Method, system and device for recognizing emotions from electroencephalogram signals

Abstract

The present invention discloses a method, system and device for recognizing emotions from electroencephalogram (EEG) signals, and relates to the field of signal processing technology. The method comprises the following steps: obtaining original EEG signals stimulated by different emotional music; preprocessing the original EEG signals; performing feature extraction and standardization on the preprocessed EEG signals to determine an original feature set; determining the importance of each feature in the original feature set; using a multi-objective gray wolf algorithm to determine a group of non-dominated solutions based on the importance of each feature; defining gray wolf individuals corresponding to non-dominated solutions in the group of non-dominated solutions as a feature subset set; and determining the classification of the original EEG signals based on the feature subset set to achieve emotion recognition. The present invention can improve recognition accuracy and search efficiency.

Description

Technical Field

[0001] The present invention relates to the field of signal processing technology, and in particular to a method, system and device for recognizing emotions from electroencephalogram (EEG) signals. Background Art

[0002] Emotions are the spontaneous emotional states that arise in response to our external environment. They play a crucial role in our daily lives, and accurately identifying emotions is a crucial step in achieving normal interactions. With the development of wearable devices, emotion recognition based on EEG signals is gaining increasing attention. To effectively identify emotions, researchers use high-dimensional features as model input. These features often include irrelevant and redundant features, which can lead to overfitting and significantly limit model performance. Therefore, feature selection methods are needed to select the optimal feature subset from these features.

[0003] In the early stages of EEG-based emotion recognition research, feature selection was often treated as a single-objective optimization problem. Consequently, previous single-objective optimization algorithms aimed to improve classification accuracy. For example, He et al. proposed a firefly ensemble optimization algorithm that achieves high accuracy with fewer features by using a sorted probability opposition function. Tahir et al. combined chaotic mapping with a genetic algorithm to propose a binary chaotic genetic algorithm for searching for the optimal feature subset.

[0004] In fact, the desired outcome in feature selection for EEG-based emotion recognition is low feature dimensionality (Goal 1) and high classification accuracy (Goal 2). However, these two objectives often conflict with each other, making this a multi-objective optimization problem. Multi-objective optimization algorithms are specifically designed to solve multi-objective optimization problems. For example, Moctezuma et al. used NSGA-II to design an optimization process for EEG channel selection to obtain a set of channel combinations with high emotion recognition accuracy.

[0005] However, there are currently few multi-objective EEG feature selection methods for studying different emotional music stimuli. At the same time, there are still the following deficiencies when using multi-objective optimization algorithms to realize feature selection based on EEG emotion recognition: (1) During the search process, it is easy to fall into local optimality, resulting in the inability to further global search; (2) The computational complexity is high and the search efficiency is low. Summary of the Invention

[0006] The purpose of the present invention is to provide a method, system and device for EEG signal emotion recognition, which can improve recognition accuracy and search efficiency.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A method for recognizing emotions from an electroencephalogram signal, comprising:

[0009] Obtain original EEG signals stimulated by different emotional music;

[0010] Preprocess the raw EEG signals;

[0011] Perform feature extraction and standardization on the preprocessed EEG signals to determine the original feature set;

[0012] Determine the importance of each feature in the original feature set;

[0013] According to the importance of each feature, a multi-objective gray wolf algorithm is used to determine a set of non-dominated solutions; the gray wolf individuals corresponding to the non-dominated solutions in the set of non-dominated solutions are the feature subset set;

[0014] The classification of the original EEG signal is determined based on the feature subset set to achieve emotion recognition.

[0015] Optionally, the preprocessing of the original EEG signal specifically includes:

[0016] Filter the original EEG signal;

[0017] Perform artifact removal on the original EEG signal after filtering.

[0018] Optionally, the feature extraction and standardization processing of the preprocessed EEG signal to determine the original feature set specifically includes:

[0019] Determine the EEG signals corresponding to specific frequency bands related to emotions in emotional music, extract the features of the EEG signals, and determine the feature set;

[0020] The feature set is standardized using the z-score to determine the original feature set.

