Cross-subject eeg emotion recognition method and system based on adaptive fuzzy domain adversarial

Abstract

The application discloses a cross-subject electroencephalogram emotion recognition method and system based on adaptive fuzzy domain confrontation. The method comprises the following steps: obtaining electroencephalogram signal features of a target domain to be recognized; inputting the electroencephalogram signal features of the target domain to be recognized into adaptive fuzzy domain confrontation networks in each source domain network branch which has been trained, mapping electroencephalogram signal feature samples into fuzzy member vectors through a fuzzy encoder, generating corresponding fuzzy feature embeddings, and then classifying the fuzzy feature embeddings through a task classifier; and obtaining a final emotion classification result according to emotion classification results of each source domain network branch. The application aims to depict the uncertainty of emotions by introducing a fuzzy logic system, so as to improve the robustness, generalization ability and recognition accuracy of an electroencephalogram emotion recognition model in a cross-subject scenario.

Description

Technical Field

[0001] This invention relates to the field of EEG signal detection and recognition technology, specifically to a cross-subject EEG emotion recognition method and system based on adaptive fuzzy domain adversarial methods. Background Technology

[0002] In recent years, significant progress has been made in feature extraction and classification based on electroencephalography (EEG). However, there are significant individual differences in EEG signals, and the distribution of EEG data among different subjects is inconsistent. Therefore, cross-individual emotion recognition still faces challenges in the field of emotion recognition. In order to develop cross-individual emotion recognition models with strong generalization performance, domain adversarial algorithms have become the mainstream research direction in recent years. In this method, the EEG data of different subjects are regarded as different domains, the training subject data is regarded as the source domain, and the test subject data is the target domain. It conducts adversarial learning by designing adversarial discriminators and task classifiers, with the goal of reducing domain-related features, thereby improving the generalization ability of the model. However, existing methods have two main limitations: (1) they ignore the inherent fuzziness and uncertainty of emotional states themselves. There are often continuous and overlapping transition areas between emotion categories, which are difficult to accurately characterize using traditional hard classification; (2) the adversarial intensity in the domain adversarial training process is usually set statically or heuristically, and cannot be dynamically adjusted according to the real-time similarity between the source domain and the target domain, resulting in poor domain alignment or unstable training. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a cross-subject EEG emotion recognition method and system based on adaptive fuzzy domain adversarial technology, which aims to improve the robustness, generalization ability and recognition accuracy of the EEG emotion recognition model in cross-subject scenarios by introducing a fuzzy logic system to characterize the uncertainty of emotions.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0005] A cross-subject EEG emotion recognition method based on adaptive fuzzy domain adversarial approaches includes the following steps:

[0006] S101, acquire the EEG signal features of the target domain to be identified;

[0007] S102, the EEG signal features of the target domain to be identified are input into the adaptive fuzzy domain adversarial networks in each trained source domain network branch. A fuzzy encoder maps the EEG signal feature samples into fuzzy member vectors and generates corresponding fuzzy feature embeddings. A task classifier then uses these fuzzy feature embeddings for emotion classification. The loss function used during training of the adaptive fuzzy domain adversarial network includes an adaptively weighted domain adversarial loss. ,in For source domain weights, For domain adversarial losses, source domain weights To measure the fuzzy domain similarity between the source and target domains The adjusted fuzzy domain similarity measure between the source and target domains is obtained. This includes the similarity between the fuzzy member vectors of the source and target domains and the fuzzy feature embeddings of the two.

[0008] S103, obtain the final emotion classification result based on the emotion classification results of each source domain network branch.

[0009] Optionally, the fuzzy encoder includes a membership network and a deep encoding module. The step of mapping EEG signal feature samples to fuzzy membership vectors and generating corresponding fuzzy feature embeddings via the fuzzy encoder includes: firstly, mapping the target domain EEG signal features to be identified to fuzzy membership vectors of the target domain using the membership network. ,in ~ This indicates that the input EEG signal feature sample belongs to the 1st to 2nd generation. Membership degree of a fuzzy sentiment category The preset number of fuzzy sets is used, and the fuzzy member vectors of the target domain are used. The input EEG signal feature samples are weighted and nonlinearly transformed according to the following formula to generate enhanced fuzzy features:

[0010] ;

[0011] in, To enhance the fuzzy features, For the input EEG signal feature samples, This represents the membership degree of the input EEG signal feature sample to the i-th fuzzy emotion category; then, the deep encoding module generates the fuzzy feature embedding of the target domain EEG signal features to be identified based on the enhanced fuzzy features. .

