Asynchronous brain-computer interface implementation method based on motion initiation evoked potential

Abstract

The invention discloses an asynchronous brain-computer interface implementation method based on motion initiation evoked potential, and belongs to the technical field of biomedical engineering brain-computer interfaces, and the method comprises the following steps: designing a visual stimulation evoked normal form, and generating an electroencephalogram signal by a user under the action of the AO visual stimulation evoked normal form; acquiring an electroencephalogram signal in real time, performing principal component analysis on the electroencephalogram signal intercepted by the time window, inputting the analyzed data into a linear judgment model, and judging whether a motion initiation evoked potential exists or not, if the motion initiation evoked potential is judged to be in an idle state, continuing sliding the time window to perform principal component analysis, and if the motion initiation evoked potential is judged to be in a brain control state, performing the next step; frequency components in the electroencephalogram signals in the brain control state are identified, the AO normal form of the user is identified, and the previous step is executed after identification is finished. Idle state and brain control state identification is carried out based on motion start evoked potential, and low-delay idle state and brain control state detection can be realized through a designed principal component analysis and linear judgment model.

Description

Technical Field

[0001] The present invention relates to the technical field of biomedical engineering brain-computer interface, and in particular to a method for realizing asynchronous brain-computer interface based on movement onset evoked potential. Background Art

[0002] Brain-computer interface (BCI) enables information exchange between the brain and external devices or computers, and has many applications in human functional rehabilitation, functional substitution and functional enhancement. With the development of artificial intelligence, BCI has developed rapidly in the past decade. Due to the non-invasiveness of non-invasive BCI, it is more widely used in engineering. In non-invasive BCI, commonly used EEG signals include steady state visual evoked potential (SSVEP), event-related potential (ERP), sensory motor rhythm (SMR), etc. Among them, BCI based on SSVEP has the advantages of small individual differences and high information transmission rate, and is mainly used in character spelling, brain-controlled wheelchairs, etc.; BCI based on SMR has the advantage of directly stimulating the motor cortex of the brain, and is mainly used in stroke motor function rehabilitation, but there are shortcomings such as large individual differences, low recognition accuracy and few recognition types. Based on the mirror neuron theory, action observation (AO) can mirror-activate the motor area of ​​the brain. At the same time, AO-based BCI recognizes intention through visual area features, which makes AO-based BCI have great application potential in the motor function rehabilitation of stroke patients.

[0003] Based on the control mode of BCI, BCI is classified into synchronous control and asynchronous control. In synchronous BCI, the user needs to perform related tasks according to the system prompts, which is a system-led control mode; in asynchronous BCI, the system can determine whether the user wants to perform brain control, that is, introduce the "brain-computer switch" mechanism, and further determine what kind of brain control is desired, which is a user-led control mode. However, the current asynchronous brain-computer interface based on SSVEP or SMR has the problem of long detection delay (the time from the generation of movement intention to successful detection), which seriously restricts the application of asynchronous brain-computer interface. Summary of the invention

[0004] The purpose of the present invention is to overcome the delay deficiency of asynchronous brain-computer interface based on SSVEP or SMR in the prior art, and to provide a method for realizing asynchronous brain-computer interface based on movement onset evoked potential.

[0005] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions:

[0006] A method for realizing an asynchronous brain-computer interface based on movement onset evoked potentials comprises the following steps:

[0007] S1: Design an AO visual stimulation induced paradigm, under which the user generates EEG signals of movement onset evoked potential (MVEP) and steady-state movement visual evoked potential (SSMVEP), and build a brain-computer interface platform to obtain the EEG signals in real time;

[0008] S2: Performing principal component analysis (PCA) on the EEG signal captured in the time window, inputting the analyzed data into a linear discriminant analysis (LDA) model and making a judgment based on the presence or absence of the movement initiation evoked potential. If it is judged to be an idle state, continuing to slide the time window to perform the principal component analysis. If it is judged to be a brain-controlled state, proceeding to step S3;

[0009] S3: Identify the frequency components in the EEG signal in the brain-controlled state, identify the AO paradigm received by the user, and return to step S2 after the identification is completed.

