A cross-domain electroencephalogram emotion recognition method for progressive target subject sample excavation

Abstract

The application discloses a cross-domain electroencephalogram emotion recognition method for progressive target subject sample excavation, which comprises the following steps: firstly, collecting electroencephalogram data under different emotion induction scenes, extracting features of different channels and frequency bands after preprocessing, and generating source domain and target domain data; secondly, aligning the marginal distribution and conditional distribution of the source domain and the target domain, determining the target domain characteristic label matrix based on the similarity, and obtaining a sample confidence weight matrix; then, measuring the error of the sample and the characteristic label by using cross entropy, combining the sample confidence weight matrix and the marginal distribution and conditional distribution alignment error, and constructing a target function; finally, performing joint iterative optimization according to the target function, and judging the emotion category of the measured person during the electroencephalogram data collection according to the target domain characteristic label matrix after the iterative optimization. The application effectively optimizes the overall performance of the model and ensures the global optimality of the result, and more accurate electroencephalogram emotion recognition is obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electroencephalogram signal processing, and particularly to a cross-domain electroencephalogram emotion recognition method for progressive target subject sample mining, a device and a storage medium thereof. BACKGROUND

[0002] Emotion is a complex psychological state that plays a crucial role in interpersonal communication. It is a natural response of an individual to external stimuli or internal psychological activities, usually generated spontaneously in an unconscious manner rather than actively controlled by will. The occurrence of emotion involves physiological activities of the central nervous system and the peripheral nervous system, reflecting the complexity of multiple levels. Generally, emotional state can be presented through external behaviors such as subjective experience, language expression, facial expression, and body movement. However, these external manifestations are easily influenced by cultural background, individual intention or social environment, and may be deliberately concealed or disguised, leading to uncertainty and inaccuracy in emotion recognition.

[0003] In contrast, electroencephalogram (EEG) as a non-invasive neuroimaging technology can directly capture the neural activity of the brain, providing more subtle and accurate physiological signals for objective assessment of emotional state. EEG signals are spontaneously generated by the brain during activity, with high temporal resolution, low cost, high reliability, and relatively high emotion recognition effect. Therefore, by analyzing EEG signals under emotional state, emotional changes can be effectively recognized, thereby playing an important role in mental health monitoring and emotional abnormality warning. In addition, in human-computer interaction systems, emotion recognition technology based on EEG can enable machines to more accurately understand users' emotional needs and preferences, thereby providing a more personalized and adaptive interaction experience.

[0004] Currently, the common semi-supervised analysis method mainly has the following five defects:

[0005] 1. The common semi-supervised learning method may have the risk of propagating errors in the initial stage, leading the model to gradually deviate to the wrong direction.

[0006] 2. Most existing models learn domain-invariant features in sequence and estimate target domain label information. This two-stage strategy breaks the internal connection between the two processes, inevitably leading to suboptimality.

[0007] 3. Different individuals may have significant differences in emotional response to the same event, which reflects the individual differences in emotional experience and expression. This difference directly affects the effectiveness of samples in the emotion classification model. However, most existing electroencephalogram emotion classification models often assign equal importance to all collected sample data, without fully considering the differences in quality and reliability of the collected data.

[0008] 4. In practical applications, the lack of clearly labeled emotional EEG signal data leads to certain limitations in analysis results. Acquiring a large number of clearly labeled emotional EEG signals is often challenging.

[0009] 5. Many sentiment classification methods rely on deep learning techniques, such as convolutional neural networks, recurrent neural networks, and transformer models. Although these models have excellent performance, they are often regarded as "black boxes" due to their complex internal mechanisms, making it difficult to intuitively understand their decision-making process. Summary of the Invention

