Low-training-cost electroencephalogram feature decoding method for steady-state visual evoked potential

Abstract

Disclosed in the present application is a low-training-cost electroencephalogram feature decoding method for a steady-state visual evoked potential. The method comprises: collecting original steady-state visual evoked potential signals of a subject as training data; using filter banks to preprocess original steady-state visual evoked potentials Xn, so as to obtain preprocessed steady-state visual evoked potentials Xn (m); dividing the preprocessed steady-state visual evoked potentials Xn (m) into Pn sub-periods (I); on the basis of (I), obtaining spatial filters (II) of m-th sub-bands; integrating the spatial filters (II) to obtain an integrated spatial filter (III); averaging (IV) to obtain (V); copying and connecting a plurality of (V), and performing synthesis to obtain a steady-state visual evoked potential template signal (VI) having a required length; and identifying an unknown steady-state visual evoked potential K by means of the integrated spatial filter (III) and the steady-state visual evoked potential template signal (VI).

Description

Low-training-cost EEG feature decoding method for steady-state visual evoked potential

[0001] This application claims priority to the Chinese patent application filed with the China Patent Office on August 7, 2023, with application number 202310981789.2, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The embodiments of the present application relate to encoding and decoding technologies for brain-computer interface signals, for example, to a low-training-cost EEG feature decoding method and a brain-computer interface system for steady-state visual evoked potentials. Background Art

[0003] In recent years, with researchers' in-depth research into electroencephalograms (EEGs), non-invasive brain-computer interfaces (BCIs) based on scalp EEG have made rapid progress. Among them, BCIs based on visually evoked potentials (VEPs) have garnered widespread attention due to their high information transmission rate and decoding reliability. The steady-state visual evoked potentials (SSVEP) BCI paradigm has become one of the mainstream BCI paradigms due to its stable evoked characteristics and high signal-to-noise ratio.

[0004] In related technologies, SSVEP decoding algorithms represented by the Task-Related Component Analysis (TRCA) algorithm require a large amount of training data to achieve high recognition accuracy. Compared with using standard sine and cosine as templates, pre-training helps the algorithm learn from individual EEG data and generate spatial filters and EEG template signals that conform to the EEG characteristics of the subjects, so that the algorithm can obtain satisfactory recognition results. Since the principle of the TRCA algorithm itself is to maximize the repeated signal components between trials, the recognition performance tends to drop significantly when there is less training data, especially when there is only a single trial training data for each frequency, the recognition performance is very poor.

[0005] Summary of the Invention

[0006] The present application provides a low-training-cost EEG feature decoding method and brain-computer interface for steady-state visual evoked potentials, so as to achieve the technical effect of improving the accuracy of identifying unknown visual stimuli.

[0007] In a first aspect, embodiments of the present application provide a low-training-cost EEG feature decoding method for steady-state visual evoked potentials, comprising:

[0008] Present N on the screen f The frequencies are f1,…f i ,…f n ,,…f Nf The original steady-state visual evoked potential signals of the subjects looking at these visual stimuli are collected as training data. The subject's gaze frequency is f n The original steady-state visual evoked potential of the subject's multi-electroencephalogram channels collected when the visual stimulation is Among them, N c is the number of EEG channels, N s is the number of sampling points, N t is the number of trials; N fb The filter bank is used to filter the original steady-state visual evoked potential X n Perform preprocessing to obtain N fb Preprocessed steady-state visual evoked potential Where m = 1, 2, ..., N fb ;

[0009] The pre-treated steady-state visual evoked potential Divided into P n Sub-cycle

[0010] according to Get the spatial filter of the mth sub-band N f The spatial filter of the mth sub-band of the frequency Integrate to obtain the integrated spatial filter of the mth sub-band

[0011] right P n Sub-periods are averaged to obtain Copy multiple connections Synthesize the steady-state visual evoked potential template signal of the required length

[0012] Collect unknown steady-state visual evoked potentials from subjects Using the N fb The unknown steady-state visual evoked potential K is preprocessed by a filter bank to obtain N fb The unknown steady-state visual evoked potential K after preprocessing of sub-bands (m) , where m = 1, 2, ..., N fb ;