[0010] By adopting the above technical scheme, the idle state and brain-controlled state are identified based on the movement onset evoked potential. The effective characteristic duration of the movement onset evoked potential after the start of the movement is about 0.4S. The designed principal component analysis and linear judgment model (PCA+LDA) can better detect the generated movement onset evoked potential, thereby realizing low-latency idle state and brain-controlled state detection.

[0011] As a preferred embodiment of the present invention, the AO visual stimulation induction paradigm in step S1 is designed by using a frame rate reduction method.

[0012] As a preferred embodiment of the present invention, the frame rate reduction method is: reducing the frame rate by increasing the duration of each picture, that is, each picture lasts for N frames, the screen refresh rate is 60Hz, and at the same time, the pictures are presented cyclically within an action cycle, and by selecting different display frame numbers, AO visual stimulation induction paradigms of different frequencies are generated.

[0013] As a preferred embodiment of the present invention, the AO visual stimulation induction paradigm includes motion frequency information and is used to induce the user's electroencephalographic signals of movement onset evoked potential and steady-state movement visual evoked potential.

[0014] As a preferred embodiment of the present invention, the construction of the brain-computer interface platform described in step S2 includes: connecting the output of the motor on the electrode cap to the input of the EEG acquisition device, and the user wearing the electrode cap sits in front of the designed AO visual stimulation induction paradigm.

[0015] As a preferred solution of the present invention, the time window described in step S2 is set to have a window length of 0.3S and a step size of 0.01S.

[0016] As a preferred embodiment of the present invention, the principal component analysis in step S2 includes: performing bandpass filtering on the multi-channel EEG information of the visual area within the time window, and reorganizing the filtered data into a column, that is, the i-th group of filtered data X i，m，n Reorganized into X i，m×n , where m is the number of EEG information points collected, n is the number of channels, the data type is voltage value, microvolt level, calculate X i，m×n Data and all X i，m×n The difference covariance matrix C of the mean is used to calculate the eigenvalues ​​and eigenvectors of the matrix C. The largest first N eigenvalues ​​constitute the principal component coefficients, and the corresponding eigenvectors are the principal components, which are used to train the linear discriminant model.

[0017] As a preferred embodiment of the present invention, the step S2 inputs the analyzed data into the linear judgment model for judgment, including: multiplying the data reorganized into a column by the principal component coefficient, and finally sending the calculated data into the LDA classifier of the linear judgment model for classification, and judging in real time whether the output result contains the movement onset evoked potential. If the results for n consecutive times all contain the movement onset evoked potential, it is judged to be a brain-controlled state; if the results for n consecutive times do not contain the movement onset evoked potential, it is judged to be an idle state, wherein the value of n is determined by the receiver operating characteristic curve (ROC curve).

[0018] As a preferred solution of the present invention, the linear judgment model is: projecting the data multiplied by the principal component coefficients onto a straight line, and the optimization target is a vector w obtained by minimizing the intra-class variance and maximizing the inter-class variance of the projected data.

[0019] As a preferred embodiment of the present invention, step S3 includes: using any one of canonical correlation analysis, extended canonical correlation analysis, task-related component analysis, task discrimination pattern analysis and convolutional neural network algorithms to identify the frequency components contained in the multi-channel EEG signals in the visual area, and the identified frequency components that are the same as the frame frequency in the AO visual stimulation induced paradigm are the identified AO paradigm of the user's gaze, that is, the user's movement intention.

[0020] Compared with the prior art, the beneficial effects of the present invention are: the idle state and the brain-controlled state are identified based on the movement onset evoked potential, the effective characteristic duration of the movement onset evoked potential after the start of the action is about 0.4S, and by designing an AO visual stimulation induction paradigm, the brain can be induced to generate movement onset evoked potentials and steady-state movement visual evoked potentials, and the designed principal component analysis and linear judgment model can better detect the generated movement onset evoked potentials, thereby achieving low-latency idle state and brain-controlled state detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of a method for implementing an asynchronous brain-computer interface based on movement onset evoked potentials as described in Example 1 of the present invention;