[0010] In response to the shortcomings of the existing technology, the present invention proposes a cross-domain EEG emotion recognition method for progressive target subject sample mining, which can fully utilize the emotional representation of EEG data. The model is based on cross-domain adaptation with joint optimization of shared subspace dual projection matrix and target label. It constructs feature alignment constraints through the bidirectional projection matrix of the source domain and the target domain, maps the target domain features to the source domain subspace, and generates high-confidence pseudo labels based on the source domain class prototype, thereby alleviating the domain shift problem in the early stages of training; further introduces an adaptive sample weight mechanism, dynamically quantifies the sample transferability according to the confidence of the pseudo label in the iteration, and focuses on high-confidence samples through weighted classification loss to suppress noise interference. Finally, the model integrates subspace alignment loss, weighted classification loss and regularization term to achieve collaborative optimization of projection matrix and classifier. This strategy effectively integrates accurate label estimation and label metric, and can accurately evaluate and utilize the EEG features unique to the target domain to reduce the impact of error propagation in the initial stage, thereby improving the accuracy of the model.

[0011] In order to solve the above technical problems, the technical solution of the present invention is:

[0012] The cross-domain EEG emotion recognition method based on progressive target subject sample mining includes the following steps:

[0013] Step 1: Collect EEG data from multiple subjects in different emotion-inducing scenarios, preprocess the collected EEG data, and extract the features of different channels and frequency bands of the preprocessed EEG data to generate source domain and target domain data.

[0014] Step 2: By aligning the marginal distribution and conditional distribution of the source domain and the target domain, each target domain sample in the subspace is calculated based on the distance similarity with the source domain subspace prototype, and the target domain feature label matrix F is determined based on the similarity. t , and according to F t The sample confidence weight matrix U, which is the sample transferability measure, is calculated using the left truncated Gaussian distribution.

[0015] Step 3, the cross-entropy is used to measure the error between the sample and the characteristic label, and then the target function is constructed by combining the sample confidence weight matrix U and the marginal distribution and conditional distribution alignment error as follows:

[0016]

[0017] wherein, represents a source domain sample set, n s represents the total number of source domain samples, represents a target domain sample set, n t represents the total number of target domain samples, n = n t +n s ; represents the estimated sample amount of the cth class, N s , N t is a diagonal matrix, and the diagonal elements are and are the projection matrices of the source domain and the target domain respectively, k is the dimension of the projection space; C is the number of sample categories, represents the number of samples belonging to the cth class in the source domain, represents the number of samples belonging to the cth class in the target domain; represents a classification matrix, b = [b1, b2, … b n represents a bias vector, represents the label information of the source domain, F t is a target domain characteristic label matrix, sample confidence weight matrix wherein v i represents the confidence weight of the ith sample; α, β and λ all represent adaptive weights, and 1 represents a vector full of 1.

[0018] Step 4, according to the constructed target function, the source domain and target domain projection matrix P s , P t , the classification matrix W and the bias vector b, the target domain characteristic label matrix F t , the sample confidence weight matrix U are jointly iteratively optimized, so as to judge the emotional category of the subject during the collection of electroencephalogram data according to the iteratively optimized target domain characteristic label matrix F t .

[0019] As preferred, since the electroencephalogram data is multi-rhythm and multi-channel, the features extracted from different frequency bands are concatenated to form a feature vector. It is assumed that a brain electrical signal data set containing 64 channels is used, wherein the Fz and Cz electrodes are selected as reference electrodes. All signals are standardized collected according to the international 10-20 system.

[0020] As preferred, the specific process of preprocessing and feature extraction of the electroencephalogram data in step 1 is as follows: after down-sampling, the electroencephalogram data is filtered by a band-pass filter to remove noise and electrooculogram artifacts. The filtered data is divided into five commonly used frequency bands of electroencephalogram, namely Delta (1-4 Hz), Theta (4-8 Hz), Alpha (8-14 Hz), Beta (14-31 Hz) and Gamma (31-50 Hz). On the basis of the data, differential entropy features of different channels and different frequency bands are extracted.

[0021] As preferred, the specific process of constructing the objective function in step 3 is as follows: the Euclidean distance is used to measure the alignment error of the subspaces after mapping of the source domain and the target domain, and the marginal distribution and the conditional distribution of each class are considered to obtain the constructed model

[0022]

[0023] The characteristic label matrix and the sample confidence matrix are obtained by distribution alignment, and the cross-entropy loss of the credible samples and the labels is combined to finally obtain the objective function formula (1).