[0021] Optionally, determining the importance of each feature in the original feature set specifically includes:

[0022] Determine the total number of categories and the number of samples per category in the original feature set;

[0023] Use formula formula Calculate the Fisher score f for each dimension feature fisher_score (i);

[0024] Using the formula Calculate feature x i and x j The mutual information value between

[0025] Using the formula Determine feature x i The average mutual information of

[0026] Using the formula Determine feature x i Score

[0027] Using the formula Determine the importance score of each dimension feature

[0028] Among them, x i represents the i-th dimension feature, x j Represents the j-th dimension feature S B (i) and S w (i) Represents the inter-class variance and intra-class variance of the i-th dimension feature, H(x i ) represents the feature x i The entropy, H(x j ) represents the feature x j The entropy, H(x i ,x j ) represents the feature x i and feature x j The joint entropy of n represents the feature dimension.

[0029] Optionally, the multi-objective grey wolf algorithm is used to determine a set of non-dominated solutions based on the importance of each feature, specifically including:

[0030] Initialize the gray wolf population using the importance of each feature;

[0031] Calculate the target value of each individual in the gray wolf population, compare the dominance relationships between individuals, and determine a set of initial non-dominated solutions;

[0032] The population is updated using a population update strategy to determine a set of non-dominated solutions.

[0033] Optionally, calculating the target value of each individual in the gray wolf population and comparing the dominance relationships between individuals to determine a set of initial non-dominated solutions specifically includes:

[0034] Calculate the ratio f of the number of selected features in each individual of the gray wolf population to the total number of features obj_1 And use the support vector machine algorithm to calculate the inverse of the AUC value of each individual in the gray wolf population obj_2 ;

[0035] Comparison of individual f in different gray wolf populations obj_1 and f obj_2 When the size of When , individual a is a non-dominated solution; where S, R = {1, 2}, a set of initial non-dominated solutions is obtained.

[0036] Optionally, updating the population using a population update strategy to determine a set of non-dominated solutions specifically includes:

[0037] Set the maximum number of iterations maxiter, define the storage unit capacity num_archive, and set the objective function f obj_1 and the objective function f obj_2 ;

[0038] Use fuzzy selection operator to select three leading wolves, namely α wolf, β wolf and δ wolf;

[0039] Binarize the wolf pack;

[0040] Using the formula Determine the adaptive mutation operator;

[0041] Among them, Thr is the adaptive mutation operator, iter is the current iteration number;

[0042] Add each non-dominated solution obtained to the storage unit. When a new non-dominated solution appears, compare the relationship between the new non-dominated solution and the non-dominated solution in the storage unit and update the storage unit.

[0043] When the number of iterations is reached, the update is stopped and a set of non-dominated solutions is finally obtained.

[0044] An EEG signal emotion recognition system, comprising:

[0045] The original EEG signal acquisition module is used to obtain the original EEG signals under different emotional music stimulation;

[0046] The original EEG signal preprocessing module is used to preprocess the original EEG signal;

[0047] The original feature set determination module is used to extract features and perform standardization on the preprocessed EEG signals to determine the original feature set;

[0048] The feature importance determination module is used to determine the importance of each feature in the original feature set;

[0049] The non-dominated solution determination module is used to determine a set of non-dominated solutions using a multi-objective gray wolf algorithm based on the importance of each feature; the gray wolf individuals corresponding to the non-dominated solutions in the set of non-dominated solutions are the feature subset set;

[0050] The recognition module is used to determine the classification of the original EEG signal based on the feature subset set to achieve emotion recognition.

[0051] A device for recognizing emotions from brain waves comprises: at least one processor, at least one memory, and computer program instructions stored in the memory, wherein the method described is implemented when the computer program instructions are executed by the processor.

[0052] Optionally, the memory is a computer-readable storage medium.