[0012] Optionally, the membership network consists of two fully connected layers and a Softmax function, and the deep coding module consists of a multilayer perceptron (MLP). The MLP contains three fully connected layers and an output layer, and the layers of the MLP use the ReLU activation function.

[0013] Optionally, the construction and training of the source domain network branch includes:

[0014] S201, Obtain the target domain EEG signal feature samples and the corresponding source domain source domain EEG signal feature samples;

[0015] S202, Construct an adaptive fuzzy domain adversarial network for this source domain network branch. The adaptive fuzzy domain adversarial network includes a fuzzy encoder, a task classifier, and a domain discriminator. The fuzzy encoder is used to map EEG signal feature samples into fuzzy member vectors and generate corresponding fuzzy feature embeddings, including obtaining fuzzy member vectors of EEG signal feature samples in the target domain. and fuzzy feature embedding , and fuzzy membership vectors of source domain EEG signal feature samples and fuzzy feature embedding The task classifier is used to classify emotions based on fuzzy feature embeddings, and the domain discriminator is used to determine the domain group between target domain EEG signal feature samples and corresponding source domain source domain EEG signal feature samples.

[0016] S203, use the EEG signal feature samples of the target domain and the source domain corresponding to the source domain network branch to train the adaptive fuzzy domain adversarial network of the source domain network branch.

[0017] Optionally, the fuzzy domain similarity measure between the source domain and the target domain The expression for the computation function is:

[0018] ;

[0019] in, Fuzzy feature embedding for source domain EEG signal feature samples Fuzzy feature embedding for target domain EEG signal feature samples for L2 norm, for L2 norm, Let be the fuzzy member vector of the source domain. For the fuzzy member vector of the target domain, for L2 norm, for The L2 norm.

[0020] Optionally, based on the fuzzy domain similarity measure between the source and target domains. Adjust source domain weights The function expression is:

[0021] ;

[0022] in, e This represents the current training progress percentage, which is the ratio of the current training round to the total number of training rounds. It is a piecewise coefficient function related to training progress, used to assign different coefficients at different training progress.

[0023] Optionally, the expression for the domain adversarial loss is:

[0024] ;

[0025] in, For domain confrontation losses, Represents the expected value in mathematics. Type labels for the EEG signal domain. Representation domain discriminator, For splicing operations, Fuzzy feature embedding for source domain EEG signal feature samples Fuzzy feature embedding for EEG signal feature samples in the target domain.

[0026] Optionally, the expression for the loss function used during training is:

[0027] ;

[0028] ;

[0029] in, The loss function used during training. For classifying losses, The adaptively weighted domain adversarial loss, For source domain weights, For domain confrontation losses, For the number of samples, Task labels for EEG signal feature samples Represents a task classifier. Fuzzy feature embedding for source domain EEG signal feature samples Fuzzy feature embedding for EEG signal feature samples in the target domain.

[0030] Optionally, obtaining the final emotion classification result based on the emotion classification results of each source domain network branch in step S103 includes:

[0031] S301, the fuzzy domain similarity between the source domain and the target domain of each source domain network branch is measured according to the following formula. Normalization is performed to obtain the integrated weights:

[0032] ;

[0033] in, For the first Integration weights of each source domain network branch For the first A fuzzy domain similarity measure between the source domain and the target domain of each source domain network branch; This is a temperature parameter used to adjust the smoothness of the weight distribution; This represents the total number of branches in the source domain network.

[0034] S302, the predicted probability distributions obtained for each source domain network branch are weighted and voted according to the following formula:

[0035] ;

[0036] in, For the final predicted probability distribution, For the first The predicted probability distribution of each source domain network branch; from the final predicted probability distribution The emotion category with the highest predicted probability is selected as the final emotion classification result.