[0024] As preferred, the method of joint iterative optimization in step 4 is as follows:

[0025] Step 4.1, initializing the projection matrix P of the source domain and the target domain s , the classification matrix W, and the offset vector b. t

[0026] Step 4.2, updating the projection matrix P of the source domain and the target domain by least squares s , the classification matrix W, and the offset vector b. t : fixing the classification matrix W, the offset vector b, and the target domain characteristic label matrix F t , the sample confidence weight matrix U, to obtain the objective function of P s , P t .

[0027]

[0028] Solving formula (3) can update the projection matrix P of the source domain and the target domain s , P t .

[0029] Step 4.3, updating the target domain characteristic label matrix F t : updating the target domain characteristic label matrix F t according to step 2.

[0030] Step 4.4, updating the sample confidence weight matrix U: updating the sample confidence weight matrix U according to step 2. ​

[0031] Step 4.5, updating the classification matrix W by least square: fixing the projection matrix P of the source domain and the target domain s t , the offset vector b, the target domain characteristic label matrix F t , the sample confidence weight matrix U, the objective function of W is

[0032]

[0033] Solving equation (4) can update the classification matrix W.

[0034] Step 4.6, updating the offset vector b by least square: fixing the projection matrix P of the source domain and the target domain s t , the classification matrix W, the target domain characteristic label matrix F t , the sample confidence weight diagonal matrix U, the objective function of b is

[0035]

[0036] Solving equation (5) can update the offset vector b.

[0037] Step 4.7, repeating the iteration of step 4.2, step 4.3, step 4.4, step 4.5, step 4.6 until the objective function converges, finally completing the joint iterative optimization of the projection matrix P s t , the classification matrix W, the offset vector b, the target domain characteristic label matrix F t , the sample confidence weight matrix U.

[0038] As a preferred, the sample confidence weight matrix U is a diagonal matrix, v i represents the confidence weight of the i-th sample.

[0039] The present application has the following characteristics and beneficial effects:

[0040] 1. The present application proposes a gradual target subject sample mining cross-domain electroencephalogram emotion recognition method, which realizes the global improvement of the model performance by jointly optimizing the shared subspace projection matrix and the target pseudo label. The loss of the target label is calculated by balancing the intra-domain and inter-domain estimation, so as to effectively optimize the overall performance of the model and ensure the global optimality of the result.

[0041] ​​​2. The application provides a mechanism for accurately evaluating target domain labels according to aligned subspace class prototypes, specifically, by analyzing the similarity between target domain subspace samples and source domain subspace class prototypes, to generate more accurate pseudo labels. On this basis, combined with the introduction of pseudo label reliability matrix for interactive learning, to improve the overall performance and robustness of the model. On the one hand, accurate estimation of pseudo labels can obtain more correct reliability matrix, on the other hand, obtaining more correct reliability matrix can promote the alignment of subspace class prototypes, so as to obtain more accurate pseudo label estimation. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, below will briefly introduce the drawings needed to be used in the embodiments or prior art description, obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0043] Figure 1 The framework diagram of the cross-domain electroencephalogram emotion recognition method of the progressive target subject sample mining of the present embodiment. DETAILED DESCRIPTION

[0044] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0045] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application will be further described in detail below with the help of the drawings.

[0046] On the contrary, the present application covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present application defined by the claims. Further, in order to make the public have a better understanding of the present application, some specific details are described in detail in the following detailed description of the present application. The present application can also be completely understood without the description of these details by those skilled in the art.