[0053] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0054] The present invention provides an EEG signal emotion recognition method, system, and device. The method first preprocesses the original EEG signal to obtain a preprocessed EEG signal. The method then extracts different features of the EEG signal and calculates the importance of the features. The method then uses the feature importance to determine a set of non-dominated solutions using a multi-objective Grey Wolf algorithm. Furthermore, the method classifies the original EEG signal to achieve emotion recognition. This method provides a multi-objective Grey Wolf EEG feature selection method for different emotional music stimuli. This method avoids the problem of being trapped in local optimality during the search process, which can lead to the inability to conduct further global searches. This method improves recognition accuracy and search efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0056] Figure 1 This is a flow chart of a method for recognizing emotions using EEG signals provided by the present invention;

[0057] Figure 2 This is a schematic diagram of the overall process of the EEG signal emotion recognition method provided by the present invention;

[0058] Figure 3 A flowchart of an embodiment of the present invention;

[0059] Figure 4 This is a flow chart of gray wolf population update according to an embodiment of the present invention. DETAILED DESCRIPTION

[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0061] The purpose of the present invention is to provide a method, system and device for EEG emotion recognition, which can obtain a set of solutions with a balance between accuracy and feature dimension, and improve the efficiency of multi-target feature search in EEG emotion recognition tasks.

[0062] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0063] The embodiment of the present invention discloses a multi-objective EEG feature selection method for EEG emotion recognition tasks, namely a multi-objective Gray Wolf EEG feature selection method for different emotional music (MGWFS_DEM). First, the raw EEG signals stimulated by different emotional music were preprocessed to obtain preprocessed EEG signals. Next, various EEG signal features were extracted, including differential entropy (DE), fractal dimension (FD), differential asymmetry (DASM), rational asymmetry (RASM), higher-order zero crossings (HOC), and Hjorth complexity (HC). Fisher scores and mutual information values were calculated for these features and used to initialize a gray wolf population. The objective value of each individual in the population was calculated, and their dominance relationships were compared to identify non-dominated solutions. A population update strategy was then implemented, using a fuzzy selection operator and feature importance to compare the objective functions of different individuals. Three leading wolves were selected, and an adaptive mutation operator was used to escape the local optimum. The algorithm was continuously updated until the maximum number of iterations was reached, ultimately obtaining a set of non-dominated solutions. The individuals corresponding to the non-dominated solutions constituted the feature subset set. This provides a new technical means for EEG signal feature selection.

[0064] like Figure 1 and Figure 2 As shown, the present invention provides an EEG signal emotion recognition method, comprising:

[0065] S101, obtain the original EEG signals under different emotional music stimulation.

[0066] S102, preprocessing the original EEG signal.

[0067] S102 specifically includes:

[0068] Filter the original EEG signal; and remove artifacts from the filtered original EEG signal. This step specifically includes:

[0069] S11, uses a 50Hz notch filter to remove power line interference.

[0070] S12, retains the signal between 0.5Hz and 45Hz through high-pass filter and low-pass filter.

[0071] S13, Electrooculogram (EOG) signal artifacts were removed by independent component analysis.

[0072] S103, performing feature extraction and standardization on the pre-processed EEG signal to determine an original feature set.

[0073] S103 specifically includes:

[0074] Determine the EEG signals corresponding to the specific frequency bands related to emotions in emotional music, extract the features of the EEG signals, and determine the feature set X; where X = {x1, x2, ..., x n}, x i =(x i,1 ,x i,2 ,...,x i,j ,...,x i,m ), (i=1,2,...,n; j=1,2,...,m), x i represents the i-th dimension feature, n represents the feature dimension, and m represents the number of samples.

[0075] The feature set is standardized using the z-score to determine the original feature set X'.

[0076] The standardization process is:

[0077] According to the formula Perform standardization to determine the original feature set X'.

[0078] Among them, x i,j ' represents the i-th dimension feature and the j-th sample standardized data; represents the average value of the i-th dimension feature, σ(x i ) represents the standard deviation of the i-th dimension feature.

[0079] As a specific embodiment, S102 specifically includes:

[0080] S21, since Alpha is related to emotions, the Alpha frequency band (8-13Hz) of the EEG signal is extracted and the EEG signal is windowed and segmented with a window length of 2s and an overlap rate of 50%.

