Stroke patient unilateral motor imagery classification research method based on electroencephalogram signals

By combining the filter group public spatial mode and multi-domain feature fusion algorithm with multi-band fuzzy entropy, the spatial and nonlinear features of EEG signals are extracted, and the problem of poor exercise imagination training effect in the existing technology is solved, and the improvement of the Recognition rate and rehabilitation effect.

CN118948299APending Publication Date: 2024-11-15NANJING UNIV OF POSTS & TELECOMM
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411282112.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-13
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The existing classic sports imagination paradigm is mainly designed for healthy people. The effect of upper limb rehabilitation training applied to stroke patients is not significant enough, and the current feature extraction algorithm is limited to a single field and cannot effectively combine multiple feature information, resulting in limited improvement in the rehabilitation training effect.

Method used

Using a multi-domain feature fusion algorithm combining the filter group public spatial mode and multi-band fuzzy entropy, the spatial characteristics of the EEG signal are extracted through the filter group public spatial mode, and combined with the fuzzy entropy algorithm to extract nonlinear features in multi-band Information, fusion is used as feature vectors for classification.

Benefits of technology

It improves the recognition rate and rehabilitation training effect of unilateral motor imagination in stroke, overcomes the limitations of a single algorithm, and enhances the decoding accuracy of EEG signals. It is especially suitable for unilateral motor imagination classification in stroke patients.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118948299A_ABST
    Figure CN118948299A_ABST
Patent Text Reader

Abstract

The invention discloses a cerebral apoplexy patient unilateral motor imagery classification research method based on electroencephalogram signals. The cerebral apoplexy patient unilateral motor imagery classification research method comprises the following specific steps that 1, a cerebral apoplexy unilateral motor imagery experiment is designed; 2, electroencephalogram signal collection, wherein original electroencephalogram signals are collected through PN-NET equipment; step 3, preprocessing the electroencephalogram signals; step 4, extracting features of the preprocessed signals by adopting a filter bank public space mode; step 5, extracting features of the signals of the same filtering frequency band by using fuzzy entropy; and step 6, fusing the spatial features and the fuzzy entropy features of different frequency bands of the signals as feature vectors, and transmitting the feature vectors to an SVM for classification. Based on the advantages of different feature extraction methods and the characteristics of fine motor imagery, a multi-feature fusion method fusing a filter bank common spatial mode algorithm in a spatial domain and a fuzzy entropy algorithm in an entropy domain is provided, and the classification accuracy is further improved by performing frequency band division on signals.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of electroencephalogram (EEG) signal processing in the field of digital signal processing, and relates to a method for classifying unilateral upper limb motor imagery in stroke patients by integrating multi-band and multi-features, and in particular to a research method for classifying unilateral motor imagery in stroke patients based on EEG signals. Background Art

[0002] Stroke, commonly known as stroke, is an acute brain dysfunction caused by abnormal cerebrovascular function, usually divided into two types: ischemic and hemorrhagic. According to statistics from the World Health Organization, millions of patients suffer from stroke every year, and a large proportion of them will have varying degrees of neurological damage after the stroke, especially motor dysfunction. These dysfunctions have a serious impact on patients' daily life and work ability, significantly reducing their quality of life. For this reason, rehabilitation therapy has become one of the main interventions to restore patients' motor function and improve their quality of life. Brain-computer interface-based motor imagery technology is a new method that has emerged in the field of stroke rehabilitation in recent years. BCI is a technology that enables direct communication between people and external devices by decoding the user's brain activity signals. In stroke rehabilitation, motor imagery technology guides patients to imagine the movement of specific limbs, activates the same brain area as the actual movement, and records and analyzes the EEG signals generated by patients during the imagination process through electroencephalograms. The BCI system can decode the patient's movement intention, thereby providing an objective basis for the medical team to evaluate the patient's rehabilitation progress and adjust rehabilitation strategies. Compared with traditional rehabilitation methods, BCI-based motor imagery training can provide stroke patients with a non-invasive, personalized rehabilitation method, especially for those patients with severe movement disorders who are unable to participate in actual sports training.

