Method for identifying delicate flavor induced electroencephalogram signals in taste sensing link

Abstract

The invention belongs to the field of freshness perception evaluation, and provides a method for identifying a delicate flavor induced electroencephalogram signal in a taste perception link, which comprises the following steps of: acquiring taste electroencephalogram data induced under the stimulation of delicate flavor by using an electroencephalogram signal acquisition system, establishing a delicate flavor electroencephalogram data set by using the acquired data, cleaning the original data by adopting an independent component analysis algorithm in sequence, including filtering, electrooculogram removal, artifact removal and re-reference, calculating frequency (Hz) and energy spectrum density, and performing vectorization and regularization arrangement on 0-13Hz signals according to a brain region; and then a minority class oversampling technology is used to amplify a data set and divide a training set and a test set, an integrated model is constructed for signal classification, and the finally trained model is used for delicate flavor identification. According to the integrated model method provided by the invention, the accuracy and the stability of the delicate flavor electroencephalogram recognition model are remarkably improved, and the delicate flavor recognition is effectively realized.

Description

Technical Field

[0001] The present invention belongs to the field of fresh sensation perception evaluation, and particularly relates to a method for identifying brain electrical signals induced by umami in the taste perception process. Background Art

[0002] At present, most taste sensory analyses rely on human senses and machine perception. However, there are many problems with the above methods. In human sensory evaluation, the judgment of taste is easily affected by the subjectivity of the testers. At the same time, consumers often express their preferences objectively and accurately, which results in low reliability and poor reproducibility of the experimental results. Collecting electroencephalogram (EEG) data of subjects for taste judgment has inherent advantages, which realizes the combination of umami perception and electroencephalogram. When using electroencephalogram to identify umami aqueous solutions, key parameters such as the diversity of umami stimulation samples, data processing methods, and pattern recognition strategies will affect the recognition results of the model. Summary of the Invention

[0003] To solve the above technical problems, the present invention provides a method for identifying brain electrical signals induced by umami in the taste perception process to solve the problems in the prior art. The technical solution adopted by the present invention is as follows:

[0004] A method for identifying brain electrical signals induced by umami in the taste perception process includes the following steps:

[0005] Step S1, configure taste stimulation samples for umami brain electrical induction, and use an electroencephalogram signal acquisition system to collect the electroencephalogram data induced by subjects under umami stimulation solutions and non-umami stimulation solutions respectively;

[0006] Step S2, sequentially use the independent component analysis algorithm to clean the original data, including filtering, removing electrooculogram, artifacts and re-referencing, and calculate the frequency and energy spectral density. Vectorize and regularize the signals from 0 to 13 Hz according to brain regions;

[0007] Step S3, then use the method of minority oversampling to amplify the data set and divide it into a training set and a test set, and perform feature screening, construct an ensemble model for signal classification, and finally use the trained model for umami recognition;

[0008] Step S4, train a classification sub-model, and then fit an ensemble judgment model to finally complete umami recognition.

[0009] Further, in step S1, the umami stimulation solutions include sodium glutamate aqueous solution, disodium 5′-inosinate aqueous solution, succinic acid aqueous solution, and a mixed aqueous solution of sodium glutamate and 5′-monophosphate purine nucleoside; the non-umami stimulation solution includes pure water.

[0010] Further, in step S1, the EEG Ag–AgCl 64-channel device is used to collect gustatory EEG data of no less than 40 subjects at a frequency of 1000 Hz. For each subject, the EEG data under umami stimulation and non-umami stimulation are collected respectively, with at least 3 parallel detections, and each parallel detection lasts for at least 30 seconds.

[0011] Further, in step S2, the collected data are all subjected to brain region signal vectorization, including: matching the relationship between the regional time points and frequencies of the forebrain region F, left brain region LT, right brain region RT, central brain region C, and parieto-occipital brain region PO, and matching from f1 to f25 to 0.976 to 12.695 Hz in sequence to obtain data of 25 points, and then merging the data in the order of the forebrain region F, left brain region LT, right brain region RT, central brain region C, and parieto-occipital brain region PO in sequence to obtain a 1*125 bits array.