[0037] The present invention also provides a cross-subject EEG emotion recognition system based on adaptive fuzzy domain adversarial methods, comprising a microprocessor and a memory interconnected thereto, wherein the microprocessor is programmed or configured to execute the cross-subject EEG emotion recognition method based on adaptive fuzzy domain adversarial methods.

[0038] Compared with the prior art, the present invention can mainly achieve the following beneficial effects:

[0039] 1. This invention involves inputting the EEG signal features of the target domain to be identified into an adaptive fuzzy domain adversarial network in each pre-trained source domain network branch. The EEG signal feature samples are mapped into fuzzy member vectors by a fuzzy encoder, and corresponding fuzzy feature embeddings are generated. A fuzzy logic system is introduced through the fuzzy encoder. The continuity and fuzziness of emotional states are systematically modeled in the domain adversarial framework, making the feature representation more consistent with the physiological nature of emotional response and significantly improving the psychological meaning and robustness of the model.

[0040] 2. This invention includes inputting the target domain EEG signal features to be identified into an adaptive fuzzy domain adversarial network in each pre-trained source domain network branch. The loss function used during training of the adaptive fuzzy domain adversarial network includes an adaptively weighted domain adversarial loss. ,in For source domain weights, For domain adversarial losses, source domain weights To measure the fuzzy domain similarity between the source and target domains The adjusted fuzzy domain similarity measure between the source and target domains is obtained. This includes the similarity between the fuzzy member vectors of the source and target domains and the fuzzy feature embeddings of the two. The adaptive fuzzy domain adversarial network adopts an adaptive adversarial weight mechanism based on fuzzy similarity, which can dynamically adjust the adversarial strength according to the training progress and domain alignment status. This solves the problem of training instability or insufficient alignment caused by fixed adversarial strength in traditional methods, and improves the efficiency and effectiveness of the domain adaptation process.

[0041] 3. This invention includes mapping EEG signal feature samples to fuzzy member vectors and generating corresponding fuzzy feature embeddings through a fuzzy encoder, realizing fuzzy rule integration, further enhancing the model's generalization ability and training stability across subject scenarios, and achieving leading recognition performance on public datasets. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the inference process of the adaptive fuzzy domain adversarial network in an embodiment of the present invention.

[0043] Figure 2 This is a schematic diagram of the training process of the adaptive fuzzy domain adversarial network in an embodiment of the present invention.

[0044] Figure 3 This is a schematic diagram of the structure and process of the adaptive fuzzy domain adversarial network in an embodiment of the present invention. Detailed Implementation

[0045] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be further described in detail below with reference to the accompanying drawings in the embodiments of the present invention.

[0046] like Figure 1 As shown, the cross-subject EEG emotion recognition method based on adaptive fuzzy domain adversarial approaches in this embodiment includes the following steps:

[0047] S101, acquire the EEG signal features of the target domain to be identified;

[0048] S102, the EEG signal features of the target domain to be identified are input into the adaptive fuzzy domain adversarial network in each of the trained source domain network branches. The EEG signal feature samples are mapped into fuzzy member vectors by the fuzzy encoder and the corresponding fuzzy feature embeddings are generated. Then, the task classifier is used to classify the fuzzy feature embeddings for emotion.

[0049] S103, obtain the final emotion classification result based on the emotion classification results of each source domain network branch.

[0050] When subjects perform an emotion-induced task, the cerebral cortex generates corresponding electrical activity in response to emotional stimulation. This embodiment selects electroencephalogram (EEG) signals, an objective physiological signal, to achieve emotion recognition, effectively improving the accuracy and generalization ability of emotion recognition. It should be noted that the required EEG signal features can be selected as needed. For example, in this embodiment, the acquisition of EEG signal features includes: denoising the acquired multi-channel EEG signals; dividing the denoised EEG signals into 1-second time windows with no overlap; and performing frequency band decomposition on the EEG signals in each time window to extract differential entropy features from five frequency bands: δ (1-3Hz), θ (4-7Hz), α (8-13Hz), β (14-30Hz), and γ (31-50Hz). It should be noted that the calculation method for the differential entropy features is a known existing method, therefore its implementation details will not be elaborated further.

