A method for constructing EEG data learning models without individual differences

Abstract

The present invention discloses a method for constructing an EEG data learning model for de-individualizing differences. EEG data with corresponding calibration values ​​are obtained through preprocessing. Data enhancement technology is used to enhance the calibrated EEG data to prevent distortion while increasing the data volume, facilitating subsequent operations and improving the accuracy of subsequent model operations. A loss function is constructed to obtain a loss baseline value in the calibration process. After obtaining the loss gain coefficient, the EEG data is adjusted. The EEG data is continuously cleaned and adjusted according to the model to present differential integration. After the data is proportionally unified, a data learning model is constructed based on the data after differential integration, thereby removing the differences between the original data and being able to uniformly incorporate the EEG data, so that the error between the output result of the final data processing model and the actual output result is controlled within a small range.

Description

Technical Field

[0001] The present invention relates to the technical field of electroencephalogram (EEG) signal processing, and in particular to a method for constructing an EEG data learning model that eliminates individual differences. Background Art

[0002] Electroencephalogram (EEG) is a method of recording brain activity using electrophysiological indicators. When the brain is active, the postsynaptic potentials generated synchronously by a large number of neurons are summed up. It records the changes in electrical waves during brain activity and is a comprehensive reflection of the electrophysiological activity of brain nerve cells on the cerebral cortex or scalp surface. EEG signals contain rich, diverse, and objective physiological information. Their research and analysis are often used in brain-computer interfaces (BCIs) to achieve a direct connection between the human or animal brain and external devices, enabling information exchange between the brain and devices. Due to their objectivity and convenience compared to other physiological signals such as magnetic resonance imaging, EEG signals are also used to assess physiological status, such as analyzing people's emotions through EEG signals and using EEG signals to determine whether the driver is in a state of fatigue in fatigue driving applications.

[0003] Existing EEG signal analysis methods include traditional signal analysis methods based on the time domain or frequency domain. Common time domain analysis methods include waveform feature description methods and autoregressive AR models. Common frequency domain analysis methods include Fourier transform, power spectral density, nonparametric spectrum estimation methods, and power spectrum estimation based on AR models. However, both time domain and frequency domain analysis methods are unable to unify the differences among individuals during data aggregation and analysis. This results in a large error between the output of the final data processing model and the actual output, making it impossible to obtain data analysis that simultaneously accounts for the actual conditions of individuals with different differences. Summary of the Invention

[0004] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of this application to avoid obscuring the purpose of this section, the abstract and the title of the invention, and such simplifications or omissions should not be used to limit the scope of the present invention.

[0005] In view of the above problems existing in the existing EEG data learning model, the present invention is proposed.

[0006] Therefore, the technical problem solved by the present invention is to solve the problem that the existing EEG data learning model is unable to unify the differences of the constituent individuals during the process of data aggregation and analysis, resulting in a large error between the output results of the final data processing model and the actual output results, and it is impossible to obtain data analysis that takes into account the actual status of individuals with different differences at the same time.

[0007] To solve the above technical problems, the present invention provides the following technical solutions: a method for constructing an EEG data learning model without individual differences, including data preprocessing to obtain EEG data with corresponding calibration values; using data enhancement technology to enhance the calibrated EEG data; constructing a loss function to obtain a loss baseline value during the calibration process; obtaining a loss gain coefficient based on the loss baseline value; adjusting the EEG data based on the loss gain coefficient; and constructing a data learning model based on the adjusted EEG data.

[0008] As a preferred solution of the method for constructing an EEG data learning model with de-individualized differences described in the present invention, data preprocessing specifically includes differential reduction of EEG data; adjusting and calibrating the EEG data after differential reduction to obtain corresponding labels; wherein, differential reduction of EEG data specifically includes: S1: obtaining the correlation between each EEG data; S2: obtaining the EEG data of each EEG data in the corresponding defined frequency band according to the frequency band definition of the correlation; S3: intercepting the EEG data of the corresponding defined frequency band and the corresponding electric wave graph, and saving the differential reduction of EEG data; wherein, adjusting and calibrating the EEG data after differential reduction specifically includes: S1: obtaining the electric wave signal fluctuation range value of each EEG data after differential reduction; S2: obtaining the average fluctuation value of the fluctuation range value of each electric wave signal; S3: obtaining the difference between the fluctuation range value of each electric wave signal and its average fluctuation value, which is defined as the calibration value, and calibrating each EEG data after differential reduction according to the calibration value to obtain EEG data with corresponding calibration value.

[0009] As a preferred solution of the method for constructing an EEG data learning model without individualized differences described in the present invention, obtaining the correlation between each EEG data specifically includes plotting each EEG data into a corresponding electric wave graph under a unified coordinate system; obtaining the frequency band with the largest fluctuation range in different electric wave graphs; and comprehensively analyzing all the obtained frequency bands to obtain the frequency band with the largest degree of intersection between different frequency bands, which is defined as the correlation.