[0003] However, the existing classic motor imagery paradigm is mainly designed for healthy people, and its effect is often not significant enough when applied to upper limb rehabilitation training for stroke patients. In addition, many current feature extraction algorithms are mostly limited to a single field and cannot effectively combine multiple feature information, resulting in limited improvement in the overall effect of rehabilitation training. In response to these problems, the present invention proposes a research method for the classification of unilateral motor imagery in stroke patients based on EEG signals, aiming to better meet the rehabilitation needs of stroke patients. Summary of the invention

[0004] 1. Technical issues to be solved

[0005] In view of the shortcomings of the existing technology, the present invention provides a research method for classifying unilateral motor imagery in stroke patients based on EEG signals, which solves the following problems: the existing classic motor imagery paradigm is mainly designed for healthy people, and its effect is often not significant enough when applied to upper limb rehabilitation training of stroke patients. In addition, many current feature extraction algorithms are mostly limited to a single field and cannot effectively combine multiple feature information, resulting in limited improvement in the overall effect of rehabilitation training.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: The present invention aims to propose a multi-domain feature fusion algorithm that combines the common spatial pattern of the filter group and the multi-band fuzzy entropy to identify unilateral motor imagery in stroke patients. In terms of recognition rate, this method has more advantages than the feature extraction algorithm of a single field, and can effectively improve the recognition rate of unilateral motor imagery in stroke.

[0008] The technical solution of the present invention is: a research method for classifying unilateral motor imagery in stroke patients based on EEG signals, and the specific steps include the following:

[0009] Step 1: Design a unilateral motor imagery experiment for stroke;

[0010] Step 2: EEG signal acquisition: using PN-NET equipment to collect raw EEG signals;

[0011] Step 3: Preprocessing of EEG signals: Preprocessing the EEG signals collected in step 1 to reduce interference from electrooculography and electromyography and improve classification recognition rate. The preprocessing mainly includes intercepting effective data of motor imagery, reducing sampling rate, and extracting data of effective channels.

[0012] Step 4, extracting features from the preprocessed signal using the filter bank common space mode;

[0013] Step 5: Use fuzzy entropy to extract features from signals in the same filtering frequency band;

[0014] Step 6: The spatial features and fuzzy entropy features of different frequency bands of the signal are fused as feature vectors and passed into SVM for classification.

[0015] As a further preferred embodiment of the present invention, in step 4, the filter bank common space mode extracts features of the pre-processed signal as follows:

[0016] First, the EEG signal x is filtered through a set of bandpass filters. Each filter extracts the signal of a specific frequency band. The frequency bands of the filter group set here are 8-10 Hz, 10-12 Hz, ..., 26-28 Hz, 28-30 Hz, a total of 11 frequency bands. Setting the bandwidth to 2 Hz can divide the frequency bands more finely, thereby capturing finer-grained frequency changes. For each frequency band, the filtered signal is expressed as x b , where b represents the bth frequency band;

[0017] For each filtered signal x b Perform CSP analysis; the steps are as follows:

[0018] Calculate the class covariance matrix. For class 1 and class 2 signals, calculate the covariance matrix and

[0019]

[0020]

[0021] Compute the total covariance matrix:

[0022] Perform eigenvalue decomposition and whitening transformation on the total covariance matrix:

[0023]

[0024]

[0025] Among them, U is the eigenvector matrix, Λ is the diagonal eigenvalue matrix, and the corresponding eigenvalues ​​are on the diagonal;

[0026] Transform the covariance matrix:

[0027]

[0028]

[0029] and is the covariance matrix after whitening transformation;

[0030] Compute the spatial filter:

[0031]

[0032] W b =P T V

[0033] Among them, V is the eigenvector matrix, which comes from the covariance matrix after whitening transformation Perform eigenvalue decomposition, Λ1 is a diagonal eigenvalue matrix, and the diagonal is the corresponding eigenvalue; use the CSP filter W for each frequency band b Transform signal:

[0034] z b =W b x b

[0035] Extract the logarithmic variance of the transformed signal as a feature:

[0036] f b = log(var(z b ))

[0037] Finally, the features of all frequency bands are combined to obtain the final feature matrix.

