A single-source domain adaptive cross-time brain pattern recognition method

Abstract

The present invention discloses a single-source domain adaptive cross-period brainprint recognition method. The cross-period brainprint recognition method processes the data of the subject through data alignment, and uses a domain discriminator adversarial module and an associated domain adaptive module to construct a cross-period brainprint recognition model. It can still recognize EEG signals of different time periods when there is only one time period of EEG data as the source domain, solving the problems of weak generalization ability of cross-period data and insufficient EEG labeling data in traditional brainprint recognition. At the same time, the data alignment method adopted by the cross-period brainprint recognition method does not require subject labels, but directly transforms the original EEG data to adapt it to these changes and extract domain-invariant features, thereby realizing single-source domain adaptive cross-period EEG signal identity prediction. In addition, the cross-period brainprint recognition method introduces a self-attention mechanism to identify key features, reduce excessive response to redundant information, and thus improve the robustness of features.

Description

Technical Field

[0001] The present invention belongs to the technical field of cross-time single-source domain adaptive brain pattern recognition, and specifically relates to a single-source domain adaptive cross-time brain pattern recognition method. Background Art

[0002] Biometric recognition technology, as an advanced means of information security, is gaining increasing attention. However, traditional biometric recognition technology still faces several challenges. Physiological features such as fingerprints, irises, and faces are susceptible to theft, copying, and tampering. Furthermore, behavioral characteristics are susceptible to imitation and attack in public surveillance environments. Compared to existing traditional biometric recognition technologies, EEG-based brainprint features offer high concealment, unforgeability, continuous authentication, and require a living individual. These characteristics make brainprint recognition technology suitable for applications requiring high security, such as those in the military, finance, and medical security.

[0003] EEG identity recognition faces multiple technical and practical challenges in practical applications. First, generalization across time periods is weak. Because the EEG cap is worn in different positions during each data collection, channel position shift occurs. This means that electrode positions can vary across time and scenarios, leading to inconsistent EEG signals. Furthermore, EEG signals vary with factors such as time, mood, and fatigue level. This can lead to significant differences in signals collected from the same subject at different time points, impacting recognition stability and accuracy. Consequently, the training and test data may not follow the same distribution across time periods, further impacting identification accuracy. Second, there is a shortage of labeled EEG data. Cross-time labeled EEG data is extremely limited, and obtaining large amounts of labeled data is prohibitively expensive and time-consuming. To cope with limited labeled training data, many attempts have been made to directly apply models trained in one source domain to an unlabeled target domain. Unfortunately, due to domain shift or dataset bias, distribution differences between the source and target domains can lead to reduced model performance in the target domain, making direct cross-domain transfer generally ineffective. Domain adaptation techniques, however, use various methods to mitigate this distribution difference, enabling models trained in the source domain to better generalize to the target domain.

[0004] Furthermore, traditional deep learning networks typically use only a fixed convolution kernel to process complex EEG signals. This approach has limitations because it ignores individual differences between subjects and changes in data distribution over time. This inattention to data distribution differences can make the model difficult to adapt to changing signals in practical applications, thus affecting recognition and classification accuracy. Therefore, designing more flexible and adaptive convolution kernels to process and capture these complex signal variations is a major challenge in current research. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention proposes a single-source domain adaptive cross-time period brain pattern recognition method to learn stable domain invariant features. Through the newly collected EEG data of the subject, the identity information of the subject is predicted from multiple pre-set target object identities.

[0006] The present invention proposes a single-source domain adaptive cross-time brain pattern recognition method, which includes the following steps:

[0007] Step 1: Collect EEG data from multiple subjects and add labels as the source domain; then collect EEG data from a different time period from the source domain as the target domain; the target domain does not require labels.

[0008] Step 2: Construct a cross-temporal brainprint recognition model. The cross-temporal brainprint recognition model includes a multi-scale feature extraction module, a domain discriminator adversarial module, an associated domain adaptation module, and a linear classification module. The feature extraction module includes a temporal convolution layer, a spatial convolution layer, a depthwise separable convolution layer, and an attention mechanism module connected in sequence. Features are extracted through the temporal convolution layer, the spatial convolution layer, and the depthwise separable convolution layer to obtain initial features and input them into the attention mechanism module to obtain the final output features. The final output features are input into the domain discriminator adversarial module, the associated domain adaptation module, and the linear classification module respectively. The domain discriminator adversarial module is used to perform adversarial training on the final output features of the source domain and the target domain. The associated domain adaptation module is used to cluster source domain samples of the same category in the feature space.