[0022] Figure 2 The visual stimulation paradigm for observing right-hand palm-to-palm action based on the method for realizing asynchronous brain-computer interface based on movement onset evoked potential described in Example 2 of the present invention;

[0023] Figure 3 This is a diagram of the design and experimental process of an AO visual stimulation induced paradigm of a method for realizing an asynchronous brain-computer interface based on movement onset evoked potentials as described in Example 2 of the present invention;

[0024] Figure 4 This is a flow chart of the PCA+LDA method of implementing a method for asynchronous brain-computer interface based on movement onset evoked potentials described in Example 2 of the present invention;

[0025] Figure 5 It is a time domain waveform diagram of the superimposed average of the EEG signals of the PO8 channels of all users in the same trial of four AO visual stimulation induced paradigms of a method for realizing asynchronous brain-computer interface based on movement onset evoked potential described in Example 2 of the present invention;

[0026] Figure 6 This is a result diagram of TPR and FPR of all users using the PCA+LDA method and the CNN method in the method for realizing asynchronous brain-computer interface based on movement onset evoked potential described in Example 2 of the present invention;

[0027] Figure 7 This is a result graph of the delay time of all users using the PCA+LDA method and the CNN method in an asynchronous brain-computer interface implementation method based on movement onset evoked potential described in Example 2 of the present invention. DETAILED DESCRIPTION

[0028] The present invention is further described in detail below in conjunction with test examples and specific implementation methods. However, this should not be understood as the scope of the above subject matter of the present invention being limited to the following embodiments, and all technologies realized based on the content of the present invention belong to the scope of the present invention.

[0029] Example 1

[0030] A method for realizing asynchronous brain-computer interface based on movement onset evoked potential, such as Figure 1 As shown, the following steps are included:

[0031] S1: Design an AO visual stimulation induced paradigm, under which the user generates EEG signals of movement onset evoked potential and steady-state movement visual evoked potential, and build a brain-computer interface platform to obtain the EEG signals in real time;

[0032] The AO visual stimulation induction paradigm in step S1 is designed by using a frame rate reduction method.

[0033] The frame rate reduction method is: reducing the frame rate by increasing the duration of each picture, that is, each picture lasts for N frames, the screen refresh rate is 60Hz, and at the same time, the pictures are cyclically presented within an action cycle, and by selecting different display frame numbers, AO visual stimulation induction paradigms of different frequencies are generated.

[0034] The AO visual stimulation induction paradigm includes motion frequency information and is used to induce the user's electroencephalographic signals of movement onset evoked potential and steady-state movement visual evoked potential.

[0035] S2: Perform principal component analysis on the EEG signal intercepted in the time window, input the analyzed data into the linear judgment model and make a judgment based on the presence or absence of the movement initiation evoked potential. If it is judged to be an idle state, continue to slide the time window to perform the principal component analysis. If it is judged to be a brain-controlled state, proceed to step S3;

[0036] The construction of the brain-computer interface platform described in step S2 includes: connecting the output of the motor on the electrode cap to the input of the EEG acquisition device, and the user wearing the electrode cap sits in front of the designed AO visual stimulation induction paradigm.

[0037] The time window described in step S2 is set to have a window length of 0.3S and a step size of 0.01S.

[0038] The principal component analysis in step S2 includes: performing bandpass filtering on the multi-channel EEG information of the visual area within the time window, and reorganizing the filtered data into a column, that is, the i-th group of filtered data X i，m，n Reorganized into X i，m×n , where m is the number of EEG information points collected, n is the number of channels, the data type is voltage value, microvolt level, calculate X i，m×n Data and all X i，m×nThe difference covariance matrix C of the mean is used to calculate the eigenvalues ​​and eigenvectors of the matrix C. The largest first N (N=30 in this embodiment) eigenvalues ​​constitute the principal component coefficients, and the corresponding eigenvectors are the principal components, which are used to train the linear discriminant model.

[0039] In step S2, the analyzed data is input into the linear judgment model for judgment, including: multiplying the data reorganized into a column by the principal component coefficient, and finally sending the calculated data into the LDA classifier of the linear judgment model for classification, and judging in real time whether the output result contains the movement onset evoked potential. If the results for n consecutive times all contain the movement onset evoked potential, it is judged to be a brain-controlled state; if the results for n consecutive times do not contain the movement onset evoked potential, it is judged to be an idle state, wherein the value of n is determined by the receiver operating characteristic curve.

