An Artificial Intelligence-Based System and Method for Detecting Flavor Components in Plant-Based Beverages

Abstract

This application provides an artificial intelligence-based system and method for detecting flavor components in plant-based beverages. It analyzes the sensory contribution spectrum of flavor compounds in a target plant-based beverage sample using a multidimensional raw data matrix. Principal component vectors of flavor compounds are extracted from the multidimensional raw data matrix, and key flavor markers are determined based on the sensory contribution spectrum. Key spectral fingerprint features of flavor components are extracted from the spectral detection results of the target beverage sample based on these key flavor markers. A flavor component prediction model is constructed for the target plant-based beverage sample, and then the flavor score of the flavor components is determined by combining the key spectral fingerprint features. Using this method, the specific contribution of each substance to the overall sensory attributes can be analyzed to identify key flavor compounds in traditional plant-based beverage flavor evaluation.

Description

Technical Field

[0001] This application relates to the field of flavor component detection technology, and more specifically, to an artificial intelligence-based system and method for detecting flavor components in plant-based beverages. Background Technology

[0002] Flavor component detection is a key technology at the intersection of food science, sensory analysis, chemistry, and artificial intelligence. It aims to identify, quantify, and analyze the chemical components in food and beverages that affect sensory experience (aroma, taste, and mouthfeel). Its core objective is to establish an objective link between chemical components and human sensory perception, and to achieve scientific, digital, and intelligent evaluation of flavor.

[0003] With the rapid development of the plant-based beverage market, consumers' demands for diversified and personalized product flavors are increasing. Flavor has become a core indicator determining the quality and market competitiveness of plant-based beverages. The flavor of plant-based beverages is jointly determined by a variety of volatile substances (such as aroma components like esters, alcohols, and aldehydes) and non-volatile substances (such as taste components like sugars, organic acids, and amino acids). However, plant-based beverages contain hundreds or even thousands of volatile (e.g., aroma) and non-volatile (e.g., taste and mouthfeel) compounds. The contributions of these substances to the overall flavor vary greatly; some are key, while others are background or interference. Traditional chemical analysis methods (such as chromatography-mass spectrometry) can detect a large number of substances, but it is difficult to efficiently and accurately screen out the core flavor substances that truly play a decisive role in sensory attributes (e.g., "rich floral aroma," "mellow taste," "balanced bitterness"). Therefore, how to analyze the specific contribution of each substance to the overall sensory attributes in order to identify the key flavor substances in the traditional evaluation of plant-based beverage flavor has become a problem faced by the industry. Summary of the Invention

[0004] This application provides an artificial intelligence-based system and method for detecting flavor components in plant-based beverages, which can analyze the specific contribution of each substance to the overall sensory attributes in order to identify key flavor substances in traditional plant-based beverage flavor evaluation.

[0005] In a first aspect, this application provides an artificial intelligence-based method for detecting flavor components in plant-based beverages, comprising the following steps:

[0006] The content of each flavor substance in the target plant beverage sample was detected, and a multidimensional raw data matrix of the flavor substance content in the target plant beverage sample was obtained.

[0007] Based on the sensory evaluation data of the target plant beverage samples and the multidimensional original data matrix, the sensory contribution spectrum of volatile and non-volatile flavor substances in the target plant beverage samples to the overall sensory attributes of the target plant beverage samples is analyzed. Principal component analysis is used to reduce the dimensionality of the multidimensional original data matrix to obtain the principal component vector of flavor substances in the target plant beverage samples. The principal component vector is used to reflect the synergistic change characteristics among flavor substances.

[0008] By combining the sensory contribution spectrum with the synergistic change characteristics among flavor substances reflected by the principal component vector, the contribution of the core components affecting the flavor of the target plant beverage sample is analyzed, and the key flavor markers of the flavor components in the target plant beverage sample are obtained.

[0009] Spectroscopic detection was performed on the target plant beverage sample, and key spectral fingerprint features of flavor components in the target plant beverage sample were extracted from the spectral detection results based on key flavor markers.

[0010] A flavor component prediction model for the target plant beverage sample is constructed based on historical spectral data of the plant beverage. The key spectral fingerprint features are input into the flavor component prediction model to output the flavor score of the flavor components in the target plant beverage sample.

[0011] In some embodiments, the sensory contribution spectrum of volatile and non-volatile flavor compounds in the target plant beverage sample to the overall sensory attributes of the target plant beverage sample is analyzed based on sensory evaluation data of the target plant beverage sample and the multidimensional raw data matrix. Specifically, this includes:

[0012] Obtain sensory evaluation data of the target plant beverage samples;

[0013] The correlation coefficients between volatile and non-volatile flavor compounds and various sensory attributes are calculated based on the sensory evaluation data and the multidimensional raw data matrix.

[0014] Based on all correlation coefficients, a sensory contribution spectrum of volatile and non-volatile flavor compounds in the target plant beverage sample to the overall sensory properties of the target plant beverage sample was constructed.

[0015] In some embodiments, principal component analysis is used to reduce the dimensionality of the multidimensional original data matrix to obtain the principal component vector of flavor substances in the target plant beverage sample, specifically including:

[0016] The multidimensional original data matrix is ​​standardized to obtain a standardized multidimensional original data matrix.

[0017] Principal component analysis algorithm was used to calculate eigenvalues ​​of the standardized multidimensional raw data matrix to obtain eigenvalue data of flavor substances in the target plant beverage sample.

[0018] Determine the characteristic thresholds of flavor compounds in target plant beverage samples;

[0019] The feature value data is judged based on the feature threshold, and then the principal component vector of flavor substances in the target plant beverage sample is obtained.

[0020] In some embodiments, the contribution of the core components affecting the flavor of the target plant beverage sample is analyzed by combining the sensory contribution spectrum with the synergistic variation characteristics among flavor substances reflected by the principal component vector, and the key flavor markers of the flavor components in the target plant beverage sample are specifically included:

[0021] Based on the sensory contribution spectrum, multiple high-contribution flavor compounds affecting the flavor of plant-based beverages were extracted from the target plant-based beverage sample;

[0022] Multiple core flavor compounds selected from the principal component loadings of the principal component vector;

[0023] Intersection analysis of each high-contribution flavor compound and each core flavor compound was performed to obtain multiple candidate core components of the target plant beverage sample.

