Motor imagery ability grading method and device based on multi-band coupling characteristics

Abstract

The application discloses a motor imagery ability grading method and device based on multi-band coupling features, and relates to the field of data processing, which comprises the following steps: obtaining original electroencephalogram signals collected by a to-be-evaluated person in motor imagery and rest stages, and performing pretreatment and data screening to obtain processed electroencephalogram signals; constructing multi-band coupling features based on the processed electroencephalogram signals; adopting a principal component analysis algorithm on the multi-band coupling features, and performing dimension reduction with the optimal principal component number to obtain corresponding dimension reduction features; inputting the dimension reduction features into a trained motor imagery ability grading model to obtain a motor imagery ability grading probability, wherein the motor imagery ability grading probability is the probability that the to-be-evaluated person belongs to a high motor imagery ability level or a low motor imagery ability level, and the motor imagery ability level to which the to-be-evaluated person belongs is determined based on the motor imagery ability grading probability. The application solves the problems of difficulty in accurately grading motor imagery ability and high-dimensional data redundancy.

Description

Technical Field

[0001] This invention relates to the field of data processing, and specifically to a method and apparatus for classifying motion imagination ability based on multi-band coupling characteristics. Background Technology

[0002] Brain-computer interface (BCI) technology establishes a direct communication and control channel between the human brain and external devices by collecting neural activity signals from the brain. Motor imagery (MI) refers to the cognitive process in which a subject mentally simulates limb movements without producing actual muscle activity. Research shows that motor imagery can activate cortical regions similar to actual movements (such as the sensorimotor cortex) and produce typical event-related desynchronization / synchronization (ERD / ERS) phenomena. Therefore, motor imagery BCI (MI-BCI) has great application potential in areas such as neurorehabilitation training for stroke patients, assistive control for people with disabilities, and neurofeedback modulation. Current motor imagery grading faces the following technical challenges:

[0003] (1) Significant individual differences (BCI illiteracy): Approximately 15%-30% of the population cannot produce sufficiently significant EEG characteristics (BCI illiteracy). Existing assessment methods mostly rely on the energy decay of a single frequency band (such as only looking at the Mu rhythm), making it difficult to accurately distinguish whether "low-ability individuals" are due to fatigue or weak physiological characteristics.

[0004] (2) Coarse frequency band division: Traditional methods often treat 8-30Hz as a wide frequency band, ignoring the different neural roles played by Alpha wave (8-13Hz), Beta1 wave (13-20Hz) and Beta2 wave (20-30Hz) in motion control.

[0005] (3) High-dimensional data redundancy: Although high-density acquisition (such as 60 leads) is rich in information, it brings huge computational burden and dimensionality curse, and is not suitable for rapid evaluation of scenarios. Summary of the Invention

[0006] The purpose of this application is to propose a method and device for classifying motion imagination ability based on multi-band coupling characteristics, so as to solve the problems of unclear correlation between cross-frequency characteristics and motion imagination ability, strong equipment dependence, and large computational load of algorithms in the prior art.

[0007] In a first aspect, the present invention provides a method for classifying motion imagination ability based on multi-band coupling features, comprising the following steps:

[0008] A motor imagination ability classification model based on a convolutional neural network is constructed and trained to obtain a trained motor imagination ability classification model. During the training process of the motor imagination ability classification model, an improved particle swarm optimization algorithm is used to determine the optimal number of principal components to be used in the principal component analysis algorithm based on the training data.

[0009] Raw EEG signals collected from the individuals to be assessed during the motor imagery and rest phases were acquired, preprocessed, and filtered to obtain processed EEG signals. Multi-band coupling features were constructed based on the processed EEG signals. Principal component analysis (PCA) was used to reduce the dimensionality of these multi-band coupling features to the optimal number of principal components, yielding the corresponding dimensionality-reduced features. These dimensionality-reduced features were then input into a trained motor imagery ability grading model to obtain the motor imagery ability grading probability. This probability represents the likelihood that the individual to be assessed belongs to either a high or low level of motor imagery ability. Based on this probability, the individual's motor imagery ability level was determined.

[0010] Preferably, multi-band coupling features are constructed based on the processed EEG signals, specifically including:

[0011] The processed EEG signal is decomposed into the amplitude of the Alpha band signal and the amplitude of the Beta band signal. The Beta band signal includes the Beta1 band signal and the Beta2 band signal.

[0012] The event-related desynchronization index for each EEG channel in each frequency band is calculated based on the amplitude of the Alpha band signal and the amplitude of the Beta band signal, as shown in the following formula:

[0013] ;

[0014] in, This represents the event-related desynchronization index of the c-th EEG channel in the f-th frequency band. This represents the mean square value of the amplitude of the c-th EEG channel under the f-th frequency band signal collected during the resting phase. This represents the mean square value of the amplitude of the c-th EEG channel under the f-th frequency band signal acquired during the motor imagery phase;

[0015] The processed EEG signal was analyzed using Hilbert transform to obtain an analytical signal. The instantaneous phase of the Alpha band signal and the instantaneous amplitude of the Beta band signal were then extracted from the analytical signal, as shown in the following formula:

[0016] ;

[0017] in, This represents the signal of each EEG channel in the processed EEG signal at time t. This represents the analytic signal at time t. Represents the Hilbert transform operator. Let represent the instantaneous amplitude of the Beta band signal at time t. This represents the instantaneous phase of the Alpha band signal at time t.

