Motor imagery recognition method based on cross-rhythm multi-layer coupled brain network of electroencephalogram signals

Abstract

The present application relates to a method for recognizing motor imagery based on cross-rhythm multilayer coupling brain network of electroencephalogram signals, and a plurality of kinds of motor imagery tasks of normal people are collected by multi-lead electroencephalogram (EEG). According to the collected signals, data preprocessing is carried out: after the node common average reference, a FIR filter band-pass filter is used to obtain the mu rhythm and beta rhythm which are active and have ERS / ERD phenomenon of motor imagery, and then an independent component analysis method is used to remove noise interference such as electrooculogram and electromyogram. On this basis, a multilayer functional brain network is established, the lead signal is used as the node, the phase-locked value is used as the connection, the adjacency matrix is established, and the brain function network features reflecting the activity characteristics of motor imagery are extracted. Three kinds of global network features in the mu rhythm and beta rhythm single-layer network and two kinds of interlayer feature parameters between networks are extracted, and after normalization, the multilayer network features for motor imagery recognition are generated, and the motor imagery is recognized by using a machine learning method such as a support vector machine and using the multilayer network features.

Description

Technical Field

[0001] This invention relates to a method for motor imagery recognition based on a multi-layer coupled brain network across rhythms of electroencephalogram (EEG) signals, and to functional brain network modeling and network feature extraction techniques. In particular, it relates to a modeling technique for constructing a multi-layer functional brain network based on EEG signals across rhythms of motor imagery, and a multi-layer composite network feature extraction method for motor imagery recognition, belonging to the field of pattern recognition. Background Technology

[0002] Motor imagery has the potential to activate the plasticity of the brain's own cells. When a person imagines limb movements but does not actually move them, the brain regions activated are the same as those activated during actual limb movement; the sensorimotor cortex is active. Therefore, motor imagery can accelerate the repair of neural connections between the brain and limbs with motor impairments, promoting rehabilitation. With the development of brain-computer interface (BCI) technology, the role of motor imagery EEG signals in human-computer interaction, rehabilitation, and brain function research is becoming increasingly prominent. Motor imagery-based BCI technology can not only help people with motor impairments effectively control external devices but also provide an effective approach to rehabilitation treatment. Motor imagery recognition is a key element of this rehabilitation method. However, the recognition rate is closely related to the method of extracting motor imagery features. Functional brain networks (FBNets), due to their excellent spatiotemporal feature representation capabilities, have become a powerful tool for analyzing brain activity and function in recent years. FBNets based on EEG signals can not only effectively characterize signal changes at EEG nodes but also express the interaction relationships between nodes, exhibiting the same mechanism as brain region activity and inter-regional interaction during brain activity. Therefore, this invention selects functional networks to build a brain model and extracts effective network features as a source of information for motor imagery recognition.

[0003] During motor imagery, two phenomena occur: event-related synchronization (ERS) and event-related desynchronization (ERD). Based on the characteristics of human physiological activity, electroencephalogram (EEG) signals are usually divided into four rhythms (some literature divides them into five), corresponding to brain functional states such as infant EEG, sleep, meditation, focus, and excitement / intense states. Due to the meditation and focus characteristics of motor imagery, the ERS / ERD phenomenon is relatively active in the μ rhythm (8-13Hz) and β rhythm (14-30Hz), and there is a strong correlation between the two.

[0004] Current research on brain functional networks is based on single-layer networks, which can only capture single interaction patterns between elements. If functional brain networks are analyzed across the entire frequency band of EEG signals, the rhythmic characteristics of EEG signals in different working states cannot be fully utilized. Considering the strong rhythmic nature of EEG signals and that different rhythms correspond to different brain activity states, it is necessary to construct different functional brain networks for different EEG rhythms. However, analyzing only a single rhythm in isolation obviously severs the inherent synergistic relationships in brain activity processes, which have already been observed in the μ and β rhythms of motor imagery. Clearly, functional brain network analysis based on single-rhythm EEG signals will result in information omissions or missing functional descriptions. This problem can be solved by constructing a dual network of μ-rhythm layers and β-rhythm layers that are interconnected. Multilayer networks are a generalization of traditional networks and an extension of single-layer networks. They retain the simplicity of single-layer networks while providing scalable flexibility in multimodal data modeling. Each network independently represents a certain time period or a certain type of brain information. The interlayer connections extended by multilayer networks preserve the structural relationships between different modalities and can describe the interactions between multi-scale patterns, enabling more effective and comprehensive analysis of EEG signal data. Summary of the Invention