[0038] As a further preferred embodiment of the present invention, in step 5, the fuzzy entropy extracts features of signals in the same filtering frequency band as follows:

[0039] Given a time series {x1,x2,...,x N}, define the embedding dimension m and delay time τ, and construct a set of embedding vectors: X i ={x i ,x i+τ ,...,x i+(m-1)τ},i=1,2,...,N-(m-1)τ;

[0040] For each pair of embedding vectors X i and X j (where i≠j), calculate the maximum difference between them:

[0041]

[0042] This distance measures the maximum difference between two embedding vectors across all dimensions;

[0043] For each pair of embedding vectors, their similarity is calculated using the fuzzy membership function:

[0044]

[0045] Where d is the maximum difference between the two embedding vectors calculated in the previous step, r is the tolerance, and n is the fuzziness index. This membership function has a larger membership value when d is less than r, indicating that the two embedding vectors are similar; when d is much larger than r, the membership value is close to 0, indicating that the two vectors are not similar;

[0046] For each embedding vector X i, calculate the sum of its fuzzy membership with all other embedded vectors, and get the average fuzzy membership:

[0047]

[0048] Then, the average fuzzy membership of all embedded vectors is calculated:

[0049]

[0050] Increase the embedding dimension m to m+1 and repeat the above steps to get the new average fuzzy membership C m+1 (r).

[0052] As a further preferred embodiment of the present invention, the fuzzy entropy is defined as:

[0053]

[0054] Calculate the fuzzy entropy at different frequency scales. Suppose there is a set of frequency bands {f1,f2,...,f L}, the frequency bands are 8-10Hz, 10-12Hz, ..., 26-28Hz, 28-30Hz, a total of 11 frequency bands, the time series can be band-pass filtered, and then the fuzzy entropy can be calculated in each frequency band;

[0055] Where the parameter m is the pre-selected embedding dimension, r is the pre-selected tolerance for the similarity between two embedding vectors, and n is the pre-selected exponent in the fuzzy membership function, where m is set to 2, r is set to 0.3, and n is set to 2.

[0056] (III) Beneficial effects

[0057] The present invention provides a research method for classifying unilateral motor imagery in stroke patients based on EEG signals. It has the following beneficial effects:

[0058] The present invention adopts a multi-domain feature fusion algorithm that combines the common spatial pattern of the filter group and the multi-band fuzzy entropy to extract more comprehensive and effective features from the EEG signal. The filter group common spatial pattern algorithm can extract spatial features of the signal by filtering in different frequency bands, capturing the spatial pattern changes related to motor imagery, while the fuzzy entropy algorithm extracts nonlinear feature information under multiple frequency bands by calculating the complexity of the signal. The combination of the two algorithms can comprehensively consider the spatial and frequency domain characteristics of the EEG signal, overcome the limitations of a single algorithm, and improve the decoding accuracy of unilateral motor imagery intentions in the EEG signals of stroke patients, thereby effectively improving the effect of rehabilitation training.

[0059] The present invention designs a fine motor imagery paradigm for unilateral wrist and elbow flexion and extension in stroke patients. Compared with bilateral limb training in stroke patients, the training of unilateral specific joints is more complex and can activate more brain areas, and therefore is more helpful for the rehabilitation of hemiplegic stroke patients.

[0060] Based on the advantages of different feature extraction methods and the characteristics of fine motion imagination, the present invention proposes a multi-feature fusion method that fuses the filter group co-spatial pattern algorithm in the spatial domain and the fuzzy entropy algorithm in the entropy domain. By dividing the signal into frequency bands, the classification accuracy is further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a paradigm specifically designed for fine motor imagery of unilateral upper limbs in stroke patients;

[0062] Figure 2 Schematic diagram of the time flow of motor imagery for this paradigm;

[0063] Figure 3 Flowchart for EEG signal feature extraction;

[0064] Figure 4 The basic information of the patient;

[0065] Figure 5 This is a schematic diagram of the channel leads of the EEG signal;

[0066] Figure 6 The original signal collected during motor imagery and the waveform after preprocessing;

[0067] Figure 7 It is the flow chart of the algorithm for fusion of filter bank common spatial pattern and fuzzy entropy features;

[0068] Figure 8 Line chart of classification accuracy of different feature extraction algorithms. DETAILED DESCRIPTION

[0069] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0070] See also Figure 1-8 The embodiment of the present invention provides a technical solution: a research method for classifying unilateral motor imagery in stroke patients based on EEG signals.