[0040] The linear judgment model projects the data multiplied by the principal component coefficients onto a straight line, and the optimization target is a vector w obtained by minimizing the intra-class variance and maximizing the inter-class variance of the projected data.

[0041] S3: Identify the frequency components in the EEG signal under the brain control state, identify the AO paradigm received by the user, store and record the AO paradigm, and return to step S2 after the identification is completed.

[0042] The step S3 includes: using any algorithm among canonical correlation analysis, extended canonical correlation analysis, task-related component analysis, task discrimination pattern analysis and convolutional neural network to identify the frequency components contained in the multi-channel EEG signals in the visual area, and the identified frequency components that are the same as the frame frequency in the AO visual stimulation induced paradigm are the identified AO paradigm of the user's gaze, that is, the user's movement intention.

[0043] The idle state and brain-controlled state are identified based on the movement onset evoked potential. The effective characteristic duration of the movement onset evoked potential after the start of the movement is about 0.4S. The designed principal component analysis and linear judgment model can better detect the generated movement onset evoked potential, thereby realizing low-latency idle state and brain-controlled state detection. At the same time, the designed AO visual stimulation evoked paradigm can quickly induce movement onset evoked potential and steady-state movement visual evoked potential.

[0044] Example 2

[0045] This embodiment is a specific embodiment of embodiment 1. This embodiment uses the 32-channel wireless EEG acquisition system g.Nautilus produced by Austrian g.tec company to collect EEG signals. The 32 channels are respectively located at Fp1, Fp2, AF3, AF4, F7, F3, Fz, F4, F8, FC5, FC1, FC2, FC4, T7, C3, Cz, C4, T8, CP5, CP1, CP2, CP6, P7, P3, Pz, P4, P8, PO7, PO3, PO4, PO8 and Oz. The sampling frequency is 500Hz, the reference is located at the right earlobe, and the ground is located at the forehead.

[0046] 18 healthy elderly users aged 60 to 75 were selected to participate in this experiment (average age 67.50±4.20 years old), including 4 males and 14 females. All users participated in the BCI experiment for the first time.

[0047] Step 1: Design the AO visual stimulation induction paradigm, such as Figure 2 As shown in the figure, the AO visual stimulation induced paradigm takes the right hand palm-to-palm action as an example. The right hand palm-to-palm action video is recorded, and 15 pictures are evenly intercepted from a movement cycle. The duration of each frame of the picture is precisely controlled to be N / SRR seconds through the Psychophysics Toolbox toolbox in Matlab, where SRR is the refresh rate of the computer frequency, such as 60Hz, and N is a positive integer. Each picture in a movement cycle is played in turn, such as Figure 3 As shown in (a), it takes a total of 15×N / SRR seconds and then is played repeatedly; the action frequency of the designed AO visual stimulation induced paradigm is SRR / 15×NHz, and the frame rate is SRR / NHz, where the selection principle of N is to ensure that the frame rate ranges from 3Hz to 30Hz.

[0048] like Figure 3 As shown in (b), this embodiment designs four AO paradigms, namely, left hand palm movement, left hand finger movement, right hand palm movement, and right hand finger movement. The frame rates of these four paradigms are 6 Hz, 5 Hz, 4.615 Hz, and 6.667 Hz, respectively. The movement frequencies are 0.6 Hz, 0.83 Hz, 0.58 Hz, and 1.11 Hz, respectively.

[0049] A brain-computer interface platform was built. Specifically, the output of the electrode on the electrode cap was connected to the input of the EEG acquisition device, and the output of the EEG acquisition device was connected to the input of the computer. Before the user wore the electrode cap to perform the fish-designed AO visual stimulation induction paradigm, the head was 80 cm away from the stimulation paradigm. The measuring electrodes were O1, O2, Oz, PO1, PO2, and POz. A reference electrode was placed on the right earlobe, and FPz was the ground electrode. Conductive paste was injected into each measuring electrode to ensure good contact between the electrode and the scalp.