[0009] Step 3: Use the data set obtained in step 1 to train the cross-time brain pattern recognition model constructed in step 2.

[0010] Step 4: Collect the subject's EEG data and input it into the trained cross-time brain pattern recognition model to identify the subject's identity.

[0011] Preferably, in step 1, the data alignment method is as follows:

[0012] Get n EEG samples X of the subject i The mean matrix R of the covariance matrix, the expression of the mean matrix R is:

[0013]

[0014] in, For EEG sample X i The transpose of ; i=1,2,...,n.

[0015] Get aligned EEG samples according to the mean matrix R :

[0016]

[0017] Preferably, in step 3, the expression of the loss function L for training the cross-period brain pattern recognition model is:

[0018]

[0019] Among them, L cls is the cross entropy loss; L domain is the domain discriminator loss; L association is the association domain adaptation loss.

[0020] Preferably, the cross entropy loss L cls , domain discriminator loss L domain and cross entropy loss L cls The expressions are:

[0021]

[0022]

[0023]

[0024] in, and are the domain discriminator losses for the source and target domains, respectively; L visit and L walker are visitor loss and pedestrian loss respectively; α is the weight of pedestrian loss, α=0.6; is the cross entropy loss; y is the true subject label; is the predicted subject label; N is the number of subjects.

[0025] Domain Discriminator Loss and The expression is:

[0026]

[0027]

[0028] Among them, n s and n t are the number of source domain samples and the number of target domain samples respectively; D s and D t are the source domain of the training set and the target domain of the test set respectively; G d is the domain identifier; G f is the feature extraction module; d i For the input sample x i The corresponding domain label.

[0029] Visitor loss L visit and the Pacers lost Lwalker The expression is:

[0030]

[0031]

[0032] in, ; ; A i and B j are the output features of the source domain samples and target domain samples after the feature extraction network respectively; class(·) is the subject category label corresponding to the output feature; From the output feature A i Transfer to output feature B j The transition probability of From the output feature A i Transfer to output feature B j Then return the output feature A i The probability of a two-step round trip.

[0033] Transition probability and the two-step round trip probability The expression is:

[0034]

[0035]

[0036] Among them, M ij Output feature A i and B j The dot product of Output feature B j Return to output feature A i probability.

[0037] Preferably, in the step three, the cross-period brainprint recognition model is trained by inputting labeled source domain samples and unlabeled target domain samples into the cross-period brainprint recognition model.

[0038] Preferably, in the step 2, the temporal convolution layer uses three convolution kernels of different lengths to extract temporal features, the heights of the three convolution kernels are all 1, and the lengths are respectively the number of sample time points × 0.5, the number of sample time points × 0.25, and the number of sample time points × 0.125.

[0039] Preferably, in step 2, the attention mechanism module includes a global pooling layer and two fully connected layers connected in sequence.

[0040] Preferably, in the step 2, a BatchNorm layer, an ELU layer, an AvgPool layer and a Dropout layer are added after the temporal convolution layer, the spatial convolution layer and the depth-separable convolution layer.

[0041] In a second aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the memory stores the computer program; and the processor executes the aforementioned single-source domain adaptive cross-period brainprint recognition method.

[0042] In a third aspect, the present invention provides a readable storage medium storing a computer program; when the computer program is executed by a processor, it is used to implement the aforementioned single-source domain adaptive cross-time period brainprint recognition method.

[0043] The present invention has the following beneficial effects:

[0044] 1. The present invention processes the subject's data through data alignment, and uses a domain discriminator adversarial module and an associated domain adaptive module to construct a cross-time brainprint recognition model. Compared with traditional brainprint recognition that requires EEG data of multiple time periods as the source domain, the present invention can still recognize EEG signals of different time periods with only one time period of EEG data as the source domain, solving the problems of weak generalization ability of cross-time data and insufficient EEG labeling data in traditional brainprint recognition.

[0045] 2. The data alignment method adopted in the present invention does not require subject labels, but directly performs transformations on the original EEG data, which can enhance the robustness of the model, enable it to adapt to these changes and extract domain-invariant features, thereby realizing identity prediction of EEG signals across time periods in a single-source domain.