[0024] The contribution score of each candidate core component to the flavor of the plant beverage is calculated, thereby obtaining the key flavor markers of the flavor components in the target plant beverage sample.

[0025] In some embodiments, performing spectral detection on a target plant beverage sample and extracting key spectral fingerprint features of flavor components in the target plant beverage sample from the spectral detection results based on key flavor markers specifically includes:

[0026] The target plant beverage sample was subjected to spectral analysis, and the spectral analysis results were obtained.

[0027] The spectral detection results are denoised to obtain denoised spectral detection results;

[0028] Based on key flavor markers, characteristic spectral ranges, absorption peak intensities, and peak shape changes that are highly correlated with the changes in the content of key flavor markers are extracted from the denoised spectral detection results.

[0029] The characteristic spectral range, absorption peak intensity, and peak shape changes are used as key spectral fingerprint features of flavor components in the target plant beverage sample.

[0030] In some embodiments, constructing a flavor component prediction model for a target plant beverage sample based on historical spectral data of the plant beverage specifically includes:

[0031] Obtain historical spectral data of plant-based beverages;

[0032] The historical spectral data is divided into a training set and a validation set;

[0033] A flavor component prediction model for the target plant beverage sample is constructed based on an artificial intelligence algorithm combined with the training and validation sets.

[0034] In some embodiments, inputting the key spectral fingerprint features into the flavor component prediction model to output a flavor score for the flavor components in the target plant beverage sample specifically includes:

[0035] The key spectral fingerprint features are standardized, and the standardized key spectral fingerprint features are input into the flavor component prediction model.

[0036] The flavor component prediction model calculates the standardized key spectral fingerprint features of the input to output the flavor score of the flavor components in the target plant beverage sample.

[0037] In some embodiments, the content of various flavor substances in the target plant beverage sample is detected by gas chromatography-mass spectrometry and high performance liquid chromatography.

[0038] In some embodiments, near-infrared spectroscopy is used to perform spectral detection on the target plant beverage sample.

[0039] Secondly, this application provides an artificial intelligence-based system for detecting flavor components in plant-based beverages, the system comprising:

[0040] The detection module is used to detect the content of various flavor substances in the target plant beverage sample and obtain a multidimensional raw data matrix of flavor substance content in the target plant beverage sample.

[0041] The processing module is used to analyze the sensory contribution spectrum of volatile and non-volatile flavor substances in the target plant beverage sample to the overall sensory attributes of the target plant beverage sample based on the sensory evaluation data of the target plant beverage sample and the multidimensional original data matrix. Principal component analysis is used to reduce the dimensionality of the multidimensional original data matrix to obtain the principal component vector of flavor substances in the target plant beverage sample. The principal component vector is used to reflect the synergistic change characteristics among flavor substances.

[0042] The processing module is also used to analyze the contribution of the core components affecting the flavor of the target plant beverage sample by combining the sensory contribution spectrum with the synergistic change characteristics between flavor substances reflected by the principal component vector, and to obtain the key flavor markers of the flavor components in the target plant beverage sample.

[0043] The processing module is also used to perform spectral detection on the target plant beverage sample and extract the key spectral fingerprint features of the flavor components in the target plant beverage sample from the spectral detection results based on key flavor markers.

[0044] The execution module is used to construct a flavor component prediction model for the target plant beverage sample based on historical spectral data of the plant beverage, and input the key spectral fingerprint features into the flavor component prediction model to output the flavor score of the flavor components in the target plant beverage sample.

[0045] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0046] The artificial intelligence-based plant beverage flavor component detection system and method provided in this application first detects the content of each flavor substance in the target plant beverage sample, obtaining a multidimensional raw data matrix of flavor substance content in the target plant beverage sample; based on the sensory evaluation data of the target plant beverage sample and the multidimensional raw data matrix, the sensory contribution spectrum of volatile and non-volatile flavor substances in the target plant beverage sample to the overall sensory attributes of the target plant beverage sample is analyzed; principal component analysis is used to reduce the dimensionality of the multidimensional raw data matrix to obtain the principal component vector of flavor substances in the target plant beverage sample; by combining the sensory contribution spectrum with the synergistic change characteristics between flavor substances reflected by the principal component vector, the contribution degree of the core components affecting the flavor of the plant beverage in the target plant beverage sample is analyzed, obtaining the key flavor markers of flavor components in the target plant beverage sample; the target plant beverage sample is subjected to spectral detection, and the key spectral fingerprint features of flavor components in the target plant beverage sample are extracted from the spectral detection results based on the key flavor markers; a flavor component prediction model of the target plant beverage sample is constructed based on the historical spectral data of the plant beverage, and the key spectral fingerprint features are input into the flavor component prediction model to output the flavor score of flavor components in the target plant beverage sample.

[0047] Therefore, in the process of detecting flavor components in plant-based beverages, this application firstly obtains a multidimensional raw data matrix through comprehensive detection of flavor substance content, providing a solid data foundation for subsequent analysis and ensuring that the analysis is based on real and comprehensive information on the substance composition. Secondly, sensory evaluation data is combined with the raw data matrix to obtain a sensory contribution spectrum, directly establishing the correlation between flavor substance content and overall sensory attributes, quantifying the specific sensory contributions of volatile and non-volatile flavor substances, and clarifying the influence weight of different substances on sensory experience. Principal component analysis is used to reduce the dimensionality of the multidimensional data to obtain principal component vectors, simplifying the complex data structure, eliminating redundant information, highlighting core influencing factors, and reducing analytical interference. Through joint analysis of sensory contribution spectrum and principal component vectors, the core components affecting flavor are accurately identified, and key flavor markers are successfully obtained, achieving accurate identification of key flavor substances. Based on the key flavor markers, spectral fingerprint features are extracted, establishing the correspondence between key substances and spectral signals, providing specific feature basis for rapid detection. Finally, a predictive model is constructed by combining artificial intelligence algorithms and historical spectral data to output flavor scores, achieving the goal of rapidly and accurately predicting flavor using spectral technology, and improving the efficiency and objectivity of flavor evaluation. Using the above method, the specific contribution of each substance to the overall sensory attributes can be analyzed to identify key flavor substances in traditional plant-based beverage flavor evaluation. Attached Figure Description

[0048] Figure 1 This is an exemplary flowchart of an artificial intelligence-based method for detecting flavor components in plant-based beverages, as shown in some embodiments of this application.