[0047] Embodiment 1

[0048] The present embodiment provides a cross-domain electroencephalogram emotion recognition method of progressive target subject sample mining, as shown in Figure 1 The specific steps include the following steps:

[0049] Step 1, acquisition and preprocessing of emotional electroencephalogram data;

[0050] In the experiment, the subjects need to wear an electroencephalogram cap to record the electroencephalogram signals generated by the subjects under the action of emotional stimuli, and store them as raw data. After the electroencephalogram signal collection is completed, the subjects need to subjectively evaluate their emotional experience, and label the corresponding electroencephalogram signals according to the emotional state perceived by themselves. Then, the emotional electroencephalogram data is preprocessed and feature extraction is performed. In this embodiment, the obtained electroencephalogram data is first down-sampled, and the sampling rate is adjusted to 200 Hz. Then, a band-pass filter is used to process the data, and the frequency range is set to 1 Hz to 50 Hz to reduce the interference of noise and eye movement artifacts. After filtering, the signal is divided into five different frequency bands, including Delta (1-4 Hz), Theta (4-8 Hz), Alpha (8-14 Hz), Beta (14-31 Hz) and Gamma (31-50 Hz). Based on these processed data, the differential entropy features of each channel in different frequency bands are further calculated.

[0051] Step 2, by aligning the marginal distribution and conditional distribution of the source domain and the target domain, distance-based similarity calculation is performed to determine the target domain characteristic label matrix F t , and according to F t , the sample confidence weight matrix U is calculated by using the left truncated Gaussian distribution.

[0052] Step 2.1, respectively obtain the feature projection of the source domain and the target domain to obtain their representations in the common subspace and On this basis, the class center value of the source domain subspace feature Z s is calculated, and these class centers represent the prototypes of each class in the subspace. Next, for each sample X (j) in the target domain, the distance between its subspace feature and each source domain prototype is calculated. In order to quantify this distance relationship and convert it into a similarity score, the reciprocal of each distance value is taken, so that a smaller distance corresponds to a higher similarity score. Based on the above similarity score, a characteristic label matrix F t is constructed, which is used to assign pseudo-labels to each sample in the target domain. Specifically, for each target domain sample X (j) , the membership probability of the jth sample to the cth class is determined according to the similarity score of the cth prototype.

[0053]

[0054] where, is an indicator function that returns 1 if , otherwise returns 0.

[0055] Step 2.2, To further evaluate the confidence of each sample, a method based on left-truncated Gaussian distribution is introduced. Specifically, for each sample According to the feature label matrix F t Obtain the probability of the maximum possible class And use the left-truncated Gaussian distribution F to calculate the confidence v of this sample j That is, the jth element in the diagonal matrix U.

[0056]

[0057] Where erf(x) represents the error function, μ is the average of the probability of the maximum possible class of all samples, and σ is the variance.

[0058] Step 3, Use cross-entropy to measure the error between the sample and the feature label, and then combine the sample confidence weight matrix U and the marginal distribution and conditional distribution alignment error to construct the objective function of the cross-domain EEG emotion recognition model for progressive target subject sample mining.

[0059] Using the method based on the maximum mean difference criterion, align the source data X s and the target data X t Marginal distribution in the latent subspace, as shown in equation (4)

[0060]

[0061] Where P s represents the original data latent subspace, P t represents the target data latent subspace, X s，i represents the i-th source domain sample, and X t，i represents the i-th target domain sample.

[0062] Equivalent to equation (5):

[0063]

[0064] Where, is a full single column vector of length n s (n t ).

[0065] Similarly, the conditional distribution of aligning the source data and the target data can also be achieved by minimizing:

[0066]

[0067] Simplifying can get:

[0068]

[0069] Where is the unknown label indicator matrix of the target domain, which is obtained by distribution alignment and satisfies the non-negativity and row normalization constraints. For example, if F t = [0.11, 0.15, 0.39, 0.35], the probabilities of the ith target sample belonging to the four emotional states are 0.11, 0.15, 0.39, and 0.35, respectively.

[0070] Since the target samples are unlabeled, the real sample size of each class cannot be obtained. Therefore, we use to estimate the real sample size. N s (N t ) is a diagonal matrix, whose diagonal elements are

[0071] By jointly considering the marginal distribution and the conditional distribution, we obtain

[0072]

[0073] where the second term limits the two projection matrices from deviating too much.