[0013] By integrating the spatial filter and the steady-state visual evoked potential template signal Calculate N fb Correlation coefficient of sub-bands N fb indivual The weighted summation is used to obtain the unknown steady-state visual evoked potential K and the original steady-state visual evoked potential X n The correlation coefficient ρ between n ;

[0014] The unknown steady-state visual evoked potential K and frequency are obtained as f1,…f i ,…f n ,,…f Nf All original steady-state visual evoked potentials [X1,…X i ,…,X n ,...,X Nf ] between the correlation coefficient [ρ1,…ρ i ,…,ρ n ,…,ρ Nf ], the i-th correlation coefficient ρ i The maximum, then the visual stimulation frequency of the unknown steady-state visual evoked potential K is f i .

[0015] In a second aspect, the present application also provides a low-training-cost EEG feature decoding device for steady-state visual evoked potentials, comprising:

[0016] Preprocessing module, set to render N on the screen f The frequencies are f1,…f i ,…f n ,,…f Nf The original steady-state visual evoked potential signals of the subjects looking at these visual stimuli are collected as training data. The subject's gaze frequency is f n The original steady-state visual evoked potential of the subject's multi-electroencephalogram channels collected when the visual stimulation is Among them, N c is the number of EEG channels, N s is the number of sampling points, N t is the number of trials; N fb The filter bank is used to filter the original steady-state visual evoked potential X n Perform preprocessing to obtain N fb Preprocessed steady-state visual evoked potential Where m = 1, 2, ..., N fb ;

[0017] Sub-period division module, set to pre-process the steady-state visual evoked potential Divided into P n Sub-cycle

[0018] Integrated spatial filter acquisition module, set according to Get the spatial filter of the mth sub-band N f The spatial filter of the mth sub-band of the frequency Integrate to obtain the integrated spatial filter of the mth sub-band

[0019] Template signal acquisition module, set to P n Sub-periods are averaged to obtain Copy multiple connections Synthesize the steady-state visual evoked potential template signal of the required length

[0020] Unknown steady-state visual evoked potential acquisition module, set to acquire the unknown steady-state visual evoked potential of the subject Using the N fb The unknown steady-state visual evoked potential K is preprocessed by a filter bank to obtain N fb The unknown steady-state visual evoked potential K after preprocessing of sub-bands (m) , where m = 1, 2, ..., N fb ;

[0021] The correlation coefficient calculation module is set to use an integrated spatial filter and steady-state visual evoked potential template signal Calculate N fb Correlation coefficient of sub-bands N fb indivual The weighted summation is used to obtain the unknown steady-state visual evoked potential K and the original steady-state visual evoked potential X n The correlation coefficient ρ between n ;

[0022] The visual stimulation frequency acquisition module is set to obtain the unknown steady-state visual evoked potential K and the frequency f1,…f i ,…f n ,,…f Nf All original steady-state visual evoked potentials [X1,…X i ,…,X n ,...,X Nf ] between the correlation coefficient [ρ1,…ρ i ,…,ρ n ,…,ρ Nf], the i-th correlation coefficient ρ i The maximum, then the visual stimulation frequency of the unknown steady-state visual evoked potential K is f i .

[0023] In a third aspect, the embodiments of the present application further provide a low-training-cost brain-computer interface system for steady-state visual evoked potentials, comprising:

[0024] at least one processor;

[0025] a memory configured to store at least one program,

[0026] When the at least one program is executed by the at least one processor, the at least one processor implements the low-training-cost EEG feature decoding method for steady-state visual evoked potentials as described in any one of the embodiments.

[0027] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium containing a computer program, which, when executed by a processor, implements a low-training-cost EEG feature decoding method for steady-state visual evoked potentials as described in any one of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a schematic diagram of the structure of the steady-state visual evoked potential (SSVEP) brain-computer interface system;

[0029] FIG2 is a schematic diagram of the signal-to-noise ratio characteristics of a typical steady-state visual evoked potential in the frequency domain;

[0030] FIG3 is a flow chart of a low-training-cost EEG feature decoding method for steady-state visual evoked potentials provided in Example 1 of the present application;

[0031] Figure 4 shows the percentage of recognition accuracy of unknown visual stimuli using different algorithms;

[0032] Figure 5(a) is a schematic diagram showing the effect of test data length on recognition accuracy for different algorithms with a trial number of 1;