[0049] Figure 2 This is an exemplary flowchart illustrating the determination of a sensory contribution spectrum according to some embodiments of this application;

[0050] Figure 3 This is an exemplary flowchart illustrating the determination of key spectral fingerprint features according to some embodiments of this application;

[0051] Figure 4 This is a schematic diagram of the structure of an artificial intelligence-based plant beverage flavor component detection system according to some embodiments of this application;

[0052] Figure 5 This is a schematic diagram of the structure of a computer device for implementing an artificial intelligence-based method for detecting flavor components in plant-based beverages, according to some embodiments of this application. Detailed Implementation

[0053] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] refer to Figure 1 The figure is an exemplary flowchart of an AI-based method for detecting flavor components in plant-based beverages, according to some embodiments of this application. This AI-based method for detecting flavor components in plant-based beverages mainly includes the following steps:

[0055] In step 101, the content of each flavor substance in the target plant beverage sample is detected to obtain a multidimensional raw data matrix of the flavor substance content in the target plant beverage sample.

[0056] In practice, representative target plant beverage samples are first selected. Gas chromatography-mass spectrometry and high-performance liquid chromatography are used to qualitatively and quantitatively detect volatile flavor compounds (such as aldehydes, esters, and terpenes) and non-volatile flavor compounds (such as sugars, organic acids, and amino acids) in the target plant beverage samples. The specific content values ​​of each detected flavor compound are recorded. Then, these data are organized into a multidimensional raw data matrix containing multiple samples, multiple flavor compounds, and their corresponding contents according to the three-dimensional relationship of "sample-flavor compound-content". In the multidimensional raw data matrix, each row represents a sample, each column represents a flavor compound, and the matrix elements are the contents of the corresponding flavor compound in the sample. For example, a detection result containing 5 samples and 8 flavor compounds will form a 5-row, 8-column multidimensional raw data matrix. The value in the 3rd row and 4th column of the matrix represents the content of the 4th flavor compound in the 3rd sample. The target plant beverage sample contains multiple samples.

[0057] In step 102, based on the sensory evaluation data of the target plant beverage sample and combined with the multidimensional original data matrix, the sensory contribution spectrum of volatile and non-volatile flavor substances in the target plant beverage sample to the overall sensory attributes of the target plant beverage sample is analyzed. Principal component analysis is used to reduce the dimensionality of the multidimensional original data matrix to obtain the principal component vector of flavor substances in the target plant beverage sample. The principal component vector is used to reflect the synergistic change characteristics among flavor substances.

[0058] In some embodiments, reference Figure 2 As shown, this figure is an exemplary flowchart for determining the sensory contribution spectrum in some embodiments of this application. In this embodiment, the sensory contribution spectrum of volatile and non-volatile flavor substances in the target plant beverage sample to the overall sensory attributes of the target plant beverage sample can be analyzed by the following steps based on the sensory evaluation data of the target plant beverage sample and the multidimensional raw data matrix:

[0059] First, in step 1021, sensory evaluation data of the target plant beverage sample are obtained;

[0060] Secondly, in step 1022, the correlation coefficients between volatile and non-volatile flavor substances and various sensory attributes are calculated based on the sensory evaluation data and the multidimensional raw data matrix.

[0061] Finally, in step 1023, a sensory contribution spectrum of volatile and non-volatile flavor compounds in the target plant beverage sample to the overall sensory properties of the target plant beverage sample is constructed based on all correlation coefficients.

[0062] The process involves extracting sensory evaluation data for the target plant-based beverage samples from a sensory evaluation database. This sensory evaluation data represents the quantitative data of the overall sensory attributes of the target plant-based beverage samples, transforming the subjective perception of beverage flavor by the human sensory system into analyzable objective data. It includes two aspects: first, an evaluation index system covering core sensory attributes such as aroma intensity, sweetness, acidity, and body, which directly reflect the flavor characteristics of the beverage at the olfactory, gustatory, and mouthfeel levels; second, specific data results based on the evaluation indexes, generated by professional evaluators through blind scoring (e.g., a 1-9 point system) or fuzzy mathematical evaluation methods to form standardized scoring data for subsequent correlation analysis of the relationship between flavor substances and sensory experience. Other methods may be used in other embodiments, which are not limited here.

[0063] In specific implementation, the correlation coefficients between volatile and non-volatile flavor compounds and various sensory attributes can be calculated based on the sensory evaluation data and the multidimensional original data matrix in the following manner: using the content data of volatile and non-volatile flavor compounds in the multidimensional original data matrix as independent variables, and the scores of various sensory attributes in the sensory evaluation data as dependent variables, the correlation coefficients between each flavor compound and each sensory attribute are calculated using statistical methods such as partial least squares regression or Pearson correlation analysis. The correlation coefficient represents the degree of correlation between flavor compounds and sensory attributes; a larger absolute value indicates a stronger correlation. Positive and negative signs are also indicated to reflect the direction of the correlation (positive or negative). Other methods can be used in other embodiments, which are not limited here.

[0064] In addition, in specific implementation, the sensory contribution spectrum of volatile and non-volatile flavor substances in the target plant beverage sample to the overall sensory attributes of the target plant beverage sample can be constructed in the following way based on all correlation coefficients: The correlation coefficients of various flavor substances and various sensory attributes are sorted out, and a matrix spectrum is constructed with flavor substance category (volatile / non-volatile) as the horizontal axis and sensory attribute as the vertical axis. The contribution intensity of different substances to each attribute is intuitively presented through color gradient (such as heat map) or numerical labeling. The position of the substance and attribute corresponding to the high absolute value coefficient is displayed with dark or highlighted marks, forming a sensory contribution spectrum that can clearly reflect the correlation strength of "flavor substance-sensory attribute". Other methods can also be used to construct it in other embodiments, which are not limited here.

[0065] It should be noted that the sensory contribution spectrum in this application represents a graph showing the specific influence of volatile and non-volatile flavor substances on the overall sensory attributes of beverages, such as aroma intensity, sweetness, and acidity. The color gradient or numerical labeling directly reflects the contribution intensity of different substances (high absolute value coefficients correspond to darker or more prominent marks), which can be used to locate key flavor driving factors, identify which substances play a major role in specific sensory characteristics (such as sweetness and aroma), and provide targeted targets for beverage flavor optimization.

