A method for analyzing the flavor preference of baijiu and a perception evaluation system

Abstract

The present application provides a kind of liquor flavor preference analysis method, comprising (1) collecting the preference sensory evaluation data and electroencephalogram data of subjects after different liquor flavor stimulation;(2) the electroencephalogram data obtained in step (1) is preprocessed, and different brain wave band data is obtained;(3) different brain wave band data obtained in step (2) is used as input feature, combined with the preference sensory evaluation data collected in step (1), XGBoost model training is carried out, and the trained XGBoost prediction model is obtained;The preference of the subject is predicted based on the XGBoost prediction model, and the prediction value X is obtained;(4) the prediction value X obtained in step (3) is combined with the ratio data between different wave band powers, and the preference result Y of the subject is calculated.The present application uses XGBoost model to learn the characteristics of electroencephalogram data, so as to explore the preference of ordinary consumers to different liquor flavors, and provide data support for the development of new liquor products.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of liquor detection, in particular to a liquor flavor preference analysis method and a perception evaluation system. BACKGROUND

[0002] The flavor of liquor affects the preference degree of consumers for the corresponding liquor. At present, the evaluation of the flavor of liquor mainly includes artificial sensory evaluation and instrument analysis. The instrument analysis cannot evaluate human perception, and the current artificial sensory evaluation is mainly artificial sensory evaluation by professional personnel. However, the evaluation by professional personnel cannot represent the evaluation of ordinary consumers. In addition, the flavor characteristics of liquor are rich and complex, and ordinary consumers are difficult to evaluate some characteristic aromas. Therefore, a method for more comprehensive evaluation of liquor flavor is needed. SUMMARY

[0003] To solve one of the above problems, on the one hand, the present application provides a liquor flavor preference analysis method, which improves the low adaptability of the prior art in liquor flavor preference analysis, and comprises the following technical solutions:

[0004] (1) Collecting the preference degree sensory evaluation data and electroencephalogram data of subjects after different liquor flavor stimuli;

[0005] (2) Preprocessing the electroencephalogram data obtained in step (1) to obtain different electroencephalogram band data;

[0006] (3) Taking the different electroencephalogram band data obtained in step (2) as input features, combining the preference degree sensory evaluation data collected in step (1), and performing XGBoost model training to obtain a trained XGBoost prediction model. The preference degree of the subjects is predicted based on the XGBoost prediction model to obtain a prediction value X.

[0007] (4) Based on the prediction value X obtained in step (3) and the ratio data between different band powers, the preference degree result Y of the subjects is calculated.

[0008] In some embodiments, in step (1), the evaluation includes providing different liquor aromas by using a digital odor player for aroma stimulation.

[0009] In some embodiments, in step (1), the method for obtaining the preference degree sensory evaluation data includes: after each odor is played, a 1-9 point scoring method is used, and the higher the score, the higher the preference degree.

[0010] In some embodiments, in step (1), the subjects are selected according to the age characteristics of the liquor consumer group.

[0011] In some embodiments, in step (1), the subjects are aged 18-65 years old, with a male: female ratio of 1:1, and an age distribution of 1:3:1 for 18-27 years old, 28-40 years old, and 41-65 years old, respectively.

[0012] In some embodiments, in step (1), the evaluation includes: different white wine aromas are divided into groups for stimulation, and multiple parallel aromas are set in each group, and an interval time is set between each two groups of flavor tests.

[0013] In some embodiments, in step (1), the stimulation method is: a digital odor player sniffer is placed 3-5 cm in front of the subject's nose, no stimulation is performed first, then the aroma is played, a period of rest is provided, and the process is repeated.

[0014] In some embodiments, in step (1), the stimulation method is: a digital odor player sniffer is placed 3-5 cm in front of the subject's nose, no stimulation is performed first, then the aroma is played, a period of rest is provided, and the process is repeated.

[0015] In some embodiments, in step (2), the preprocessing includes: detrending, band-pass filtering, and power frequency interference removal processing are performed on the electroencephalogram data obtained in step (2); according to the frequency range of different electroencephalogram bands, the bandwidth, attenuation coefficient, and filter order of the digital filter are designed, the electroencephalogram signal after power frequency interference removal processing is filtered with the corresponding digital filter, the corresponding frequency band signal features are extracted, and different electroencephalogram band data are obtained.

[0016] In some embodiments, in step (2), the preprocessing includes: detrending, band-pass filtering, and power frequency interference removal processing are performed on the electroencephalogram data obtained in step (2); according to the frequency range of different electroencephalogram bands, the bandwidth, attenuation coefficient, and filter order of the digital filter are designed, the electroencephalogram signal after power frequency interference removal processing is filtered with the corresponding digital filter, the corresponding frequency band signal features are extracted, and different electroencephalogram band data are obtained.