[0033] Figure 5(b) is a schematic diagram showing the effect of test data length on information transmission efficiency for different algorithms with a trial number of 1;

[0034] Figure 6 is a comparison of the recognition accuracy of the ePRCA algorithm and the eTRCA algorithm with different numbers of trials;

[0035] FIG7 is a schematic diagram of the structure of a low-training-cost EEG feature decoding device for steady-state visual evoked potentials provided in Example 2 of the present application;

[0036] FIG8 is a schematic structural diagram of a low-training-cost brain-computer interface system for steady-state visual evoked potentials provided in Example 3 of the present application. DETAILED DESCRIPTION

[0037] Example 1

[0038] Scalp EEG-based BCIs can noninvasively establish an information transmission channel between the brain and external devices, helping patients with severe movement disorders or other healthy individuals in need to communicate with the external environment. Figure 1 shows the structural diagram of a steady-state visual evoked potential (SSVEP) brain-computer interface system. The system primarily comprises a stimulation module, an acquisition module, and a processing module. The stimulation module is a display that encodes visual stimuli at a fixed frequency. When a subject gazes at the visual stimuli, steady-state visual evoked potentials (SVSEPs) are induced in the occipital region of the brain. The acquisition module uses electrodes to acquire electroencephalographic (EEG) signals representing brain thinking and cognitive activity from the subject's cerebral cortex or intracranial cavity. The processing module then extracts corresponding EEG signatures from the EEG signals through a series of signal processing methods and performs pattern recognition, thereby converting the EEG signatures into machine language instructions.

[0039] Figure 2 is a schematic diagram of the signal-to-noise ratio characteristics of a typical steady-state visual evoked potential in the frequency domain. The steady-state visual evoked potential (SSVEP) signal is induced by a visual stimulus of a fixed frequency. When the visual stimulus is presented periodically at a specific frequency (such as flickering, graphic flipping, graphic scaling, etc.), the subject's visual system will be affected and produce an evoked response with stable frequency domain characteristics. This response contains the frequency component of the visual stimulus and its higher harmonic components. Figure 2 shows the result of averaging the steady-state visual evoked potential of multiple trials. Averaging multiple trials can offset random noise and improve the signal-to-noise ratio. In the case of a single trial, the high signal-to-noise ratio effect at the visual stimulus frequency and higher harmonic frequency shown in Figure 2 may not be achieved. The steady-state visual evoked potential also has spatial characteristics. For example, the different frequency components of the SSVEP signal come from different brain regions of the subject. Therefore, when the SSVEP signal is transmitted to different parts of the subject's scalp surface, the signal-to-noise ratio of each harmonic is also different.

[0040] FIG3 is a flow chart of a low-training-cost EEG feature decoding method for steady-state visual evoked potentials provided in Example 1 of the present application. This embodiment can be applied to provide a spatial filter to decode the EEG features of steady-state visual evoked potentials. The training of the spatial filter can be based on experimental data from a small number of trials, or even based on data from a single experiment. The method can be executed by a computer, for example, including the following steps:

[0041] Among them, steps S310 to S340 are the process of training to obtain the spatial filter, and steps S350 to S370 use the trained spatial filter to filter and analyze the unknown steady-state visual evoked potential, so as to obtain the meaning represented by the unknown steady-state visual evoked potential, that is, to obtain the frequency information of the visual stimulus that the subject is looking at.

[0042] S310, presenting N on the screen f The frequencies are f1,…f i ,…f n ,,…f Nf The original steady-state visual evoked potential signals of the subjects looking at these visual stimuli are collected as training data. The subject's gaze frequency is f n The original steady-state visual evoked potential of the subject's multi-electroencephalogram channels collected when the visual stimulation is Among them, N c is the number of EEG channels, N s is the number of sampling points, N t is the number of trials; N fb The filter bank is used to filter the original steady-state visual evoked potential X n Perform preprocessing to obtain N fb Preprocessed steady-state visual evoked potential Where m = 1, 2, ..., N fb .

[0043] Assume that the subject's attention frequency is f n Visual stimulation is given, and the EEG signals of the subjects at this time are collected to obtain the original steady-state visual evoked potential Where n = 1, 2, ..., N f ; N f is the number of frequency types of visual stimulation, N c 、N s 、N t are the number of EEG channels, the number of sampling points, and the number of trials, respectively.