[0005] To overcome the shortcomings of existing research, this invention provides a method for motor imagery recognition based on a multi-layered coupled brain network across rhythms of EEG signals. First, based on the established leads, different types of motor imagery EEG signals are collected from the subjects. The collected signals undergo data preprocessing and rhythm segmentation. Second, the μ and β rhythms, which exhibit the most active motor phenomena and the most pronounced ERS / ERD phenomena, are selected to establish a two-layered multiplexed functional brain network. Finally, relevant features of the two-layered multiplexed network are extracted to construct feature vectors, which are then input into a machine learning algorithm for motor imagery recognition.

[0006] A method for recognizing motor imagery based on cross-rhythm, multi-layer coupled brain networks using electroencephalogram (EEG) signals, characterized by the following steps:

[0007] S1: Collect EEG signals of multiple target categories of motor imagery from an individual in a resting state and perform preprocessing.

[0008] S1 specifically includes:

[0009] S1.1: Preparation: The subject agrees with the type of motor imagery to be collected in this experiment. After being fully relaxed, the subject remains at rest, wears an EEG cap and sits in front of the monitor. The EEG cap uses 41-channel electrodes, which conforms to the international 10 / 20-lead standard, and the EEG signal sampling frequency is 100 Hz.

[0010] S1.2: Start collecting EEG signals. The monitor remains black for the first 2 seconds. In the 3rd second, the monitor displays the task for this collection. The subject performs different types of motor imagery according to the text prompts for 4 seconds. In the 6th second, the monitor returns to black and the motor imagery ends.

[0011] S1.3: After completing one signal acquisition, repeat S1.1 and S1.2. After 5 sets of signal acquisitions, the subject rests and relaxes for 30 seconds before repeating 1-1 and 1-2. 150 sets of signals are collected for each category of motor imagery task before the acquisition ends.

[0012] S1.4: Import the obtained EEG signals into the EEGLab tool on the MATLAB platform, define the area covered by the electrode corresponding to each EEG lead as a node, complete the localization of all electrode nodes, retain the 41 nodes required for modeling, and remove invalid nodes.

[0013] S2: Extract the μ-rhythm and β-rhythm EEG signals from the signal, remove physiological artifacts, μ-rhythm is 8-14 Hz, β-rhythm is 14-30 Hz;

[0014] S2 specifically includes:

[0015] S2.1: A common average reference is used for preprocessed EEG signals in eeglab;

[0016] S2.2: Using an FIR filter, the signal is averaged after bandpass filtering to extract the μ rhythm and β rhythm. Independent component analysis is performed on each rhythm to remove physiological artifacts such as electromyography and electrooculography from the signal.

[0017] S2.3: The final modeling signal x is obtained. μ (t)∈R N*K With x β (t)∈R N*K Where N is the number of leads and K is the number of sampling points.

[0018] S3: Based on the multi-layer reuse network modeling structure and prior knowledge of motor imagery, a two-layer functional brain network based on μ rhythm and β rhythm is constructed.

[0019] S3 specifically includes:

[0020] S3.1: Each lead corresponds to a network node. The phase-locked value between any two nodes in the same layer is used to quantify the strength between nodes as the network connection to construct a single-layer network. Single-layer networks for μ rhythm and β rhythm are constructed separately. The phase-locked value between nodes with the same name in different layers is used to quantify the inter-layer connection strength between nodes with the same name in different rhythms in μ rhythm and β rhythm networks to construct inter-layer connections. Two single-layer network correlation matrices and one inter-layer connection correlation matrix can be obtained.

[0021] If the functional brain network is analyzed across the entire frequency band of EEG signals, the rhythmic characteristics of EEG signals in different working states cannot be fully utilized. Considering the strong rhythmic nature of EEG signals and the fact that different rhythms correspond to different brain activity states, the μ and β rhythms in the motor imagery EEG signals are selected as the basic data for network construction. Among them, the μ rhythm, as the most active rhythm in the human brain during meditation and concentration, contains a large amount of information exchange in the human brain during motor imagery. On the other hand, analyzing only the μ rhythm in isolation obviously severs the inherent synergistic relationship in the brain activity process, which has been observed in both the μ and β rhythms of motor imagery. Event correlation / discorrelation, as a significant feature of motor imagery, contains a large amount of inter-brain connection interaction information and is widely present in the μ and β rhythms. Therefore, in order to fully express the information exchange and synergistic structure between brain regions, the μ and β rhythms are selected to perform a two-layer network modeling of the functional brain network for motor imagery.