[0046] 3. The self-attention mechanism introduced in this invention can identify key features in time series data, help the model dynamically adjust the weights of different convolution kernels to adapt to different domain data with different distribution differences, and through the compression incentive mechanism, help the model identify and amplify the channel information and time domain information features related to the task, reduce excessive response to redundant information, and thus improve the robustness of the features.

[0047] 4. This invention introduces a domain discriminator adversarial module, which maps source and target domain features into the same feature space through the feature extractor. The domain discriminator attempts to distinguish between source and target domain features, while the feature extractor attempts to trick the domain discriminator into being unable to distinguish between the two types of features. Through this adversarial learning approach, the feature distributions of the source and target domains are made as similar as possible, thereby reducing the distribution difference between them.

[0048] 5. The present invention calculates the association matrix between source domain and target domain samples through the association domain adaptation module, and minimizes the KL divergence between the distribution between source domain samples and the target domain category distribution, so that source domain samples of the same category are more clustered together and match the category distribution of target domain samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is the overall flow chart of the present invention.

[0050] Figure 2 Schematic diagram of the overall architecture of the cross-time brain pattern recognition model in the present invention.

[0051] Figure 3 Schematic diagram of the feature extraction module in the present invention.

[0052] Figure 4 Schematic diagram of the attention mechanism module in the present invention.

[0053] Figure 5 Schematic diagram of the domain discriminator adversarial module in the present invention.

[0054] Figure 6 Schematic diagram of the association domain adaptation module in the present invention. DETAILED DESCRIPTION

[0055] The present invention will be further described below with reference to the accompanying drawings.

[0056] like Figure 1 As shown, a single-source domain adaptive cross-time brain pattern recognition method includes the following steps:

[0057] Step 1: Get the dataset

[0058] The dataset uses a cross-temporal dataset for collaborative brain-computer interface based on rapid serial visual presentation (RSVP) from Tsinghua University. In this dataset, 14 subjects participated in two identical experiments, with an average interval of approximately 23 days. The 14 subjects were divided into 14 groups, each consisting of two people. Each trial consisted of 100 images (10 seconds at a rate of 10 Hz), including four target images. The images displayed in the first and last seconds of a trial were non-target images to prevent the target images from appearing during the onset or offset of the RSVP-induced steady-state visual evoked potential (SSVEP). The interval between target images was at least 500 milliseconds to reduce the effects of attentional flicker. Subjects were instructed to press a key immediately after detecting a target. This key-pressing task was designed to focus subjects on target detection. Due to the time delay between the target image and the key press, a key press within 500 milliseconds of a target image was considered a correct response to the target image during the experiment. In the experiment, subjects were required to find four targets from the 100 images and perform four key presses. If a subject missed any of the targets, the system displayed the missed targets at the end of the trial. The experiment consisted of two experiments conducted on different days, using identical stimulation paradigms. EEG data from both subjects were recorded simultaneously using two Neuroscan Synamps2 systems. A 64-electrode EEG cap based on the 10-20 system was used to record 62 channels of EEG data from both subjects (M1 and M2 were not used). The reference electrode was located at the vertex. The electrode impedance was kept below 10kΩ, and the sampling rate was 1,000 Hz. Dataset preprocessing included filtering, re-referencing, and data segmentation. A 50 Hz notch filter was used to remove common power line noise. The amplifier passband was set to 0.15 Hz–200 Hz. All event triggers were transmitted via a parallel port and labeled on the EEG data. EEG data corresponding to all target images were selected, using all channels, filtered within a 1 Hz–40 Hz band, and 0.5 s of data as a sample, with 500 time steps per sample. The samples obtained from both experiments were used as the training and test sets, respectively. The training and test sets were used as the source and target domains, respectively.

[0059] Step 2: Data alignment

[0060] Each subject has n preprocessed EEG samples X i , calculate the covariance matrix of each EEG sample point, and obtain the mean matrix R of n covariance matrices. The expression of the mean matrix R is:

[0061]

[0062] in, For EEG sample X iThe transpose of ; i=1,2,...,n.

[0063] Get aligned EEG samples according to the mean matrix R :

[0064]

[0065] Aligned EEG samples With the original EEG sample X i The dimensions of any subject are the same. The mean matrix of the covariance matrix They are all unit matrices, and the overall distribution is more consistent.

[0066]

[0067] Where I is the identity matrix.

