Neural rehabilitation training and evaluation method based on electroencephalogram and myoelectricity coupling analysis

Abstract

The invention discloses a neurological rehabilitation training and evaluation method based on electroencephalogram and electromyography coupling analysis, which comprises the following steps: designing an electroencephalogram and electromyography signal acquisition experiment normal form, and acquiring an electroencephalogram signal and an electromyography signal of a subject; preprocessing the electroencephalogram signals and the electromyographic signals; electroencephalogram signal features and electromyographic signal features are extracted, and the signal features are classified by adopting a classification model; carrying out frequency band decomposition on the classified electroencephalogram signal features and electromyographic signal features to obtain intrinsic mode function components of different frequency bands, and calculating the coupling strength between the intrinsic mode function components by adopting a transfer entropy function; based on the coupling strength obtained through calculation as a feedback signal, the electroencephalogram signal beta rhythm and the electromyographic signal coupling strength value are analyzed, rehabilitation training of a patient is assisted, rehabilitation evaluation is carried out according to the transfer entropy value before and after rehabilitation training, electroencephalogram and electromyographic coupling analysis is introduced into a rehabilitation training evaluation link, and rehabilitation evaluation is carried out according to the transfer entropy value before and after rehabilitation training. The current neuromuscular rehabilitation effect is judged, and the next round of rehabilitation training is assisted.

Description

Technical Field

[0001] The present invention relates to the technical field of hybrid brain-computer interface, and specifically to a neurorehabilitation training and evaluation method based on brain-myoelectric coupling analysis. Background Art

[0002] Evaluation and judgment, and the formulation of the next training plan are not only highly subjective and inefficient, but also ignore the subjectivity of people. In recent years, due to the rapid development of computer technology and the increasingly mature research on brain functional areas, the current simple motor imagery (MI) brain-computer interface technology still has limitations in recognition rate, number of pattern recognition, control accuracy, etc.

[0003] At present, hybrid Brain-Computer Interface (hBCI) technology has become a hot topic in research. The brain-computer interface system that integrates EEG signals with another modality of EEG signals or other physiological signals and multimodal signal brain-computer interface technology has made certain progress, but the hybrid brain-computer interface based on EEG, electrooculogram or electrocardiogram signals cannot reflect the muscle state related to hand motor function of stroke patients. In comparison, the hybrid brain-computer interface system with brain-myoelectric fusion can better reflect the motor rehabilitation status of stroke patients, and the EEG signals of the motor cortex of the brain and the EMG signals of the body's muscle tissue respectively reflect the motor control information and the functional response information of the muscles to the brain's control intention. Therefore, the synchronous feature analysis of EEG and EMG signals can reflect the functional connection characteristics between the cerebral cortex and muscle tissue.

[0004] Based on the above-mentioned advantages of hybrid brain-computer interface, it can be seen that the development of EEG and EMG rehabilitation training system methods has great theoretical and practical application value. Summary of the invention

[0005] The purpose of the present invention is to provide a neurofeedback rehabilitation training method based on brain-myoelectric coupling analysis, to carry out neurofeedback training and motor rehabilitation training, and to evaluate the rehabilitation effect by coupling analysis based on real-time collected brain-myoelectric signals.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solution: a neurorehabilitation training and evaluation method based on brain myoelectric coupling analysis, comprising the following steps:

[0007] S1. Design the experimental paradigm for collecting EEG and EMG signals, and collect the subjects’ EEG and EMG signals;

[0008] S2, preprocessing the EEG signal and the EMG signal; and

[0009] Extract EEG signal features and EMG signal features, and classify the signal features using a classification model;

[0010] S3, performing frequency band decomposition on the classified EEG signal features and EMG signal features to obtain intrinsic mode function components of different frequency bands, and using transfer entropy function to calculate the coupling strength between the intrinsic mode function components;

[0011] S4. Based on the calculated coupling strength as a feedback signal, the coupling strength value of the EEG signal β rhythm and the EMG signal is analyzed to assist the patient in rehabilitation training, and rehabilitation evaluation is performed based on the transfer entropy value before and after rehabilitation training.

[0012] Preferably, in step S1, the experimental paradigm includes three types of tasks: the subject performs low-amplitude movement, movement attempt, and movement imagery, and each type of task includes lower limb knee joint and ankle joint experimental trials; and

[0013] The subjects performed the corresponding tasks according to the prompts on the interface and collected the corresponding EEG and EMG signals.

