in 150 mL of ultrapure water, and 30 mL was taken as the sample to be detected. After setting the acquisition frequency, the sensor signal amplification factor, and the detection time, the sample was immersed in the bottom of the electrode for detection, and the data was exported for analysis and processing. Each sample was repeated 6 times, and each repetition was detected 5 times, and a total of 180 sets of data sets were obtained. As Figure 3 shown, the schematic diagram of the results of the electronic tongue for different types of instant green tea is given. Figure 3 (A) is the PCA result of tea soup and Figure 3 (B) is the LDA result of tea soup.

[0086] The determination of the quality indexes of the tea soup and tea powder materials includes the following process:

[0087] The components in tea such as TP (tea polyphenols), FAA (total free amino acids), TP / FAA (the ratio of tea polyphenols to total free amino acids), CAF (caffeine), EGC (epigallocatechin), C (catechin), EC (epicatechin), EGCG (epigallocatechin gallate) were determined by liquid chromatography. The tea samples were ground into powder, homogenized, and then ultrasonic extraction and centrifugation treatment were carried out with ultrapure water or specific solvents to obtain the extract. The supernatant was filtered and then used for detection. A C-18 reversed-phase chromatographic column was selected, and the contents of tea polyphenols, free amino acids, caffeine, and catechins were determined according to the national standard, and the data was recorded.

[0088] Pearson correlation analysis was carried out on 9 color characteristic parameters and quality indexes of instant green tea soup and tea powder materials. At the same time, 14 aroma characteristic parameters of tea soup and tea powder materials were extracted by electronic nose technology, and the correlations between these aroma characteristics and 10 quality indexes were deeply analyzed. In addition, the correlation analysis was also carried out on 18 taste characteristic parameters extracted by the electronic tongue sensor and 10 quality indexes of instant green tea. Among them, the determination of TP content was carried out with reference to GB / T 8313-2018 "Determination Method for the Contents of Tea Polyphenols and Catechins in Tea". The determination of catechins and CAF content was carried out with reference to GB / T 8313-2018 "Determination Method for the Contents of Tea Polyphenols and Catechins in Tea". The determination of FAA content was carried out with reference to GB / T 8314-2013 "Determination of Total Free Amino Acids in Tea".

[0089] During the modeling process, two methods, namely the classification model and the regression model, were adopted. The classification model (regression classification model) includes Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Random Forest (RF), which are used for the qualitative discrimination of instant green tea varieties; the prediction model includes Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), and Random Forest Regression, which are used for the rapid quantitative prediction of quality indicators. As Figure 4 shown, it illustrates the contents of catechins EGC, C, EC, EGCG, ECG, and TC in different varieties of instant green tea.

[0090] The model evaluation indicators and calculation methods in the evaluation criteria are as follows:

[0091] Classification accuracy

[0092] Coefficient of determination R 2 value

[0093] Root Mean Square Error of Cross-Validation

[0094] Root Mean Square Error of Prediction

[0095] Standard deviation

[0096] Relative percentage deviation

[0097] The dataset was subjected to z-score standardization before modeling to eliminate the influence of dimensions. The data was divided into a 70% training set and a 30% test set, and 10-fold cross-validation was used to evaluate the model performance. The specific approach was to evenly divide the original dataset into 10 mutually exclusive subsets. Each time, 9 subsets were used for training and 1 subset was used for testing. This was repeated 10 times, and the average result was taken as the final evaluation result to improve the reliability and generalization ability of the evaluation.

[0098] The performance of the classification model was evaluated using classification accuracy, while the prediction model used the coefficient of determination (R2), root mean square error of cross-validation (RMSEC), root mean square error of prediction (RMSEP), and relative percentage deviation (RPD) as evaluation indicators. The larger the R2 and RPD, and the smaller the RMSECV and RMSEP, the more reliable the model. When RPD is greater than or equal to 3.0, the model prediction effect is significant; when it is between 2.5 and 3.0, the prediction performance is acceptable; when it is less than or equal to 2.5, further optimization and improvement are required. The main model evaluation indicators are as described above to ensure the comprehensiveness and accuracy of model evaluation.

[0099] The classification fusion strategy is to comprehensively analyze the data information extracted by computer vision, electronic nose, and electronic tongue using feature-level fusion and decision-level fusion to construct a classification model. In the feature-level fusion strategy, first, color features (9 dimensions), aroma features (14 dimensions), and taste features (18 dimensions) that can represent instant green tea powder and tea soup are extracted from the original information of computer vision, electronic nose, and electronic tongue respectively. Next, the features of tea powder and tea soup are merged pairwise to generate 4 new two-source fusion feature sets. Finally, SVM and RF classification models are constructed based on the fused feature datasets respectively. In the feature extraction process, the same as the feature-level fusion strategy, SVM and RF classification models are independently established based on the features of each data source respectively to obtain their respective decision results. By integrating the model decision results from computer vision, electronic nose, and electronic tongue, and then inputting them into the MLR model and D-S evidence theory, decision-level data fusion is achieved, and finally, classification results are generated.

[0100] The results of multi-feature fusion are shown in Tables 1-4 below:

[0101] Table 1 Classification results of SVM and RF for instant green tea categories based on the feature-level fusion strategy.

[0102]

[0103] Table 2 Classification results of the MLR model based on the decision-level fusion strategy

[0104]

[0105]

[0106] Table 3 Classification results of the D-S evidence theory based on the decision-level fusion strategy

[0107]

[0108]

[0109] Table 4 Prediction results of quality index contents based on the fused feature sets

[0110]

[0111] As can be seen from Tables 1 to 4, in this embodiment, three intelligent evaluation technologies of computer vision, electronic nose and electronic tongue are organically combined to construct a multi-source data fusion platform. Through image acquisition and processing, computer vision technology can accurately extract the color features of instant green tea powder and tea soup, which intuitively reflect key information such as the oxidation degree and pigment distribution of instant green tea. The electronic nose technology simulates the olfactory system of mammals and objectively evaluates the aroma quality of tea by detecting the aroma components of instant green tea. The electronic tongue technology further simulates the human taste system and can accurately analyze the taste characteristics of instant green tea, providing more comprehensive data support for quality evaluation.