[0081] S22, according to the channel, extract different features of the EEG signal for each segment, including differential entropy, fractal dimension, differential asymmetry, rational asymmetry, high-order zero crossing, and Hjorth complexity; and take the average of the extracted EEG features to form a feature set X, where X = {x1, x2, ..., x n}, x i =(x i,1 ,x i,2 ,...,x i,j ,...,x i,m ), (i=1,2,...,n; j=1,2,...,m), x i represents the i-th dimension feature, n represents the feature dimension, and m represents the number of samples.

[0082] S23, using z-score to analyze the feature set X according to the formula Standardize and form the original feature set X'. Among them, x i,j ' represents the i-th dimension feature and the j-th sample standardized data; represents the average value of the i-th dimension feature, σ(x i ) represents the standard deviation of the i-th dimension feature.

[0083] S104, determining the importance of each feature in the original feature set.

[0084] S104 specifically includes:

[0085] Determine the total number of categories r(w1,w2,...,w r ) and the number of samples per category k M ,in

[0086] Use formula formula Calculate the Fisher score f for each dimension feature fisher_score (i).

[0087] Using the formula Calculate feature x i and x j The mutual information value between

[0088] Using the formula Determine feature x i The average mutual information of

[0089] Using the formula Determine feature x i Score

[0090] Using the formula Determine the importance score of each dimension feature

[0091] Among them, x i represents the i-th dimension feature, x j Represents the j-th dimension feature S B (i) and S w (i) Represent the between-class variance and within-class variance of the i-th dimension feature, respectively. represents the average value of the i-th dimension feature in the M-th class, H(x i ) represents the feature x i The entropy, H(x j ) represents the feature x j The entropy, H(x i ,x j ) represents the feature x i and feature x j The joint entropy of n represents the feature dimension.

[0092] S105, according to the importance of each feature, a multi-objective gray wolf algorithm is used to determine a group of non-dominated solutions; the gray wolf individuals corresponding to the non-dominated solutions in the group of non-dominated solutions are the feature subset set.

[0093] S105 specifically includes:

[0094] S501, initialize the gray wolf population using the importance of each feature.

[0095] S501 specifically includes:

[0096] According to the formula Select the feature. If the value is greater than a randomly given value, the feature is selected and recorded as "1"; otherwise, it is not selected and recorded as "0".

[0097] S502, calculating the target value of each individual in the gray wolf population, and comparing the dominance relationships between individuals to determine a set of initial non-dominated solutions.

[0098] S502 specifically includes:

[0099] Calculate the ratio f of the number of selected features in each individual of the gray wolf population to the total number of features obj_1 And use the support vector machine algorithm to calculate the inverse of the AUC value of each individual in the gray wolf population obj_2 .

[0100] Comparison of individual f in different gray wolf populations obj_1 and fobj_2 When the size of When , individual a is a non-dominated solution; where S, R = {1, 2}, a set of initial non-dominated solutions is obtained.

[0101] S503: Update the population using a population update strategy to determine a set of non-dominated solutions.

[0102] S503 specifically includes:

[0103] Set the maximum number of iterations maxiter, define the storage unit capacity num_archive, and set the objective function f obj_1 and the objective function f obj_2 .

[0104] Use fuzzy selection operator to select three leading wolves, namely α wolf, β wolf and δ wolf; define a function u c K :

[0105]

[0106] where f c max 、f c min 、f c K They represent the maximum value, minimum value and target value of the K-th solution of the c-th objective respectively. The membership value of the K-th solution of the objective function can be expressed as:

[0107]

[0108] Among them, N F Indicates the number of objective functions, N P is the number of non-dominated solutions. Rank the membership values of each non-dominated solution and select the top three solutions as the leader wolves. When performing fuzzy selection operations, there may be situations where two or more solutions have the same membership value. In this case, feature importance is used for evaluation. The calculation formula is as follows:

[0109] f sig_sum =sum(f sig_#nsfeat ).