[0018] A phase-amplitude distribution histogram is constructed based on the instantaneous phase of the Alpha band signal and the instantaneous amplitude of the Beta band signal. The range of instantaneous phase values ​​of the Alpha band signal is divided into N phase intervals. The average instantaneous amplitude of the Beta band signal falling within each phase interval is calculated. The amplitude distribution probability within each phase interval is calculated based on the average instantaneous amplitude of the Beta band signal within each phase interval, as shown in the following formula:

[0019] ;

[0020] in, This represents the average instantaneous amplitude of the Beta band signal within the j-th phase interval. This represents the average instantaneous amplitude of the Beta band signal within the l-th phase interval. This represents the amplitude distribution probability within the j-th phase interval;

[0021] By calculating the Kullback-Leibler distance between the amplitude distribution probability and the uniform distribution, and then calculating the modulation index of the Kullback-Leibler divergence in each EEG channel, as shown in the following formula:

[0022] ;

[0023] ;

[0024] in, Shannon entropy represents the probability distribution of amplitude across all phase intervals. The maximum entropy value under a uniform distribution is expressed as follows: = log N, where MI represents the modulation index of the Kullback-Leibler divergence;

[0025] Multi-band coupling features were constructed by combining the event-related desynchronization index and the modulation index of Kullback-Leibler divergence in three frequency bands for each EEG channel.

[0026] As a preferred method, an improved particle swarm optimization algorithm is used to determine the optimal number of principal components to be used in the principal component analysis algorithm based on the training data, specifically including:

[0027] M particles are randomly generated, and the position of each particle represents a potential principal component number k.

[0028] In each iteration, the multi-band coupling features in the training data are input into the principal component analysis algorithm, the k principal components corresponding to the position of each particle are retained, and the corresponding dimensionality reduction matrix is ​​constructed. The fitness value of each particle is calculated based on the dimensionality reduction matrix.

[0029] If the fitness value calculated in the i-th iteration exceeds the individual's historical best value, then the fitness value calculated in the i-th iteration will be updated to the individual's historical best value.

[0030] If the fitness value calculated in the i-th iteration exceeds the global optimum, then the fitness value calculated in the i-th iteration will be updated to the global optimum.

[0031] In each iteration, the velocity and position of each particle are updated based on the individual historical best value and the global best value;

[0032] Once the iteration count meets the iteration termination condition, the position of the particle corresponding to the global optimal value is output as the optimal principal component number k'.

[0033] As a preferred method, the fitness value of each particle is calculated as follows:

[0034] ;

[0035] in, Represents the trace of a matrix. This represents the fitness value of the dimensionality-reduced matrix constructed with principal component number k. The inter-class scatter matrix, The formulas for calculating the intra-class scatter matrix are as follows:

[0036] ;

[0037] ;

[0038] in, The eigenvectors of each row in the dimensionality-reduced matrix are represented as individual samples. These eigenvectors include the event-related desynchronization index for all three frequency bands across all EEG channels and the modulation index of the Kullback-Leibler divergence. μ is the mean of the entire sample. Classification is based on the motor imagery ability grading labels in the training data, and P represents the total number of classes. Let be the mean of the p-th class of samples. Let p be the number of samples in class p.

[0039] The inertia weight in the particle velocity update formula adopts a linearly decreasing inertia weight, as shown in the following formula:

[0040] ;

[0041] in, and These represent the initial weight and the final weight, respectively. Indicates the maximum number of iterations. This represents the inertial weight used to update the particle's velocity during the i-th iteration.

[0042] As a preferred option, the preprocessing process includes downsampling, bandpass filtering, interpolation repair of channels with substandard signal quality, and whole-brain average rereference.

[0043] The data filtering process is as follows:

[0044] Extract m-second core EEG signals from the preprocessed EEG signals corresponding to each frequency of motor imagery stage;

[0045] The m-second core EEG signals corresponding to multiple frequencies of motor imagery stages are removed to obtain the processed EEG signals, which correspond to frequencies with severe eye movement artifacts or inattention.

[0046] As a preferred embodiment, the motor imagination ability classification model includes a convolutional layer, a ReLU activation function layer, a max pooling layer, a fully connected layer, and a softmax function layer connected in sequence.

[0047] Secondly, the present invention provides a motion imagination ability grading device based on multi-band coupling characteristics, comprising:

[0048] The model building module is configured to build and train a motor imagination ability classification model based on a convolutional neural network to obtain a trained motor imagination ability classification model. During the training process of the motor imagination ability classification model, an improved particle swarm algorithm is used to determine the optimal number of principal components to be used in the principal component analysis algorithm based on the training data.

[0049] The grading module is configured to acquire raw EEG signals collected from the individuals to be assessed during the motor imagery and rest phases, perform preprocessing and data filtering to obtain processed EEG signals; construct multi-band coupling features based on the processed EEG signals, and use principal component analysis (PCA) algorithm on the multi-band coupling features to reduce dimensionality to the optimal number of principal components, obtaining the corresponding dimensionality-reduced features; input the dimensionality-reduced features into a trained motor imagery ability grading model to obtain the motor imagery ability grading probability, which is the probability that the individual to be assessed belongs to the high or low level of motor imagery ability, and determine the motor imagery ability level to which the individual to be assessed belongs based on the motor imagery ability grading probability.

[0050] Thirdly, the present invention provides an electronic device including one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any implementation of the first aspect.

[0051] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any of the implementations of the first aspect.

[0052] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the method as described in any of the implementations in the first aspect.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] (1) The motor imagery ability grading method based on multi-band coupling features proposed in the embodiments of this application can effectively remove noise caused by fatigue and attention distraction, and significantly improve the robustness of training the motor imagery ability grading model. By introducing the multi-band coupling features of Alpha, Beta1 and Beta2 signals, the differences in brain activity among subjects with different motor imagery abilities can be successfully captured.