[0066] In some embodiments, principal component analysis is used to reduce the dimensionality of the multidimensional original data matrix to obtain the principal component vector of flavor substances in the target plant beverage sample. This can be achieved through the following steps:

[0067] The multidimensional original data matrix is ​​standardized to obtain a standardized multidimensional original data matrix.

[0068] Principal component analysis algorithm was used to calculate eigenvalues ​​of the standardized multidimensional raw data matrix to obtain eigenvalue data of flavor substances in the target plant beverage sample.

[0069] Determine the characteristic thresholds of flavor compounds in target plant beverage samples;

[0070] The feature value data is judged based on the feature threshold, and then the principal component vector of flavor substances in the target plant beverage sample is obtained.

[0071] In specific implementation, firstly, considering the dimensional differences in the content data of different flavor substances in the multidimensional original data matrix, the Z-score standardization method is used to calculate the mean and standard deviation of each flavor substance content, converting the original data into standardized values ​​with a mean of 0 and a standard deviation of 1, ultimately obtaining a standardized multidimensional original data matrix that eliminates the influence of dimensions. Secondly, the standardized multidimensional original data matrix is ​​imported into a statistical product and service solution or a data analysis tool in R language, and principal component analysis functions are called to perform eigenvalue decomposition or singular value decomposition on the standardized multidimensional original data matrix, calculating the eigenvalues, eigenvectors (i.e., principal component loading coefficients, reflecting the correlation strength between each flavor substance and the principal component) and the score of each sample on each principal component for each principal component. The set of all eigenvalues ​​is used as the eigenvalue data of flavor substances in the target plant beverage sample. The eigenvalues ​​in the eigenvalue data reflect the explanatory power of the principal component for the total data variation and are a core indicator for measuring the importance of the principal component. The larger the eigenvalue, the stronger the explanatory power of the principal component for the overall data variation and the richer the information content of the original flavor substance content data. Other methods can also be used in other embodiments, which are not limited here.

[0072] In addition, the specific implementation method for determining the characteristic threshold of flavor substances in the target plant beverage sample is as follows: based on the eigenvalue data obtained from principal component analysis, the classic criterion of "eigenvalue greater than 1" is adopted (i.e., retain principal components with eigenvalue ≥ 1, because their explanatory power exceeds the average level of a single original variable), or combined with the cumulative variance contribution rate threshold (e.g., setting the cumulative variance contribution rate ≥ 85% to ensure the retention of the main information of the data). By drawing a scree plot (the horizontal axis is the principal component index and the vertical axis is the eigenvalue), the downward trend of the eigenvalue is intuitively judged to determine the characteristic threshold used to screen the core principal components. The characteristic threshold represents the quantitative boundary of whether the principal component has sufficient information retention value and can be used as a critical judgment criterion for screening the core principal components. Other methods can also be used to determine it in other embodiments, which are not limited here.

[0073] Furthermore, the specific implementation of judging the feature value data based on the feature threshold to obtain the principal component vector of flavor substances in the target plant beverage sample is as follows: the feature value data is compared with the set feature threshold, and the principal components with feature values ​​greater than the feature threshold are selected as core principal components. The flavor substances corresponding to these core principal components (i.e., substances that significantly contribute to the core principal components) are extracted. Then, combined with the score distribution of the sample on the core principal components, a principal component vector is formed with the core principal components as the dimension and key flavor substances as the core. The principal component vector specifically includes three aspects: first, the core principal components, that is, the principal components with feature values ​​greater than the threshold selected by the feature threshold (such as the first 2-3 principal components). These principal components collectively carry the original flavor... The main variation information of flavor substances (usually cumulative variance contribution rate ≥85%), each principal component represents a core flavor dimension (such as "aroma-dominant dimension" or "sweet and sour taste dimension"); secondly, key flavor substances, that is, flavor substances with high absolute values ​​of loading coefficients corresponding to the core principal components. They are the core contributors to each core flavor dimension and directly determine the flavor characteristics of that dimension (such as aldehydes and esters with high loading in a core principal component representing aroma-related characteristics); thirdly, sample score distribution, that is, the score data of each sample on the core principal components. The score distribution can intuitively reflect the differences or commonalities of different samples on the core flavor dimensions, helping to distinguish the flavor characteristic types of samples; other methods can also be used to determine this in other embodiments, which are not limited here.

[0074] It should be noted that the principal component vector in this application reflects the synergistic variation characteristics among flavor substances. This vector not only retains the main distribution characteristics of flavor substances in the original data, but also intuitively reflects the flavor differences and commonalities among samples.

[0075] It should be noted that the principal component vector is a comprehensive feature vector obtained by reducing the dimensionality of the multidimensional original data matrix through principal component analysis. Its core function is to capture the most representative change patterns in the flavor substance content data, thereby intuitively reflecting the synergistic change characteristics between different flavor substances. For example, when certain volatile aroma components (such as esters and alcohols) and non-volatile taste components (such as sugars and organic acids) show a significant synchronous increase or decrease trend in content, the principal component vector will condense this correlation into a few comprehensive dimensions in the form of eigenvalues ​​and loadings. This not only preserves the intrinsic correlation information between flavor substances in the original data, but also eliminates the interference of redundant noise, so that subsequent analysis can more clearly focus on the core component combinations that truly drive the synergistic change of flavor.

[0076] In step 103, the contribution of the core components affecting the flavor of the target plant beverage sample is analyzed by combining the sensory contribution spectrum with the synergistic change characteristics among flavor substances reflected by the principal component vector, thereby obtaining the key flavor markers of the flavor components in the target plant beverage sample.

[0077] In some embodiments, the contribution of the core components affecting the flavor of the target plant beverage sample to the sensory contribution spectrum combined with the synergistic variation characteristics among flavor substances reflected by the principal component vector can be analyzed to obtain the key flavor markers of the flavor components in the target plant beverage sample. This can be achieved through the following steps:

[0078] Based on the sensory contribution spectrum, multiple high-contribution flavor compounds affecting the flavor of plant-based beverages were extracted from the target plant-based beverage sample;

[0079] Multiple core flavor compounds were screened from the target plant beverage sample based on the principal component vector reflecting the synergistic variation characteristics among flavor compounds;

[0080] Intersection analysis of each high-contribution flavor compound and each core flavor compound was performed to obtain multiple candidate core components of the target plant beverage sample.