[0022] S3.2: Use a threshold to determine the node connection relationship and construct a two-layer multiplexed functional brain network. The threshold K should satisfy K>2lnN≈7.7, where N is the number of network nodes. If the strength between nodes is greater than the set threshold, a connection is established between the two leads and the corresponding element of the superadjacency matrix is ​​set to 1. Otherwise, no connection is established and the corresponding element of the superadjacency matrix is ​​set to 0. Finally, a two-layer functional network of μ rhythm and β rhythm can be obtained.

[0023] S4: Extract the features of the two-layer functional brain network that reflect different categories of motor imagination, and construct a composite feature vector based on the multi-rhythm information of EEG signals;

[0024] S4 specifically includes:

[0025] S4.1: The extracted feature vectors from the dual-layer network exhibit time-varying characteristics, encompassing the temporal information of EEG signals during a 6-second duration of motor imagery in a single acquisition task. Simultaneously, the acquisition leads cover various brain regions associated with motor imagery, and two rhythms are extracted for composite modeling. This allows the extracted feature vectors to reflect the interactive information of motor imagery in the spatial dimension, covering the spatial structural relationships of information changes during brain activity.

[0026] S4.2: The specific feature vectors are as follows:

[0027] Single-layer feature vector Fs, inter-layer feature vector Fm, and two-layer feature vector F all

[0028] (1)

[0029] The calculation methods for each eigenvalue in the eigenvector are as follows:

[0030] 1) Node degree:

[0031] (2)

[0032] In equation (2), K L The average node degree of layer L. K is the number of connections established between node i and its neighboring nodes in the same layer, calculated over layer μ. μ It is the average degree of all network nodes in a μ-rhythmic single-layer network. Similarly, K is obtained by performing the calculation on the β layer. β This is the average degree of all nodes in the β-layer network;

[0033] 2) Clustering coefficient

[0034] As shown in equation (3), in a single-layer network, the node clustering coefficient It is the proportion of adjacent nodes connected to node i;

[0035] (3)

[0036] In equation (3), It is the degree of a single-layer network node i. The clustering coefficient CC of the network can be obtained by averaging the clustering coefficients of the network nodes, where N is the number of nodes, as shown in equation (3):

[0037] (4)

[0038] 3) Average path length

[0039] Average path length It expresses the average distance between any two nodes within a certain layer, characterizing the information transmission capability of a certain layer in a single-layer network, as shown in equation (5):

[0040] (5)

[0041] 4) Multilevel clustering coefficient

[0042] In multi-layered multiplexing networks, the same node has different clustering coefficients in different layers. In a multi-layered network where nodes are aligned (i.e., only duplicate nodes exist as connections), the multi-layered clustering coefficient of node i can be expressed as:

[0043] (6)

[0044] In equation (6), m is the number of layers in the multilayer multiplexing network. It is the connection between nodes i and j within the same level. Let i be a node and m be the connection between different layers. If the edge centered at node i can form a triangle between layers, then the multiplexing clustering coefficient of node i is 1, otherwise it is 0. The multi-layer clustering coefficient of the network can be obtained by averaging the MCC of all nodes in the multi-layer network, as shown in equation (7):

[0045] (7)

[0046] 5) Multi-level participation coefficient

[0047] In multilayer networks, the equivalent node degree is measured by overlap, which is the weighted sum of the node degrees of each layer. A multilayer participation coefficient is introduced to quantify the degree of a node's participation in information exchange across different layers, which can be expressed as:

[0048] (8)

[0049] In equation (8), The overlap of node i is the sum of the degrees of node i in each layer. It represents the proportion of node degree in each layer. This index characterizes whether the connectivity of node i is uniform across layers. =1 means that i has the same degree in all layers, otherwise it is concentrated in one layer. =0, similar to the multi-level clustering coefficient of a network, the multi-level participation coefficient of a network can be expressed as:

[0050] (9)

[0051] S5: Apply machine learning and other methods to perform motion image recognition using the obtained composite feature vectors. Specifically, this includes selecting a specific machine learning method for the recognition task. 75 samples from each of the different types of motion image tasks are selected as the training set, and the remaining 75 samples are used as the test set. Fall is used as the feature vector input for training, and recognition is performed on the test set. The final output is the motion image recognition accuracy.