[0074] The data represented by the subspace are combined with the label information obtained by the subspace information through the classification matrix W, and the regularization of W is added to prevent the model from overfitting, and we get

[0075]

[0076] On this basis, in order to reduce the risk of propagating errors in the initial stage, the confidence weight of each sample is considered; thus, the model of the progressive target subject sample mining cross-domain electroencephalogram emotion recognition method is obtained

[0077]

[0078] where, denotes the source domain sample set, n s denotes the total number of source domain samples, denotes the target domain sample set, n t denotes the total number of target domain samples, n = n t + n s ; and are the projection matrices of the source domain and the target domain, respectively, k is the dimension of the projection space; c is the number of sample categories, denotes the number of samples belonging to the cth category in the source domain, denotes the number of samples belonging to the cth category in the target domain; denotes the label information of the source domain, denotes the classification matrix, b = [b1, b2, … b nrepresents the offset vector; v i represents the confidence weight of the ith sample; β represents the adaptive weight, and 1 represents a vector full of 1.

[0079] In general, by combining formula (8) and formula (10), the target formula (11) of the cross-domain electroencephalogram emotion recognition model of the progressive target subject sample mining is finally obtained:

[0080]

[0081] Step 4, according to the constructed target function, the source domain and target domain projection matrix P s , P t , the classification matrix W and the offset vector b, the target domain characteristic label matrix F t , the sample confidence weight matrix U are jointly iteratively optimized;

[0082] Step 4.1, before optimizing the target function, the data needs to be initialized. Among them, the sample confidence weight matrix U is initialized to 1; the target domain characteristic label matrix F t is initialized to The source domain and target domain projection matrix P s , P t , the classification matrix W are all initialized according to the principal component analysis; the offset vector b is randomly initialized.

[0083] Step 4.2, update the projection matrix P s , P t of the source domain and the target domain: fix the classification matrix W, the offset vector b, the target domain characteristic label matrix F t , the sample confidence weight matrix U, and obtain the target function of P s , P t .

[0084] For convenience of solution, let M=[I d , -I d ; -I d , I d ], where I d is a dxd unit matrix, N s (N t ) is a diagonal matrix, and the cth diagonal element is Then record Rewrite as λtr(P T TP)+αtr(P T MP), where In the following discussion, it is also the same.

[0085] Let G = UX T , H = U1, J = UY, we have P = (G

[0086]

[0087] The partial derivative of P is taken and set to zero, we have

[0088]

[0089] Simplifying, we have

[0090] (G T G)-1(λT+αM)P+PWW T = (G T G) -1 (G T JW T -G T Hb T W T ). (15)

[0091] which is equivalent to AP+PB=C, where A=(G T G) -1 (λT+αM), B=WW T , C=(G T G) -1 (G T JW T -G T Hb T W T ). Obviously, equation (15) is a standard Sylvester equation, where P is the variable, then εI T is introduced on the diagonal of (G d G) to avoid possible singularities.

[0092] Step 4.3, update the target domain property label matrix F t : update the target domain property label matrix F t according to step 2.1.

[0093] Step 4.4, update the sample confidence weight matrix U: update the sample confidence weight matrix U according to step 2.2.

[0094] Step 4.5, update the classification matrix W: fix the projection matrix P s , P t of the source domain and the target domain, the offset vector b, the target domain property label matrix F t , the sample confidence weight matrix U, the objective function of W is:

[0095]

[0096] Let G = UX T , H = U1, J = UY, equation (16) can be rewritten as:

[0097]

[0098] Equivalently:

[0099] min(tr(W T P T G T +bH T -J T )(GPW+Hb T -J)+βtr(W T W)). (18)

[0100] Taking the partial derivative of W and setting it to 0, we get:

[0101]

[0102] Solving, we get:

[0103] W = (β1+P T G T GP) -1 P T G T (J-Hb T ). (20)

[0104] Step 4.6, update the offset vector b: fix the projection matrix P of the source domain and the target domain s , P t , the classification matrix W, the target domain characteristic label matrix F t , the sample confidence weight diagonal matrix U, get the objective function of b:

[0105]

[0106] Equation (21) can be rewritten as:

[0107]

[0108] Equivalently:

[0109]

[0110] Taking the partial derivative of b and setting it to 0, we get:

[0111]

[0112] Solving, we get:

[0113] b = (H T H)-1 (J-W T P T G T )H. (25)

[0114] Step 4.7, repeat the iteration of Step 4.2, Step 4.3, Step 4.4, Step 4.5, Step 4.6 until the objective function converges, finally complete the projection matrix P of the fixed source domain and target domain s , P t , offset vector B, target domain characteristic label matrix F t , sample confidence weight matrix U, classification matrix W joint iterative optimization, so as to obtain the corresponding prediction value label.