[0051] EEG signal features can be denoised using feasible methods as needed. For example, as an optional implementation, end-to-end denoising can include: downsampling the original EEG signal to 200Hz to reduce the amount of data and improve the efficiency of subsequent processing; performing baseline correction on the downsampled EEG signal, specifically by removing the channel mean, that is, subtracting the mean of the signal of each channel to eliminate the effects of DC offset and low-frequency drift; and applying a 0.5-50Hz bandpass filter to the baseline-corrected EEG signal to remove noise interference, thus obtaining the denoised EEG signal.

[0052] like Figure 2 As shown, the fuzzy encoder in this embodiment includes a membership network and a deep coding module. The step of mapping EEG signal feature samples to fuzzy membership vectors and generating corresponding fuzzy feature embeddings through the fuzzy encoder includes: firstly, mapping the target domain EEG signal features to be identified to fuzzy membership vectors of the target domain through the membership network. :

[0053] ;

[0054] in, ~ This indicates that the input EEG signal feature sample belongs to the 1st to 2nd generation. Membership degree of a fuzzy sentiment category The preset number of fuzzy sets is used, and the fuzzy member vectors of the target domain are used. The input EEG signal feature samples are weighted and nonlinearly transformed according to the following formula to generate enhanced fuzzy features:

[0055] ;

[0056] in, To enhance the fuzzy features, For the input EEG signal feature samples, This represents the membership degree of the input EEG signal feature sample to the i-th fuzzy emotion category; then, the deep encoding module generates the fuzzy feature embedding of the target domain EEG signal features to be identified based on the enhanced fuzzy features. .

[0057] As an optional implementation, the membership network consists of two fully connected layers and a Softmax function. The deep coding module is composed of a multilayer perceptron (MLP), which contains three fully connected layers and one output layer. The inter-layer activation function of the MLP is ReLU. Figure 2 As shown, the task classifier consists of a multilayer perceptron (MLP), containing multiple fully connected layers and a softmax output layer, which receives the fuzzy feature embeddings from the fuzzy encoder. Through forward propagation calculation, the Softmax output layer finally generates the probability distribution of samples belonging to each specific emotion category.

[0058] like Figure 3 As shown, the construction and training of the source domain network branch in this embodiment includes:

[0059] S201, Obtain target domain EEG signal feature samples and corresponding source domain. Source domain EEG signal feature samples;

[0060] S202, Construct an adaptive fuzzy domain adversarial network for this source domain network branch. The adaptive fuzzy domain adversarial network includes a fuzzy encoder, a task classifier, and a domain discriminator. The fuzzy encoder is used to map EEG signal feature samples into fuzzy member vectors and generate corresponding fuzzy feature embeddings, including obtaining fuzzy member vectors of EEG signal feature samples in the target domain. and fuzzy feature embedding , and fuzzy membership vectors of source domain EEG signal feature samples and fuzzy feature embedding The task classifier is used to classify emotions based on fuzzy feature embeddings, and the domain discriminator is used to determine the domain grouping between target domain EEG signal feature samples and corresponding source domain source domain EEG signal feature samples; for example... Figure 2 As shown, in this embodiment, the domain discriminator is also composed of a multilayer perceptron (MLP) containing multiple fully connected layers (FC) and an output layer. Its input is the fuzzy feature embedding of source domain EEG signal feature samples. Fuzzy feature embedding of target domain EEG signal feature samples splicing pairs The domain discriminator's task is to determine the domain group (source domain or target domain) to which the sample belongs, and output an R-dimensional probability distribution vector. ;

[0061] S203, use the EEG signal feature samples of the target domain and the source domain corresponding to the source domain network branch to train the adaptive fuzzy domain adversarial network of the source domain network branch.