[0052] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0053] Based on prior knowledge of neurophysiology and motor imagery, this invention extracts the μ and β rhythms, which are prominent and exhibit active ERS / ERD phenomena, from motor imagery to construct a multi-layered reusable functional brain network. This avoids the problem of poor analysis results caused by excessive redundant information in full-frequency modeling, while preserving the collaborative working information of the brain during motor imagery. Compared to existing motor imagery recognition methods that extract features from single-layer networks, this invention's two-layer brain network modeling method fully considers the mutual influence between different brain regions, supplements the spatiotemporal correlation between different rhythmic signals in EEG, and completes the functional structure and information transmission between layers, resulting in a higher recognition rate. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a map showing the location distribution of the acquired electroencephalogram (EEG) signals in the motor imagery recognition method based on a multi-layered coupled brain network across rhythms, as described in this invention.

[0056] Figure 2 This is a schematic diagram of the principle of the present invention;

[0057] Figure 3 This is the experimental data collection paradigm of the present invention;

[0058] Figure 4 is a schematic diagram of the two-layer network of the present invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] like Figure 2 As shown, this embodiment includes the following steps:

[0061] S1: Collect EEG signals from individuals exhibiting motor imagery in a resting state across multiple (target) categories and perform preprocessing. The specific process is as follows:

[0062] S1.1: Subjects were informed in advance that the sampling task involved left and right hand motor imagery. Ten minutes before the sampling, subjects A and B remained at rest, wearing EEG caps, sitting in chairs, and adjusting their mood to maintain steady breathing. The EEG caps had 41 electrodes, distributed according to the 10 / 20 international standard, and the sampling frequency was 100Hz.

[0063] S1.2: Data collection begins. The monitor remains black for 2 seconds, followed by a random motor imagery instruction displayed for 4 seconds. The subject performs motor imagery according to the instruction for 4 seconds. Five sets of data are collected in this cycle, followed by a 30-second rest period. After the rest period, data collection is repeated, with 150 sets of motor imagery collected for each hand, for a total of 300 sets of signals.

[0064] S1.3: Save the obtained data as a .mat file and input it into EEGLAB, a MATLAB-based application, for processing.

[0065] S1.4: Import data, locate the 10 / 20 standard lead electrodes, retain the 41 leads required for data analysis, and remove invalid electrodes.

[0066] S2: Extract the μ rhythm (8-14Hz) and β rhythm (14-30Hz) EEG signals from the signal, and remove physiological artifacts. The specific steps are as follows:

[0067] S2.1: Perform common averaging on the original signal.

[0068] S2.2: Using an FIR filter, bandpass filtering is performed to obtain the μ rhythm (8-14Hz) and β rhythm (14-30Hz). Independent component analysis is then performed on the obtained filters to remove significant noise signals such as electromyography and electrooculography, resulting in the processed signal xμ(t)∈R. 41*90000 With xβ(t)∈R 41*90000 .

[0069] S3: Based on the multi-layer reuse network modeling structure and prior knowledge of motor imagery, a two-layer functional brain network based on μ rhythm and β rhythm is constructed. The specific process is as follows:

[0070] S3.1: Export the processed signal, calculate the connection strength using phase-locked loop (PLL) values, and calculate the PLL values ​​between each pair of leads in each rhythm as intra-layer connections, resulting in two single-layer networks, μ and β. Calculate the PLL values ​​between the same lead in μ and β rhythms as inter-layer connections. This yields two 41*41 single-layer correlation matrices and one 41*41 inter-layer correlation matrix.

[0071] S3.2: The correlation matrix is ​​transformed into a functional network based on the threshold K. The threshold K is chosen to satisfy K>2ln41≈7.7, where 41 is the number of network nodes. If the strength between nodes is greater than the set threshold, a connection is established between the two leads, and the corresponding element of the superadjacency matrix is ​​set to 1; otherwise, no connection is established, and the corresponding element of the superadjacency matrix is ​​set to 0. Finally, a two-layer functional network of μ rhythm and β rhythm can be obtained.

[0072] S4: Extract the features of the two-layer functional brain network that reflect different types of motor imagery, and construct a composite feature vector based on the multi-rhythm information of EEG signals.