[0055] (2) The motor imagery ability classification method based on multi-band coupling features proposed in the embodiments of this application solves the noise interference problem in the data through a specific screening mechanism. Multi-band coupling features are constructed by three frequency band signals: Alpha, Beta1, and Beta2. The corresponding dimensionality reduction features obtained by using the principal component analysis algorithm to reduce the dimensionality of the data with the optimal number of principal components are used as the input features of the motor imagery ability classification model. Compared with single ERD features, it can capture the differences in the neural control ability of the subjects better, and realize the rapid and accurate classification of motor imagery ability.

[0056] (3) The motor imagination ability classification method based on multi-band coupling features proposed in the embodiments of this application introduces a motor imagination ability classification model with a lightweight convolutional neural network structure. This not only solves the problem of computational redundancy of high-dimensional EEG signals, but also completely solves the problem of misjudging "threshold individuals" in traditional methods, and can provide an absolutely reliable classification basis for rehabilitation training. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0058] Figure 1 This is a flowchart illustrating a motion imagination ability grading method based on multi-band coupling features, as an embodiment of this application.

[0059] Figure 2 A sequence trend graph of the subject scores of the subject in the motor imagery ability grading method based on multi-band coupling features, which is an embodiment of this application;

[0060] Figure 3 A scatter plot of the subject scores of the subject in the motor imagery ability grading method based on multi-band coupling features, which is an embodiment of this application.

[0061] Figure 4 Box plot of subject scores for a motor imagery ability grading method based on multi-band coupling features, as an embodiment of this application;

[0062] Figure 5 Histogram of subject scores for a motor imagery ability grading method based on multi-band coupling features, as an embodiment of this application;

[0063] Figure 6 A PCA projection scatter plot visualizing the motion imagination ability classification method based on multi-band coupling characteristics, which is an embodiment of this application.

[0064] Figure 7 The convergence curve of the training of the motion imagination ability classification model of the motion imagination ability classification method based on multi-band coupling features in the embodiments of this application;

[0065] Figure 8 The feature distribution difference diagram of the motion imagination ability classification model based on multi-band coupling feature in the embodiment of this application is shown in the test set between the high ability group and the low ability group.

[0066] Figure 9 The classification confusion matrix diagram of the motion imagination ability classification model based on multi-band coupling features in the embodiments of this application on the test set;

[0067] Figure 10 This is a schematic diagram of a motion imagination ability grading device based on multi-band coupling characteristics, as an embodiment of this application.

[0068] Figure 11 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0070] Figure 1 The embodiment of this application illustrates a method for classifying motion imagination ability based on multi-band coupling characteristics, comprising the following steps:

[0071] S1. Construct and train a motor imagination ability classification model based on a convolutional neural network to obtain a trained motor imagination ability classification model. During the training process of the motor imagination ability classification model, the optimal number of principal components to be used in the principal component analysis algorithm is determined based on the training data using an improved particle swarm optimization algorithm.

[0072] In a specific embodiment, the preprocessing process includes downsampling, bandpass filtering, interpolation repair of channels with substandard signal quality, and whole-brain average rereference.

[0073] The data filtering process is as follows:

[0074] Extract m-second core EEG signals from the preprocessed EEG signals corresponding to each frequency of motor imagery stage;

[0075] The m-second core EEG signals corresponding to multiple frequencies of motor imagery stages are removed to obtain the processed EEG signals, which correspond to frequencies with severe eye movement artifacts or inattention.

[0076] Specifically, in the embodiments of this application, a motor imagination ability classification model based on a convolutional neural network is first constructed. The process of constructing the training data used in the training of this motor imagination ability classification model is as follows:

[0077] S11: 20 subjects were selected from the subject database, and electrodes with 60 EEG channels were arranged in strict accordance with the international 10-20 system for data collection. The initial sampling rate was set to 1000Hz.

[0078] The experimental paradigm was set up, which included four motor imagery tasks: left hand, right hand, foot, and tongue.

[0079] The training process for a single frequency of motor imagery is as follows: screen prompts to start, enters a 7-second imagery phase, prompts to rest, enters a 5-second rest phase, and proceeds to the next training step.

[0080] Experimental procedure: A total of 5 groups (Blocks) were conducted, each group containing 40 motor imagery training sessions. In each motor imagery training session, 4 motor imagery tasks appeared randomly 10 times. Each subject completed a total of 200 motor imagery training sessions. The frequency of each motor imagery task was 50 times. During each frequency of motor imagery training, raw EEG signals from 60 EEG channels were collected simultaneously as training samples.

[0081] S12 involves preprocessing and filtering the raw EEG data from the training samples. The preprocessing process in one example is as follows:

[0082] The raw EEG data was downsampled to 250Hz and bandpass filtered from 1 to 40Hz to cover the core rhythms. Channels with substandard signal quality were repaired by interpolation and the signal-to-noise ratio was improved by whole-brain average rereference (CAR) to obtain the preprocessed EEG signal.

[0083] The data filtering process in one example is as follows:

[0084] The core 3-second EEG signal of the imagination phase was extracted from the preprocessed EEG signal, forming a data segment of 3s × 250Hz = 750 time points.

[0085] The quality of 3-second core EEG signals at 50 frequencies for each motor imagery task for each subject was assessed. Frequencies with severe eye movement artifacts or inattention were removed. Finally, 35 high-quality frequencies of 3-second core EEG signals were manually selected and retained to obtain the processed EEG signals.