[0050] Step 2: Design of a "brain-computer switch" identification method based on movement onset evoked potential. Specifically, multiple AO visual stimulation evoked paradigms are simultaneously presented on the computer screen. Different visual stimulation evoked paradigms have different frame rates. The user chooses what action he wants to perform according to his autonomous movement intention and then looks at the corresponding AO visual stimulation evoked paradigm. The user's EEG signal is acquired from the EEG acquisition device in real time, and a training model and "brain-computer switch" identification are established. A time window with a window length of 0.3S and a step of 0.01S is slid in the real-time acquired EEG signal stream, and the EEG signal in the time window is band-pass filtered. The filtered multi-channel signal is reorganized into a column, which is then multiplied by the principal component coefficient obtained in the training model. Finally, the calculated data is sent to the LDA classifier in the training model to determine in real time whether the output contains movement onset evoked potential; if the results of n consecutive times all contain the movement onset evoked potential, it is judged as a brain-controlled state; if the results of n consecutive times do not contain the movement onset evoked potential, it is judged as an idle state, wherein the n value is determined by the receiver operating characteristic curve. If it is in the idle state, the system does not make any input and continues to slide the time window to perform the above analysis; if it is in the brain-controlled state, the system enters step 3.

[0051] The steps to build the model include: Figure 3 As shown in (c), in each round of experiments, one of the numbers "1", "2", "3" or "4" appears in a corner of the screen, suggesting that the user is ready, which lasts for 2S; then the number disappears, and the AO visual stimulation induced paradigm appears in the same position, which lasts for 5S; the user concentrates on annotating the paradigm, and then the AO visual stimulation induced paradigm disappears, and the Chinese character "rest" appears in the center of the screen, suggesting that the user rest, which lasts for 2S; after a set of experiments is completed, it automatically cycles back and forth, and the user rests for 1 minute after every 20 sets of experiments. There are four different AO visual stimulation induced paradigms, and in these 20 sets of experiments, the four numbers appear randomly 5 times each. After each round of experimental tasks, the user rests for 5 minutes. During the experiment, the user should avoid moving the head, avoiding leg and foot movements, and avoiding swallowing. A total of four rounds of experiments were conducted, with frame rates of 10Hz, 12Hz, 8.57Hz, and 7.5Hz, respectively. Each AO visual stimulation induced paradigm appeared 20 times, and the EEG signals collected accordingly were training data. The EEG signals from -0.5s to -0.2s before and 0.1s to 0.4s after the AO visual stimulation paradigm were intercepted. During the training phase, the EEG signals of the six channels of the visual area were recombined into a column of X i，m×n , then the mean value of the EEG signal in the training set is for:

[0052]

[0053] Among them, m is the number of sampling points, n is the number of channels, i is the number of experimental trials, and T is the total number of experiments.

[0054] The difference between each group of data and the mean is Then the covariance matrix C can be obtained by the following formula:

[0055]

[0056] Among them, i is the experimental trial, T is the total number of experiments;

[0057] Calculate the eigenvalues ​​and eigenvectors of the covariance matrix, and select the largest first N (in this embodiment, N = 30) eigenvalues ​​to form the principal component coefficients P m×n，N The corresponding eigenvector is the principal component These principal components are used to train the LDA model.

[0058] In the test phase, the EEG signal x of a single trial m，n First rearrange to x 1，m×n . Then the dimension reduction is performed according to the following formula:

[0059] Finally, the reduced-dimensional test data is fed into the trained LDA model to identify whether it contains movement-onset evoked potentials. The calculation method is as follows: Figure 4 shown.

[0060] like Figure 5 As shown in the figure, the time domain waveform of the EEG signal induced by the four AO paradigms is the result of the superposition and average of the EEG signals of the PO8 channel of all users in all the same trials. The 0 in the horizontal axis represents the start time of the movement in the AO paradigm. In general, the four AO paradigms can induce mVEP, which mainly includes a positive peak P1 with a delay of 100-160ms and a negative trough N2 with a delay of 160-240ms. This waveform was not found before the start of the movement. After the transient response, i.e. mVEP, the EEG response enters a steady state.