[0062] The target domain EEG signal feature samples obtained in step S201 include labeled target domain EEG signal feature samples. and unlabeled target domain EEG signal feature samples These are used for training and testing, respectively. For each source domain... Compare it with labeled target domain EEG signal feature samples By combining these components, an independent source domain network branch is constructed for training. Thus, a total of [number] branches are constructed. K Each source domain network branch, with identical structure but independently initialized parameters, acquires labeled target domain EEG signal feature samples. and each source domain The system takes EEG signal feature samples as input; combines the samples and inputs them into a branch network to calculate the membership degree of each sample to multiple fuzzy emotion categories, generating a fuzzy feature representation of uncertainty perception; inputs the fuzzy feature representation into a task classifier to obtain emotion category prediction, and simultaneously inputs it into a domain discriminator for domain relation discrimination; during the adversarial training of the branch network, the weights of the adversarial loss are dynamically adjusted based on the calculated fuzzy domain similarity metric; after training the branch network using each source domain and target domain, the prediction results of each branch network are integrated using a fuzzy ensemble strategy to analyze unlabeled target domain EEG signal feature samples. Perform emotion recognition.

[0063] In this embodiment, the loss function used during training of the adaptive fuzzy domain adversarial network includes an adaptively weighted domain adversarial loss. ,in For source domain weights, For domain adversarial losses, source domain weights To measure the fuzzy domain similarity between the source and target domains The adjusted fuzzy domain similarity measure between the source and target domains is obtained. This includes the similarity between the fuzzy membership vectors of the source and target domains and the fuzzy feature embeddings. In this embodiment, the method introduces a fuzzy logic system to characterize emotional uncertainty and combines it with an adaptive adversarial training mechanism, effectively improving the accuracy, robustness, and generalization ability of cross-subject EEG emotion recognition.

[0064] The similarity between the fuzzy member vectors and fuzzy feature embeddings of the source and target domains can be determined using a desired similarity algorithm. For example, as an optional implementation, this embodiment measures the fuzzy domain similarity between the source and target domains. The expression for the computation function is:

[0065] ;

[0066] in, Fuzzy feature embedding for source domain EEG signal feature samples Fuzzy feature embedding for target domain EEG signal feature samples for L2 norm, for L2 norm, Let be the fuzzy member vector of the source domain. For the fuzzy member vector of the target domain, for L2 norm, for The L2 norm. This metric comprehensively evaluates the alignment between the feature layer and the fuzzy semantic layer.

[0067] Based on the fuzzy domain similarity measurement between the source and target domains Adjust source domain weights The appropriate function or association method can be selected as needed. For example, as an optional implementation, this embodiment measures the fuzzy domain similarity between the source and target domains. Adjust source domain weights The function expression is:

[0068] ;

[0069] in, e This represents the current training progress percentage, which is the ratio of the current training round to the total number of training rounds. This is a piecewise function with coefficients related to training progress, used to assign different coefficients at different training stages. As an optional implementation, the piecewise function can be designed according to the domain adaptation training principle of "stabilizing the foundation in the early stage, strengthening alignment in the middle stage, and maintaining stability in the later stage." The training progress is divided into early, middle, and late stages, with the coefficient values ​​increasing sequentially from early to late. This mechanism enhances alignment pressure in the early stage or when there are large differences between domains, and reduces pressure in the later stage or when alignment is good, thus achieving stable and efficient adaptive domain adaptation. As an optional implementation, the piecewise function can be:

[0070] .

[0071] In this embodiment, the expression for the calculation function of the domain adversarial loss is:

[0072] ;

[0073] in, For domain confrontation losses, Represents the expected value in mathematics. Type labels for the EEG signal domain. Representation domain discriminator, For splicing operations, Fuzzy feature embedding for source domain EEG signal feature samples Fuzzy feature embedding for EEG signal feature samples in the target domain.

[0074] In this embodiment, the expression for the loss function used during training is as follows:

[0075] ;

[0076] ;

[0077] in, The loss function used during training. For classifying losses, The adaptively weighted domain adversarial loss, For source domain weights, For domain confrontation losses, For the number of samples, Task labels for EEG signal feature samples Represents a task classifier. Fuzzy feature embedding for source domain EEG signal feature samples Fuzzy feature embedding for EEG signal feature samples in the target domain. By simultaneously optimizing the parameters of the fuzzy encoder, task classifier, and domain discriminator through backpropagation, the encoder is driven to learn feature representations that can accurately classify emotions while making it difficult for the domain discriminator to distinguish key cross-domain relationships.