[0071] Embodiment 1, as Figure 1 , Figure 2 and Figure 3 As shown in the figure, this is the experimental paradigm and overall plan of this experiment, which mainly includes the following steps:

[0072] (1) Collecting EEG signals: Using PN-NET equipment to collect EEG data, 11 channels (F3, F4, FC3, FC4, C3, C4, CP3, CP4, FZ, FCZ, CZ) related to movement were selected. The experimental subjects were 13 hemiplegic stroke patients (aged 30-65 years old) with impaired left and right hand functions, including 9 males and 4 females, 8 with left-side paralysis and 5 with right-side paralysis. The EEG signal data of these patients were mainly collected within 30 days after the onset of stroke. The basic information of the patients is as follows: Figure 4 As shown. Before the experiment, the subjects were arranged to sit in a comfortable position on a chair about 1 meter away from the computer screen and wear EEG equipment. The sampling rate of the experimental setting was 1024Hz, and each trial lasted 14 seconds, including 2 seconds of watching the video, 10 seconds of motor imagination, and 2 seconds of relaxation. Bandpass filtering and notch filters were used to preprocess the signal during the acquisition process to remove interference noise.

[0073] (2) The common spatial pattern of the filter bank is used to extract features from the preprocessed signal.

[0074] (3) Use fuzzy entropy to extract features from signals in the same filtering frequency band.

[0075] (4) The spatial features and fuzzy entropy features of the signal in different frequency bands are fused as feature vectors and passed into SVM for classification.

[0076] The steps are described below:

[0077] Step 1: Experimental paradigm design and EEG signal acquisition:

[0078] The experimental paradigm was designed, and the entire experiment was divided into three main stages: video demonstration stage, closed-eye motor imagery stage, and rest stage. In the video demonstration stage, videos of wrist and elbow flexion and extension movements were randomly played on the screen to guide the subjects to focus their attention and understand the upcoming motor imagery task. In the closed-eye motor imagery stage, the subjects closed their eyes and imagined the corresponding movements based on the content of the video they had just watched, including unilateral wrist and elbow flexion and extension. This stage aims to activate relevant brain areas for subsequent EEG signal acquisition and analysis. In the rest stage, the subjects were instructed to relax and avoid any form of motor imagery to ensure that brain activity returned to the baseline state.

[0079] Scalp electrode placement in brain regions Figure 5As shown, the present invention adopts the 10-20 electrode lead positioning standard calibrated by the International Electroencephalography Association, injects conductive paste to enhance the conductivity of the electrode, and wears the EEG cap correctly. The collected EEG signal is preprocessed, a low-pass filter is used to remove baseline drift, and independent component analysis ICA is used to remove eye artifacts. Figure 6 It is a graph after preprocessing the collected original EEG signal.

[0080] After obtaining the preprocessed data, 11 channels, 4 in the left frontal lobe, 4 in the right frontal lobe, and 3 in the central channel, were selected according to the relevant channel positions generated by emotional motor imagery for the next step of data processing.

[0081] Step 2: Extract the features of the common spatial pattern of the filter bank:

[0082] First, the EEG signal x is filtered through a set of bandpass filters. Each filter extracts the signal in a specific frequency band. The frequency bands of the filter group set here are 8-10Hz, 10-12Hz, ..., 26-28Hz, 28-30Hz, a total of 11 frequency bands. Setting the bandwidth to 2Hz can divide the frequency bands more finely, thereby capturing more fine-grained frequency changes. For each frequency band, the filtered signal is represented as x b , where b represents the bth frequency band.

[0083] For each filtered signal x b Perform CSP analysis. The steps are as follows:

[0084] Calculate the class covariance matrix. For class 1 and class 2 signals, calculate the covariance matrix and

[0085]

[0086]

[0087] Compute the total covariance matrix:

[0088] Perform eigenvalue decomposition and whitening transformation on the total covariance matrix:

[0089]

[0090]

[0091] Among them, U is the eigenvector matrix, Λ is the diagonal eigenvalue matrix, and the corresponding eigenvalues ​​are on the diagonal.

[0092] Transform the covariance matrix:

[0093]

[0094]

[0095] and is the covariance matrix after whitening transformation.

[0096] Compute the spatial filter:

[0097]

[0098] W b =P T V

[0099] Among them, V is the eigenvector matrix, which comes from the covariance matrix after whitening transformation Perform eigenvalue decomposition. Λ1 is a diagonal eigenvalue matrix, and the corresponding eigenvalues ​​are on the diagonal.