[0112] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Those skilled in the art should understand that although the present invention has been described in detail with reference to the foregoing embodiments, they can still modify the technical solutions of the foregoing embodiments, or perform equivalent replacements for some of the technical features, and these modifications or replacements do not exceed the protection scope of the present invention. The protection scope of the present invention is determined by the appended claims.

Claims

1. An intelligent evaluation method for instant green tea quality based on multi-source data fusion, characterized in that: The intelligent evaluation method comprises the following steps: Step 1: Instant green tea is selected as the raw material, and a portion of the instant green tea is brewed with boiling water to form tea soup, and the other portion of the instant green tea is used as tea powder material. A computer vision device is used to collect visual images of the tea soup and the tea powder material, an electronic nose is used to obtain the aroma information of the tea soup and the tea powder material, an electronic tongue is used to collect the taste information of the tea soup, and a high performance liquid chromatography is used to measure the quality component information in the tea soup and the tea powder; Step 2: The aroma, visual, and taste information are used as feature parameter data for dimensionality reduction. After dimensionality reduction, the data set is divided, the quality indicators of tea soup and tea powder materials are measured, and the correlation between the quality indicators and feature parameters is analyzed to form a fusion feature set; Step 3: Perform feature selection on the fusion feature set formed in step 2 to screen out the optimal feature subset. Based on the optimal feature subset, a regression classification model and a prediction model are constructed to predict the quality index content of instant green tea. The classification accuracy, R of the training set and the prediction set are used to compare the quality index content of instant green tea. 2 Value, RMSEC, RMSEP and RPD indicators were used as evaluation criteria; Step 4: Based on the optimal feature subset formed in step 3, the decision-level and feature-level data fusion was used. The multivariate linear regression model and DS evidence theory were used in the decision-level model architecture. The feature selection method was introduced in the feature level to fuse the data from multiple sources. According to the evaluation results, the fusion model was obtained to improve the accuracy and stability of classification and prediction. The decision-level and feature-level model architectures were used for the variety identification and quality prediction of instant green tea. The feature-level model architecture was used for the quality identification of instant green tea, and R was used to analyze the data. 2 The value, RMSEC, RMSEP and RPD were used as evaluation indicators to select the best fusion model.

2. The method for intelligent evaluation of instant green tea quality based on multi-source data fusion according to claim 1, characterized in that: The computer vision device comprises: Dark box, stand, light source, camera and container; The support, light source, camera and container are all arranged inside the dark box; The light source and the camera are respectively connected to the bracket; The camera, light source and container are arranged in sequence from top to bottom; The container is used for containing tea soup or tea powder material.

3. The method for intelligent evaluation of instant green tea quality based on multi-source data fusion according to claim 2, characterized in that: The dark box includes a black matte acrylic plate in a cubic shape, and side sliding grooves and bottom grooves are arranged on the front side of the dark box body to enable the front panel to slide up and down and insert into the bottom groove, so that a closed space is formed inside the dark box to prevent the external light environment from affecting the sample collection; The bracket includes a bearing platform, a support rod and an adjustment frame, the bearing platform is connected to the support rod, the adjustment frame is connected to the support rod, a stepped circular groove is designed at the center point of the bearing platform for fixing the container to ensure that the acquisition position is consistent each time, and the adjustment frame is used to adjust the distance between the light source and the camera and the container; The light source is a ring-shaped shadowless light source; The camera is connected to a computer via a data line and the camera parameters are adjusted by software to achieve image acquisition.

4. The method for intelligent evaluation of instant green tea quality based on multi-source data fusion according to claim 1, characterized in that: The olfactory sensing detection system includes an electronic nose; The taste sensing system includes an electronic tongue.

5. The method for intelligent evaluation of instant green tea quality based on multi-source data fusion according to claim 1, characterized in that: The regression classification model includes one or more of SVM, KNN and RF; The prediction model includes one or more of PLSR, SVR and RF.

6. The method for intelligent evaluation of instant green tea quality based on multi-source data fusion according to claim 1, characterized in that: The model evaluation indicators and calculation methods in the evaluation criteria are as follows: Classification accuracy Coefficient of determination R 2 value Cross-validation root mean square error Prediction root mean square error Standard Deviation Relative percentage deviation 7. The method for intelligent evaluation of instant green tea quality based on multi-source data fusion according to claim 1, characterized in that: The determination of the quality indexes of the tea soup and tea powder materials includes the following process: Tea polyphenols, total free amino acids, the ratio of tea polyphenols to total free amino acids, caffeine, epigallocatechin, catechin, epicatechin, and epigallocatechin gallate were determined by liquid chromatography.

8. The method for intelligent evaluation of instant green tea quality based on multi-source data fusion according to claim 1, characterized in that: In step 2, PCA dimensionality reduction is used for dimensionality reduction.

9. The method for intelligent evaluation of instant green tea quality based on multi-source data fusion according to claim 1, characterized in that: The feature selection method includes one or more of four feature selection methods: Pearson score, recursive feature elimination, particle swarm optimization and Lasso regression.

10. An application of the instant green tea quality intelligent evaluation method based on multi-source data fusion as claimed in claims 1 to 9, characterized in that: The method is used for intelligent identification and prediction of instant green tea quality.

Images