[0017] In some embodiments, in step (3), the preprocessing comprises: in step (3), the following preprocessing is performed on the electroencephalogram data obtained in step (2): first, the original electroencephalogram data is subjected to detrending processing to remove linear trends or baseline drift in the data, so as to eliminate possible low-frequency interference; then, the data after detrending is subjected to band-pass filtering, and the frequency band range is set to 0.5-80 Hz, so as to retain meaningful electroencephalogram activity and remove high-frequency noise and very low-frequency components; subsequently, power frequency interference is removed to eliminate the influence of power frequency on electroencephalogram signals, so as to obtain electroencephalogram data that has removed main environmental noise; next, according to the frequency ranges of different electroencephalogram bands, the frequency range of delta waves is 0.5-4 Hz, the frequency range of theta waves is 4-8 Hz, the frequency range of alpha waves is 8-12 Hz, the frequency range of beta waves is 12-30 Hz, and the frequency range of gamma waves is 30-100 Hz, the bandwidth, attenuation coefficient and filter order of the digital filter are designed, and the digital filter suitable for each frequency band is constructed; finally, the preprocessed electroencephalogram signals are filtered using these filters, the corresponding frequency band signal features are extracted, and different electroencephalogram band data is obtained to provide input for subsequent feature extraction and model training.

[0018] In some embodiments, in step (4), the calculation of the preference result Y of the subject based on the prediction value X obtained in step (3) and the ratio data between different wave band powers comprises: adding the prediction value X and the ratio between the corresponding different wave band powers according to the corresponding weights to obtain the preference result Y.

[0019] In some embodiments, the ratio between the corresponding different wave band powers comprises at least one of the ratio of delta wave power to beta wave power, the ratio of theta wave power to beta wave power, and the ratio of alpha wave power to beta wave power.

[0020] In some embodiments, the preference result Y is calculated by the following formula:

[0021] , wherein , , , , , , , , are the power response values of delta waves, beta waves, alpha waves, and theta waves, respectively.

[0022] In some embodiments, , , , .

[0023] In some embodiments, in step (4), the objective function of the XGBoost model consists of a loss function and a regularization term, and the loss function is the sum of squares of the difference between the predicted value and the true value.

[0024] In some embodiments, in step (1), the flavor of the different Baijiu includes at least one dimension of grassy, fruity, floral, sweet, nutty, dry plant, sour, and other flavors.

[0025] In some embodiments, in step (1), the flavor of the different Baijiu includes 8 dimensions of grassy, fruity, floral, sweet, nutty, dry plant, sour, and other flavors; the grassy flavor includes green grass, raw grain, and bamboo; the fruity flavor includes apple, pineapple, and banana; the floral flavor includes rose, gardenia, and violet; the sweet flavor includes vanilla and milk candy; the nutty flavor includes cocoa and paste; the dry plant flavor includes traditional Chinese medicine; the sour flavor includes vinegar; and the other flavor includes mushroom aroma, cellar aroma, sulfur aroma, yogurt aroma, and oil aroma.

[0026] In some embodiments, the different brain wave band data includes at least one of the ratio data of the power corresponding to different wave bands and the signal complexity data.

[0027] In some embodiments, the ratio data of the power corresponding to different wave bands includes at least one of the ratio of the power of the theta band to the power of the beta band, the ratio of the power of the alpha band to the power of the beta band, the ratio of the power of the delta band to the power of the beta band, the ratio of the inverse of the power of the alpha band to the sum of the power of the beta band and the power of the alpha and theta bands.

[0028] In some embodiments, the signal complexity includes C0 complexity, Hjorth activity parameter, and cross-frequency phase-amplitude coupling.

[0029] In some embodiments, the calculation method of the C0 complexity includes: converting the electroencephalogram signal to the frequency domain by fast Fourier transform, calculating the average value of the frequency domain amplitude, filtering the time domain signal after the inverse Fourier transform of the components with an amplitude greater than the average value, calculating the sum of the absolute difference values between the filtered time domain signal and the original time domain signal at each time point, and then dividing the total sum by the total sum of the absolute values of all points of the original signal.