[0044] Optionally, the original steady-state visual evoked potential X n Before preprocessing, the collected multi-channel EEG signals X n Perform bandpass filtering and power frequency notching to filter out obvious noise signals. For example, the frequency of visual stimulation and high-order harmonics is usually between 8Hz and 80Hz. n Perform a wide range of bandpass filtering to obtain a signal with a frequency between 8Hz and 80Hz.

[0045] Using N fb The filter bank is used to filter the original steady-state visual evoked potential Xn Preprocessing is performed, that is, the original steady-state visual evoked potential X n Perform frequency division to obtain different sub-band signals and get N fb Sub-band signals in a specific frequency band Where m = 1, 2, ..., N fb ; N fb is the number of filter banks, and the filters in the filter bank can be set according to actual needs. For example, there are 8 filters in the filter bank with m=2, and the passbands are 16Hz~80Hz, 24Hz~80Hz, 32Hz~80Hz, 40Hz~80Hz, 48Hz~80Hz, 56Hz~80Hz, 64Hz~80Hz and 72Hz~80Hz respectively. n The pre-processing operations will not be described in detail here.

[0046] S320, pre-processed steady-state visual evoked potential Divided into P n Sub-cycle

[0047] The subject's gaze frequency is f n When the visual stimulation is given, the original steady-state visual evoked potential X of the subject's multiple EEG channels is collected. n For example, the visual stimulus is presented in the form of flashing, and one trial is a flash lasting 2 seconds (the appropriate duration can also be selected according to the situation). Within this 2 seconds, the flash is repeated multiple times to induce the original steady-state visual evoked potential X of the subject. n Corresponding to multiple flickers, the steady-state visual evoked potential (SSVEP) is a periodically repeated sinusoidal signal, and the SSVEP is formed by the temporal superposition of transient visual evoked potentials. Taking the above facts into consideration, the present application proposes a method for maximizing the periodically repeated component (PRC) of the SSVEP, namely the periodically repeated component analysis (PRCA) algorithm, which improves the performance of the TRCA algorithm.

[0048] Optionally, pre-processed steady-state visual evoked potentials Divided into P n Sub-cycle Including: converting the original steady-state visual evoked potential to X n Divided into P n sub-period, P n =floor(N s f n / F s ), the number of sampling points in each sub-period is L n =round(F s / f n ), where F s is the sampling frequency, floor is the rounding function, and round is the rounding function; according to the number of sub-period sampling points L n segmentation get

[0049] Since the period length of high-order harmonics is too short to be affected by noise when calculating the covariance matrix, for all The number of sub-period sampling points is f n Calculate the original steady-state visual evoked potential X n The number of sub-period sampling points L n The same can be applied to steady-state visual evoked potentials after pretreatment According to the number of sub-period sampling points L n segmentation Get a new three-dimensional tensor It contains N t P n The data of each cycle can be called The periodically repeating component (PRC) matrix of .

[0050] By applying the floor function and the round function, the steady-state visual evoked potential after preprocessing is given. A scheme for sub-period division, and then get N t P n cycle data.

[0051] S330, according to Get the p-th sub-period spatial filter of the m-th sub-band N f The spatial filter of the mth sub-band of the frequency Integrate to obtain the integrated spatial filter of the mth sub-band

[0052] Due to the influence of volume conductor effect and closed electric field, the spatial resolution of the collected subject's scalp EEG signal is not high, only reaching the centimeter level. At the same time, the collected EEG signal is also very weak, generally at the microvolt level, and the noise signal is larger than the useful signal amplitude, so the signal-to-noise ratio of the collected scalp EEG signal is relatively low.

[0053] The signal-to-noise ratio of scalp EEG signals will affect the subsequent extraction and pattern recognition of EEG feature signals, and then affect the performance of brain-computer interface. Studies have shown that spatial filtering plays an important role in reducing the signal-to-noise ratio of SSVEP EEG signals. The basic principle of spatial filtering is to convert EEG signals recorded by different EEG channels, enhance the intensity of specific signal components, and reduce the common noise in EEG channels to obtain a projection method that can better distinguish useful signals from noise signals, that is, a spatial filter. For example, the spatial filter in this application is N c dimensional vector, each component corresponds to the weight of an EEG channel.