[0078] In step S103, the final emotion classification result obtained based on the emotion classification results of each source domain network branch can be obtained using random or voting methods as needed. As an optional implementation, this embodiment uses unlabeled samples from the target domain... A fuzzy rule integration strategy is used for the final decision. Step S103, which obtains the final emotion classification result based on the emotion classification results of each source domain network branch, includes:

[0079] S301, the fuzzy domain similarity between the source domain and the target domain of each source domain network branch is measured according to the following formula. Normalization is performed to obtain the integrated weights:

[0080] ;

[0081] in, For the first Integration weights of each source domain network branch For the first A fuzzy domain similarity measure between the source domain and the target domain of each source domain network branch; This is a temperature parameter used to adjust the smoothness of the weight distribution; This represents the total number of branches in the source domain network.

[0082] S302, the predicted probability distributions obtained for each source domain network branch are weighted and voted according to the following formula:

[0083] ;

[0084] in, For the final predicted probability distribution, For the first The predicted probability distribution of each source domain network branch; from the final predicted probability distribution The emotion category with the highest predicted probability is selected as the final emotion classification result, such as positive, neutral and negative, or neutral, sad, fear and happy, etc.

[0085] To validate the method in this embodiment, cross-subject identification was performed using the publicly available SEED and SEED-IV datasets. The SEED dataset meticulously records the EEG signals of 15 participants when receiving three types of emotional stimuli (positive, neutral, and negative). The EEG signals were recorded using 62 electrodes at a high sampling rate of 1000Hz. Each participant underwent three trials (one week apart), each trial consisting of one session, and each trial involved watching 15 different videos. The SEED-IV dataset contains EEG emotional data from 15 participants, categorized into four states: neutral, sad, fearful, and happy. Similar to the SEED dataset, the SEED-IV dataset was recorded using 62 electrodes at a high sampling rate of 1000Hz. Each participant underwent three trials (one week apart), each trial consisting of one session, and the stimuli were derived from 24 different videos they watched. Results of cross-subject emotion recognition experiment: In this embodiment, each subject was used as the target domain in turn, and the remaining 14 subjects were used as the source domain. The classification experiment was carried out. The experimental results of the SEED and SEED-IV datasets are shown in Table 1 and Table 2, respectively.

[0086] Table 1. Cross-subject emotion recognition results based on the SEED dataset (%)

[0087]

[0088] Table 2. Cross-subject emotion recognition results based on the SEED-IV dataset (%)

[0089]

[0090] As shown in Tables 1 and 2, the average accuracy of the method in this embodiment for each subject in three sessions in the SEED dataset is above 90%, and the average recognition accuracy for each subject in three sessions in the SEED-IV dataset is above 80%. This fully demonstrates that the method in this embodiment has superior generalization ability in cross-subject emotion recognition tasks.

[0091] In summary, this embodiment of the cross-subject EEG emotion recognition method based on an adaptive fuzzy domain adversarial network includes: acquiring multi-channel EEG signals and performing preprocessing and feature extraction; constructing an adaptive fuzzy domain adversarial network comprising a fuzzy encoder, a task classifier, and a domain discriminator; employing an adaptive adversarial training strategy based on fuzzy domain similarity measurement; and fusing prediction results from multiple source domain models using a fuzzy ensemble strategy during the testing phase. This embodiment addresses the problems of large individual differences, fuzzy emotion boundaries, and limited domain adaptation in cross-subject EEG emotion recognition by innovatively combining fuzzy logic systems with domain adversarial learning. It introduces fuzzy membership calculation to characterize the uncertainty of emotional states and designs an adaptive adversarial weight adjustment mechanism to optimize the domain alignment process. The fuzzy encoder proposed in this embodiment can effectively learn feature representations with uncertainty perception capabilities, and the adaptive training strategy can dynamically adjust the adversarial intensity according to training progress and domain similarity. The comprehensive application of these innovative technologies enables the model to simultaneously capture the fuzzy characteristics of emotional expression and achieve efficient cross-subject knowledge transfer. Experimental results on public datasets show that the obtained EEG emotion recognition method based on adaptive fuzzy domain adversarial network achieves leading performance in cross-subject emotion recognition tasks, with advantages such as high recognition accuracy, strong generalization ability, and good model interpretability. The method in this embodiment effectively improves the robustness and practicality of the EEG emotion recognition system in practical applications by systematically integrating fuzzy learning and domain adversarial training.