[0086] In a specific embodiment, multi-band coupling features are constructed based on the processed EEG signals, specifically including:

[0087] The processed EEG signal is decomposed into the amplitude of the Alpha band signal and the amplitude of the Beta band signal. The Beta band signal includes the Beta1 band signal and the Beta2 band signal.

[0088] The event-related desynchronization index for each EEG channel in each frequency band is calculated based on the amplitude of the Alpha band signal and the amplitude of the Beta band signal, as shown in the following formula:

[0089] ;

[0090] in, This represents the event-related desynchronization index of the c-th EEG channel in the f-th frequency band. This represents the mean square value of the amplitude of the c-th EEG channel under the f-th frequency band signal collected during the resting phase. This represents the mean square value of the amplitude of the c-th EEG channel under the f-th frequency band signal acquired during the motor imagery phase;

[0091] The processed EEG signal was analyzed using Hilbert transform to obtain an analytical signal. The instantaneous phase of the Alpha band signal and the instantaneous amplitude of the Beta band signal were then extracted from the analytical signal, as shown in the following formula:

[0092] ;

[0093] in, This represents the signal of each EEG channel in the processed EEG signal at time t. This represents the analytic signal at time t. Represents the Hilbert transform operator. Let represent the instantaneous amplitude of the Beta band signal at time t. This represents the instantaneous phase of the Alpha band signal at time t.

[0094] A phase-amplitude distribution histogram is constructed based on the instantaneous phase of the Alpha band signal and the instantaneous amplitude of the Beta band signal. The range of instantaneous phase values ​​of the Alpha band signal is divided into N phase intervals. The average instantaneous amplitude of the Beta band signal falling within each phase interval is calculated. The amplitude distribution probability within each phase interval is calculated based on the average instantaneous amplitude of the Beta band signal within each phase interval, as shown in the following formula:

[0095] ;

[0096] in, This represents the average instantaneous amplitude of the Beta band signal within the j-th phase interval. This represents the average instantaneous amplitude of the Beta band signal within the l-th phase interval. This represents the amplitude distribution probability within the j-th phase interval;

[0097] The Kullback-Leibler distance between the amplitude distribution probability and the uniform distribution is calculated, and the modulation index of the Kullback-Leibler divergence in each EEG channel is calculated as follows:

[0098] ;

[0099] ;

[0100] in, Shannon entropy represents the probability distribution of amplitude across all phase intervals. The maximum entropy value under a uniform distribution is expressed as follows: = log N, where MI represents the modulation index of the Kullback-Leibler divergence;

[0101] Multi-band coupling features were constructed by combining the event-related desynchronization index and the modulation index of Kullback-Leibler divergence for each EEG channel in three frequency bands.

[0102] Specifically, the processed EEG signals are decomposed into three specific sub-frequency bands: Alpha band, Beta1 band, and Beta2 band. The Alpha band, with a frequency range of 8-13 Hz, is primarily used to characterize inhibition or idle states; the Beta1 band, with a frequency range of 13-20 Hz, is primarily used to characterize low-frequency motor activation; and the Beta2 band, with a frequency range of 20-30 Hz, is primarily used to characterize high-frequency motor commands. Both the Beta1 and Beta2 bands belong to the Beta band.

[0103] For 60 collected EEG channels, the Event-Related Desynchronization (ERD) index was calculated in three frequency bands (Alpha, Beta1, and Beta2) to quantify the activation level of the cerebral cortex. The formula for calculating the Event-Related Desynchronization Index is as follows: The corresponding reference period energy specifically refers to the mean square value of the amplitude of the EEG channel in the corresponding frequency band during the rest phase (the baseline period before the stimulus appears, such as t=-1s ~ 0s) in the experimental paradigm. The energy corresponding to the imagination phase specifically refers to the value within the key time window (i.e., the 3-second core EEG signal) selected during the imagination phase. Its value is the mean square value of the amplitude of the EEG channel in the corresponding frequency band. In the calculation result of this formula, a negative value indicates a decrease in energy relative to the baseline (i.e., the ERD phenomenon), indicating a reduction in synchronous firing of the neuronal population in the motor cortex, indicating an activated state; a positive value indicates an increase in energy (i.e., the ERS phenomenon), indicating that the cortex is in an inhibited or idle state. Through calculation, a 60×3 = 180-dimensional energy feature vector is finally constructed, where 60 represents the number of channels and 3 represents the three frequency band signals.

[0104] Cross-frequency coupling characteristics: To quantify the modulatory effect of low-frequency rhythms on high-frequency activity, embodiments of this application focus on calculating the phase-amplitude coupling (PAC) strength between the phase of the Alpha band signal and the amplitude of the Beta1 and Beta2 band signals of key EEG channels (C3, C4, Cz, etc.) in the motor cortex. Specifically, the modulation index (MI) based on Kullback-Leibler divergence is used for calculation, and the steps are as follows:

[0105] First, the instantaneous phase of the Alpha band signal is extracted using the Hilbert Transform. Instantaneous amplitude of Beta band signals .

[0106] Secondly, a phase-amplitude distribution histogram is constructed to represent the instantaneous phase of the Alpha band signal. The range of values ​​for [- , Divide the signal into N uniform phase intervals (Bin). Calculate the average instantaneous amplitude of the Beta band signal falling within the j-th phase interval, denoted as […]. Subsequently, the average instantaneous amplitude of the Beta band signal within the j-th phase interval is normalized to obtain the amplitude distribution probability P(j).

[0107] Finally, the modulation index MI of the Kullback-Leibler divergence is obtained by calculating the Kullback-Leibler distance between this distribution and the uniform distribution.