[0010] As a preferred solution of the method for constructing an EEG data learning model without individualized differences described in the present invention, when obtaining the frequency band with the largest fluctuation range in different electric wave graphs, the frequency band selection difference range is 10 Hz.

[0011] As a preferred solution of the method for constructing an EEG data learning model without individualized differences described in the present invention, the calibrated EEG data is enhanced using data enhancement technology, specifically including obtaining the calibration value and each EEG data after differential reduction; and the calibrated EEG data is enhanced based on the absolute content of the calibration value to increase the corresponding fluctuation range content.

[0012] As a preferred solution of the method for constructing an EEG data learning model without individualized differences described in the present invention, the loss function constructed is specifically:

[0013]

[0014] Wherein, L represents the loss reference value of the mth sample; E s represents the number of EEG data samples after differential reduction; X m represents the calibration value of the mth sample; z represents the average fluctuation value;

[0015] Moreover, L 基 =(L1+L2+…+L m ) / m.

[0016] As a preferred solution of the method for constructing an EEG data learning model without individualized differences described in the present invention, the formula for obtaining the loss gain coefficient based on the loss reference value specifically includes:

[0017]

[0018] Where ρ represents the loss gain coefficient, x L基-1 represents the loss gain coefficient of the mth sample, i represents the equilibrium constant coefficient, E s represents the number of EEG data samples after difference reduction; represents the equalization loss gain coefficient of the mth sample;

[0019] The equilibrium constant coefficient i is defined as 1 or 2 or ㏑2.

[0020] As a preferred embodiment of the method for constructing a de-individualized EEG data learning model according to the present invention, the method further comprises: adjusting the EEG data according to the loss-gain coefficient specifically comprising constructing a difference function; incorporating the loss-gain coefficient into the difference function to obtain a corresponding loss-gain difference function; and performing difference adjustment on each EEG data according to the output value of each loss-gain difference function.

[0021] Among them, the constructed difference function is specifically:

[0022]

[0023] Among them, ψ m is the difference function of the mth sample; ρ represents the loss gain coefficient; E s represents the number of EEG data samples after differential reduction; L represents the loss baseline value of the mth sample; L 基 Represents the loss benchmark value; X m represents the calibration value of the mth sample; z represents the average fluctuation value; xdx is the integration operation.

[0024] As a preferred solution of the method for constructing an EEG data learning model with de-individualized differences described in the present invention, the constructed data learning model includes a CNN layer and an encoding layer; the CNN layer includes a convolutional layer, a pooling layer and a fully connected layer; the encoding layer includes a convolutional encoding layer and a convolutional decoding layer.

[0025] Beneficial effects of the present invention: The present invention provides a method for constructing an EEG data learning model without individualized differences, wherein EEG data with corresponding calibration values ​​are obtained through preprocessing, wherein the calibrated EEG data is enhanced by using data enhancement technology to prevent distortion while increasing the amount of data, facilitate subsequent operations and improve the accuracy of subsequent model operations, and obtain the loss baseline value in the calibration process by constructing a loss function, and adjust the EEG data after obtaining the loss gain coefficient, and continuously clean and adjust the EEG data according to the model to make it present differential integration, and construct a data learning model based on the data after differential unification after data proportionalization and unification, thereby removing the differences between the original data and being able to uniformly incorporate the EEG data, so that the error between the output result of the final data processing model and the actual output result is controlled within a small range, thereby solving the problem that the existing EEG data learning model cannot unify the differences of the constituent individuals in the process of data aggregation and analysis, resulting in a large error between the output result of the final data processing model and the actual output result, and failing to obtain data analysis that takes into account the actual status of individuals with different differences at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them:

[0027] Figure 1 This is a flowchart of the overall method for constructing an EEG data learning model without individualized differences provided by the present invention.

[0028] Figure 2Flowchart of the method for differential reduction of EEG data provided by the present invention.

[0029] Figure 3 This is a flow chart of the method for adjusting and calibrating EEG data after differential reduction provided by the present invention.

[0030] Figure 4 This is a flow chart of the method for obtaining the correlation between EEG data provided by the present invention.

[0031] Figure 5 This is a flow chart of the method for adjusting EEG data based on the loss-gain coefficient provided by the present invention. DETAILED DESCRIPTION

[0032] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0033] The existing EEG data learning model is unable to unify the differences among individuals during data aggregation and analysis, resulting in a large error between the output results of the final data processing model and the actual output results, and it is impossible to obtain data analysis that takes into account the actual status of individuals with different differences at the same time.