[0061] This example uses the true positive rate (TPR), false positive rate (FPR) and accuracy (ACC) to evaluate the quality of the algorithm. An asynchronous BCI with excellent performance needs to be able to achieve a high true positive rate and a low false positive rate. The calculation formula is:

[0062]

[0063]

[0064]

[0065] Among them: TP is the number of times the test results are correctly accepted, that is, the IC state is correctly identified as the IC state; FN represents the number of times the test results are incorrectly rejected, that is, the IC state is not identified as the IC state; FP represents the number of false alarms, that is, the NC state is not identified as the NC state; TN represents the number of times the test results are correctly rejected, that is, the NC state is correctly identified as the NC state.

[0066] Step 3: Intention recognition of multiple actions based on steady-state motion visual evoked potentials: When the system identification result is brain-controlled state, canonical correlation analysis (CCA), extended canonical correlation analysis (eCCA), task-related component analysis (TRCA), task-discriminant component analysis (TDCA), convolutional neural network (CNN) and other methods are used to identify the frequency components in the EEG signal. Taking CCA as an example, the intention recognition steps of multiple actions are as follows:

[0067] Design template function: Template function Y is a sine and cosine function consisting of the frame rate and the double frequency of the frame rate in the AO visual stimulation paradigm;

[0068]

[0069] Among them, fi is the frame rate of the AO visual stimulation paradigm, i is the number of AO visual stimulation paradigms

[0070] Calculate the correlation coefficient ρ between the multi-channel EEG signal and the template function:

[0071]

[0072] Among them, x, X are EEG signals; y, Y are template functions, wx and wy are filter coefficients

[0073] The obtained correlation coefficient value ρ i , where the stimulation unit to which the frequency corresponding to the maximum value belongs is identified as the target that the user is looking at, and after the identification is completed, return to step 2.

[0074] Table 1 shows the peak amplitudes and latency of P1 and N2 in mVPE induced by the four AO paradigms, where 1 is left hand palm movement, 2 is left hand finger movement, 3 is right hand palm movement, and 4 is right hand finger movement. ANOVA was used to test the differences in P1 amplitude (Amp_P1), N2 amplitude (Amp_N2), P1 latency (Lat_P1), N2 latency (Lat_N2), and the difference between N2 and P1 latency (Lat_dif = Lat_N2-Lat_P1) in mVEP induced by the four different AO paradigms. The results showed that the type of AO paradigm had no significant effect on Amp_P1 (p = 0.553) and Amp_N2 (p = 0.576), had a significant effect on Lat_P1 (p = 0.002) and Lat_N2 (p = 0.002), but had no significant effect on Lat_dif (p = 0.317). This shows that the type of AO paradigm has a certain effect on the latency of induced mVEP, but has little effect on the shape of mVEP.

[0075] Table 1 Peak and latency times of P1 and N2 in EEG signals induced by four AO paradigms

[0076]

[0077] Figure 6 and Figure 7 The results of TPR, FPR and delay time of all users using PCA+LDA method and CNN method are shown respectively. It can be seen that the FPR obtained by PCA+LDA method is smaller and the delay is shorter. The average TPR, FPR and delay obtained by PCA+LDA method are 74.31%±7.27%, 20.72%±8.82% and 266.1ms±28.6ms respectively; the average TPR, FPR and delay obtained by CNN method are 6.13%±3.88%, 48.25%±10.21%, and 583.7ms±44.6ms respectively. The t-test showed that the FPR obtained using the PCA+LDA method was significantly lower than that obtained using the CNN method (t=-9.18, p<0.001), and the delay time obtained using the PCA+LDA method was also significantly shorter than that obtained using the CNN method (t=-25.1, p<0.001), but there was no significant difference in the TPR obtained by the two methods (t=-0.87, p=0.398).