[0108] The modulation index MI of the Kullback-Leibler divergence ranges from [0, 1]. MI=0 indicates that the instantaneous amplitude of the Beta band signal is uniformly distributed across the instantaneous phase of the Alpha band signal, and there is no coupling. The closer MI is to 1, the more concentrated the instantaneous amplitude of the Beta band signal is on a specific instantaneous phase (such as a peak or trough) of the Alpha band signal, indicating a significant cross-frequency modulation effect.

[0109] By combining the event-related desynchronization index of 60 EEG channels in three frequency bands with the modulation index of Kullback-Leibler divergence, a multi-band coupling feature with a dimension of 60×4 was constructed.

[0110] The following process was used to obtain the motor imagery ability rating labels for each subject in the training data:

[0111] Six discrimination pairs (left hand-right hand, left hand-tongue, tongue-foot, etc.) were constructed for four types of motor imagery tasks.

[0112] The average classification accuracy of each subject across 6 discriminant pairs was calculated, and the 20 subjects were divided into a "high-ability group" and a "low-ability group" based on this standard.

[0113] like Figure 2 and Figure 3 As shown, although the performance of the subjects fluctuated, there was a clear center of distribution overall. The embodiments of this application used the average accuracy of all subjects (shown by the red dashed line in the figure, approximately 0.6319) as an adaptive threshold.

[0114] Subjects with an accuracy rate greater than 0.6319 were labeled as “High Ability”;

[0115] Subjects with an accuracy of 0.6319 or less were labeled as “Low Ability”.

[0116] Figure 4 and Figure 5 Histograms and box plots further validated the normal distribution of the data, demonstrating the statistical rationality of using the mean as the grading limit for motor imagery ability in the embodiments of this application. Training data was constructed using the multi-band coupling features of each subject obtained through the above steps and the corresponding motor imagery ability grading labels.

[0117] In a specific embodiment, an improved particle swarm optimization algorithm is used to determine the optimal number of principal components to be used in the principal component analysis algorithm based on the training data, specifically including:

[0118] M particles are randomly generated, and the position of each particle represents a potential principal component number k.

[0119] In each iteration, the multi-band coupling features in the training data are input into the principal component analysis algorithm, the k principal components corresponding to the position of each particle are retained, and the corresponding dimensionality reduction matrix is ​​constructed. The fitness value of each particle is calculated based on the dimensionality reduction matrix.

[0120] If the fitness value calculated in the i-th iteration exceeds the individual's historical best value, then the fitness value calculated in the i-th iteration will be updated to the individual's historical best value.

[0121] If the fitness value calculated in the i-th iteration exceeds the global optimum, then the fitness value calculated in the i-th iteration will be updated to the global optimum.

[0122] In each iteration, the velocity and position of each particle are updated based on the individual historical best value and the global best value;

[0123] Once the iteration count meets the iteration termination condition, the position of the particle corresponding to the global optimal value is output as the optimal principal component number k'.

[0124] In a specific embodiment, the fitness value of each particle is calculated as follows:

[0125] ;

[0126] in, Represents the trace of a matrix. This represents the fitness value of the dimensionality-reduced matrix constructed with principal component number k. The inter-class scatter matrix, The formulas for calculating the intra-class scatter matrix are as follows:

[0127] ;

[0128] ;

[0129] in, The eigenvectors of each row in the dimensionality-reduced matrix are represented as individual samples. These eigenvectors include the event-related desynchronization index for all three frequency bands across all EEG channels and the modulation index of the Kullback-Leibler divergence. μ is the mean of the entire sample. Classification is based on the motor imagery ability grading labels in the training data, and P represents the total number of classes. Let be the mean of the p-th class of samples. Let p be the number of samples in class p.

[0130] The inertia weight in the particle velocity update formula adopts a linearly decreasing inertia weight, as shown in the following formula:

[0131] ;

[0132] in, and These represent the initial weight and the final weight, respectively. Indicates the maximum number of iterations. This represents the inertial weight used to update the particle's velocity during the i-th iteration.

[0133] Specifically, for multi-band coupling features, traditional Principal Component Analysis (PCA) methods typically determine the number of principal components based on a fixed cumulative variance contribution rate (e.g., 95%), often retaining a large amount of noise irrelevant to the classification task. This application proposes using an improved particle swarm optimization (IPSO) algorithm to adaptively optimize the number of principal components k in PCA. The specific steps are as follows:

[0134] Constructing the optimization objective (fitness function): The core of the improved particle swarm optimization algorithm is to find the optimal number of principal components k', maximizing the discriminative power between the "high-ability group" and the "low-ability group" in the dimensionality-reduced feature space. In this application, the fitness function Fitness(k) in the improved particle swarm optimization algorithm is defined as the ratio of inter-class divergence to intra-class divergence under the Fisher discriminant criterion. The larger the Fitness(k) value, the greater the distance between the cluster centers of the two classes of samples in that dimension. Larger), and the more compact the distribution within each class ( Smaller features are more separable.

[0135] To avoid the improved particle swarm optimization algorithm getting trapped in local optima, the embodiments of this application introduce a linear decreasing inertia weight strategy in the improved particle swarm optimization algorithm, that is, the inertia weight w(i) in the particle velocity update formula changes dynamically with the number of iterations i.

[0136] In one example, the initial weights in the particle velocity update formula =0.9, ensuring strong global search capabilities in the initial stage; termination weight =0.4, to ensure fine-grained local searches in the later stages.