[0030] In some embodiments, the calculation method of the C0 complexity includes: FFT calculation: performing fast Fourier transform on the signal to obtain the frequency domain representation of the signal fft_data; then calculating the average value of the frequency domain amplitude fft_average; filtering high amplitude components: selecting components with an amplitude greater than the average value fft_average from fft_data, setting other components to 0, and forming a new frequency domain signal inverse FFT: inverse Fourier transform of the filtered frequency domain signal

[0031] ;

[0032] x (n) represents the original signal at the n-th sampling point; x (n) represents the filtered signal (i.e. ifft_data) at the n-th sampling point; N is the number of data points. In some embodiments, the Hjorth activity parameter is obtained by calculating the ratio of the variance of the electroencephalogram signal to the square of the mean, to analyze the change in amplitude of the signal in the time domain.

[0033] In some embodiments, the formula for calculating the Hjorth activity parameter is:

[0034]

[0035] ;

[0036] N is the number of data points; x (n) represents the electroencephalogram signal to be analyzed at the n-th data point; is the mean of the signal, and Activity is the Hjorth activity parameter.

[0037] In some embodiments, the cross-frequency phase-amplitude coupling is obtained by applying Hilbert transform to the signal in the gamma band to obtain the instantaneous power, applying Hilbert transform to the theta signal to obtain the phase, and then calculating the phase of the gamma power, and the synchronization index SI is represented by the average of the complex exponential of the phase difference.

[0038] In some embodiments, the calculation of the cross-frequency phase-amplitude coupling includes the following method: applying Hilbert transform to the signal in the gamma band to obtain the envelope information in complex form, calculating the square of the envelope to obtain the instantaneous power in the gamma band; applying Hilbert transform to the theta signal to obtain the phase θ phase , applying Hilbert transform to the gamma power and obtaining its phase γ power phase ; the cross-frequency phase-amplitude coupling is the synchronization index SI represented by the average of the complex exponential of the phase difference, and the calculation formula is as follows:

[0039] ;

[0040] ; ​​​​​

[0041] wherein PAC is the cross-frequency phase-amplitude coupling, SI is the synchrony index; θ phase is the phase of the theta band, γ power phase is the power phase of the gamma band.

[0042] In some embodiments, the different brain wave band data obtained in step (2) is taken as an input feature, and an XGBoost model training is performed to obtain a trained XGBoost prediction model, and the XGBoost model training comprises: calculating the EEG signal features corresponding to the no odor stimulation and the odor stimulation respectively and inputting them into the XGBoost model, setting an initial prediction value based on the preference sensory evaluation data, gradually reducing the error between the initial prediction value and the actual target value in an iterative manner until the model performance reaches a predetermined condition to stop training, and obtaining the XGBoost prediction model.

[0043] In some embodiments, the error between the initial prediction value and the actual target value is gradually reduced in an iterative manner until the model performance reaches a predetermined condition to stop training. This includes gradually reducing the error between the initial prediction value and the actual target value by iteratively adding decision trees, and the prediction result of each tree is added to the prediction value of the previous round to form a new prediction value, which is taken as the basis for the next round of iteration, and the iteration is continuously performed until the model performance reaches a predetermined condition to stop training.

[0044] In another aspect, the present application also provides a liquor flavor perception evaluation system, which is based on the aforementioned liquor flavor preference analysis method to evaluate the perception of liquor flavor.

[0045] In some embodiments, the liquor flavor perception evaluation system comprises:

[0046] A liquor flavor perception evaluation module: the liquor flavor perception evaluation module is used to obtain the sensory evaluation data of the subjects on different liquor flavors, and to obtain the EEG signal data of the subjects during the sensory evaluation of different liquor flavors;

[0047] An EEG signal data processing module: the EEG signal data processing module analyzes and processes the EGG signal data and the sensory evaluation data collected and transmitted by the liquor flavor perception evaluation module to obtain the preference results of the subjects on different liquor flavors;

[0048] A preference result output module: the preference result output module outputs the preference results of the subjects obtained by the EEG signal data processing module.

[0049] The above liquor flavor preference degree analysis method collects brain electrical data and preference degree sensory evaluation data of subjects after being stimulated by different aroma type liquors, pre-processes the obtained brain electrical data, uses an XGBoost model to perform feature learning on the pre-processed brain electrical data to predict the preference degree, optimizes the predicted preference degree by using the wave band response value as a reference standard, and makes the final preference degree prediction value Y closer to the actual value. Since the present application mainly uses brain electrical data to analyze the preference degree, the time series data recorded by each channel of the brain electrical data can contain a large number of sampling points, so the brain electrical data has the characteristics of high dimensionality. XGBoost uses parallel computing technology and is suitable for processing the high dimensionality characteristics of brain electrical data. In addition, the features in the brain electrical signal have complex nonlinear relationships, and XGBoost can capture these nonlinear relationships through the combination of gradient boosting and decision trees. The feature importance evaluation mechanism of XGBoost can help identify which brain electrical features are most critical to predicting the preference degree. Since the preference degree of consumers is usually continuous rather than binary or multi-class classification, the XGBoost model can predict continuous preference degree scores rather than simple classification results. This means that it can more finely reflect the actual preferences of users and more accurately predict the true preference degree of the subjects. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 FIG. 1 is a structural schematic diagram of a liquor flavor perception evaluation system in one embodiment;