[0092] Those skilled in the art will understand that the technical solutions provided by this invention can take the form of methods, systems, or computer program products. For example, this invention can provide a cross-subject EEG emotion recognition system based on adaptive fuzzy domain adversarial methods, including a microprocessor and a memory interconnected, wherein the microprocessor is programmed or configured to execute the cross-subject EEG emotion recognition method based on adaptive fuzzy domain adversarial methods. This invention can provide a computer-readable storage medium storing a computer program or instructions programmed or configured to execute the cross-subject EEG emotion recognition method based on adaptive fuzzy domain adversarial methods via a processor. This invention can provide a computer program product including a computer program or instructions programmed or configured to execute the cross-subject EEG emotion recognition method based on adaptive fuzzy domain adversarial methods via a processor. Furthermore, this invention can also take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this invention can take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0093] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A cross-subject electroencephalogram emotion recognition method based on adaptive blur domain adversarial, characterized in that, Includes the following steps: S101, acquire the EEG signal features of the target domain to be identified; S102, the EEG signal features of the target domain to be identified are input into the adaptive fuzzy domain adversarial networks in each trained source domain network branch. A fuzzy encoder maps the EEG signal feature samples into fuzzy member vectors and generates corresponding fuzzy feature embeddings. A task classifier then uses these fuzzy feature embeddings for emotion classification. The loss function used during training of the adaptive fuzzy domain adversarial network includes an adaptively weighted domain adversarial loss. ,in For source domain weights, For domain adversarial losses, source domain weight To measure the fuzzy domain similarity between the source and target domains The adjusted fuzzy domain similarity measure between the source and target domains is obtained. This includes the similarity between the fuzzy member vectors of the source and target domains and the fuzzy feature embeddings of the two. S103, obtain the final emotion classification result based on the emotion classification results of each source domain network branch.

2. The cross-subject EEG emotion recognition method based on adaptive fuzzy domain adversarial analysis according to claim 1, characterized in that, The fuzzy encoder includes a membership network and a deep encoding module. The process of mapping EEG signal feature samples to fuzzy membership vectors and generating corresponding fuzzy feature embeddings via the fuzzy encoder includes: firstly, mapping the target domain EEG signal features to be identified to fuzzy membership vectors of the target domain using the membership network. ,in ~ This indicates that the input EEG signal feature sample belongs to the 1st to 2nd generation. Membership degree of a fuzzy sentiment category The preset number of fuzzy sets is used, and the fuzzy member vectors of the target domain are used. The input EEG signal feature samples are weighted and nonlinearly transformed according to the following formula to generate enhanced fuzzy features: ； in, To enhance the fuzzy features, For the input EEG signal feature samples, This represents the membership degree of the input EEG signal feature sample to the i-th fuzzy emotion category; then, the deep encoding module generates the fuzzy feature embedding of the target domain EEG signal features to be identified based on the enhanced fuzzy features. .

3. The cross-subject EEG emotion recognition method based on adaptive fuzzy domain adversarial analysis according to claim 2, characterized in that, The membership network consists of two fully connected layers and a Softmax function. The deep coding module is composed of a multilayer perceptron (MLP), which contains three fully connected layers and one output layer. The layers of the MLP use the ReLU activation function.