[0081] The contribution score of each candidate core component to the flavor of the plant beverage is calculated, thereby obtaining the key flavor markers of the flavor components in the target plant beverage sample.

[0082] It is important to note that the sensory contribution spectrum clearly quantifies the specific contribution of different flavor compounds (including volatile and non-volatile) to the overall sensory attributes of plant-based beverages (such as sweetness, acidity, and aroma intensity), clarifying which substances are directly related to sensory experience. Meanwhile, principal component vector analysis (PCA) captures the synergistic changes among flavor compounds, revealing the intrinsic correlation patterns between them (e.g., which substances always change synchronously, and which show inverse correlations). When combining both for core component contribution analysis, the sensory contribution spectrum is first used to identify candidate substances with substantial sensory impact. Then, PCA analyzes the role of these candidate substances in synergistic changes. Substances that are in strong synergistic combinations and significantly contribute to sensory experience are identified as core components driving the overall flavor characteristics. By quantifying and ranking the contributions of these core components, the most representative substances that stably reflect the flavor characteristics of the beverage are ultimately selected as key flavor markers.

[0083] The specific implementation method for extracting multiple high-contribution flavor substances affecting the flavor of plant beverages from the target plant beverage sample based on the sensory contribution spectrum is as follows: Based on the correlation strength matrix of "flavor substance-sensory attribute" in the sensory contribution spectrum, flavor substances with the highest absolute values ​​of correlation coefficients for various sensory attributes (such as aroma, sweetness, acidity, etc.) are selected (e.g., substances with the top 30% of absolute values ​​of coefficients). These flavor substances make significant contributions to the flavor of the beverage at the sensory perception level. Each of the selected flavor substances is regarded as a high-contribution flavor substance. The high-contribution flavor substance refers to the flavor substance that has a significant impact on various sensory attributes of the plant beverage in the sensory contribution spectrum. Other extraction methods can also be used in other embodiments, which are not limited here.

[0084] The specific implementation of selecting multiple core flavor substances from the target plant beverage sample based on the synergistic change characteristics reflected by the principal component vector is as follows: First, based on the magnitude and sign of the loading values ​​of each flavor substance in the principal component vector, flavor substances that significantly contribute to the principal components are selected (usually substances with the highest absolute loading values). These substances are potential carriers of synergistic change characteristics. Next, the consistency of the signs of these substances in the principal component vector is analyzed. If multiple substances have the same loading sign, it indicates that they show a synergistic increase or decrease trend in content change, that is, flavor association is formed through positive synergistic change characteristics of synchronous fluctuation. If the signs are opposite, it reflects a negative synergistic change characteristic, which is a dynamic balance relationship of one increasing while the other decreases. Finally, combined with the proportion of total variation explained by different principal components, the combination of substances that shows strong synergy in the principal components with high explanatory power is preferentially selected. The synergistic change characteristics of such combinations can more accurately reflect the overall fluctuation law of the plant beverage flavor, thus determining them as the core flavor substances driving the overall flavor characteristics of the plant beverage. The core flavor substances refer to flavor substances that significantly contribute to the core principal components in the target plant beverage sample. Other screening methods can also be used in other embodiments, which are not limited here.

[0085] The specific implementation method for obtaining multiple candidate core components of the target plant beverage sample by performing intersection analysis on each high-contribution flavor substance and each core flavor substance is as follows: by comparing the substance names or chemical identifiers of each high-contribution flavor substance and each core flavor substance, flavor substances that appear in both sets are screened out. These flavor substances have a significant impact on sensory attributes in the sensory contribution spectrum and are key components of the core principal components in the principal component vector, thus being identified as candidate core components, thereby obtaining multiple candidate core components. The candidate core components refer to flavor substances that possess both "high sensory contribution" and "core chemical characteristics" by integrating the sensory contribution spectrum and the principal component vector. Other methods can also be used for analysis in other embodiments, which are not limited here.

[0086] The specific implementation method for calculating the contribution score of each candidate core component to the flavor of the plant beverage, and then obtaining the key flavor markers of the flavor components in the target plant beverage sample, is as follows: A weighted summation algorithm (such as weighting the sensory contribution weight with the principal component loading coefficient according to a preset ratio) is used to calculate the comprehensive contribution score of each candidate core component. After sorting the scores from high to low, flavor substances with scores exceeding a set threshold (such as the top 20% or a score ≥ 0.6) are selected. These flavor substances demonstrate outstanding performance in both "sensory influence" and "chemical characteristics," and are ultimately determined as key flavor markers. In other embodiments, other methods can also be used for calculation, which are not limited here.

[0087] It should be noted that the key flavor markers in this application refer to the characteristic flavor substances that have a decisive influence on flavor, screened from the flavor components of the target plant beverage sample. They can characterize the core flavor features of the beverage and accurately link sensory experience with chemical composition. They are the core indicators for analyzing the flavor formation mechanism, optimizing product flavor, and constructing flavor detection models.

[0088] In step 104, the target plant beverage sample is subjected to spectral detection, and the key spectral fingerprint features of the flavor components in the target plant beverage sample are extracted from the spectral detection results based on key flavor markers.

[0089] In some embodiments, reference Figure 3 As shown, this figure is an exemplary flowchart for determining key spectral fingerprint features in some embodiments of this application. In this embodiment, the key spectral fingerprint features of flavor components in the target plant beverage sample are extracted from the spectral detection results based on key flavor markers using the following steps:

[0090] First, in step 1041, the target plant beverage sample is subjected to spectral detection to obtain the spectral detection results;

[0091] Secondly, in step 1042, the spectral detection result is denoised to obtain the denoised spectral detection result;

[0092] Furthermore, in step 1043, based on the key flavor markers, characteristic spectral ranges, absorption peak intensities, and peak shape changes that are highly correlated with the changes in the content of the key flavor markers are extracted from the denoised spectral detection results;

[0093] Finally, in step 1044, the characteristic spectral range, absorption peak intensity, and peak shape changes are used as key spectral fingerprint features of flavor components in the target plant beverage sample.