[0051] Figure 2 FIG. 2 is a structural schematic diagram of a digital odor player in one embodiment;

[0052] Figure 3 FIG. 3 is an odor playing flowchart of the digital odor player in one embodiment. DETAILED DESCRIPTION

[0053] The technical solutions of the present application are further illustrated by specific embodiments below, and the specific embodiments do not represent a limitation on the scope of protection of the present application. Some non-essential modifications and adjustments made by others according to the concept of the present application still fall within the scope of protection of the present application.

[0054] Embodiment 1: A liquor flavor preference degree analysis method

[0055] I. Collecting preference degree sensory evaluation data and brain electrical data of subjects after being stimulated by different aroma type liquors:

[0056] (1) Subjects

[0057] The experiment involved 30 participants, with a male-to-female ratio of 1:1; their ages ranged from 18 to 65. Considering the baijiu (Chinese liquor) consumer group, the ratio of participants was set to 18-27 years old: 28-40 years old: 41-65 years old as 1:3:1.

[0058] (2) The flavor profiles of different aroma types of baijiu evaluated are shown in Table 1:

[0059] Table 1: Flavor Types of Baijiu

[0060]

[0061] (3) Testing process

[0062] use Figure 2 The digital odor player shown is used to evaluate different types of baijiu flavors on test subjects. The digital odor player includes a controller, an air valve, and a smelling tube. The controller is electrically connected to the air pump and air valve and can control the opening and closing of the air pump and air valve. The air pump is connected to multiple air valves, each of which controls a different odor. When an air valve is opened, the corresponding odor controlled by that air valve flows into the smelling tube.

[0063] The specific testing process is as follows:

[0064] 1) Preparation before the experiment: Before the experiment, the subjects were asked to abstain from alcohol for eight hours and not eat for two hours. The windows and exhaust fans were opened to ventilate the laboratory, and the central air conditioning was turned on to maintain the laboratory temperature at 21±2°C.

[0065] 2) During the experiment, the subjects were asked to relax, breathe evenly, and follow the prompts on the software interface.

[0066] 3) Fill in the subject's information, including gender, age, and drinking habits (favorite brands and frequency of drinking).

[0067] 4) Wear the brain patch correctly with the assistance of staff; during the test, the EEG data will be transmitted to the device's EEG analysis processor in real time. It is essential to ensure that the subject wears the brain patch correctly throughout the test.

[0068] 5) Place the olfactometer of the digital odor player 3-5cm directly in front of the subject's nose to begin the test.

[0069] 6) As shown in Table 1, there are 20 odors, and each stimulus for each odor is... Figure 3The flow is shown, i.e. start the test, no odor test for 10s, odor test for 20s, rest for 5min. After each odor is played, the subject scores the sensory preference on the software interface, using a 1-9 scoring method, the higher the score, the higher the preference, the score obtained by scoring is used as the preference sensory evaluation data, then click to play the next odor, repeat the above odor playing process, the interval between the two odor tests is 5min, which aims to let the subjects rest fully and avoid olfactory fatigue. During the rest period, open the door and fan to increase air circulation. The residual odor in the olfactory evoking instrument is discharged to prevent the odor of the subsequent experiment from being contaminated.

[0070] 7) After all eight aroma dimensions are tested, the test is completed.

[0071] Through the above method, the preference sensory evaluation data and EEG data of 30 subjects after stimulation by 20 kinds of odors in eight dimensions of grassy, fruity, floral, sweet, nutty, dry plant, sour and other odors are obtained.