[0034] Therefore, please refer to Figure 1 The present invention provides a method for constructing an EEG data learning model that eliminates individual differences, comprising:

[0035] S1: Data preprocessing, obtaining EEG data with corresponding calibration values;

[0036] Furthermore, data preprocessing specifically includes:

[0037] Perform difference reduction on EEG data;

[0038] Adjust and calibrate the EEG data after differential reduction to obtain corresponding labels;

[0039] Among them, see Figure 2 , the difference reduction of EEG data specifically includes:

[0040] 1: Obtain the correlation between each EEG data;

[0041] For further information, see Figure 4 , obtaining the correlation between each EEG data specifically includes:

[0042] Each EEG data is plotted into a corresponding wave graph in a unified coordinate system;

[0043] Get the frequency band with the largest fluctuation range in different radio wave graphs;

[0044] All the obtained frequency bands are integrated to obtain the frequency band with the largest degree of intersection between different frequency bands, which is defined as the correlation degree.

[0045] It should be noted that when obtaining the frequency band with the largest fluctuation range in different radio wave graphs, the frequency band selection difference range is 10 Hz.

[0046] 2: Obtaining EEG data of each EEG data in the corresponding defined frequency band according to the frequency band definition of the correlation degree;

[0047] 3: intercept the EEG data of the corresponding defined frequency band and the corresponding electrogram, save and complete the differential reduction of the EEG data;

[0048] Among them, see Figure 3 , the adjustment and calibration of the EEG data after differential reduction specifically includes:

[0049] 1: Obtain the fluctuation range of the EEG signal after differential reduction;

[0050] 2: Get the average fluctuation value of the fluctuation range of each radio wave signal;

[0051] 3: Obtain the difference between the fluctuation range value of each radio wave signal and its average fluctuation value, which is defined as the calibration value. Calibrate each EEG data after differential reduction according to the calibration value to obtain EEG data with corresponding calibration value.

[0052] S2: Use data enhancement technology to enhance the calibrated EEG data;

[0053] It should be noted that data enhancement technology is a conventional signal processing technology and will not be described in detail again.

[0054] Furthermore, the data enhancement technology is used to enhance the calibrated EEG data, specifically including:

[0055] Obtain the EEG data after calibration and difference reduction;

[0056] The calibrated EEG data is enhanced based on the absolute content of the calibration value to increase the corresponding fluctuation range content.

[0057] S3: Construct a loss function to obtain the loss benchmark value during the calibration process;

[0058] Specifically, the constructed loss function is:

[0059]

[0060] Where L represents the loss benchmark value of the mth sample; E s represents the number of EEG data samples after differential reduction; X m represents the calibration value of the mth sample; z represents the average fluctuation value;

[0061] Moreover, L 基 =(L1+L2+…+L m ) / m.

[0062] S4: Obtaining the loss gain coefficient based on the loss baseline value;

[0063] Furthermore, the formula for obtaining the loss gain coefficient based on the loss baseline value specifically includes:

[0064]

[0065] Where ρ represents the loss gain coefficient, x L基-1 represents the loss gain coefficient of the mth sample, i represents the equilibrium constant coefficient, E s represents the number of EEG data samples after difference reduction; represents the equalization loss gain coefficient of the mth sample;

[0066] The equilibrium constant coefficient i is defined as 1 or 2 or ㏑2.

[0067] S5: Adjust the EEG data according to the loss-gain coefficient;

[0068] For further information, see Figure 5 , adjusting EEG data according to the loss-gain coefficient specifically includes:

[0069] Constructing a difference function;

[0070] Incorporating the loss-gain coefficient into the difference function to obtain the corresponding loss-gain difference function;

[0071] The difference of each EEG data is adjusted in turn according to the output value of each loss-gain difference function;

[0072] Among them, the constructed difference function is specifically:

[0073]

[0074] Among them, ψ m is the difference function of the mth sample; ρ represents the loss gain coefficient; E s represents the number of EEG data samples after differential reduction; L represents the loss baseline value of the mth sample; L 基 Indicates the loss benchmark value; X mrepresents the calibration value of the mth sample; z represents the average fluctuation value; xdx is the integration operation.

[0075] S6: Build a data learning model based on the adjusted EEG data.

[0076] It should be noted that the constructed data learning model includes CNN layer and encoding layer;

[0077] CNN layers include convolutional layers, pooling layers, and fully connected layers;

[0078] The coding layer includes a convolutional coding layer and a convolutional decoding layer.

[0079] Among them, the method of constructing a data learning model is an existing conventional data processing method, which will not be elaborated here.