[0054] Optional, according to Get the spatial filter of the mth sub-band include:

[0055] The calculation is performed according to the following generalized eigenvalue decomposition

[0056] Among them, here is the sum of the cross-period covariance matrices, is the sum of the periodic autocovariances, and express The function Cov represents the covariance operation. In order to make equation (1) have a finite solution, the denominator of equation (1) is restricted to 1, then the spatial filter obtained is It's the matrix The eigenvector corresponding to the largest eigenvalue of .

[0057] The above calculation process gives the Solve the spatial filter of the mth sub-band calculation process.

[0058] N f The spatial filter of the mth sub-band of the frequency Integrate to obtain the integrated spatial filter of the mth sub-band That is, N with the same m f spatial filters Used in combination, we get a combined spatial filter

[0059] S340, yes P n Sub-periods are averaged to obtain Copy multiple connections Synthesize the steady-state visual evoked potential template signal of the required length

[0060] Calculate the template of the sub-cycle of the steady-state visual evoked potential (SSVEP) signal, i.e., the periodic repetitive component In order to apply the template matching method based on SSVEP template as the TRCA algorithm, you can copy and connect To reconstruct the SSVEP template of the required length Reconstruction acquisition The theoretical basis of this is that, without considering possible transient responses, SSVEP is essentially composed of PRC.

[0061] Optionally, synthesize a steady-state visual evoked potential template signal of the desired length include:

[0062] Copy multiple connections Synthesize the steady-state visual evoked potential template signal of the required length in, for part of The number of sampling points is N s .

[0063] is a matrix The part from the first sampling point to the lth sampling point. The number of sampling points in the subsequent unknown steady-state visual evoked potential K is N s In the case of The number of sampling points is also synthesized into N s , where n = 1, 2, ..., N f ; m=1,2,…,N fb .

[0064] Synthesize the steady-state visual evoked potential template signal of the required length according to the number of sampling points of the subsequent unknown steady-state visual evoked potential K The appropriate number of sampling points can be selected according to the situation, which increases Scope of application.

[0065] S350, collecting the unknown steady-state visual evoked potential of the subject Using the N fb The unknown steady-state visual evoked potential K is preprocessed by a filter bank to obtain N fb The unknown steady-state visual evoked potential K after preprocessing of sub-bands (m) , where m = 1, 2, ..., N fb .

[0066] Adopt and train ensemble spatial filter for unknown steady-state visual evoked potential K The same N fb The filter groups are preprocessed to obtain the unknown steady-state visual evoked potential K after preprocessing. (m) The meaning of this preprocessing is detailed in S310 and will not be repeated here.

[0067] S360, through integrated spatial filter and steady-state visual evoked potential template signal Calculate N fb Correlation coefficient of sub-bands N fb indivual The weighted summation is used to obtain the unknown steady-state visual evoked potential K and the original steady-state visual evoked potential X n The correlation coefficient ρ between n .

[0068] Optionally, by integrating spatial filters and steady-state visual evoked potential template signal Calculate N fb Correlation coefficient of sub-bands include:

[0069] The correlation coefficient is calculated according to the following formula

[0070] Where ρ represents the calculated two-dimensional correlation coefficient. is the mth sub-band signal K of the unknown steady-state visual evoked potential K (m) With frequency f n The mth sub-band signal of the EEG signal induced by the visual stimulation The correlation coefficient between .

[0071] Optionally, the unknown steady-state visual evoked potential K is obtained and the original steady-state visual evoked potential X n The correlation coefficient ρ between n ,include:

[0072] The correlation coefficient ρ is obtained according to the following formula n

[0073] Where a(m) is defined as a(m)=m -s The optimal parameters for +β, α, and β can be determined by grid search, or empirical values ​​s = 1.25 and β = 0.25 can be used.

[0074] N fbindivual The weighted summation is used to obtain the unknown steady-state visual evoked potential K and the original steady-state visual evoked potential X n The correlation coefficient ρ between n , Among them, a(m) is the correlation coefficient of each specific frequency band The weighting coefficient can be determined by grid search based on the final recognition accuracy as the objective function.