[0072] II. Data preprocessing:

[0073] Based on the complete EEG data collected in step one, EEG bioindicators with obvious dominance effect are designed based on the EEG analysis model, and EEG segments with and without odor stimulation are preprocessed, and the specific steps are as follows:

[0074] Firstly, the original EEG signal is detrended to remove the linear trend or baseline drift in the data to eliminate possible low-frequency interference;

[0075] Next, the detrended data is band-pass filtered with a frequency band range of 0.5-80 Hz to retain meaningful EEG activity and remove high-frequency noise and very low-frequency components;

[0076] Subsequently, power frequency interference is removed to eliminate the influence of power frequency on EEG signals, and EEG data that has been removed from the main environmental noise is obtained;

[0077] Next, according to the frequency range of different specific EEG bands, Delta (δ) wave: 0.5-4 Hz, Theta (θ) wave: 4-8 Hz, Alpha (α) wave: 8-12 Hz, Beta (β) wave: 12-30 Hz, Gamma (γ) wave: 30-100 Hz), the bandwidth, attenuation coefficient and filter order of the digital filter are designed, and the digital filter suitable for each frequency band is constructed;

[0078] Finally, these filters are used to filter the preprocessed EEG signals, extract the corresponding frequency band signal features, and obtain different EEG band data to provide input for subsequent feature extraction and model training.

[0079] III. Preference level model construction

[0080] The following EEG signal features are input into the XGBoost model:

[0081] 1. Ratio of different brain rhythm powers

[0082] 1) TBR: ratio of theta and beta power;

[0083] 2) ABR: ratio of alpha and beta power;

[0084] 3) DBR: ratio of delta and beta power;

[0085] 4) ARR: inverse of alpha power;

[0086] 5) BATR: ratio of beta and sum of alpha and theta power.

[0087] 2. Signal complexity

[0088] 1) C0: C0 complexity of the signal, C0 complexity is used to describe the proportion of irregular components in the electroencephalogram signal; its calculation method is:

[0089] FFT calculation: perform fast Fourier transform (FFT) on the signal to obtain the frequency domain representation of the signal fft_data. Then calculate the average of its frequency domain amplitude fft_average;

[0090] Filter high amplitude components: filter out components with amplitude greater than the average value fft_average from fft_data, set other components to 0, and form a new frequency domain signal

[0091] Inverse FFT: perform inverse Fourier transform (IFFT) on the filtered frequency domain signal to obtain the filtered time domain signal ifft_data

[0092] Calculate the C0 complexity by the following formula:

[0093] ;

[0094] denotes the original signal at the th sampling point; denotes the filtered signal (i.e. ifft_data) at the th sampling point; is the number of data points.

[0095] 2) Hjorth Activity: usually used to analyze the amplitude change of the EEG signal in the time domain; the calculation method is:

[0096] ;

[0097] is the number of data points; represents the EEG signal to be analyzed of the th data point; is the mean value of the signal,

[0098] 3) PAC: extract the cross-frequency phase-amplitude coupling (PAC) in the specified signal, PAC is used to analyze the interaction between different frequency components, the calculation method is: apply the Hilbert transform to the signal of the γ band to get the envelope information in complex form, calculate the square of the envelope to get the instantaneous power of the γ band; apply the Hilbert transform to the θ signal to get the phase θ phase , apply the Hilbert transform to the γ power and get its phase γ power phase ; the PAC value is the synchronization index (SI) represented by the complex exponential of the phase difference, the calculation formula is as follows:

[0099] ;

[0100] ;

[0101] The EEG signal features of multiple samples are input into the XGBoost model for training, based on the XGBoost (Extreme Gradient Boosting) algorithm, the input EEG signal features, XGBoost can effectively capture the complex nonlinear relationship between these features and the preference degree, and predict the preference degree.

[0102] The objective function of XGBoost is composed of a loss function and a regularization term, the objective function of XGBoost is:

[0103] ;

[0104] The objective function will be summed for all n samples of the loss function , and the regularization term is added to minimize the overall loss. Wherein, represents each sample in the data set, is the loss function, represents the predicted value and the true value The error between the predicted value of the model and the true value The degree of preference predicted by the model during the training process for the The true value of the degree of preference of the subject for the The true value of the degree of preference of the subject for the The EEG signal features constructed using the power and relative power of the frequency bands, and the complexity of the signal, are obtained by inputting The predicted value obtained by the model , The regularization term is used to control the complexity of the model and prevent overfitting.

[0105] The Mean Squared Error (MSE) is used as the loss function. MSE optimizes the model by minimizing the error between the predicted value and the actual degree of preference, measures the accuracy of the model's prediction, and helps to adjust the model parameters to improve the prediction performance. The squared error is the sum of the squares of the differences between the predicted value and the true value , and the formula is as follows:

[0106] ;

[0107] XGBoost uses a second-order Taylor expansion to approximate the objective function to utilize gradient (Gradient) and Hessian information to update the model. For a given loss function , we perform a second-order Taylor expansion of the loss function for each sample point :

[0108] ;