[0094] It should be noted that key flavor markers, as specific chemical substances, produce characteristic absorption, scattering, or vibrational signals (such as absorption peaks at specific wavelengths) in their molecular structures during spectral detection. These signals are quantitatively correlated with the content of the markers. Because the differences in molecular structures of different chemical substances lead to their unique "fingerprints" in the spectrum, by focusing on the known characteristic spectral ranges of key flavor markers, signals directly related to these markers can be screened from the complex overall spectrum, eliminating interference from other components. This allows for the extraction of key spectral fingerprint features that specifically reflect the target flavor components, establishing a direct correlation between spectral signals and flavor components.

[0095] The specific implementation method for obtaining the spectral detection results of the target plant beverage sample is as follows: an appropriate amount of the target plant beverage sample is placed in a quartz cuvette, and a suitable spectral detection technology (such as near-infrared spectroscopy, mid-infrared spectroscopy, or Raman spectroscopy) is selected according to the chemical characteristics of the key flavor markers. Reasonable detection parameters (including wavelength range, number of scans, resolution, etc.) are set, and the sample is scanned using a spectrometer. The absorbance, transmittance, or scattering intensity data of the sample at different wavelengths are recorded to form a spectral detection result containing two-dimensional information of wave-signal intensity. Other detection methods can also be used in other embodiments, which are not limited here.

[0096] The specific implementation method for denoising the spectral detection results to obtain denoised spectral detection results is as follows: For baseline drift and random noise caused by instrument noise, environmental interference or sample scattering in the original spectrum, the Savitzky-Gauley smoothing filter method is used to eliminate high-frequency random noise. Baseline drift and scattering effects are corrected by multivariate scattering correction or standard normal variable transformation, or the signal and noise components are separated by wavelet transform threshold denoising method. After the above processing, a denoised spectral detection result with stable baseline, reduced noise, and more realistic reflection of sample component information is obtained.

[0097] The specific implementation method for extracting characteristic spectral intervals, absorption peak intensities, and peak shape changes that are highly correlated with the content changes of key flavor markers from the denoised spectral detection results based on key flavor markers is as follows: Based on the known molecular structure of the key flavor markers (such as the functional groups of hydroxyl, carbonyl, and benzene rings) and their characteristic absorption wavelengths (such as the characteristic absorption of hydroxyl in the 3200-3600 cm⁻¹ range in the mid-infrared region), combined with partial least squares regression or correlation analysis, the correlation coefficient between the signal intensity of each wavelength point in the denoised spectrum and the actual content of the marker is calculated. Then, a threshold for the absolute value of the correlation coefficient is set (such as ≥0.8), and the wavelength range that continuously meets this threshold is defined as the candidate characteristic interval. Next, the moving window method (such as a window size of 10-20 wavelength points) is used to calculate the average correlation coefficient within the candidate interval, and the continuous interval with the highest average correlation coefficient is retained as the characteristic spectral interval. For the identification of absorption peaks within a specific spectral range, the spectral curve can be processed using either the first or second derivative method. The absorption peaks in the original spectrum can be located by analyzing the peak positions of the derivative spectrum. Then, a linear regression analysis is performed on the intensity value of each absorption peak and the content of the marker. The coefficient of determination (R²) is used as an indicator (e.g., R² ≥ 0.7) to screen out absorption peaks with a significant linear relationship. At the same time, parameters such as peak position and half-width are recorded to fully characterize the peak shape. The intensity value, peak position shift, and half-width are also recorded. The characteristic spectral range refers to a specific wavelength range that is highly correlated with the chemical structure or content changes of key flavor markers in the target plant beverage during spectral detection. The absorption peak intensity refers to the signal strength of the absorption peak in the spectral curve within the characteristic spectral range (usually expressed as absorbance, transmittance, or scattering intensity). The peak shape change refers to the morphological characteristics and changing trends of the absorption peaks within the characteristic spectral range.

[0098] It should be noted that the key spectral fingerprint features in this application represent a set of spectral signals extracted from the spectral detection results of the target plant beverage that can specifically characterize the core features of its flavor components. They serve as a "bridge" between spectral data and the chemical composition and sensory characteristics of flavor substances.

[0099] In step 105, a flavor component prediction model for the target plant beverage sample is constructed based on historical spectral data of the plant beverage. The key spectral fingerprint features are input into the flavor component prediction model to output the flavor score of the flavor components in the target plant beverage sample.

[0100] In some embodiments, constructing a flavor component prediction model for a target plant beverage sample based on historical spectral data of the plant beverage can be achieved through the following steps:

[0101] Obtain historical spectral data of plant-based beverages;

[0102] The historical spectral data is divided into a training set and a validation set;

[0103] A flavor component prediction model for the target plant beverage sample is constructed based on an artificial intelligence algorithm combined with the training and validation sets.

[0104] In practice, firstly, historical spectra of the plant-based beverages are obtained from the spectral database corresponding to the target plant beverage samples. Secondly, the historical spectral data is divided into training and validation sets. This is achieved using stratified sampling or random sampling to divide the historical spectral dataset, with the training set typically comprising 70%-80% for model parameter learning and construction, and the validation set comprising 20%-30% for evaluating the model's generalization ability. During the division, it is crucial to ensure consistency in the content distribution of key flavor markers and the batch distribution of samples between the training and validation sets to avoid impacting model performance due to data distribution deviations. The rationality of the division can be verified by calculating the mean, standard deviation, and content range of the two sets of data. If necessary, k-fold cross-validation can be used to further optimize the data division.

[0105] The specific implementation of constructing a flavor component prediction model for target plant beverage samples based on artificial intelligence algorithms combined with the training and validation sets is as follows: A suitable artificial intelligence algorithm (such as partial least squares regression, support vector regression, random forest, or deep learning models such as convolutional neural networks) is selected. Feature spectral data from the training set (such as extracted key spectral fingerprint features) is used as input, and the content and flavor score of the corresponding key flavor markers are used as output. The model parameters are optimized through iterative training (such as adjusting the number of layers in the neural network and the kernel function parameters of the support vector machine). The performance of the trained model is evaluated using the validation set, with root mean square error and coefficient of determination (R²) as the core indicators. If the model accuracy does not reach a preset threshold (such as R² ≥ 0.85), the algorithm parameters are adjusted or feature engineering steps are added (such as spectral preprocessing optimization) for retraining until a stable flavor component prediction model with strong generalization ability is obtained.