[0068] Step 3: Build a cross-time brain pattern recognition model

[0069] like Figure 2 、 3 As shown in Figures 4 and 5, the cross-temporal brainprint recognition model is a multi-scale attention feature fusion network, comprising a multi-scale feature extraction module, a domain discriminator adversarial module, a correlation domain adaptation module, and a linear classification module. The feature extraction module is used to extract EEG features and includes a temporal convolutional layer, a spatial convolutional layer, a depthwise separable convolutional layer, and an attention mechanism module based on a compression-excitation mechanism. The temporal convolutional layer uses three convolution kernels of different lengths to extract temporal features. The depthwise separable convolutional layer inputs the initial features into the attention mechanism module to obtain corresponding feature weights. The attention mechanism module consists of a global pooling layer and two fully connected layers connected in sequence. The global pooling layer globally pools the features input to the attention mechanism module, compressing the input features into a feature vector whose length is equal to the number of channels to achieve a larger receptive field. The two fully connected layers generate a feature weight matrix composed of multiple channels. The initial features are multiplied by the feature weight matrix and flattened to obtain the final output features. The final output features are input to the domain discriminator adversarial module, the correlation domain adaptation module, and the linear classification module. The linear classification module uses the Kan classifier to classify the subject's identity based on the final features.

[0070] like Figure 5 As shown in the figure, the domain discriminator adversarial module inputs the final output features of the source domain and the final output features of the target domain respectively for adversarial training. In the domain adversarial training stage, gradient reversal is used to achieve domain confusion, making it impossible for the feature extractor to distinguish that the source domain and target domain samples are from two different domains, thereby extracting stable domain invariant features.

[0071] like Figure 6As shown in the figure, the association domain adaptation module calculates the association matrix between the source domain and target domain samples, inputs the labeled source domain output features and the unlabeled target domain final output features, and calculates the conditional probability distribution P from the source domain to the target domain by performing matrix multiplication on the source domain and target domain features. st And the conditional probability distribution P from the target domain to the source domain ts , generate the correlation matrix P between source domain samples sts And the category distribution P of the target domain samples t By minimizing the correlation matrix between source domain samples and target domain category distribution P t The KL divergence (‌Kullback-Leibler Divergence)‌ between them is used to achieve the aggregation of source domain samples of the same category in the feature space.

[0072] Step 4: Training the model

[0073] 4-1. Input labeled source domain samples and unlabeled target domain samples into the temporal convolution layer, spatial convolution layer, and depthwise separable convolution layer to obtain initial features. After inputting these initial features into the attention mechanism module, a weight matrix is ​​obtained. After weighting the initial features with this weight matrix, the final features are obtained.

[0074] 4-2. Input the final feature map into the domain discriminator adversarial module, the associated domain adaptation module and the linear classification module to obtain the domain discriminator loss L domain , association domain adaptation loss L association and cross entropy loss L cls , whose expressions are:

[0075]

[0076]

[0077]

[0078] in, and are the domain discriminator losses for the source and target domains, respectively; L visit and L walker are visitor loss and pedestrian loss respectively; α is the weight of pedestrian loss, α=0.6; is the cross entropy loss; y is the true subject label; is the predicted subject label; N is the number of subjects.

[0079] Domain Discriminator Loss and The expression is:

[0080]

[0081]

[0082] Among them, n s and n t are the number of source domain samples and the number of target domain samples respectively; D s and D t are the source domain of the training set and the target domain of the test set respectively; G d is the domain identifier; G f is the feature extraction module; d i For the input sample x i The corresponding domain label.

[0083] Visitor loss L visit Ensure that the category distribution of the target domain samples is as uniform as possible by minimizing the target domain category distribution (predicted by the source domain samples) and the uniform distribution p visit KL divergence between realization; Walker loss L walker It aims to enable the model to align features between different domains by minimizing the KL divergence between the predicted category distribution and the target category distribution. visit and the Pacers lost L walker The expression is:

[0084]

[0085]

[0086] in, ; ; A i and B j are the output features of the source domain samples and target domain samples after the feature extraction network respectively; class(·) is the subject category label corresponding to the output feature; From the output feature A i Transfer to output feature B j The transition probability of From the output feature A i Transfer to output feature B j Then return the output feature A i The probability of a two-step round trip.

[0087] Transition probability and the two-step round trip probability The expression is:

[0088]

[0089]

[0090] Among them, M ij Output feature A i and B j The dot product of Output feature B j Return to output feature A i probability.