[0100] Use CSP filter W for each frequency band b Transform signal:

[0101] z b =W b x b

[0102] Extract the logarithmic variance of the transformed signal as a feature:

[0103] f b = log(var(z b ))

[0104] Finally, the features of all frequency bands are combined to obtain the final feature matrix.

[0105] Step 3: The fuzzy entropy extracts features of signals in the same filtering frequency band as follows:

[0106] Given a time series {x1,x2,...,x N}, define the embedding dimension m and delay time τ, and construct a set of embedding vectors: X i ={x i ,x i+τ ,...,x i+(m-1)τ},i=1,2,...,N-(m-1)τ.

[0107] For each pair of embedding vectors X i and X j (where i≠j), calculate the maximum difference between them:

[0108]

[0109] This distance measures the maximum difference between two embedding vectors across all dimensions.

[0110] For each pair of embedding vectors, their similarity is calculated using the fuzzy membership function:

[0111]

[0112] Among them, d is the maximum difference between the two embedding vectors calculated in the previous step, r is the tolerance, and n is the fuzziness index. This membership function has a larger membership value when d is less than r, indicating that the two embedding vectors are similar; when d is much larger than r, the membership value is close to 0, indicating that the two vectors are not similar.

[0113] For each embedding vector X i , calculate the sum of its fuzzy membership with all other embedded vectors, and get the average fuzzy membership:

[0114]

[0115] Then, the average fuzzy membership of all embedded vectors is calculated:

[0116]

[0117] Increase the embedding dimension m to m+1 and repeat the above steps to get the new average fuzzy membership C m+1 (r).

[0119] The fuzzy entropy is defined as:

[0120]

[0121] In this invention, we calculate the fuzzy entropy at different frequency scales, and there is a set of frequency bands {f1, f2, ..., f L}, the frequency bands are 8-10Hz, 10-12Hz, …, 26-28Hz, 28-30Hz, a total of 11 frequency bands. The time series can be band-pass filtered, and then the fuzzy entropy can be calculated in each frequency band.

[0122] where the parameter m is the pre-selected embedding dimension, r is the pre-selected tolerance for similarity between two embedding vectors, and n is the pre-selected exponent in the fuzzy membership function

[0123] In the present invention, m is set to 2, r is set to 0.3, and n is set to 2.

[0124] Figure 7This is a flowchart of the algorithm for fusion of filter bank common spatial pattern and fuzzy entropy features. The EEG signal is processed by the filter bank and divided into multiple frequency bands. The figure shows the frequency bands of 8-10Hz, 10-12Hz, 26-28Hz and 28-30Hz. These frequency bands roughly correspond to common EEG rhythms, such as mu band and beta band, which are usually related to motor imagery tasks. The spatial features and entropy features of each frequency band are calculated separately. The CSP feature can extract the spatial features related to the task, while the fuzzy entropy can characterize the nonlinear complexity of the signal. The two complement each other and can more comprehensively reflect the characteristics of the EEG signal. The CSP and fuzzy entropy features extracted from each frequency band are fused together. This feature fusion method can combine spatial and nonlinear features together to achieve better performance in classification tasks. The final fused features are input into the SVM classifier for classification.

[0125] Figure 8 The classification accuracy trends of different feature extraction algorithms are shown. As can be seen from the figure, for each subject, the classification accuracy based on multi-domain feature extraction is better than that of fuzzy entropy features. In addition, compared with the classification results of filter group common space pattern features, the classification accuracy of this method is also improved. The feature extraction method of filter group common space pattern combined with fuzzy entropy showed consistent advantages in all subjects. This shows that the two feature extraction methods are complementary and can effectively improve the robustness and accuracy of classification, indicating its potential application value in stroke rehabilitation training.

[0126] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the attached claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any figure mark in the claims should not be regarded as limiting the claims involved.