[0014] Preferably, in step S2, the preprocessing includes: using a fourth-order IRR digital filter to perform 5-45Hz filtering on the EEG signal, using an independent component analysis method to remove physiological artifacts, using a Butterworth filter to perform 20-250Hz filtering on the EMG signal, and using a 50Hz notch filter to eliminate power frequency interference.

[0015] Preferably, in step S2, the EMG signal features select the time domain feature mean absolute value and the nonlinear feature wavelet singular entropy, and the eigenvalues ​​extracted from the two channels are constructed into a feature matrix as the input of the classification model; the EEG signal features select the frequency domain feature power spectral density and the nonlinear feature differential entropy, and the EEG signal channels are screened for feature calculation to construct a feature matrix as the input of the classification model.

[0016] Preferably, in step S2, the classification model adopts a support vector machine model, the electromyographic signals are divided into four categories, namely, knee movement attempt, ankle movement attempt, knee low-amplitude movement and ankle low-amplitude movement, and the electroencephalographic signals are divided into two categories, namely, knee movement imagination and ankle movement imagination.

[0017] Preferably, in step S3, the decomposition is specifically as follows: using improved adaptive noise complete set empirical mode decomposition to adaptively decompose the EEG C4 channel signal and the electromyographic signal to obtain IMF1, IMF2, ..., IMFn components respectively, where n depends on the number of adaptive decomposition frequency bands; and each frequency band component of the EEG signal and the electromyographic signal is calculated as an input variable of the transfer entropy function.

[0018] Preferably, in step S3, transfer entropy is used to characterize the amount of information transferred between two time series signals, and in the transfer entropy, the data is discretized into P parts, P=200; the time series length J used for calculation is 50.

[0019] Preferably, in step S4, the rehabilitation training is: inputting the classification results into the rehabilitation robot host computer, displaying dynamic pictures for visual stimulation to perform neural training or driving the rehabilitation robot to perform lower limb rehabilitation training on the subject, and performing coupling analysis on the real-time collected EEG signals and EMG signals of the subject to evaluate the rehabilitation training effect, and using this as a basis to assist the next round of rehabilitation training tasks.

[0020] Preferably, in step S4, the rehabilitation training includes the following two stages:

[0021] Motor imagery training stage: The host computer presents dynamic pictures of lower limb knee or ankle joint movements based on the results of EEG signal classification to assist the subjects in performing the corresponding imagination tasks;

[0022] Muscle rehabilitation training stage: The results of electromyographic signal classification are transmitted to the rehabilitation robot host computer, and the rehabilitation machine assists the patient in corresponding rehabilitation training.

[0023] Compared with the existing technical solutions, the beneficial effects of the present invention are: in the present invention, an offline EEG and EMG signal acquisition experimental paradigm is designed to extract features for classification, and motor imagery training and rehabilitation training are carried out; the EEG and EMG signals of the training process are analyzed to evaluate the rehabilitation effect, wherein EEG and EMG coupling analysis is introduced into the rehabilitation training evaluation link to judge the current neuromuscular rehabilitation effect and assist the next round of rehabilitation training plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0025] In the attached picture:

[0026] Figure 1 This is a flowchart of the online rehabilitation training process of the present invention;

[0027] Figure 2 (a) is a schematic diagram of the offline application scenario and acquisition equipment, and (b) is a schematic diagram of the EEG electrode coverage area and the EMG electrode position;

[0028] Figure 3 This is a flowchart of the rehabilitation training phase of the present invention;

[0029] Figure 4 This is a schematic diagram of the rehabilitation assessment module of the present invention. DETAILED DESCRIPTION

[0030] The embodiments of the present invention are described below in conjunction with the drawings in the embodiments of the present invention. The terms used in the implementation mode of the present invention are only used to explain the specific embodiments of the present invention, and are not intended to limit the present invention. The embodiments of the present application are described below in conjunction with the drawings.

[0031] like Figure 1 As shown, the present invention provides a neurofeedback rehabilitation training method based on brain-myoelectric coupling, and its technical process is: designing an offline brain-myoelectric signal acquisition experimental paradigm and building a brain-myoelectric signal synchronization acquisition platform; acquiring signals offline and processing data; extracting features from EEG and EMG signals online for pattern recognition, and conducting motor imagery training and rehabilitation training; analyzing the brain-myoelectric signals during the training process to evaluate the rehabilitation effect.