[0073] S4.1: Based on the preferred content mentioned in S4 of the invention, multi-layer functional brain network features are extracted. The network features of subjects A and B are shown in Table 1.

[0074] Table 2. Characteristic functional parameters of single-layer and multi-layer networks for the subjects under different rhythms.

[0075]

[0076] Table 1 shows that the network features were not identical when subjects performed different motor imagery actions with their left and right hands. Furthermore, the network features exhibited even more significant differences under different rhythms for the same motor imagery action. This demonstrates that different motor imagery actions have a certain degree of discriminative power on the two-layer functional brain network established in this paper. Combining the obtained network features yields a single-layer feature vector Fs, an inter-layer feature vector Fm, and a multi-layer feature vector Fall.

[0077] S5: The extracted features were used for a motion imagery recognition task. This experiment employed a Support Vector Machine (SVM) classifier and a 5-fold cross-validation method for verification. From the sample data of each participant (A and B), 75 sets of left-hand and right-hand motion imagery samples were randomly selected, totaling 150 sets as the training set, and the remaining 150 sets as the test set. The RBF kernel function was used, and the penalty factor C and parameter g were selected using a traversal method to achieve the highest average recognition rate. The classification was repeated 20 times, and the average correct recognition rate of the 20 experiments was recorded. The 5-fold cross-validation results are shown in Table 2.

[0078] Table of average correct recognition rate and variance (%) for different feature vectors

[0079]

[0080] In use When classifying using feature vectors, the average recognition rate reached 86.59%, with a variance of 1.66%, and the highest recognition rate reached 87.95%.

[0081] Step Six: Validate and compare this method with commonly used methods on the same data, selecting the same Support Vector Machine classifier. The results are shown in Table 3.

[0082] Comparison of average correct recognition rates of different feature extraction methods

[0083]

[0084] As shown in Table 3, the accuracy rates of the power spectral density algorithm, the common space mode, the filter bank common space mode, and the Hilbert-Huang transform method are 73.13%, 70.90%, 80.88%, and 84.70%, respectively, all lower than the multilayer network feature extraction method proposed in this paper, proving the effectiveness of the proposed multilayer network feature extraction method. The construction of the multilayer network and the completion of inter-layer information can effectively improve the recognition accuracy of left and right hand motor imagery.

[0085] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and these variations still fall within the protection scope of the present invention.

Claims

1. A method for recognizing motor imagery based on a multi-layered brain network with cross-rhythm coupling of EEG signals, characterized in that: Includes the following steps: S1: Collect EEG signals of multiple target categories of motor imagery from an individual in a resting state and perform preprocessing. S2: Extract the μ-rhythm and β-rhythm EEG signals from the signal, remove physiological artifacts, μ-rhythm is 8-14 Hz, β-rhythm is 14-30 Hz; S3: Based on the multi-layer reuse network modeling structure and prior knowledge of motor imagery, a two-layer functional brain network based on μ rhythm and β rhythm is constructed. S4: Extract the features of a two-layer functional brain network that reflect different types of motor imagery, and construct a composite feature vector based on multi-rhythm information of EEG signals; S4 includes: S4.1: The extracted feature vectors of the two-layer network exhibit time-varying characteristics; the collected leads cover various brain regions associated with motor imagery in the human brain, and two rhythms are extracted for composite modeling. S4.2: The specific feature vectors are as follows: Single-layer feature vector Fs, inter-layer feature vector Fm, and two-layer feature vector F all , （1） The calculation methods for each eigenvalue in the eigenvector are as follows: 1) Node degree: （2） K L The average node degree of layer L. K is the number of connections established between node i and its neighboring nodes in the same layer, calculated over layer μ. μ It is the average degree of all network nodes in a μ-rhythmic single-layer network. Similarly, K is obtained by performing the calculation on the β layer. β This is the average degree of all nodes in the β-layer network; 2) Clustering coefficient In a single-layer network, the node clustering coefficient It is the proportion of adjacent nodes connected to node i; （3） It is the degree of a single-layer network node i. For the edges connecting nodes i and j within the same layer, the average of the clustering coefficients of the network nodes yields the network's clustering coefficient CC, where N is the number of nodes. （4） 3) Average path length Average path length It expresses the average distance between any two nodes within a certain layer, characterizing the information transmission capability of a certain layer in a single-layer network, as shown in equation (5): （5） 4) Multilevel clustering coefficient In a multi-layered multiplexing network, in a multi-layered network where nodes are aligned (i.e., only replicated nodes exist as connections), the multi-layered clustering coefficient of node i is: （6） In equation (6), m is the number of layers in the multilayer multiplexing network. It is the connection between nodes i and j within the same level. Let node i be the connection between different layers and m be the connection between different layers. If the edge centered at node i forms a triangle between layers, then the multiplexing clustering coefficient of node i is 1, otherwise it is 0. The multi-layer clustering coefficient of the network can be obtained by averaging the MCC of all nodes in the multi-layer network, as shown in equation (7): （7） 5) Multi-level participation coefficient In multi-layer networks, the equivalent node degree is measured by overlap. A multi-layer participation coefficient is introduced to quantify the degree of node participation in information exchange across different layers, as follows: （8） In equation (8), The overlap of node i is the sum of the degrees of node i in each layer. It represents the proportion of node degree in each layer, and the index characterizes whether the connectivity of node i is uniform across layers. =1 means that i has the same degree in all layers, otherwise it is concentrated in one layer. =0, similar to the multi-level clustering coefficient of the network, the multi-level participation coefficient of the network is: （9） S5: Apply machine learning and other methods to the obtained composite feature vectors for motion image recognition.