[0110] Where, f sig_#nsfeat Indicates the importance of the selected features, f sig_sum The larger the number, the higher the ranking.

[0111] Binarize the wolf pack; and use the following formula to binarize the wolf pack:

[0112]

[0113] Using the formula Determine the adaptive mutation operator to avoid the algorithm from falling into local optimality and premature maturation when searching for the optimal features for emotion recognition. To solve this problem, a random approach is used to ensure that at least one element of each solution is selected.

[0114] Among them, Thr is the adaptive mutation operator, iter is the current iteration number;

[0115] Add each non-dominated solution obtained to the storage unit. When a new non-dominated solution appears, compare the relationship between the new non-dominated solution and the non-dominated solution in the storage unit and update the storage unit.

[0116] When the number of iterations is reached, the update is stopped and a set of non-dominated solutions is finally obtained.

[0117] During the iteration process, when a new non-dominated solution enters the archive unit, it is necessary to compare the relationship between the non-dominated solution in the storage unit and the new non-dominated solution. At this time, several different situations will occur:

[0118] S51: When the new non-dominant can be dominated by any member in the storage unit, the new non-dominant is rejected as a member of the archive unit.

[0119] S52, when the new non-dominated solution dominates the solution in the archive unit, the solution in the archive unit should be removed and the new non-dominated solution should be entered into the archive unit.

[0120] S53, when the new non-dominated solution and the solution in the archive unit do not dominate each other, the new non-dominated solution should be entered into the archive unit.

[0121] S106, determining the classification of the original EEG signal based on the feature subset set to achieve emotion recognition.

[0122] The method first preprocesses the raw EEG signals to obtain preprocessed EEG signals. It then extracts different features of the EEG signals and calculates the Fisher scores and mutual information values of the features. The scores are then combined to initialize the population. The target value of each individual in the population is calculated, and their dominance relationships are compared to determine non-dominated solutions. A population update strategy is then implemented, including comparing the objective functions of different individuals using a fuzzy selection operator and feature importance. Three leading wolves are then selected, and an adaptive mutation operator is used to escape the local optimum. The algorithm is continuously updated until the maximum number of iterations is reached, ultimately obtaining a set of non-dominated solutions. The gray wolf individuals corresponding to the non-dominated solutions constitute the feature subset set.

[0123] The following uses the above-mentioned specific implementation of multi-target gray wolf EEG feature selection for different emotional music stimuli as an example to illustrate the superiority of the technical solution provided by the present invention. In the specific application process, changes to the numerical values are all within the scope of protection of the present invention.

[0124] like Figure 3 As shown in Figure 2, the multi-target EEG feature selection process is as follows:

[0125] S2-1: Preprocessing of raw EEG signals from two emotional music EEG datasets (happy and sad). A 50Hz notch filter was used to remove power line interference. Signals between 0.5Hz and 45Hz were retained using high-pass and low-pass filters. Independent component analysis was used to remove electrooculogram (EOG) artifacts.

[0126] S2-2: Extract the alpha frequency band of the preprocessed EEG signal data, then perform feature extraction and normalization to form the original feature set. Window the EEG signal for segmentation with a 2-second window length and a 50% overlap. Average the extracted EEG features to form feature set X, which contains 78 features. Normalize feature set X using the z-score to form the original feature set X'.

[0127] S2-3: Calculate the Fisher score and mutual information value for each feature in the standardized feature set and multiply them together to obtain the importance of each feature. Initialize the gray wolf population based on the importance of each feature.

[0128] S2-4: Calculate the ratio f of the number of selected features to the total number of features in each individual of the gray wolf population obj_1 And use the support vector machine algorithm to calculate the inverse of the AUC value of each individual in the gray wolf population obj_2 , compare the dominance relationship between them and obtain the initial dominance solution.

[0129] S2-5: Update the population using a population update strategy. Fuzzy selection operators and feature importance are used to compare the objective functions of different individuals, selecting three leading wolves. Adaptive mutation operators are then used to escape local optima. The population is continuously updated until the maximum number of iterations is reached, ultimately obtaining a set of non-dominated solutions. The gray wolf individuals corresponding to these non-dominated solutions constitute the feature subset set.