[0091] 4-3. Constructing the loss function L of the cross-time brain pattern recognition model:

[0092]

[0093] The loss function L is minimized through iterative cycles, and the final cross-time brain pattern recognition model is obtained.

[0094] Step 5: Evaluate the model

[0095] To compare with existing technologies, top-1 classification accuracy was used as the performance metric. All experimental results are the average of five test runs using different random seeds. The temporal, spatial, and depthwise separable convolutional layers are all 2D convolutional layers. The kernel size for the temporal convolutional layer is (1,250), the kernel size for the spatial convolutional layer is (62, 1), and the kernel size for the depthwise separable convolutional layer is (1, 1). Each convolutional layer is followed by a BatchNorm layer, an ELU layer, an AvgPool layer, and a Dropout layer with a dropout coefficient of 0.25. The model was trained using the Adam optimizer with a learning rate of 5e-4 and a batch size of 8 for 200 epochs.

[0096] The models used in the comparative experiments with the present invention are divided into the following three categories: (1) traditional deep learning models, which refer to deep learning methods that do not include domain adaptation structures; (2) single-source domain adaptation models, which have only one source domain; and (3) multi-source domain adaptation models, which align the data distributions of different source domains and target domains. The specific comparative models are introduced below. The EEGNet model first uses a one-dimensional convolution operation to extract the temporal features of the EEG, and then uses deep separable convolution to extract the EEG frequency-space features. This model is a traditional deep learning model and has been applied to brain pattern recognition. The CNNRNN model uses a two-dimensional convolutional neural network (CNN) and a bidirectional recurrent neural network (BiLSTM) structure to extract the frequency-space features and time-dependent temporal features of the EEG. This model is a traditional deep learning model and has been proposed for brain pattern recognition. BrainNet uses a one-dimensional convolution operation to extract the temporal features of the EEG, and then uses triplet loss (TripletLoss) to make samples of the same type closer to each other and samples of different types farther away from each other. MEERNet is a multi-source domain adaptation model that considers both domain-invariant and domain-specific features. It has been proposed for EEG emotion recognition across time periods and subjects. MTDANN is a multi-source domain adaptation method for EEG signals. This method consists of an EEGNet and a domain discriminant network, aiming to learn domain-invariant representations between the source and target domains.

[0097] Since traditional models lack single-source domain adaptation, to better compare them with the model of the present invention, the method of the present invention was used to transform traditional models into single-source domain adaptation models for comparison. The comparative experimental results are shown in Table 1 below. On a collaborative brain-computer interface cross-period dataset based on rapid serial visual presentation (RSVP), the accuracy of the present invention reached 84.37% in the test set (second period). This represents a significant advantage over traditional deep learning models and multi-source domain adaptation models. In the single-source domain adaptation model, the present invention achieved an average accuracy improvement of 13.86% compared to EEGNet and 3.42% compared to BrainNet, demonstrating that the present invention provides more accurate recognition results.

[0098] Table 1 Test results of different models

[0099]

[0100] We conducted multiple ablation experiments on the overall model to evaluate the impact of each module on model performance. By removing or modifying different modules and observing the changes in model accuracy, we can understand the role of each module in the overall model. The results of the ablation experiments are shown in Table 2.

[0101] Table 2 Ablation experiment

[0102] Model Accuracy Overall model 84.37% Domain Discriminator Adversarial Module 34.15% Unrelated domain adaptation module 77.34% No attention mechanism module 83.65%

[0103] The accuracy of the overall model is 84.37%, and the model performance is better than the existing single-source domain adaptation method and multi-source domain adaptation method, which proves the effectiveness of the transferable feature learning of the present invention.

[0104] After removing the domain discriminator adversarial module, the model accuracy dropped significantly to 34.15%. This means that domain discriminator adversarial learning can learn rich stable and invariant features, narrow the distribution gap between the source domain and the target domain through adversarial training, help the model better adapt to the distribution differences between different domains, and enhance the model's generalization ability.

[0105] After removing the association domain adaptation module, the model accuracy dropped to 77.34%. This means that the association domain adaptation module can enhance the feature association between the source domain and the target domain, thereby helping the model better adapt to the data distribution of the target domain, more effectively transfer the knowledge of the source domain to the target domain, and improve the performance of the model on the target domain.

[0106] After removing the attention mechanism module, the model accuracy dropped to 83.65%, which means that the introduction of the attention mechanism provides the model with more fine-grained information processing capabilities, which can effectively identify and highlight important features related to the task.