[0109] where is the output value of the model in the mth round (i.e., the output of the current weak learner), and are the first-order (Gradient) and second-order derivatives (Hessian) of the loss function with respect to the predicted value . Gradient is used to guide the update direction of the model parameters, and Hessian is used to determine the step size and amplitude of the model parameter update. The definitions are as follows:

[0110] Gradient is the first-order derivative of the loss function with respect to the predicted value , which is used to guide the update direction of the model parameters. The gradient calculation formula of the squared error loss function is as follows:

[0111] ;

[0112] Hessian is the second-order derivative of the loss function with respect to the predicted value​ second derivative of the square error loss function, used to determine the step size and magnitude of the model parameter updates. The Hessian of the square error loss function is computed as follows:

[0113] ;

[0114] The grade of these indexes represents a degree, and the division of the degree is based on the collection and manual labeling of the user's electroencephalogram (EEG) for model training. The training process is as follows:

[0115] Remove irrelevant data before odor stimulation, and calculate the corresponding EEG signal features under odor stimulation and no odor stimulation respectively; input the above EEG signal features into the XGBoost model, and the model processes the above EEG signal features as follows:

[0116] 1) Initial prediction value setting: The XGBoost model sets an initial prediction value based on the preference sensory evaluation data at the beginning of training, which is usually the mean value or other simple statistical value of the sample. This initial prediction value is only a baseline prediction and has no direct relationship with the EEG signal input features, reflecting the basic prediction of the model without learning any signal features.

[0117] 2) Error fitting and adjustment: Next, the model gradually introduces decision trees to reduce the difference between the initial prediction and the actual target value (preference sensory evaluation data). The model builds a series of decision trees, and each tree tries to fit the error between the current prediction value and the actual target value.

[0118] 3) Cumulative prediction value: When a new decision tree is established, it outputs a prediction value for each sample, and the prediction results of each decision tree are gradually accumulated to the initial prediction value, and the updated prediction value is used as the baseline prediction for the next round to continuously reduce the error between the prediction value and the target value.

[0119] 4) Training stopping condition: The model stops training when one of the following conditions is met: when the latest prediction value meets the iteration number reaches the preset maximum number of rounds (default is 100 times) or the reduction amplitude of the training error or the validation error is lower than the preset threshold (1e-4), the training is stopped, and the model preference prediction value X is output.

[0120] During the above training process, the calculation of the corresponding EEG signal features under odor stimulation and no odor stimulation refers to using the model to learn the relationship between the EEG signal features under odor stimulation and no odor stimulation based on the state without odor stimulation as the reference. Through this processing process, the processed data mainly reflects the state change before and after stimulation, avoiding the influence of individual differences caused by different emotions and attention of different subjects during testing.

[0121] IV. Optimization of preference prediction value

[0122] Based on the GBoost model obtained by training according to step III, the model preference prediction value X is obtained, and the final prediction result is further optimized based on the obtained model preference prediction value X. In the optimization process, multi-dimensional features are comprehensively considered, and in addition to the initial value output by the model, the ratios of delta waves, theta waves, alpha waves and beta waves are also combined for adjustment. The optimization method is as follows:

[0123] 1) Calculate the ratio of power of each wave band, and map the data to [1, 5]:

[0124] ;

[0125] Wherein, represents the power response value of the delta wave, represents the power response value of the beta wave, represents the power response value of the theta wave, represents the power response value of the beta wave, represents the power response value of the alpha wave,

[0126] 2) Adjust the preference prediction value according to the following formula:

[0127] ;

[0128] is the model preference prediction value, is the final preference prediction value, and the final preference prediction value is obtained by adjusting and optimizing according to the real data test: , , , .

[0129] V. Verification of the prediction and analysis method of preference

[0130] Taking the EGG signal characteristics corresponding to the flavor of the Baijiu of the subjects as the input characteristics, the preference of the consumers is predicted based on the XGBoost model obtained by training according to step III and the prediction formula of the preference prediction value obtained by optimization according to step IV, and compared with the preference sensory evaluation data as the true value of the preference. The specific results are shown in Table 2 below. The results show that the error between all the aroma preference prediction values and the true values is less than 10%, indicating that the method is accurate and applicable.

[0131] Table 2: Prediction value and true value of preference

[0132]

[0133] As can be seen from the above, the above liquor flavor preference analysis method provided by the present application can better realize the prediction of the preference degree of consumers for different liquor flavors. Therefore, the analysis results obtained by the method can provide data support for the development of new liquor products, and for the new products or samples that have been developed, the analysis method of the method can be used as a preference prediction method to provide data support for the market promotion of subsequent products.