[0137] The improved particle swarm optimization algorithm in the embodiments of this application mainly redefines the fitness function and the inertia weight in the particle velocity update formula in the existing particle swarm optimization algorithm. The steps of the improved particle swarm optimization algorithm are as follows:

[0138] 1. Initialization: Randomly generate M particles, the position of each particle represents a potential principal component number k.

[0139] 2. Projection and Evaluation: In each iteration, based on the particle values, the original data is projected into a multidimensional space using PCA, and the fitness value is calculated according to the Fitness(k) formula mentioned above.

[0140] 3. Update: Update the velocity and position of each particle based on the individual historical best value and the global best value.

[0141] 4. Output: After the iteration terminates, the output of the global optimal solution is the optimal principal component number k'.

[0142] Specifically, in the embodiments of this application, the particle values, as individual samples, contain the feature vector of each row in the dimensionality reduction matrix. This feature vector includes the event-related desynchronization index of all three frequency bands of the EEG channels and the modulation index of the Kullback-Leibler divergence. After determining the optimal number of principal components k' using an improved particle swarm optimization algorithm, the optimal number of principal components k' is used to perform final PCA dimensionality reduction on all subject data, generating dimensionality-reduced features for input into the motor imagery ability grading model.

[0143] like Figure 6 As shown, by optimizing the principal component count of the PCA algorithm using the improved particle swarm optimization (IPSO) algorithm, and without label training, the "high-ability group" (green dots) and the "low-ability group" (red dots) exhibit a clear clustering separation trend in two-dimensional space after dimensionality reduction using the multi-band coupling features of the embodiments of this application combined with IPSO-PCA. The data points of the "high-ability group" are mainly concentrated in the central region, while the distribution of the "low-ability group" is more scattered. This directly proves that the dimensionality reduction features extracted by the embodiments of this application can effectively characterize the differences in motor imagination ability.

[0144] In a specific embodiment, the motor imagination ability classification model includes a convolutional layer, a ReLU activation function layer, a max pooling layer, a fully connected layer, and a softmax function layer connected in sequence.

[0145] Specifically, in one embodiment of this application, the motor imagery ability grading model consists of a convolutional layer, a ReLU activation function layer, a max pooling layer, a fully connected layer, and a softmax function layer connected sequentially. In other embodiments, other convolutional neural network structures can also be selected. The input of this motor imagery ability grading model is the dimensionality-reduced features obtained by principal component analysis (PCA) with the optimal number of principal components. The output is the motor imagery ability grading probability, which characterizes the probability that an individual belongs to a high or low level of motor imagery ability. The motor imagery ability grading probability determines whether an individual belongs to a high or low level of motor imagery ability. Subjects with a high level of motor imagery ability are classified into the high-ability group, and subjects with a low level of motor imagery ability are classified into the low-ability group.

[0146] The above training data was used to train the motor imagery ability grading model, resulting in a trained motor imagery ability grading model. For example... Figure 7As shown, thanks to the lightweight CNN network structure design of the motor imagination ability ranking model mentioned in one embodiment of this application, the model exhibits excellent convergence during training. Around the 50th epoch, the cross-entropy loss rapidly decreases and stabilizes (close to 0), indicating that the motor imagination ability ranking model can quickly learn the discrimination rules from the features without significant oscillations. Figure 8 As shown, the high-ability and low-ability groups exhibit distinctly different distribution patterns (peak misalignment between gray and dark gray regions) in terms of PCA principal component intensities. This significant statistical difference demonstrates that the dimensionality reduction features selected in the embodiments of this application have clear physical meaning and can serve as biomarkers for assessing brain control capabilities.

[0147] Combining IPSO-PCA with a lightweight CNN, the motor imagination ability ranking model performs excellently on the test set. For example... Figure 9 As shown, in the classification of 560 test samples (268 low-ability samples and 292 high-ability samples), the model achieved a 100% recognition accuracy rate, with 0 false positives in the low-ability group and 0 false negatives in the high-ability group. This indicates that the method proposed in the embodiments of this application not only solves the computational redundancy problem of high-dimensional EEG signals, but also completely solves the problem of misjudging "borderline individuals" in traditional methods, and can provide an absolutely reliable grading basis for rehabilitation training.

[0148] S2. Obtain the raw EEG signals collected from the person to be assessed during the motor imagery and rest phases, and perform preprocessing and data filtering to obtain the processed EEG signals. Construct multi-band coupling features based on the processed EEG signals, and use principal component analysis algorithm on the multi-band coupling features to reduce the dimensionality with the optimal number of principal components to obtain the corresponding dimensionality-reduced features. Input the dimensionality-reduced features into the trained motor imagery ability grading model to obtain the motor imagery ability grading probability. The motor imagery ability grading probability is the probability that the person to be assessed belongs to the high or low level of motor imagery ability. Determine the motor imagery ability level of the person to be assessed based on the motor imagery ability grading probability.

[0149] Specifically, the motor imagery ability grading model of this application is deployed. In the inference phase, the raw EEG signals collected from the person to be evaluated during the motor imagery and rest phases are preprocessed and data filtered to obtain processed EEG signals. The preprocessing and data filtering methods are the same as in the training process. Multi-band coupling features are constructed from the processed EEG signals in the same way as in the training process. Principal component analysis algorithm is used on the multi-band coupling features to reduce the dimensionality with the optimal number of principal components to obtain the corresponding dimensionality-reduced features. The dimensionality-reduced features are input into the trained motor imagery ability grading model to predict the probability of motor imagery ability grading. The motor imagery ability grading probability is used to determine whether the person to be evaluated belongs to the high or low level of motor imagery ability.