[0012] Further, in step S2, a percentage conversion of the individualized difference PSD value is performed: using the key frequency as the value, calculating the percentage of each value in the 25 values and the maximum value as the PSD percentage, which is used as the input data for the model in step S3:

[0013]

[0014] where F i represents the percentage of the current data point; PSD i represents the energy spectral density of the current data point; PSD1 to PSD 25 represents the maximum value of the energy spectral density of 25 data points, (PSD1 to PSD 25 ) max represents the maximum value among them.

[0015] Further, in step S2, the collected data are standardized, and the data are made absolute and then subjected to percentage conversion.

[0016] Further, in step S3, the Kolmogorov-Smirnov test and T test are used for feature screening, and the features with significant P≤0.05 are selected as the eigenvalue for modeling in step S4.

[0017] Further, in step S4, based on at least 5 binary classification algorithms, sub-models are trained, and accuracy and the receiver operating characteristic curve are used as the criteria for optimization. The calculation formulas for the evaluation indicators are as follows:

[0018]

[0019] Among them, ACC represents accuracy, PRE represents precision, MCC represents Matthews correlation coefficient, AP represents average precision, TP is true positive, TN is true negative, FP is false positive, and FN is false negative.

[0020] Furthermore, in step S4, the integrated model uses the SVC algorithm for fitting.

[0021] The present invention has the following beneficial effects:

[0022] (1) While establishing the umami EEG experimental paradigm, the present invention also establishes a method for obtaining umami EEG signals, including the acquisition process of EEG signals under umami stimulation and the preprocessing method of EEG signals;

[0023] (2) The present invention proposes a brand-new integrated model that can realize the recognition of umami EEG signals based on the integrated model. This model identifies the characteristic EEG region recognition signals and can achieve high-accuracy classification by rejecting the signals that can be realized;

[0024] (3) The umami EEG signal recognition strategy proposed by the present invention can be used in brain-computer interface technologies related to taste and has important value in sensory evaluation, auxiliary judgment of taste disorder-related diseases, and the application of virtual taste. Description of the Drawings

[0025] Figure 1 is a flowchart of the umami signal recognition method based on electroencephalogram in the embodiment of the present invention;

[0026] Figure 2 is a schematic diagram of the sample stimulus recording details;

[0027] Figure 3 is a diagram of the parameters and evaluation results of the sub-model;

[0028] Figure 4 is a schematic diagram of the pre- and post-distribution of the EEG PSD percentage;

[0029] Figure 5 is a schematic diagram of the effect of the data augmentation sampling method;

[0030] Figure 6 is a diagram of the integration mode and evaluation results of the model. Detailed Embodiments

[0031] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the Figures 1-6 in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. If not specifically specified, the technical means used in the embodiments are conventional means well known to those skilled in the art.

[0032] A recognition method based on umami-induced brain electrical signals in the umami perception process, comprising the following specific steps:

[0033] S1. Configure taste stimulus samples for umami brain electrical induction, and use an electroencephalogram signal acquisition system to collect the electroencephalogram data induced in subjects under the stimulation of umami / non-umami solutions respectively.

[0034] S2. Successively use the independent component analysis algorithm to clean the original data, including filtering, removing electrooculogram, artifacts and re-referencing, and calculate the frequency in Hz and the power spectral density. Vectorize and regularize the signals from 0 to 13 Hz according to brain regions.

[0035] S3. Subsequently, use the method of minority oversampling to amplify the data set and divide it into a training set and a test set, and perform feature screening. Use the idea of an ensemble model to construct an ensemble model for signal classification, and finally the trained model is used for taste recognition.

[0036] S4. Train the classification sub-model, and then fit the ensemble judgment model to finally complete umami taste recognition.

[0037] In step S1, 40 subjects were recruited. These subjects have all received strict electroencephalogram and sensory training, are familiar with umami, are in good health, have a good mental state and can accurately distinguish umami. The experiment was carried out in a safe, quiet and odorless electroencephalogram test room, and the appropriate temperature was 21±1°C. To avoid the influence of negative emotions or satiety during the test, electroencephalogram measurements were carried out in the morning (from 10:00 to 11:00 am) and in the afternoon (from 3:00 to 4:00 pm).