4. The cross-subject electroencephalogram emotion recognition method based on adaptive blur domain adversarial of any one of claims 1-3, characterized in that, The construction and training of the source domain network branches include: S201, Obtain the target domain EEG signal feature samples and the corresponding source domain source domain EEG signal feature samples; S202, Construct an adaptive fuzzy domain adversarial network for this source domain network branch. The adaptive fuzzy domain adversarial network includes a fuzzy encoder, a task classifier, and a domain discriminator. The fuzzy encoder is used to map EEG signal feature samples into fuzzy member vectors and generate corresponding fuzzy feature embeddings, including obtaining fuzzy member vectors of EEG signal feature samples in the target domain. and fuzzy feature embedding , and fuzzy membership vectors of source domain EEG signal feature samples and fuzzy feature embedding The task classifier is used to classify emotions based on fuzzy feature embeddings, and the domain discriminator is used to determine the domain group between target domain EEG signal feature samples and corresponding source domain source domain EEG signal feature samples. S203, use the EEG signal feature samples of the target domain and the source domain corresponding to the source domain network branch to train the adaptive fuzzy domain adversarial network of the source domain network branch.

5. The cross-subject electroencephalogram emotion recognition method based on adaptive blur domain adversarial of claim 1, characterized in that, A fuzzy domain similarity measure between the source domain and the target domain The computational function expression is: ； in, Fuzzy feature embedding for source domain EEG signal feature samples Fuzzy feature embedding for target domain EEG signal feature samples for L2 norm, for L2 norm, Let be the fuzzy member vector of the source domain. For the fuzzy member vector of the target domain, for L2 norm, for The L2 norm.

6. The cross-subject electroencephalogram emotion recognition method based on adaptive blur domain adversarial of claim 1, characterized in that, According to a fuzzy domain similarity measure between the source domain and the target domain Adjusting the source domain weight The function expression is: ； in, e This represents the current training progress percentage, which is the ratio of the current training round to the total number of training rounds. It is a piecewise coefficient function related to training progress, used to assign different coefficients at different training progress.

7. The cross-subject electroencephalogram emotion recognition method based on adaptive blur domain adversarial of claim 1, characterized in that, The expression for the domain adversarial loss is as follows: ； in, For domain confrontation losses, Represents the expected value in mathematics. Type labels for the EEG signal domain. Representation domain discriminator, For splicing operations, Fuzzy feature embedding for source domain EEG signal feature samples Fuzzy feature embedding for EEG signal feature samples in the target domain.

8. The cross-subject EEG emotion recognition method based on adaptive fuzzy domain adversarial analysis according to claim 1, characterized in that, The expression for the loss function used during training is as follows: ； ； in, The loss function used during training. For classifying losses, The adaptively weighted domain adversarial loss, For source domain weights, For domain confrontation losses, For the number of samples, Task labels for EEG signal feature samples Represents a task classifier. Fuzzy feature embedding for source domain EEG signal feature samples Fuzzy feature embedding for EEG signal feature samples in the target domain.

9. The cross-subject electroencephalogram emotion recognition method based on adaptive blur domain adversarial of claim 1, characterized in that, Step S103, which obtains the final emotion classification result based on the emotion classification results of each source domain network branch, includes: S301, obtaining the fuzzy domain similarity measure between the source domain and the target domain of each source domain network branch according to the following formula Normalization is performed to obtain the integrated weight: ； in, For the first Integration weights of each source domain network branch For the first A fuzzy domain similarity measure between the source domain and the target domain of each source domain network branch; This is a temperature parameter used to adjust the smoothness of the weight distribution; This represents the total number of branches in the source domain network. S302, the predicted probability distributions obtained for each source domain network branch are weighted and voted according to the following formula: ； in, For the final predicted probability distribution, For the first The predicted probability distribution of each source domain network branch; from the final predicted probability distribution The emotion category with the highest predicted probability is selected as the final emotion classification result.

10. An adaptive blur domain adversarial based cross-subject electroencephalogram emotion recognition system comprising a microprocessor and a memory interconnected, characterized in that, The microprocessor is programmed or configured to perform the cross-subject EEG emotion recognition method based on adaptive fuzzy domain adversarial methods as described in any one of claims 1 to 9.

Images