[0150] Further reference Figure 10 As an implementation of the methods shown in the above figures, this application provides an embodiment of a motion imagination ability grading device based on multi-band coupling characteristics. This device embodiment is similar to... Figure 1 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0151] This application provides a motion imagination ability grading device based on multi-band coupling characteristics, including:

[0152] Model building module 1 is configured to build and train a motor imagination ability classification model based on a convolutional neural network to obtain a trained motor imagination ability classification model. During the training process of the motor imagination ability classification model, an improved particle swarm algorithm is used to determine the optimal number of principal components to be used in the principal component analysis algorithm based on the training data.

[0153] The grading module 2 is configured to acquire the raw EEG signals collected from the individuals to be assessed during the motor imagery and rest phases, perform preprocessing and data filtering to obtain processed EEG signals; construct multi-band coupling features based on the processed EEG signals, and use principal component analysis (PCA) algorithm on the multi-band coupling features to reduce dimensionality to the optimal number of principal components, obtaining the corresponding dimensionality-reduced features; input the dimensionality-reduced features into the trained motor imagery ability grading model to obtain the motor imagery ability grading probability, which is the probability that the individual to be assessed belongs to the high or low level of motor imagery ability, and determine the motor imagery ability level to which the individual to be assessed belongs based on the motor imagery ability grading probability.

[0154] Figure 11 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. For example... Figure 11As shown, the electronic device in this embodiment includes a processor 1101 and a memory 1102; wherein the memory 1102 is used to store computer execution instructions; and the processor 1101 is used to execute the computer execution instructions stored in the memory to implement the various steps performed by the electronic device in the above embodiment. For details, please refer to the relevant descriptions in the foregoing method embodiments.

[0155] Alternatively, the memory 1102 can be either standalone or integrated with the processor 1101.

[0156] When the memory 1102 is set up independently, the electronic device also includes a bus 1103 for connecting the memory 1102 and the processor 1101.

[0157] This invention also provides a computer storage medium storing computer execution instructions, which, when executed by the processor 1101, implement the above method.

[0158] This invention also provides a computer program product, including a computer program that, when executed by a processor 1101, implements the above-described method.

[0159] In the embodiments provided by this invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0160] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.

[0161] Furthermore, the functional modules in the various embodiments of this invention can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit formed by the above modules can be implemented in hardware or in the form of hardware plus software functional units.

[0162] The integrated modules implemented as software functional modules described above can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor 1101 to execute some steps of the methods of the various embodiments of this application.

[0163] It should be understood that the processor 1101 described above can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor, or the processor 1101 can be any conventional processor 1101. The steps of the method disclosed in this invention can be directly manifested as the hardware processor 1101 executing the steps, or as a combination of hardware and software modules within the processor 1101 executing the steps.

[0164] The memory 1102 may include high-speed RAM memory, and may also include non-volatile memory NVM, such as at least one disk storage, and may also be a USB flash drive, portable hard drive, read-only memory, disk or optical disc, etc.

[0165] Bus 1103 can be an Industry Standard Architecture (ISA), a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Bus 1103 can be divided into address bus, data bus, control bus, etc. For ease of illustration, the bus 1103 in the accompanying drawings of this application is not limited to only one bus 1103 or one type of bus 1103.

[0166] The aforementioned storage medium can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium can be any available medium accessible to general-purpose or special-purpose computers.

[0167] An exemplary storage medium is coupled to a processor 1101, enabling the processor 1101 to read information from and write information to the storage medium. Alternatively, the storage medium can be an integral part of the processor 1101. The processor 1101 and the storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor 1101 and the storage medium can exist as discrete components in an electronic device or a host device.

[0168] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0169] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for classifying motor imagination ability based on multi-band coupling characteristics, characterized in that, Includes the following steps: A motor imagination ability classification model based on a convolutional neural network is constructed and trained to obtain a trained motor imagination ability classification model. During the training process of the motor imagination ability classification model, M particles are randomly initialized and generated, and the position of each particle represents a potential principal component number k. In each iteration, the multi-band coupling features in the training data are input into the principal component analysis algorithm, the k principal components corresponding to the position of each particle are retained, and the corresponding dimensionality reduction matrix is ​​constructed. The fitness value of each particle is calculated based on the dimensionality reduction matrix. If the fitness value calculated in the i-th iteration exceeds the individual's historical best value, then the fitness value calculated in the i-th iteration will be updated to the individual's historical best value. If the fitness value calculated in the i-th iteration exceeds the global optimum, then the fitness value calculated in the i-th iteration will be updated to the global optimum. In each iteration, the velocity and position of each particle are updated based on the individual historical best value and the global best value; Once the iteration count meets the iteration termination condition, the position of the particle corresponding to the global optimum is output as the optimal principal component number. The raw EEG signals collected from the individuals to be evaluated during the motor imagery and rest phases were acquired, preprocessed, and filtered to obtain the processed EEG signals. Based on the processed EEG signals, a multi-band coupling feature is constructed, specifically including: The processed EEG signal is decomposed into the amplitude of the Alpha band signal and the amplitude of the Beta band signal, wherein the Beta band signal includes the Beta1 band signal and the Beta2 band signal. The event-related desynchronization index of each EEG channel in each frequency band signal was calculated based on the amplitude of the Alpha band signal and the amplitude of the Beta band signal, respectively. The processed EEG signal is used to obtain an analytical signal using Hilbert transform, and the instantaneous phase of the Alpha band signal and the instantaneous amplitude of the Beta band signal are extracted from the analytical signal respectively. Construct a phase-amplitude distribution histogram based on the instantaneous phase and instantaneous amplitude, divide the range of instantaneous phase values ​​into N phase intervals, calculate the average value of the instantaneous amplitude falling within each phase interval, and calculate the amplitude distribution probability within each phase interval based on the average value of the instantaneous amplitude within each phase interval. The modulation index of the Kullback-Leibler divergence of each EEG channel is obtained by calculating the Kullback-Leibler divergence between the amplitude distribution probability and the uniform distribution. Multi-band coupling features are constructed by combining the event-related desynchronization index and the modulation index of Kullback-Leibler divergence in three frequency bands for each EEG channel. Principal component analysis is applied to the multi-band coupling features to reduce the dimensionality of the features to the optimal number of principal components, resulting in corresponding dimensionality-reduced features. The dimensionality-reduced features are then input into the trained motor imagery ability grading model to obtain the motor imagery ability grading probability. Based on the motor imagery ability grading probability, the motor imagery ability level of the person to be assessed is determined.