2. The method for motor imagery recognition based on a multi-layered, cross-rhythm coupled brain network of electroencephalogram signals according to claim 1, characterized in that: S1 specifically includes: S1.1: Preparation: The subject agrees with the type of motor imagery to be collected in this experiment. After being fully relaxed, the subject remains at rest, wears an EEG cap and sits in front of the monitor. The EEG cap uses 41-channel electrodes, which conforms to the international 10 / 20-lead standard, and the EEG signal sampling frequency is 100 Hz. S1.2: Start collecting EEG signals. The monitor remains black for the first 2 seconds. In the 3rd second, the monitor displays the task for this collection. The subject performs different types of motor imagery according to the text prompts for 4 seconds. In the 6th second, the monitor returns to black and the motor imagery ends. S1.3: After completing one signal acquisition, repeat S1.1 and S1.

2. After 5 sets of signal acquisitions, the subject rests and relaxes for 30 seconds before repeating 1-1 and 1-2. 150 sets of signals are collected for each category of motor imagery task before the acquisition ends. S1.4: Import the obtained EEG signals into the EEGLab tool on the MATLAB platform, define the area covered by the electrode corresponding to each EEG lead as a node, complete the localization of all electrode nodes, retain the 41 nodes required for modeling, and remove invalid nodes.

3. The method for motor imagery recognition based on a multi-layered, cross-rhythm coupled brain network of electroencephalogram signals according to claim 1, characterized in that: S2 specifically includes: S2.1: A common average reference is used for preprocessed EEG signals in eeglab; S2.2: Use an FIR filter to perform bandpass filtering on the preprocessed signal to extract μ-rhythm and β-rhythm motor imagery EEG signals; S2.3: The final modeling signal x is obtained. μ (t)∈R N*K With x β (t)∈R N*K Where N is the number of leads and K is the number of sampling points.

4. The method for motor imagery recognition based on a multi-layered, cross-rhythm coupled brain network of electroencephalogram signals according to claim 1, characterized in that: S3 specifically includes: S3.1: Each lead corresponds to a network node. The phase-locked value between any two nodes in the same layer is used to quantify the strength between nodes as the network connection to construct a single-layer network. Single-layer networks for μ rhythm and β rhythm are constructed respectively. The phase-locked value between nodes with the same name in different layers is used to quantify the inter-layer connection strength between nodes with the same name in different rhythms in μ rhythm and β rhythm networks to construct inter-layer connections, resulting in two single-layer network correlation matrices and one inter-layer connection correlation matrix. S3.2: Use a threshold to determine the node connection relationship and construct a two-layer multiplexed functional brain network. The threshold K should satisfy K>2lnN≈7.7, where N is the number of network nodes. If the strength between nodes is greater than the set threshold, a connection is established between the two leads and the corresponding element of the superadjacency matrix is ​​set to 1. Otherwise, no connection is established and the corresponding element of the superadjacency matrix is ​​set to 0. Finally, a two-layer functional network of μ rhythm and β rhythm can be obtained.

Images