[0115] The benchmark emotional electroencephalogram dataset SEED-IV is used in the experiment. SEED-IV is a brain electrical data set collected by Shanghai Jiaotong University, which is dedicated to affective computing research. The dataset recruited 15 subjects, each of whom participated in 3 experiments. In each experiment, the subjects watched 24 movie clips, 6 clips for each emotional state (happy, angry, sad, and neutral) to evoke specific emotions. During the viewing process, 62-channel EEG equipment was used to record the brain electrical signals with a sampling rate of 1000 Hz, and data preprocessing was performed to remove noise. SEED-IV is widely used in affective computing, brain-computer interface (BCI), and machine learning research, providing high-quality data support for multi-modal emotion recognition.

[0116] Table 1 Cross-subject emotion recognition results of one-to-one transfer paradigm on SEED-IV dataset

[0117]

[0118]

[0119] Through the comparison experiment, it can be known that the result of the embodiment is higher than the semi-supervised method of the popular joint electroencephalogram feature transfer and semi-supervised cross-subject emotion recognition model JTSR in the recognition accuracy of three scenes, the first scene is improved by an average of 7.02%, the second scene is improved by an average of 2.29%, and the third scene is improved by an average of 3.97%.

[0120] In the embodiment, the projection matrix P s , P t , offset vector B, target domain characteristic label matrix F tThe joint iterative optimization of the sample confidence weight matrix U and the classification matrix W, and the final model can effectively distinguish the label of the unlabeled sample; and the weight coefficient quantitatively describing the sample confidence is obtained. In the embodiment, on the one hand, the model combines the domain subspace feature and the target domain label information, avoiding the suboptimal solution problem caused by the separation of the two processes and the destruction of the internal relationship; on the other hand, the model gradually labels the confident target sample by using the aligned subspace class prototype, fuses the accurate label estimation and the label confidence measurement, and effectively reduces the risk of error propagation in the initial stage.

[0121] The embodiments of the application are described in detail above with reference to the drawings, but the application is not limited to the described embodiments. For those skilled in the art, various changes, modifications, replacements and variations of the embodiments including components are made without departing from the principles and spirits of the application, and still fall within the protection scope of the application.

Claims

1. A cross-domain EEG emotion recognition method based on progressive target subject sample mining, characterized by: It includes the following steps: Step 1: Collect EEG data from multiple subjects in different emotion-inducing scenarios, pre-process the EEG data, extract features of different channels and frequency bands of the EEG data, and generate source domain and target domain data; Step 2: By aligning the marginal distribution and conditional distribution of the source domain and the target domain, the target domain feature label matrix F is determined based on the similarity. t , and according to F t Obtain the sample confidence weight matrix U. The specific implementation process is as follows: Step 2.1: Obtain the source domain and target domain respectively for feature projection and obtain their representation in the common subspace and Calculate the source domain subspace feature Z s The class center value; for each sample X in the target domain (j) , calculate its subspace features The distance between each source domain prototype is calculated by taking the inverse of each distance value to obtain a similarity score; based on the similarity score, a feature label matrix F is constructed. t , used to assign pseudo labels to each sample in the target domain, for each target domain sample X (j) , determine the membership probability of the jth sample to the cth class based on its similarity score with the cth class prototype in, is an indicator function, which means if If yes, it returns 1, otherwise it returns 0; Step 2.2: For each sample According to F t Get the probability of its maximum possible class And use the left truncated Gaussian distribution F to calculate the sample confidence v j , the sample confidence weight matrix U is a diagonal matrix, v j That is, the j-th element in the diagonal matrix U: erf(x) represents the error function, μ is the average probability of the maximum possible class of all samples, and σ is the variance; Step 3: Use cross entropy to measure the error between the sample and the feature label, and construct the objective function by combining the sample confidence weight matrix, marginal distribution, and conditional distribution alignment error. Step 4: Based on the objective function, perform joint iterative optimization on the source and target domain projection matrices, classification matrices and offset vectors, target domain feature label matrix, and sample confidence weight matrix. Determine the emotional category of the subject during EEG data collection based on the iteratively optimized target domain feature label matrix.