[0078] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for realizing an asynchronous brain-computer interface based on action observation movement onset evoked potential, characterized in that: The following steps are involved: S1: Design an AO visual stimulation induced paradigm, under which the user generates EEG signals of movement onset evoked potential and steady-state movement visual evoked potential, and build a brain-computer interface platform to obtain the EEG signals in real time; S2: Perform principal component analysis on the EEG signal intercepted in the time window, input the analyzed data into the linear judgment model and make a judgment based on the presence or absence of the movement initiation evoked potential. If it is judged to be an idle state, continue to slide the time window to perform the principal component analysis. If it is judged to be a brain-controlled state, proceed to step S3; S3: Identify the frequency components in the EEG signal in the brain-controlled state, identify the AO paradigm received by the user, and return to step S2 after the identification is completed.

2. The method for realizing an asynchronous brain-computer interface based on movement onset evoked potential according to claim 1, characterized in that: The AO visual stimulation induction paradigm in step S1 is designed by using a frame rate reduction method.

3. The method for realizing an asynchronous brain-computer interface based on movement onset evoked potential according to claim 2, characterized in that: The frame rate reduction method is: reducing the frame rate by increasing the duration of each picture. At the same time, the pictures are presented cyclically within an action cycle, and AO visual stimulation induction paradigms with different frequencies are generated by selecting different display frame numbers.

4. The method for realizing an asynchronous brain-computer interface based on movement onset evoked potential according to claim 3, characterized in that: The AO visual stimulation induction paradigm includes motion frequency information and is used to induce the user's electroencephalographic signals of movement onset evoked potential and steady-state movement visual evoked potential.

5. The method for realizing an asynchronous brain-computer interface based on movement onset evoked potential according to claim 1, characterized in that: The construction of the brain-computer interface platform described in step S2 includes: connecting the output of the motor on the electrode cap to the input of the EEG acquisition device, and the user wearing the electrode cap sits in front of the designed AO visual stimulation induction paradigm.

6. The method for realizing an asynchronous brain-computer interface based on movement onset evoked potential according to claim 1, characterized in that: The time window described in step S2 is set to have a window length of 0.3S and a step size of 0.01S.

7. The method for realizing asynchronous brain-computer interface based on movement onset evoked potential according to claim 1, characterized in that: The principal component analysis in step S2 includes: performing bandpass filtering on the multi-channel EEG information of the visual area within the time window, and reorganizing the filtered data into a column, that is, the i-th group of filtered data X i，m，n Reorganize into X i，m×n , where m is the number of EEG information points collected, n is the number of channels, the data type is voltage value, microvolt level, calculate X i，m×n Data and all X i，m×n The difference covariance matrix C of the mean is used to calculate the eigenvalues ​​and eigenvectors of the matrix C. The largest first N eigenvalues ​​constitute the principal component coefficients, and the corresponding eigenvectors are the principal components, which are used to train the linear discriminant model.

8. The method for realizing asynchronous brain-computer interface based on movement onset evoked potential according to claim 7, characterized in that: In step S2, the analyzed data is input into the linear judgment model for judgment, including: multiplying the data reorganized into a column by the principal component coefficient, and finally sending the calculated data into the LDA classifier of the linear judgment model for classification, and judging in real time whether the output result contains the movement onset evoked potential. If the results for n consecutive times all contain the movement onset evoked potential, it is judged to be a brain-controlled state; if the results for n consecutive times do not contain the movement onset evoked potential, it is judged to be an idle state, wherein the value of n is determined by the receiver operating characteristic curve.

9. The method for realizing asynchronous brain-computer interface based on movement onset evoked potential according to claim 8, characterized in that: The linear judgment model is as follows: the data multiplied by the principal component coefficients are projected onto a straight line, and the optimization target is a vector w obtained by minimizing the intra-class variance and maximizing the inter-class variance of the projected data.

10. The method for realizing asynchronous brain-computer interface based on movement onset evoked potential according to claim 1, characterized in that: The step S3 includes: using any algorithm among canonical correlation analysis, extended canonical correlation analysis, task-related component analysis, task discrimination pattern analysis and convolutional neural network to identify the frequency components contained in the multi-channel EEG signals in the visual area, and the identified frequency components that are the same as the frame frequency in the AO visual stimulation induced paradigm are the identified AO paradigm of the user's gaze, that is, the user's movement intention.

Images