[0032] The technical solution of the present invention is described in detail as follows:

[0033] Design an experimental paradigm for collecting EEG and myoelectric signals, and collect the subjects' EEG and EMG signals:

[0034] Firstly, an experimental paradigm for collecting EEG and EMG signals was designed. The experimental paradigm required the subjects to perform three types of tasks: low-amplitude movement, movement attempt, and movement imagination. Each type of task included lower limb knee and ankle joint experimental trials. The subjects performed the corresponding tasks according to the prompts on the interface and collected the corresponding EEG and EMG signals.

[0035] In a specific embodiment, each of the three types of tasks contains 20 trials, including 20 knee joint trials and 20 ankle joint trials. Nine subjects are recruited to conduct the experiment according to the experimental plan. The subjects sign the informed consent form and experimental agreement before the experiment.

[0036] The hardware equipment used for signal acquisition in this embodiment includes wireless electromyography acquisition equipment, OpenBci platform wireless EEG cap, EEG signal amplifier, a computer, and a display screen. The experimental scene and equipment are as follows: Figure 2 As shown in (a), the EEG electrode coverage area and the EMG electrode position are shown in Figure 2 As shown in (b) in the figure; the duration of an EEG experiment is 10s, the preparation and prompt time is 5s, the task execution time is 5s, the sampling frequency of the EMG acquisition device is 1024Hz, two channels, the sampling frequency of the EEG acquisition device is 250Hz, eight channels, and one trial experiment can obtain EMG signals of size 5120×2 and EEG signal data of size 1250×8; This embodiment collects a total of 1080 sample data from 9 subjects, and the sizes of the obtained EMG and EEG data are 5120×2×1080 and 1250×8×1080, respectively.

[0037] Preprocess the EEG signal and the EMG signal; extract the EEG signal features and the EMG signal features, and classify the signal features using a classification model:

[0038] EEG and EMG signals were preprocessed. The effective EMG signal was between 20-250Hz and processed using a Butterworth filter. A 50Hz notch filter was used to eliminate power frequency interference. EEG signals are nonlinear and non-stationary. First, a fourth-order IRR digital filter was used to filter the EEG signals at 5-45Hz, and independent component analysis (ICA) was used to remove physiological artifacts such as electrooculogram.

[0039] The sample signal features are extracted, the time domain feature mean absolute value (MAV) and the nonlinear feature wavelet singular entropy (WSE) are selected as the features of the electromyogram signal, and the eigenvalues ​​extracted from the two channels are constructed into a feature matrix as the input of the classification model. The classification model adopts the support vector machine (SVM) model; the frequency domain feature power spectral density (PSD) is selected as the feature of the EEG signal, and the differential entropy (DE) is selected as the nonlinear feature. The EEG signal channels are screened for feature calculation, and the feature matrix is ​​constructed as the input of the classification model SVM.

[0040] The basic formula for the eigenvalues ​​used in EEG and EMG signals is as follows:

[0041] Time domain feature mean absolute value (MAV):

[0042]

[0043] Wherein, N is the number of collected surface electromyography signal data points, x(i) is the i-th data in the signal data sequence, indicating the intensity amplitude of the muscle action signal.

[0044] Nonlinear characteristic wavelet singular entropy (WSE):

[0045] In singular value theory, the matrix can be decomposed as follows:

[0046] W m×n =U m×l ∧V l×n ;

[0047] Where: m = 2 s , n = N / 2 s , where m and n are the number of rows and columns that can be decomposed into the matrix of the moment sequence signal, and l is the number of columns in the process of matrix diagonalization. m×l Matrix columns and V l×n The value determined by the same number of matrix rows, s is an intermediate parameter, which changes with the length of the sequence, ∧=diag(λ1,λ2,…,λ i ) is a singular value diagonal matrix, and the singular value λ satisfies λ1≥λ2…≥λ i, the wavelet singular entropy data processing method combining wavelet transform, singular value decomposition and information entropy is as follows:

[0048]

[0049] W l is the calculated wavelet singular entropy value, Δp i is the i-th order incremental wavelet singular entropy, defined as:

[0050]

[0051] Frequency domain eigenvalue power spectral density (PSD):

[0052]

[0053] Where: P(f) is the power spectrum density, which represents the power of each extreme unit frequency component. yes After Fourier transformation, X(t) represents the time series signal, t0 is a defined cutoff threshold, and the energy of the time series X(t) process is:

[0054]

[0055] According to Parseval's theorem, the power of a frequency component of one limit unit under the t0 cutoff threshold is for:

[0056]

[0057] To find the overall average power, the expectation of the random variable is:

[0058]

[0059] From this we can get Find the limit The value of is the desired power spectral density of the time series signal.

[0060] Nonlinear characteristic differential entropy (DE):

[0061]

[0062] Where p(x) represents the probability density function of continuous information, [a, b] represents the information value interval, and for a specific length, it approximately follows the Gaussian distribution. The differential entropy of the EEG signal can be calculated as:

[0063]

[0064] The collected EEG and EMG signals were classified and identified. The EMG signals were divided into four categories, namely, knee and ankle movement attempts, and knee and ankle low-amplitude movements. The EEG signals were divided into two categories, namely, knee movement imagination and ankle movement imagination. The specific parameters of the SVM model were set as follows: the radial basis kernel function was used as the kernel function, the penalty coefficient was set to 0.8, the gamma parameters of the EMG and EEG training models were set to 1 / EMGfeatures and 1 / EEGfeatures, respectively. EMGfeatures and EEGfeatures respectively represent the feature dimensions of the EMG and EEG signal extraction features, and other parameters were set to the default values.

[0065] The classification results are used as output labels. Based on the above classification, the EMG signal results are 1, 2, 3, 4, and the EEG signal results are a, b.

[0066] The classified EEG signal features and EMG signal features are decomposed by frequency band to obtain the intrinsic mode function (IMF) components of different frequency bands. The transfer entropy function is used to calculate the coupling strength between the intrinsic mode function components:

[0067] The collected lower limb electromyography and electroencephalography sample data were decomposed into frequency bands to obtain the IMF components in the time domain and frequency domain, and the transfer entropy function was used to calculate the coupling strength between the IMF components.

[0068] like Figure 4 As shown, the signal is decomposed using the improved adaptive noise complete ensemble empirical mode decomposition (ICEEMDAN) method, and the decomposition method is as follows:

[0069] The signal is decomposed by EMD, and the obtained intrinsic mode function (IMF) is used to remove the noise in each IMF through an adaptive noise algorithm. The denoised IMFs are fully integrated to obtain the decomposed signal, which can reduce the problem of modal aliasing in the decomposition process.

[0070] The specific principles are as follows:

[0071] First, define x(t) as the time series signal to be decomposed, add k Gaussian white noise with a mean of 0 to form a sequence x of k experiments i (t), where i = 1, 2, ..., k, E k (·) represents the k-order modal component generated by EMD decomposition, N(·) represents the local mean of the generated signal, and w (i) represents Gaussian white noise.

[0072] x i (t) = x(t) + εδ i (t);

[0073] Where ε is the Gaussian white noise weight coefficient; δ i(t) is the Gaussian white noise generated by the i-th processing.

[0074] x i After (t) is decomposed once, the first IMF component IMF1(t) is obtained.

[0075] R1(t)=x(t)-IMF1(t);

[0076]

[0077] R j (t) = R j-1 (t)-IMF j (t);

[0078] IMF j (t) represents the jth modal component obtained by decomposition, j = 2, 3, ..., k; E j-1 (·) represents the j-1th IMF component after EMD decomposition of the sequence; R j (t) represents the residual signal of the jth stage.

[0079] The calculation steps are as follows:

[0080] 1) Add a set of white noise δ to the original time series signal x i (t), construct sequence x i (t) = x(t) + εδ i (t), and obtain the first set of residuals R1 = N(x i (t));

[0081] 2) Calculate the first modal component R1(t);

[0082] 3) Continue to add white noise and use local mean decomposition to calculate the second set of residuals R1+εδ i (t), define the second modal component R2(t);

[0083] 4) Calculate the kth component R k (t);

[0084] 5) Until the calculation and decomposition are completed, all modes and residual numbers are obtained.