[0038] In step S1, the electroencephalogram signals were collected by a non-gel electrode cap and a device with 64 brain regions of EEG Ag–AgCl, such as a 64-channel non-invasive electroencephalograph Brain Products GmbH, Munich, Germany. The Fcz channel was used as the reference recording. The resistance was 5 kΩ and the sampling frequency was 1000 Hz.

[0039] During the experiment preparation stage, the subject put on the cap and was ready in a resting state to ensure a good fit between the cap and the scalp. When the instructor gave the subject the "start" instruction, the subject put the sample into the mouth (held for 10 s to perceive the taste), and at the same time the instructor started to record the EEG data. During the test, the subject closed his eyes and kept his head in one position during the measurement to minimize motion artifacts. After that, the subject was required to empty the mouth, rinse with ultrapure water and restore the electrical signal to the nearby baseline. For specific details, see Figure 2 .

[0040] In step S2, the brain signals are processed using Neuracle system software. Through data processing, the independent component analysis (ICA) algorithm in EEGLAB software is used to clean the waveforms of the original EEG signals, including filtering, removing electrooculogram (EOG), artifacts, and re-referencing. And the frequency (Hz) and power spectral density (PSD) are calculated. The present invention selects the signals of 0-13 Hz related to taste perception, and the specific matching relationship is shown in Table 1.

[0041] Table 1 Point-Time relationship table

[0042]

[0043] Furthermore, in step S2, percentage conversion of the PSD values of individual differences is performed: taking the key frequency as the value, calculating the percentage of each value in the 25 values and the maximum value as the PSD percentage, which serves as the input data for the model in step S3:

[0044]

[0045] where, F i represents the percentage of the current data point; PSD i represents the power spectral density of the current data point; PSD1 to PSD 25 represents the maximum value of the power spectral density of 25 data points, (PSD1 to PSD 25 ) max represents the maximum value among them.

[0046] Furthermore, the collected data is standardized, and the data is made absolute and then percentage-converted.

[0047] Furthermore, the Kolmogorov-Smirnov test and T-test are used for feature screening, and the features with significant P≤0.05 are selected as the eigenvalue for modeling in step S4.

[0048] Furthermore, in step S4, sub-models are trained based on at least 5 binary classification algorithms, and accuracy and receiver operating characteristic curve are used as the criteria for optimization. The calculation formulas for the evaluation indicators are as follows:

[0049]

[0050] where, ACC represents accuracy, PRE represents precision, MCC represents Matthews correlation coefficient, AP represents average precision, TP is true positive, TN is true negative, FP is false positive, and FN is false negative.

[0051] Furthermore, the integrated model is fitted using the SVC algorithm

[0052] In step S3, considering that the input data with more than two hundred samples and 125 features is prone to overfitting, and low-quality features are not conducive to model training. The present invention uses the Kolmogorov-Smirnov test and the T-test for feature screening. Under the condition that the significance screening threshold P≤0.05, out of a total of 125 features, 52 and 19 features remain after the KS-Test and T-Test respectively. Taking the intersection of the results of the two screening conditions gives 14 common features, which are (f2-F, f18-F, f22-F, f23-F, f2-LT, f11-LT, f2-RT, f18-RT, f2-C, f17-C, f23-C, f2-PO, f18-PO, f20-PO) and will be used for model training.

[0053] In the data augmentation sampling of step S3, considering the difference in the number of positive and negative data samples, oversampling of the minority samples is necessary. Figure 5 The results of the original data and the benchmark model built after data augmentation using the support vector machine / minority oversampling technique are shown respectively. The results show that the data augmentation method is the minority oversampling technique and the effect is the best when K-Neighbour is 4, and the results of several evaluation indicators are at relatively high points.

[0054] In step S4, the algorithm shown in Table 2 is used to construct the sub-models. Figure 3 The evaluation of the top four sub-models with the highest scores is shown.

[0055] Table 2 Binary classification algorithm parameter table

[0056]

[0057]

[0058]

[0059] In the model integration stage of step S4, Figure 6 The accuracy of the model when the voting decision threshold is 1 - 4 votes is recorded. It can be seen that the best case is when N = 2 (N = 1 means that if one of the four sub-models considers it umami, it is umami, and it has better performance when the number of positive samples in the Test Set has an advantage, so the generalization performance is poor). The accuracy of the optimal model (N = 2) of the voting method in the Validation and Test Set is 0.9 and 0.75 respectively. Although the voting method has achieved good results, the integrated model based on fitting the confidence probabilities of the outputs of the four sub-models using the support vector machine algorithm can achieve an accuracy of 0.78 and an MCC of 0.39, and its effect is better.