[0134] In summary: the present application, by collecting the EEG data and the preference sensory evaluation data of different types of liquor after aroma stimulation, preprocessing the obtained EEG data, using the XGBoost model to learn the features of the preprocessed EEG data to predict the preference degree, taking the preference sensory evaluation data as the reference standard, and optimizing the predicted preference degree by the ratio of the band response value, the final preference prediction value Y is closer to the actual value. The ratio of different EEG rhythm power and signal complexity reflects the activity characteristics of the brain in different states. By inputting these features into the preference degree model, the influence of odor stimulation on the subjects can be more comprehensively understood, providing stronger explanatory power. The scores of multiple indicators can reduce the influence of individual differences, environmental factors or other noise; make the EEG response analysis of odor stimulation more comprehensive and in-depth.

[0135] Since the present application mainly uses EEG data for preference degree analysis, the time series data recorded by each channel of the EEG data can contain a large number of sampling points, so the EEG data has the characteristics of high dimensionality, and XGBoost uses parallel computing technology, which is suitable for processing the high dimensionality of EEG data. In addition, the features in the EEG signal have complex nonlinear relationships, XGBoost can capture these nonlinear relationships through the combination of gradient boosting and decision trees, and through the feature importance evaluation mechanism of XGBoost, it can help identify which EEG features are most critical to predicting preference degree; Since the preference degree of consumers is usually continuous rather than binary or multi-class classification, the XGBoost model can predict continuous preference scores rather than simple classification results. This means it can reflect users' actual preferences more finely, helping to more accurately predict the true preference degree of the subjects.

[0136] At the same time, taking the comprehensive sensitivity score considering the ratio of multiple EEG frequency bands and its change as one of the model input features, the predicted value of the model can more comprehensively evaluate the influence of odor on brain activity.

[0137] Embodiment 2 A liquor flavor perception evaluation system

[0138] This embodiment is based on the liquor flavor preference analysis method provided in embodiment 1, and provides a liquor flavor perception evaluation system, which is described with reference to Figure 1, the evaluation system comprises:

[0139] The Baijiu flavor perception evaluation module includes sensory evaluation tests of different Baijiu flavors on the subjects, and collects EEG signal data of the subjects during the sensory evaluation tests of different Baijiu flavors, and transmits the collected EEG signal data to the EEG signal data module; wherein, based on the method in embodiment 1, the sensory evaluation tests of different Baijiu flavors on the subjects are performed, and when the subjects are subjected to the sensory evaluation tests of different Baijiu flavors by using the digital odor player, the subjects need to wear the brain stickers, and the EEG signal data generated by the subjects during the sensory evaluation tests of different Baijiu flavors based on the digital odor player are collected through the brain stickers, and the collected EEG signal data are transmitted to the EEG analysis processor;

[0140] The EEG signal data processing module uses the EEG analysis processor to analyze and process the EGG signal data collected and transmitted based on the Baijiu flavor perception evaluation module, to obtain the preference degree results of the subjects for different Baijiu flavors, and transmits the preference degree results of the subjects obtained by analysis and processing to the result output module;

[0141] The preference degree result output module uses a software display interface to output the preference degree results of the subjects obtained by the EEG signal data processing module.

[0142] The above description is a detailed description of the preferred embodiments of the present application, but the embodiments are not intended to limit the scope of the patent application of the present application, and any equivalent changes or modifications made under the technical spirit disclosed by the present application should belong to the patent scope covered by the present application.

Claims

1. A method for analyzing the flavor preference of white liquor, characterized in that, The method comprises the following steps: (1) collecting sensory evaluation data of subjects' preference degree and electroencephalogram data after being stimulated by different Baijiu flavor stimuli; (2) preprocessing the electroencephalogram data obtained in step (1) to obtain different electroencephalogram band data; (3) taking the different electroencephalogram band data obtained in step (2) as input features, combining the sensory evaluation data of the subjects' preference degree collected in step (1), and performing XGBoost model training to obtain a trained XGBoost prediction model, and predicting the subjects' preference degree based on the XGBoost prediction model to obtain a prediction value X; the different electroencephalogram band data comprises at least one of the ratio data of power corresponding to different bands and signal complexity data; the signal complexity data comprises at least one of C0 complexity, Hjorth activity parameter, and cross-frequency phase-amplitude coupling; (4) combining the prediction value X obtained in step (3) with the ratio data of power corresponding to different bands, and adding the prediction value X and the ratio of power corresponding to different bands according to the corresponding weight to calculate the subjects' preference degree result Y; the ratio of power corresponding to different bands comprises at least one of the ratio of delta wave power to beta wave power, the ratio of theta wave power to beta wave power, and the ratio of alpha wave power to beta wave power.