[0085] The EMG signal is decomposed by selecting the electromyographic signal channel corresponding to the lower limbs, and the EEG signal is decomposed by selecting the C4 channel to obtain two groups of decomposed signals, EMG and EEG. k} and {B k}, where k = 1, 2, ..., j, j represents the number of decomposed frequency bands, A k and B k represents the kth group of time series signals, and the time series lengths are m and n respectively.

[0086] Transfer Entropy Calculation and Analysis,Transfer Entropy (TE) is used to characterize the amount of information transferred,between two time series signals.

[0087] The above calculation results in {A k} and {B k}The transfer entropy value is calculated for each group of signals, and the calculation process is as follows:

[0088] Calculate the respective entropy values ​​h, A k The formula for calculating sequence entropy is as follows:

[0089]

[0090] B k The formula for calculating sequence entropy is as follows:

[0091]

[0092] Further calculate the EMG to EEG signal transfer entropy, denoted as

[0093]

[0094] The value of the transfer entropy indicates the strength of the cortical muscle coupling from the specific frequency band of the EMG signal to the specific frequency band of the EEG signal. Similarly, B can be calculated. k →A k Transfer entropy value.

[0095] Based on the calculated coupling strength as the feedback signal, the coupling strength value of the EEG signal β rhythm and the EMG signal is analyzed to assist the patient in rehabilitation training, and rehabilitation evaluation is performed based on the transfer entropy value before and after rehabilitation training:

[0096] like Figure 1-Figure 4 To more clearly illustrate the online rehabilitation training process, each module is described in detail below:

[0097] Build an EEG and EMG signal acquisition platform. On the basis of the offline EEG and EMG signal acquisition steps, set a fixed time window to acquire signals and perform the next step of processing at the same time.

[0098] Online feature extraction pattern recognition: when the subject performs a task, the signal in the time window is subjected to real-time feature calculation to construct a feature matrix, which is input into the offline training model for classification and recognition, and the corresponding label is output;

[0099] Rehabilitation training: The process includes two stages of rehabilitation, such as Figure 3As shown, it is divided into two stages. When the paralyzed patient has no active muscle activity in the lower limbs, motor imagery training is performed. The host computer presents dynamic pictures of the lower limb knee or ankle joint movements according to the label a or b recognized by the electroencephalogram (EEG) signal to assist the subject in performing the corresponding imagination task. After the patient's lower limb muscles recover a certain function, they can perform trial movements or low-amplitude movements and perform muscle rehabilitation training. The labels 1, 2, 3, and 4 recognized by the electromyography signals are transmitted to the host computer of the rehabilitation robot, and the rehabilitation machine assists the patient in performing the corresponding rehabilitation training.

[0100] The whole process collects the patient's EEG and EMG signals, and performs corresponding neurofeedback training or muscle rehabilitation training according to the patient's condition. The technical steps framework is shown in the figure below: Figure 1 shown.

[0101] Evaluation of rehabilitation effectiveness, such as Figure 1 As shown in the rehabilitation evaluation module, the patient's EEG and EMG signals are collected throughout the whole process, the signals within the training time are saved, and online frequency band decomposition and coupling strength calculation are performed. Since the β rhythm (12-30Hz) will have certain changes in different motor tasks, the coupling strength value of the EEG signal β rhythm and the EMG signal when performing the task is analyzed when evaluating the rehabilitation effect. Rehabilitation evaluation is performed according to the size of the transfer entropy value before and after rehabilitation training. The larger the value, the stronger the cortical muscle coupling relationship. This is used as a basis to assist the next round of rehabilitation training tasks. The present invention combines a hybrid brain-computer interface and a rehabilitation robot to perform online neurofeedback rehabilitation training and lower limb movement rehabilitation training, and performs rehabilitation effect evaluation, which can provide help for the application of hybrid brain-computer interface rehabilitation training.

[0102] The embodiments of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments. For ordinary technicians in this technical field, after knowing the contents recorded in the present invention, they can make several equivalent changes and substitutions without departing from the principle of the present invention. These equivalent changes and substitutions should also be regarded as belonging to the protection scope of the present invention.