[0060] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations, variations, modifications, and substitutions made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for identifying umami-induced electroencephalogram signals in the taste perception process, characterized in that, Including the following steps: Step S1, configure taste stimulus samples for umami EEG induction, and use an EEG signal acquisition system to collect the EEG data induced by the subject under umami stimulus solution and non-umami stimulus solution respectively; Step S2, successively use the independent component analysis algorithm to clean the original data, including filtering, removing electrooculogram, artifacts and rereferencing, and calculate the frequency and energy spectral density. Vectorize and regularize the signals from 0 to 13 Hz according to brain regions; Step S3, then use the method of minority oversampling to amplify the dataset and divide it into a training set and a test set, and perform feature screening, construct an ensemble model for signal classification, and finally use the trained model for umami recognition; Step S4, train the classification sub-model, and then fit the ensemble judgment model to finally complete umami recognition.

2. The method for identifying the umami-induced electroencephalogram signal in the taste perception link according to claim 1, wherein In step S1, the umami stimulus solution includes aqueous solutions of sodium glutamate, disodium 5′-inosinate, succinic acid, and a mixed aqueous solution of sodium glutamate and 5′-monophosphate purine nucleoside; the non-umami stimulus solution includes pure water.

3. The method for identifying umami-induced electroencephalogram signals in the taste perception link according to claim 1, wherein, In step S1, use a device with 64 brain regions of EEG Ag–AgCl to collect the gustatory EEG data of no less than 40 subjects at a frequency of 1000 Hz. For each subject, collect the EEG data under umami stimulus and non-umami stimulus respectively, with at least 3 parallel detections, and each parallel detection time is at least 30 seconds.

4. The method for identifying the umami-induced brain electrical signal in the taste perception link according to claim 1, wherein In step S2, the collected data are all vectorized by brain region signals, including: matching the relationship between the regional time points and frequencies of the frontal brain region F, left brain region LT, right brain region RT, central brain region C, and parieto-occipital brain region PO, and matching from f1 to f25 to 0.976 to 12.695 Hz in sequence to obtain 25 data points, and then merge the data in the order of the frontal brain region F, left brain region LT, right brain region RT, central brain region C, and parieto-occipital brain region PO in sequence to obtain a 1*125 bits array.

5. The method for identifying umami-induced electroencephalogram signals in the taste perception link according to claim 1, characterized in that, In step S2, perform percentage conversion of the PSD values of individual differences: using the key frequency as the value, calculate the percentage of each value in the 25 values and the maximum value as the PSD percentage, which is used as the input data for the model in step S3; Among them, F i represents the percentage of the current data point; PSD i represents the energy spectral density of the current data point; PSD1 to PSD 25 represents the maximum value of the energy spectral density of 25 data points, (PSD1 to PSD 25 ) max represents the maximum value among them.

6. The method for identifying the umami-induced brain electrical signal in the taste perception link according to claim 1, wherein In step S2, perform standardization processing on the collected data, and perform absolute value conversion on the data and then percentage conversion.

7. The method for identifying umami-induced electroencephalogram signals in the taste perception link according to claim 1, characterized in that, In step S3, use the Kolmogorov-Smirnov test and T test for feature screening, and select the features with significant P≤0.05 as the eigenvalue for modeling in step S4.

8. The method for identifying the umami-induced brain electrical signal in the taste perception link according to claim 1, wherein In step S4, train the sub-model based on at least 5 binary classification algorithms, and use accuracy and receiver operating characteristic curve as the criteria for optimization. The calculation formula of the evaluation index is as follows: Among them, ACC represents accuracy, PRE represents precision, MCC represents Matthews correlation coefficient, AP represents average precision, TP is true positive, TN is true negative, FP is false positive, and FN is false negative.

9. The method for identifying the umami-induced electroencephalogram signal in the taste perception link according to claim 1, characterized in that, In step S4, the ensemble model is fitted using the SVC algorithm.

Images