[0127] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A research method for classifying unilateral motor imagery in stroke patients based on EEG signals. It is characterized in that The specific steps include the following: Step 1: Design a unilateral motor imagery experiment for stroke; Step 2: EEG signal acquisition: using PN-NET equipment to collect raw EEG signals; Step 3: Preprocessing of EEG signals: Preprocessing the EEG signals collected in step 1 to reduce interference from electrooculography and electromyography and improve classification recognition rate. The preprocessing mainly includes intercepting effective data of motor imagery, reducing sampling rate, and extracting data of effective channels. Step 4, extracting features from the preprocessed signal using the filter bank common space mode; Step 5: Use fuzzy entropy to extract features from signals in the same filtering frequency band; Step 6: The spatial features and fuzzy entropy features of different frequency bands of the signal are fused as feature vectors and passed into SVM for classification.

2. According to claim 1, a research method for classifying unilateral motor imagery in stroke patients based on EEG signals, Features: In step 4, the filter bank common space mode extracts features from the pre-processed signal as follows: First, the EEG signal x is filtered through a set of bandpass filters. Each filter extracts the signal of a specific frequency band. The frequency bands of the filter group set here are 8-10 Hz, 10-12 Hz, ..., 26-28 Hz, 28-30 Hz, a total of 11 frequency bands. Setting the bandwidth to 2 Hz can divide the frequency bands more finely, thereby capturing finer-grained frequency changes. For each frequency band, the filtered signal is expressed as x b , where b represents the bth frequency band; For each filtered signal x b Perform CSP analysis; the steps are as follows: Calculate the class covariance matrix. For class 1 and class 2 signals, calculate the covariance matrix and Compute the total covariance matrix: Perform eigenvalue decomposition and whitening transformation on the total covariance matrix: Among them, U is the eigenvector matrix, Λ is the diagonal eigenvalue matrix, and the corresponding eigenvalues ​​are on the diagonal; Transform the covariance matrix: and is the covariance matrix after whitening transformation; Compute the spatial filter: W b =P T V Among them, V is the eigenvector matrix, which comes from the covariance matrix after whitening transformation Perform eigenvalue decomposition, Λ 1 It is a diagonal eigenvalue matrix, and the corresponding eigenvalues ​​are on the diagonal; use the CSP filter W for each frequency band b Transform signal: z b =W b x b Extract the logarithmic variance of the transformed signal as a feature: f b =log(var(z b )) Finally, the features of all frequency bands are combined to obtain the final feature matrix.

3. According to claim 2, a method for classifying unilateral motor imagery in stroke patients based on EEG signals, Features: In step 5, the fuzzy entropy extracts features of signals in the same filtering frequency band as follows: Given a time series {x 1 ,x 2 ,...,x N }, define the embedding dimension m and delay time τ, and construct a set of embedding vectors: X i ={x i ,x i+τ ,...,x i+(m-1)τ },i=1,2,...,N-(m-1)τ; For each pair of embedding vectors X i and X j (where i≠j), calculate the maximum difference between them: This distance measures the maximum difference between two embedding vectors across all dimensions; For each pair of embedding vectors, their similarity is calculated using the fuzzy membership function: Where d is the maximum difference between the two embedding vectors calculated in the previous step, r is the tolerance, and n is the fuzziness index. This membership function has a larger membership value when d is less than r, indicating that the two embedding vectors are similar; when d is much larger than r, the membership value is close to 0, indicating that the two vectors are not similar; For each embedding vector X i , calculate the sum of its fuzzy membership with all other embedded vectors, and get the average fuzzy membership: Then, the average fuzzy membership of all embedded vectors is calculated: Increase the embedding dimension m to m+1 and repeat the above steps to get the new average fuzzy membership C m+1 (r).

4. According to claim 3, a method for classifying unilateral motor imagery in stroke patients based on EEG signals, Features: The fuzzy entropy is defined as: Calculate the fuzzy entropy at different frequency scales. Suppose there is a set of frequency bands {f 1 ,f 2 ,...,f L }, the frequency bands are 8-10Hz, 10-12Hz, ..., 26-28Hz, 28-30Hz, a total of 11 frequency bands, the time series can be band-pass filtered, and then the fuzzy entropy can be calculated in each frequency band; Where the parameter m is the pre-selected embedding dimension, r is the pre-selected tolerance for the similarity between two embedding vectors, and n is the pre-selected exponent in the fuzzy membership function, where m is set to 2, r is set to 0.3, and n is set to 2.

Citation Information

Cited By

  • Lower limb movement brain-computer interface signal acquisition method based on target guidance

    CN121694772A