2. The method for classifying motion imagination ability based on multi-band coupling characteristics according to claim 1, characterized in that, The formula for calculating the event-related desynchronization index is as follows: ； in, This represents the event-related desynchronization index of the c-th EEG channel in the f-th frequency band. This represents the mean square value of the amplitude of the c-th EEG channel under the f-th frequency band signal collected during the resting phase. This represents the mean square value of the amplitude of the c-th EEG channel under the f-th frequency band signal acquired during the motor imagery phase; The analytical signal is obtained by calculating the following formula: in, Let represent the signal of each EEG channel in the processed EEG signal at time t. This represents the analytic signal at time t. Represents the Hilbert transform operator; The instantaneous phase of the Alpha band signal and the instantaneous amplitude of the Beta band signal are calculated using the following formula: ； in, Let represent the instantaneous amplitude of the Beta band signal at time t. This represents the instantaneous phase of the Alpha band signal at time t. The formula for calculating the amplitude distribution probability within each phase interval is: ； in, Indicates the first The average instantaneous amplitude of the Beta band signal within each phase interval. This represents the average instantaneous amplitude of the Beta band signal within the l-th phase interval. Indicates the first The amplitude distribution probability within each phase interval; The formula for calculating the modulation index of the Kullback-Leibler divergence for each EEG channel is: ； ； in, Shannon entropy represents the probability distribution of amplitude across all phase intervals. The maximum entropy value under a uniform distribution is expressed as follows: = log N, where MI represents the modulation index of the Kullback-Leibler divergence.

3. The method for classifying motion imagination ability based on multi-band coupling characteristics according to claim 1, characterized in that, The fitness value of each particle is calculated as follows: ； in, Represents the trace of a matrix. This represents the fitness value of the dimensionality-reduced matrix constructed with principal component number k. The inter-class scatter matrix, The formulas for calculating the intra-class scatter matrix are as follows: ； ； in, Let μ represent the feature vector of each row in the dimensionality reduction matrix, representing a single sample. The feature vector includes the event-related desynchronization index and the modulation index of the Kullback-Leibler divergence for three frequency bands across all EEG channels. μ is the mean of the entire sample. Classification is performed based on the motor imagery ability grading labels in the training data, and P represents the total number of classes. Let be the mean of the p-th class of samples. Let p be the number of samples in class p. The inertial weight in the particle velocity update formula adopts a linearly decreasing inertial weight, as shown in the following formula: ； in, and These represent the initial weight and the final weight, respectively. Indicates the maximum number of iterations. This represents the inertial weight used to update the particle's velocity during the i-th iteration.

4. The method for classifying motion imagination ability based on multi-band coupling characteristics according to claim 1, characterized in that, The preprocessing process includes downsampling, bandpass filtering, interpolation repair of channels with substandard signal quality, and whole-brain average rereference. The data filtering process is as follows: Extract m-second core EEG signals from the preprocessed EEG signals corresponding to each frequency of motor imagery stage; The m-second core EEG signals corresponding to multiple frequencies of motor imagery stages are removed to obtain the processed EEG signals, which correspond to frequencies with severe eye movement artifacts or inattention.

5. The method for classifying motion imagination ability based on multi-band coupling characteristics according to claim 1, characterized in that, The motor imagination ability classification model includes a convolutional layer, a ReLU activation function layer, a max pooling layer, a fully connected layer, and a softmax function layer connected in sequence.

6. A motion imagination ability grading device based on multi-band coupling characteristics, characterized in that, The motion imagination ability classification method based on multi-band coupling characteristics according to any one of claims 1-5 includes: The model building module is configured to build and train a motor imagination ability classification model based on a convolutional neural network to obtain a trained motor imagination ability classification model; during the training process of the motor imagination ability classification model, an improved particle swarm optimization algorithm is used to determine the optimal number of principal components to be used in the principal component analysis algorithm based on the training data. The grading module is configured to acquire raw EEG signals collected from the person to be assessed during the motor imagery and rest phases, perform preprocessing and data filtering to obtain processed EEG signals; construct multi-band coupling features based on the processed EEG signals, apply principal component analysis (PCA) to the multi-band coupling features to reduce dimensionality to the optimal number of principal components, and obtain corresponding dimensionality-reduced features; input the dimensionality-reduced features into the trained motor imagery ability grading model to obtain motor imagery ability grading probabilities, which are the probabilities that the person to be assessed belongs to the high or low level of motor imagery ability, and determine the motor imagery ability level of the person to be assessed based on the motor imagery ability grading probabilities.

7. An electronic device, comprising: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-5.

Images