[0080] The present invention provides a method for constructing an EEG data learning model without individualizing differences. The method obtains EEG data with corresponding calibration values ​​through preprocessing, wherein the calibrated EEG data is enhanced by using data enhancement technology to prevent distortion while increasing the data volume, thereby facilitating subsequent operations and improving the accuracy of subsequent model operations. A loss function is constructed to obtain a loss baseline value in the calibration process, and the EEG data is adjusted after obtaining a loss gain coefficient. The EEG data is continuously cleaned and adjusted according to the model to present differential integration. After the data is proportionally unified, a data learning model is constructed based on the differentially unified data, thereby removing the differences between the original data and being able to uniformly incorporate the EEG data, so that the error between the output result of the final data processing model and the actual output result is controlled within a small range. This solves the problem that the existing EEG data learning model cannot perform corresponding unification of the differences of the constituent individuals during data aggregation processing and analysis, resulting in a large error between the output result of the final data processing model and the actual output result, and cannot obtain data analysis that takes into account the actual status of individuals with different differences at the same time.

[0081] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for constructing a learning model for EEG data that eliminates individual differences, characterized by: include, Data preprocessing to obtain EEG data with corresponding calibration values; Data preprocessing specifically includes: Perform difference reduction on EEG data; Adjust and calibrate the EEG data after differential reduction to obtain corresponding labels; Among them, the differential reduction of EEG data specifically includes: S1: Obtaining the correlation between each EEG data, including: plotting each EEG data into a corresponding wave graph under a unified coordinate system; obtaining the frequency band with the largest fluctuation range in different wave graphs; and combining all the obtained frequency bands to obtain the frequency band with the largest degree of intersection between different frequency bands, which is defined as the correlation; S2: acquiring EEG data of each EEG data in a corresponding defined frequency band according to the frequency band definition of the correlation degree; S3: intercepting the EEG data of the corresponding defined frequency band and the corresponding electrogram, saving and completing the differential reduction of the EEG data; Among them, the adjustment and calibration of the EEG data after difference reduction specifically includes: S1: Obtain the fluctuation range value of the electric wave signal of each EEG data after differential reduction; S2: Obtain the average fluctuation value of the fluctuation range of each radio wave signal; S3: Obtaining the difference between the fluctuation range value of each electric wave signal and its average fluctuation value, defining it as the calibration value, calibrating each EEG data after the difference reduction according to the calibration value, and obtaining EEG data with the corresponding calibration value; The calibrated EEG data is enhanced using data enhancement technology, including: Obtaining the calibration value and each EEG data after differential reduction; Performing data enhancement on the calibrated EEG data according to the absolute content of the calibration value, increasing the corresponding fluctuation range content to construct a loss function, and obtaining a loss baseline value during the calibration process; Obtaining a loss gain coefficient based on the loss baseline value; Adjusting the EEG data according to the loss-gain coefficient includes constructing a difference function; incorporating the loss-gain coefficient into the difference function to obtain a corresponding loss-gain difference function; and performing difference adjustment on each EEG data according to the output value of each loss-gain difference function; Build a data learning model based on the adjusted EEG data.

2. The method for constructing a de-individualized EEG data learning model according to claim 1, characterized in that: When obtaining the frequency band with the largest fluctuation range in different radio wave graphs, the frequency band selection difference range is 10Hz.

3. The method for constructing a de-individualized EEG data learning model according to claim 1, characterized in that: The loss function constructed is specifically: Wherein, L represents the loss reference value of the mth sample; E s represents the number of EEG data samples after differential reduction; X m represents the calibration value of the mth sample; z represents the average fluctuation value.

4. The method for constructing a de-individualized EEG data learning model according to claim 3, characterized in that: The formula for obtaining the loss gain coefficient based on the loss reference value specifically includes: Where ρ represents the loss-gain coefficient, represents the loss gain coefficient of the mth sample, i represents the equilibrium constant coefficient, E s represents the number of EEG data samples after difference reduction; represents the equalization loss gain coefficient of the mth sample; Moreover, L 基 =(L1+L2+…+L m ) / m; The equalization constant coefficient i is defined as 1 or 2 or In2.

5. The method for constructing a de-individualized EEG data learning model according to claim 4, characterized in that: The constructed difference function is specifically: Among them, ψ m is the difference function of the mth sample; ρ represents the loss gain coefficient; E s represents the number of EEG data samples after differential reduction; L represents the loss baseline value of the mth sample; L 基 =(L1+L2+…+L m ) / m;X m represents the calibration value of the mth sample; z represents the average fluctuation value; xdx is the integration operation.

6. The method for constructing a de-individualized EEG data learning model according to claim 5, characterized in that: The constructed data learning model includes a CNN layer and a coding layer; the CNN layer includes a convolutional layer, a pooling layer and a fully connected layer; the coding layer includes a convolutional coding layer and a convolutional decoding layer.

Images