[0130] Figure 4 This is the population update process, the specific operations are:

[0131] S3-1, using fuzzy selection operator and feature importance to compare the objective functions of different individuals, screen out three leader wolves and update their positions.

[0132] S3-2, using the positions of the three leading wolves, updates the positions of the non-leader wolves. The positions of the wolf pack are updated through the adaptive mutation operator.

[0133] S3-3, recalculate the target values of the updated wolf pack individuals, compare the dominance relationship of the target values between them, and update the non-dominated solution.

[0134] S3-4, determine whether the maximum number of iterations has been reached. If so, output the final non-dominated solution set; otherwise, repeat steps S3-1, S3-2, and S3-3 until the maximum number of iterations has been reached.

[0135] The method MGWFS_DEM of the present invention is compared with another multi-objective feature selection algorithm BMOPSO, and the average AUC, average feature dimension and average duration results are shown in Table 1.

[0136] Table 1

[0137]

[0138] As shown in Table 1, the MGWFS_DEM method achieved an average AUC of 0.646, an average feature dimension of 3.682 (with an initial feature dimension of 78), and an average search time of 777.814 seconds in the alpha band of the EEG signal. In contrast, the BMOPSO method achieved an average AUC of 0.565, an average feature dimension of 18.556, and an average search time of 1671.734 seconds. These results demonstrate that the MGWFS_DEM method outperforms the BMOPSO method, effectively reducing feature dimensionality while maintaining good search efficiency and reducing search time.

[0139] In response to the above-mentioned method, the present invention further provides an EEG signal emotion recognition system, comprising:

[0140] The original EEG signal acquisition module is used to obtain the original EEG signals under different emotional music stimulation;

[0141] The original EEG signal preprocessing module is used to preprocess the original EEG signal;

[0142] The original feature set determination module is used to extract features and perform standardization on the preprocessed EEG signals to determine the original feature set;

[0143] The feature importance determination module is used to determine the importance of each feature in the original feature set;

[0144] The non-dominated solution determination module is used to determine a set of non-dominated solutions using a multi-objective gray wolf algorithm based on the importance of each feature; the gray wolf individuals corresponding to the non-dominated solutions in the set of non-dominated solutions are the feature subset set;

[0145] The recognition module is used to determine the classification of the original EEG signal based on the feature subset set to achieve emotion recognition.

[0146] In order to execute the methods corresponding to the above embodiments to achieve corresponding functions and technical effects, the present invention also provides an EEG signal emotion recognition device, including: at least one processor, at least one memory, and computer program instructions stored in the memory, and when the computer program instructions are executed by the processor, the described method is implemented.

[0147] The memory is a computer-readable storage medium.

[0148] Based on the above description, the technical solution of the present invention, or the portion that contributes to the prior art, or the portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes instructions for causing a computer device (such as a personal computer, server, or network device) to execute all or part of the steps of the various embodiments of the method of the present invention. The aforementioned computer storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory, a random access memory, a magnetic disk, or an optical disk.

[0149] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0150] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A method for recognizing emotions from EEG signals, characterized in that: include: Obtain original EEG signals stimulated by different emotional music; Preprocess the raw EEG signals; Perform feature extraction and standardization on the preprocessed EEG signals to determine the original feature set; Determine the importance of each feature in the original feature set; According to the importance of each feature, a multi-objective gray wolf algorithm is used to determine a set of non-dominated solutions; the gray wolf individuals corresponding to the non-dominated solutions in the set of non-dominated solutions are the feature subset set; Determine the classification of the original EEG signal based on the feature subset set to achieve emotion recognition; Determining the importance of each feature in the original feature set specifically includes: Determine the total number of categories and the number of samples per category in the original feature set; Use formula formula Calculate the Fisher score f for each dimension feature fisher_score (i); Using the formula Calculate feature x i and x j The mutual information value between Using the formula Determine feature x i The average mutual information of Using formula f score_xi =f MI_xi ·f fisher_score (i) Determine the feature x i Score Using the formula Determine the importance score of each dimension feature Among them, x i represents the i-th dimension feature, x j Represents the j-th dimension feature S B (i) and S w (i) Represents the inter-class variance and intra-class variance of the i-th dimension feature, H(x i ) represents the feature x i The entropy, H(x j ) represents the feature x j The entropy, H(x i ,x j ) represents the feature x i and feature x j The joint entropy of n represents the feature dimension; According to the importance of each feature, a multi-objective grey wolf algorithm is used to determine a set of non-dominated solutions, specifically including: Initialize the gray wolf population using the importance of each feature; Calculate the target value of each individual in the gray wolf population, compare the dominance relationships between individuals, and determine a set of initial non-dominated solutions; The population is updated using a population update strategy to determine a set of non-dominated solutions.