2. The method of claim 1, wherein, In step (2), the preprocessing comprises: detrending, band-pass filtering, and power line interference removal processing on the electroencephalogram data obtained in step (2); designing the bandwidth, attenuation coefficient, and filter order of a digital filter according to the frequency range of different electroencephalogram bands, filtering the electroencephalogram signal after power line interference removal processing with the corresponding digital filter, extracting the corresponding frequency band signal features, and obtaining different electroencephalogram band data.

3. The method of claim 1, wherein, The preference degree result Y is calculated by the following formula: wherein , , , , , , , , are the power response values of delta, beta, alpha, theta waves, respectively.

4. The method of claim 3, wherein, ， ， ， 。 5. The method of claim 1, wherein, In step (1), the flavor of different Baijiu comprises at least one dimension of grassy aroma, fruity aroma, floral aroma, sweet aroma, nutty aroma, dry plant aroma, sour aroma, and other aroma; the grassy aroma comprises at least one of green grass aroma, raw grain aroma, and bamboo aroma; the fruity aroma comprises at least one of apple aroma, pineapple aroma, and banana aroma; the floral aroma comprises rose aroma, gardenia aroma, and violet aroma; the sweet aroma comprises at least one of vanilla aroma and milk candy aroma; the nutty aroma comprises cocoa aroma and paste aroma; the dry plant aroma comprises at least one of traditional Chinese medicine aroma; the sour aroma is vinegar aroma; and the other aroma comprises at least one of mushroom aroma, cellar aroma, sulfur aroma, yogurt aroma, and oil aroma.

6. The method of claim 1, wherein, In step (3), The ratio data of power corresponding to different bands comprises at least one of the ratio of theta band power to beta band power, the ratio of alpha band power to beta band power, the ratio of delta band power to beta band power, the reciprocal of alpha band power, and the ratio of beta band power to the sum of alpha band power and theta band power.

7. The method of claim 6, wherein, In step (3), The Co complexity is obtained by converting the electroencephalogram signal into a frequency domain through a fast Fourier transform, calculating the average value of the frequency domain amplitude, screening out components with an amplitude greater than the average value, and then performing an inverse Fourier transform to obtain a filtered time domain signal, calculating the sum of the absolute differences between the filtered time domain signal and the original time domain signal at each time point, and then dividing the sum by the sum of the absolute values of all points of the original signal. The Hjorth activity parameter is obtained by calculating the ratio of the variance of the electroencephalogram signal to the square of the mean value to analyze the amplitude change of the signal in the time domain. The cross-frequency phase-amplitude coupling is obtained by applying Hilbert transform to the signal of the gamma band to obtain the instantaneous power, applying Hilbert transform to the theta signal to obtain the phase, and then calculating the phase of the gamma power, and the synchronization index represented by the average of the complex exponential of the phase difference.

8. The method of claim 1, wherein, The XGBoost prediction model is obtained by inputting the different electroencephalogram wave band data obtained in step (2) as input features into the XGBoost model, and training the XGBoost model, wherein the XGBoost model training comprises: inputting the EEG signal features corresponding to the no odor stimulation and the odor stimulation into the XGBoost model, respectively, setting an initial prediction value based on the preference sensory evaluation data, gradually reducing the error between the initial prediction value and the actual target value through iteration, and stopping training until the model performance reaches the predetermined condition to obtain the XGBoost prediction model.

9. The method of claim 8, wherein, The method of gradually reducing the error between the initial prediction value and the actual target value through iteration, and stopping training until the model performance reaches the predetermined condition comprises: gradually reducing the error between the initial prediction value and the actual target value by iteratively adding decision trees, accumulating the prediction results of each tree to the prediction value of the previous round to form a new prediction value, taking the new prediction value as the basis for the next iteration, and continuously iterating until the model performance reaches the predetermined condition to stop training.

10. A system for the sensory evaluation of white spirits flavor, characterized by, The evaluation system performs the perception evaluation of the flavor of the liquor based on the method of any one of claims 1-9. The evaluation system comprises: The liquor flavor perception evaluation module is configured to obtain the sensory evaluation data of the subjects on different liquor flavors, and obtain the EEG signal data of the subjects during the sensory evaluation of different liquor flavors. The EEG signal data processing module is configured to analyze and process the EGG signal data and the sensory evaluation data collected and transmitted by the liquor flavor perception evaluation module to obtain the preference results of the subjects on different liquor flavors. The preference result output module is configured to output the preference results of the subjects obtained by the EEG signal data processing module.

Images