[0106] It should be noted that the flavor component prediction model in this application is a mathematical model built on artificial intelligence algorithms. It aims to quickly predict the flavor component-related characteristics of plant beverages through spectral data. It can directly predict the content of key flavor markers and further output comprehensive or detailed flavor scores, thereby realizing rapid detection, quality assessment and optimization guidance of flavor components of plant beverages, avoiding the cumbersome and time-consuming traditional chemical analysis methods.

[0107] In some embodiments, inputting the key spectral fingerprint features into the flavor component prediction model to output flavor scores for the flavor components in the target plant beverage sample can be achieved through the following steps:

[0108] The key spectral fingerprint features are standardized, and the standardized key spectral fingerprint features are input into the flavor component prediction model.

[0109] The flavor component prediction model calculates the standardized key spectral fingerprint features of the input to output the flavor score of the flavor components in the target plant beverage sample.

[0110] The specific implementation of standardizing the key spectral fingerprint features and inputting the standardized key spectral fingerprint features into the flavor component prediction model is as follows: The extracted key spectral fingerprint features (including signal values, absorption peak intensities, and peak shape parameters of the characteristic spectral range) are processed using the same standardization method (such as Z-score standardization or minimum-maximum standardization) as the historical spectral data used when building the model. By calculating the feature mean and standard deviation or the range of extreme values, the feature values ​​are converted into standardized data with uniform dimensions, eliminating scale differences between different features. Subsequently, according to the input format requirements of the flavor component prediction model, the standardized key spectral fingerprint features are imported into the model in vector or matrix form, ensuring that the input data structure is consistent with the feature dimensions during model training.

[0111] The specific implementation of calculating the flavor score of the flavor components in the target plant beverage sample using the flavor component prediction model is as follows: The flavor component prediction model calls the internally trained algorithm parameters (such as the weight matrix of the neural network and the kernel function parameters of support vector regression) to perform a series of operations such as feature mapping, weight calculation, and nonlinear transformation on the input standardized key spectral fingerprint features. Based on the correlation between the key spectral fingerprint features and flavor components, and combined with the mapping relationship between spectral features and flavor marker content in historical data, the model outputs a quantitative score for the corresponding flavor component (such as a comprehensive score that comprehensively considers the predicted content and contribution weight of each key flavor marker, or a detailed score for a single flavor attribute such as aroma, sweetness, or sourness). This comprehensive score is used as the flavor score of the flavor components in the target plant beverage sample. In other embodiments, other methods can also be used for calculation, which are not limited here.

[0112] It should be noted that the flavor score in this application represents the overall characteristics and quality of the flavor components of the sample. The value of the score is directly related to the actual performance of the flavor of the sample: a high score usually corresponds to a sample with appropriate content of key flavor substances, prominent and harmonious flavor characteristics, while a low score may reflect the absence of flavor substances, imbalance of proportions, or the presence of undesirable flavor interference. This provides a quantifiable reference standard for the evaluation of the flavor quality of plant beverages, product optimization, and quality control.

[0113] In another aspect, in some embodiments, this application provides an artificial intelligence-based system for detecting flavor components in plant-based beverages, referencing... Figure 4 The figure is a schematic diagram of the structure of an artificial intelligence-based plant beverage flavor component detection system according to some embodiments of this application. The artificial intelligence-based plant beverage flavor component detection system 400 includes: a detection module 401, a processing module 402, and an execution module 403, which are described below:

[0114] The detection module 401 in this application is mainly used to detect the content of various flavor substances in the target plant beverage sample and obtain a multidimensional raw data matrix of the content of flavor substances in the target plant beverage sample.

[0115] Processing module 402 in this application is used to analyze the sensory contribution spectrum of volatile and non-volatile flavor substances in the target plant beverage sample to the overall sensory attributes of the target plant beverage sample based on the sensory evaluation data of the target plant beverage sample and the multidimensional original data matrix, and to perform dimensionality reduction processing on the multidimensional original data matrix using principal component analysis to obtain the principal component vector of flavor substances in the target plant beverage sample.

[0116] It should be noted that the processing module 402 in this application is also used to analyze the contribution of the core components affecting the flavor of the plant beverage in the target plant beverage sample by combining the sensory contribution spectrum with the synergistic change characteristics between flavor substances reflected by the principal component vector, and to obtain the key flavor markers of the flavor components in the target plant beverage sample.

[0117] In addition, it should be noted that the processing module 402 in this application is also used to perform spectral detection on the target plant beverage sample, and extract the key spectral fingerprint features of the flavor components in the target plant beverage sample from the spectral detection results based on key flavor markers.

[0118] The execution module 403 in this application is mainly used to construct a flavor component prediction model for a target plant beverage sample based on historical spectral data of the plant beverage, and input the key spectral fingerprint features into the flavor component prediction model to output the flavor score of the flavor components in the target plant beverage sample.

[0119] In addition, this application also provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described artificial intelligence-based method for detecting flavor components in plant beverages.

[0120] In some embodiments, reference Figure 5The figure is a schematic diagram of the structure of a computer device implementing an artificial intelligence-based method for detecting flavor components in plant-based beverages, according to some embodiments of this application. The artificial intelligence-based method for detecting flavor components in plant-based beverages described in the above embodiments can... Figure 5 The computer device shown is used to implement this, and the computer device 500 includes at least one processor 501, a communication bus 502, a memory 503, and at least one communication interface 504.

[0121] Processor 501 can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0122] The communication bus 502 can be used to transmit information between the aforementioned components.

[0123] Memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 503 may exist independently and be connected to processor 501 via communication bus 502. Memory 503 may also be integrated with processor 501.

[0124] The memory 503 stores program code for executing the scheme of this application, and its execution is controlled by the processor 501. The processor 501 executes the program code stored in the memory 503. The program code may include one or more software modules. The method used in the above embodiments can be implemented by the processor 501 and one or more software modules in the program code in the memory 503.

[0125] Communication interface 504 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0126] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0127] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0128] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described artificial intelligence-based method for detecting flavor components in plant-based beverages.