2. The cross-domain EEG emotion recognition method for progressive target subject sample mining according to claim 1 is characterized in that: The specific process of preprocessing and feature extraction of EEG data in step 1 is as follows: downsampling the EEG data and filtering out noise and electrooculogram artifacts through a bandpass filter; dividing the filtered data into five EEG frequency bands, namely Delta: 1-4Hz, Theta: 4-8Hz, Alpha: 8-14Hz, Beta: 14-31Hz, and Gamma: 31-50Hz, and extracting differential entropy features of different channels and frequency bands based on this data.

3. The cross-domain EEG emotion recognition method for progressive target subject sample mining according to claim 2 is characterized in that: The objective function is as follows: in, represents the source domain sample set, n s represents the total number of source domain samples, represents the target domain sample set, n t Represents the total number of target domain samples, n=n t +n s ; Indicates the estimated sample size of class c, N s 、N t is a diagonal matrix, The diagonal elements are and are the projection matrices of the source domain and target domain respectively, k is the dimension of the projection space; C is the number of sample categories, represents the number of samples belonging to the cth class in the source domain, Indicates the number of samples belonging to the cth class in the target domain; Represents the classification matrix, b=[b1,b2,…b n ] represents the offset vector, Indicates the label information of the source domain, F t is the target domain feature label matrix, Sample confidence weight matrix where v i Represents the confidence weight of the i-th sample; α, β, and λ are all expressed as adaptive weights, and 1 represents a vector of all 1s.

4. The cross-domain EEG emotion recognition method for progressive target subject sample mining according to claim 3 is characterized in that: The specific process of constructing the objective function is: using Euclidean distance to measure the alignment error of the subspaces after the source domain and the target domain are mapped respectively, and considering the marginal distribution and conditional distribution of each class to obtain the constructed model: The feature label matrix and sample confidence matrix are obtained through distribution alignment, and the objective function is finally obtained by combining the cross entropy loss of the reliable samples and labels.

5. The cross-domain EEG emotion recognition method for progressive target subject sample mining according to claim 4 is characterized in that: The method of joint iterative optimization in step 4 is: Step 4.

1. Initialize the projection matrix P of the source domain and target domain s and P t , classification matrix W, offset vector b; Step 4.2: Update the projection matrix P of the source and target domains by least squares s 、P t : Fixed classification matrix W, offset vector b, target domain feature label matrix F t , sample confidence weight matrix U, get P s 、P t The objective function is: Solve the above equation to update the projection matrix P of the source domain and target domain s 、P t ; Step 4.3: Update the target domain feature label matrix F according to step 2 t ; Step 4.4: Update the sample confidence weight matrix U according to step 2; Step 4.5: Update the classification matrix W by least squares: Fix the projection matrix P of the source domain and target domain s and P t , offset vector b, target domain feature label matrix F t , the sample confidence weight matrix U, and the objective function of W is: Update the classification matrix W by solving the above formula; Step 4.

6. Update the offset vector b by least squares: Fix the projection matrix P of the source domain and the target domain s and P t , classification matrix W, target domain feature label matrix F t , the sample confidence weight diagonal matrix U, the objective function of b is: Calculate the above formula to update the offset vector b; Step 4.7, repeat steps 4.2, 4.3, 4.4, 4.5, and 4.6 until the objective function converges, and finally complete the projection matrix P of the source domain and the target domain s and P t , classification matrix W, offset vector b, target domain feature label matrix F t , joint iterative optimization of the sample confidence weight matrix U.

Images