Claims

1. A single-source domain adaptive cross-time brainprint recognition method, characterized by: The following steps are involved: Step 1: Collect EEG data from multiple subjects and add labels as the source domain; then collect EEG data from different time periods as the target domain; Step 2: Construct a cross-time brain pattern recognition model; The cross-temporal brainprint recognition model includes a multi-scale feature extraction module, a domain discriminator adversarial module, a correlation domain adaptation module, and a linear classification module. The feature extraction module includes a temporal convolution layer, a spatial convolution layer, a depthwise separable convolution layer, and an attention mechanism module connected in sequence. Features are extracted through the temporal convolution layer, the spatial convolution layer, and the depthwise separable convolution layer to obtain initial features and input them into the attention mechanism module to obtain the final output features. The final output features are input into the domain discriminator adversarial module, the associated domain adaptation module, and the linear classification module respectively; the domain discriminator adversarial module is used to perform adversarial training on the final output features of the source domain and the target domain; The associated domain adaptation module is used to aggregate source domain samples of the same category in the feature space; Step 3: Use the data set obtained in step 1 to train the cross-time brain pattern recognition model constructed in step 2; In step 3, the loss function L for training the cross-period brain pattern recognition model is expressed as: ； Among them, L cls is the cross entropy loss; L domain is the domain discriminator loss; L association is the association domain adaptive loss; The cross entropy loss L cls , domain discriminator loss L domain and cross entropy loss L cls The expressions are: ； ； ； in, and are the domain discriminator losses for the source and target domains, respectively; L visit and L walker are visitor loss and pedestrian loss respectively; α is the weight of pedestrian loss, α=0.6; is the cross entropy loss; y is the true subject label; is the predicted subject label; N is the number of subjects; Domain Discriminator Loss and The expression is: ； ； Among them, n s and n t are the number of source domain samples and the number of target domain samples respectively; D s and D t are the source domain of the training set and the target domain of the test set respectively; G d is the domain identifier; G f is the feature extraction module; d i For the input sample x i The corresponding domain label; Visitor loss L visit and the Pacers lost L walker The expression is: ； ； in, ; ; A i and B j are the output features of the source domain samples and target domain samples after the feature extraction network respectively; class(·) is the subject category label corresponding to the output feature; From the output feature A i Transfer to output feature B j The transition probability of is the two-step round trip probability; Transition probability and the two-step round trip probability The expression is: ； ； Among them, M ij Output feature A i and B j The dot product of Output feature B j Return to output feature A i probability; Step 4: Collect the subject's EEG data and input it into the trained cross-time brain pattern recognition model to identify the subject's identity.

2. The single-source domain adaptive cross-time brainprint recognition method according to claim 1 is characterized by: In step 1, the data alignment method is as follows: Get n EEG samples X of the subject i The mean matrix R of the covariance matrix, the expression of the mean matrix R is: ； in, For EEG sample X i transpose of ; i=1,2,...,n; Get aligned EEG samples according to the mean matrix R : 。 3. The single-source domain adaptive cross-time brainprint recognition method according to claim 1 is characterized by: In the step three, the cross-period brainprint recognition model is trained, and the input to the cross-period brainprint recognition model is labeled source domain samples and unlabeled target domain samples.

4. The single-source domain adaptive cross-time brain pattern recognition method according to claim 1 is characterized by: In the step 2, the temporal convolution layer uses three convolution kernels of different lengths to extract temporal features. The heights of the three convolution kernels are all 1, and the lengths are respectively the number of sample time points × 0.5, the number of sample time points × 0.25, and the number of sample time points × 0.

125.

5. The single-source domain adaptive cross-time brain pattern recognition method according to claim 1 is characterized by: In the step 2, the attention mechanism module includes a global pooling layer and two fully connected layers connected in sequence.

6. The single-source domain adaptive cross-time brain pattern recognition method according to claim 1 is characterized by: In the step 2, a BatchNorm layer, an ELU layer, an AvgPool layer, and a Dropout layer are added after the temporal convolution layer, the spatial convolution layer, and the depthwise separable convolution layer.

7. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the memory stores the computer program; and the processor executes the single-source domain adaptive cross-time period brainprint recognition method according to claim 1.

8. A readable storage medium, characterized in that: A computer program is stored; when the computer program is executed by a processor, it is used to implement a single-source domain adaptive cross-time period brainprint recognition method as shown in claim 1.

Images