[0129] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0130] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for detecting flavor components in plant-based beverages based on artificial intelligence, characterized in that, Includes the following steps: The content of each flavor substance in the target plant beverage sample was detected, and a multidimensional raw data matrix of the flavor substance content in the target plant beverage sample was obtained. Based on the sensory evaluation data of the target plant beverage samples and the multidimensional original data matrix, the sensory contribution spectrum of volatile and non-volatile flavor substances in the target plant beverage samples to the overall sensory attributes of the target plant beverage samples is analyzed. Principal component analysis is used to reduce the dimensionality of the multidimensional original data matrix to obtain the principal component vector of flavor substances in the target plant beverage samples. The principal component vector is used to reflect the synergistic change characteristics among flavor substances. By combining the sensory contribution spectrum with the synergistic change characteristics among flavor substances reflected by the principal component vector, the contribution of the core components affecting the flavor of the target plant beverage sample is analyzed, and the key flavor markers of the flavor components in the target plant beverage sample are obtained. Spectroscopic detection was performed on the target plant beverage sample, and key spectral fingerprint features of flavor components in the target plant beverage sample were extracted from the spectral detection results based on key flavor markers. A flavor component prediction model for a target plant beverage sample is constructed based on historical spectral data of plant beverages. The key spectral fingerprint features are input into the flavor component prediction model to output the flavor score of the flavor components in the target plant beverage sample. Specifically, the sensory contribution spectrum of volatile and non-volatile flavor compounds in the target plant beverage samples to the overall sensory attributes of the target plant beverage samples, analyzed based on sensory evaluation data of the target plant beverage samples and the multidimensional raw data matrix, includes: Obtain sensory evaluation data of the target plant beverage samples; The correlation coefficients between volatile and non-volatile flavor compounds and various sensory attributes are calculated based on the sensory evaluation data and the multidimensional raw data matrix. Based on all correlation coefficients, a sensory contribution spectrum of volatile and non-volatile flavor compounds in the target plant beverage sample to the overall sensory properties of the target plant beverage sample was constructed. Specifically, the process of performing spectral analysis on target plant-based beverage samples and extracting key spectral fingerprint features of flavor components from the spectral analysis results based on key flavor markers includes: The target plant beverage sample was subjected to spectral analysis, and the spectral analysis results were obtained. The spectral detection results are denoised to obtain denoised spectral detection results; Based on key flavor markers, characteristic spectral ranges, absorption peak intensities, and peak shape changes that are highly correlated with the changes in the content of key flavor markers are extracted from the denoised spectral detection results. The characteristic spectral range, absorption peak intensity, and peak shape changes are used as key spectral fingerprint features of flavor components in the target plant beverage sample.

2. The method as described in claim 1, characterized in that, Principal component analysis was used to reduce the dimensionality of the original multidimensional data matrix, resulting in the principal component vectors of flavor compounds in the target plant beverage sample. Specifically, these vectors include: The multidimensional original data matrix is ​​standardized to obtain a standardized multidimensional original data matrix. Principal component analysis algorithm was used to calculate eigenvalues ​​of the standardized multidimensional raw data matrix to obtain eigenvalue data of flavor substances in the target plant beverage sample. Determine the characteristic thresholds of flavor compounds in target plant beverage samples; The feature value data is judged based on the feature threshold, and then the principal component vector of flavor substances in the target plant beverage sample is obtained.

3. The method as described in claim 1, characterized in that, By combining the sensory contribution spectrum with the synergistic variation characteristics among flavor substances reflected by the principal component vector, the contribution of the core components affecting the flavor of the target plant beverage sample is analyzed, and the key flavor markers of the flavor components in the target plant beverage sample are obtained, specifically including: Based on the sensory contribution spectrum, multiple high-contribution flavor compounds affecting the flavor of plant-based beverages were extracted from the target plant-based beverage sample; Multiple core flavor compounds selected from the principal component loadings of the principal component vector; Intersection analysis of each high-contribution flavor compound and each core flavor compound was performed to obtain multiple candidate core components of the target plant beverage sample. The contribution score of each candidate core component to the flavor of the plant beverage is calculated, thereby obtaining the key flavor markers of the flavor components in the target plant beverage sample.

4. The method as described in claim 1, characterized in that, The specific steps involved in constructing a flavor component prediction model for target plant-based beverage samples based on historical spectral data include: Obtain historical spectral data of plant-based beverages; The historical spectral data is divided into a training set and a validation set; A flavor component prediction model for the target plant beverage sample is constructed based on an artificial intelligence algorithm combined with the training and validation sets.

5. The method as described in claim 1, characterized in that, The key spectral fingerprint features are input into the flavor component prediction model to output the flavor score of the flavor components in the target plant beverage sample. Specifically, this includes: The key spectral fingerprint features are standardized, and the standardized key spectral fingerprint features are input into the flavor component prediction model. The flavor component prediction model calculates the standardized key spectral fingerprint features of the input to output the flavor score of the flavor components in the target plant beverage sample.

6. The method as described in claim 1, characterized in that, The content of various flavor substances in the target plant beverage sample was detected by gas chromatography-mass spectrometry and high performance liquid chromatography.

7. The method as described in claim 1, characterized in that, Near-infrared spectroscopy was used to perform spectral detection on the target plant beverage sample.

8. An artificial intelligence-based system for detecting flavor components in plant-based beverages, comprising detecting flavor components in plant-based beverages using the method described in any one of claims 1 to 7, characterized in that, The system includes: The detection module is used to detect the content of various flavor substances in the target plant beverage sample and obtain a multidimensional raw data matrix of flavor substance content in the target plant beverage sample. The processing module is used to analyze the sensory contribution spectrum of volatile and non-volatile flavor substances in the target plant beverage sample to the overall sensory attributes of the target plant beverage sample based on the sensory evaluation data of the target plant beverage sample and the multidimensional original data matrix. Principal component analysis is used to reduce the dimensionality of the multidimensional original data matrix to obtain the principal component vector of flavor substances in the target plant beverage sample. The principal component vector is used to reflect the synergistic change characteristics among flavor substances. The processing module is also used to analyze the contribution of the core components affecting the flavor of the target plant beverage sample by combining the sensory contribution spectrum with the synergistic change characteristics between flavor substances reflected by the principal component vector, and to obtain the key flavor markers of the flavor components in the target plant beverage sample. The processing module is also used to perform spectral detection on the target plant beverage sample and extract the key spectral fingerprint features of the flavor components in the target plant beverage sample from the spectral detection results based on key flavor markers. The execution module is used to construct a flavor component prediction model for the target plant beverage sample based on historical spectral data of the plant beverage, and input the key spectral fingerprint features into the flavor component prediction model to output the flavor score of the flavor components in the target plant beverage sample.

Images