Claims

1. A neurorehabilitation training and assessment method based on brain-myoelectric coupling analysis, It is characterized in that The steps include: S1. Design the experimental paradigm for collecting EEG and EMG signals, and collect the subjects’ EEG and EMG signals; S2, preprocessing the EEG signal and the EMG signal; as well as Extract EEG signal features and EMG signal features, and classify the signal features using a classification model; S3, performing frequency band decomposition on the classified EEG signal features and EMG signal features to obtain intrinsic mode function components of different frequency bands, and using transfer entropy function to calculate the coupling strength between the intrinsic mode function components; S4. Based on the calculated coupling strength as a feedback signal, the coupling strength value of the EEG signal β rhythm and the EMG signal is analyzed to assist the patient in rehabilitation training, and rehabilitation evaluation is performed based on the transfer entropy value before and after rehabilitation training.

2. A neurorehabilitation training and evaluation method based on brain myoelectric coupling analysis according to claim 1, Features: In step S1, the experimental paradigm includes three types of tasks: the subject performs low-amplitude movement, movement attempt, and movement imagery, and each type of task includes lower limb knee and ankle joint experimental trials; and The subjects performed the corresponding tasks according to the prompts on the interface and collected the corresponding EEG and EMG signals.

3. A neurorehabilitation training and evaluation method based on brain myoelectric coupling analysis according to claim 2, Features: In step S2, the preprocessing includes: using a fourth-order IRR digital filter to perform 5-45Hz filtering on the EEG signal, using an independent component analysis method to remove physiological artifacts, using a Butterworth filter to perform 20-250Hz filtering on the EMG signal, and using a 50Hz notch filter to eliminate power frequency interference.

4. A neurorehabilitation training and assessment method based on brain myoelectric coupling analysis according to claim 3, Features: In step S2, the EMG signal features are selected as the mean absolute value of the time domain features and the nonlinear feature wavelet singular entropy, and the eigenvalues ​​extracted from the two channels are constructed into a feature matrix as the input of the classification model; brain The power spectrum density of the frequency domain features and the differential entropy of the nonlinear features are selected for the electrical signal features, and the EEG signal channels are screened for feature calculation to construct a feature matrix as the input of the classification model.

5. A neurorehabilitation training and evaluation method based on brain myoelectric coupling analysis according to claim 4, Features: In step S2, the classification model adopts a support vector machine model, and the electromyographic signals are divided into four categories, namely, knee joint movement attempt, ankle joint movement attempt, knee joint low-amplitude movement and ankle joint low-amplitude movement. The electroencephalographic signals are divided into two categories, namely, knee joint movement imagination and ankle joint movement imagination.

6. A neurorehabilitation training and assessment method based on brain myoelectric coupling analysis according to claim 1, Features: In step S3, the decomposition is specifically as follows: the EEG C4 channel signal and the EMG signal are adaptively decomposed using the improved adaptive noise complete set empirical mode decomposition to obtain IMF1, IMF2, ..., IMFn components respectively, where n depends on the number of adaptive decomposition frequency bands; and each frequency band component of the EEG signal and the EMG signal is calculated as an input variable of the transfer entropy function.

7. A neurorehabilitation training and assessment method based on brain myoelectric coupling analysis according to claim 1, Features: In step S3, the transfer entropy is used to characterize the amount of information transferred between two time series signals, and in the transfer entropy, the data is discretized into P parts, P=200; the time series length J used for calculation is 50.

8. A neurorehabilitation training and assessment method based on brain myoelectric coupling analysis according to claim 1, Features: In step S4, the rehabilitation training is as follows: inputting the classification results into the rehabilitation robot host computer, displaying dynamic picture visual stimulation for neural training or driving the rehabilitation robot to perform lower limb rehabilitation training on the subject, and performing coupling analysis on the real-time collected EEG signals and EMG signals of the subject, evaluating the rehabilitation training effect, and using this as a basis to assist the next round of rehabilitation training tasks.

9. A neurorehabilitation training and assessment method based on brain myoelectric coupling analysis according to claim 8, Features: In step S4, the rehabilitation training includes the following two stages: Motor imagery training stage: The host computer presents dynamic pictures of lower limb knee or ankle joint movements based on the results of EEG signal classification to assist the subjects in performing the corresponding imagination tasks; Muscle rehabilitation training stage: The results of electromyographic signal classification are transmitted to the rehabilitation robot host computer, and the rehabilitation machine assists the patient in corresponding rehabilitation training.

Images