2. The method for recognizing emotions from an EEG signal according to claim 1, wherein: The preprocessing of the original EEG signal specifically includes: Filter the original EEG signal; Perform artifact removal on the original EEG signal after filtering.

3. The method for recognizing emotions from an EEG signal according to claim 1, wherein: The feature extraction and standardization of the pre-processed EEG signal to determine the original feature set specifically includes: Determine the EEG signals corresponding to specific frequency bands related to emotions in emotional music, extract the features of the EEG signals, and determine the feature set; The feature set is standardized using the z-score to determine the original feature set.

4. The method for recognizing emotions from an EEG signal according to claim 1, wherein: The target value of each individual in the gray wolf population is calculated, and the dominance relationship between individuals is compared to determine a set of initial non-dominated solutions, specifically including: Calculate the ratio f of the number of selected features in each individual of the gray wolf population to the total number of features obj_1 And use the support vector machine algorithm to calculate the inverse of the AUC value of each individual in the gray wolf population obj_2 ; Comparison of individual f in different gray wolf populations obj_1 and f obj_2 The size of When , individual a is a non-dominated solution; where S, R = {1, 2}, a set of initial non-dominated solutions is obtained.

5. The method for recognizing emotions from an EEG signal according to claim 1, wherein: The updating of the population using the population updating strategy to determine a set of non-dominated solutions specifically includes: Set the maximum number of iterations maxiter, define the storage unit capacity num_archive, and set the objective function f obj_1 and the objective function f obj_2 ; Use fuzzy selection operator to select three leading wolves, namely α wolf, β wolf and δ wolf; Binarize the wolf pack; Using the formula Determine the adaptive mutation operator; Among them, Thr is the adaptive mutation operator, iter is the current iteration number; Add each non-dominated solution obtained to the storage unit. When a new non-dominated solution appears, compare the relationship between the new non-dominated solution and the non-dominated solution in the storage unit and update the storage unit. When the number of iterations is reached, the update is stopped and a set of non-dominated solutions is finally obtained.

6. An EEG signal emotion recognition system, used to implement the EEG signal emotion recognition method according to any one of claims 1 to 5, characterized in that: include: The original EEG signal acquisition module is used to obtain the original EEG signals under different emotional music stimulation; The original EEG signal preprocessing module is used to preprocess the original EEG signal; The original feature set determination module is used to extract features and perform standardization on the preprocessed EEG signals to determine the original feature set; The feature importance determination module is used to determine the importance of each feature in the original feature set; The non-dominated solution determination module is used to determine a set of non-dominated solutions using a multi-objective gray wolf algorithm based on the importance of each feature; the gray wolf individuals corresponding to the non-dominated solutions in the set of non-dominated solutions are the feature subset set; The recognition module is used to determine the classification of the original EEG signal based on the feature subset set to achieve emotion recognition.

7. An EEG signal emotion recognition device, characterized in that: include: At least one processor, at least one memory, and computer program instructions stored in the memory, when the computer program instructions are executed by the processor, implement the EEG signal emotion recognition method according to any one of claims 1 to 5.

8. The EEG signal emotion recognition device according to claim 7, characterized in that: The memory is a computer-readable storage medium.

Images