[0075] The above unknown steady-state visual evoked potential K and the original steady-state visual evoked potential X n The correlation coefficient ρ between n The method is called Ensemble Periodically Repeated Component Analysis (ePRCA) algorithm.

[0076] S370, get the unknown steady-state visual evoked potential K and frequency f1,…f i ,…f n ,,…f Nf All original steady-state visual evoked potentials [X1,…X i ,…,X n ,...,X Nf ] between the correlation coefficient [ρ1,…ρ i ,…,ρ n ,…,ρ Nf ], the i-th correlation coefficient ρ i The maximum, then the visual stimulation frequency of the unknown steady-state visual evoked potential K is f i .

[0077] Steady-state visual evoked potential identification is essentially to determine which frequency of visual stimulation the subject is looking at when the unknown steady-state visual evoked potential is collected. Therefore, when the correlation coefficient ρ is obtained, n Afterwards, according to all N f ρ n The maximum value in determines which frequency visual stimulus the unknown steady-state visual evoked potential K corresponds to, that is, the recognition result is τ=max n ρ n , the correlation coefficient between the unknown steady-state visual evoked potential K and the original steady-state visual evoked potential of which frequency is the largest, then the unknown steady-state visual evoked potential K is the signal induced by the visual stimulus of which frequency the subject pays attention, that is, the recognition of the unknown visual stimulus is achieved.

[0078] Figure 4 is a percentage graph of the recognition accuracy of unknown visual stimuli of different algorithms. As shown in Figure 4, the comparison results of the recognition accuracy of unknown visual stimuli of the ePRCA algorithm described in this application and the ensemble task-related component analysis (ensemble TRCA, eTRCA) algorithm of the related art are as follows: when the trial length Ttest is 0.5s and 1s, the recognition accuracy of the ePRCA algorithm of this application is always better than that of the eTRCA algorithm under different numbers of EEG channels and different numbers of trials. And as the number of training trials and the number of EEG channels decrease, the recognition accuracy advantage of this application compared to the eTRCA algorithm is more prominent. It is worth noting that the method of this application can obtain a recognition accuracy equivalent to 5 training trials of the eTRCA algorithm when the number of training trials is 1. Among them, the number of training trials is the number of unit durations of a single frequency visual stimulus applied to the subject when the spatial filter is trained. The number of EEG channels is the number of electrodes worn at different positions on the subject's head.

[0079] Figure 5(a) is a schematic diagram showing the effect of the test data length on the recognition accuracy of different algorithms with a trial number of 1. Figure 5(b) is a schematic diagram showing the effect of the test data length on the information transmission efficiency of different algorithms with a trial number of 1, where T train is the data length of the training trial. The results in Figure 5 show that the recognition accuracy and information transmission efficiency of the ePRCA algorithm of this application are consistently higher than those of the eTRCA algorithm. Combining Figures 4, 5(a), and 5(b), it can be seen that the ePRCA algorithm of this application can solve a more reliable spatial filter with fewer trials, thereby achieving better recognition results for SSVEP-BCI.

[0080] Figure 6 compares the recognition accuracy of the ePRCA algorithm and the eTRCA algorithm at different trial numbers. As shown in Figure 6, the recognition accuracy of the ePRCA algorithm described in this application is superior at all trial numbers. In particular, when the trial number is 1, the average recognition accuracy of the ePRCA algorithm exceeds 90%, while the average recognition accuracy of the eTRCA algorithm does not exceed 40%.

[0081] The technical solution of this embodiment is to present N f The frequencies are f1,…f i ,…f n ,,…f Nf The original steady-state visual evoked potential signals of the subjects looking at these visual stimuli are collected as training data. The subject's gaze frequency is f n The original steady-state visual evoked potential of the subject's multi-electroencephalogram channels collected when the visual stimulation is Using N fb The filter bank is used to filter the original steady-state visual evoked potential Xn Perform preprocessing to obtain N fb Preprocessed steady-state visual evoked potential Steady-state visual evoked potential Divided into P n Sub-cycle according to Get the spatial filter of the mth sub-band N f The spatial filter of the mth sub-band of the frequency Integrate to obtain the integrated spatial filter of the mth sub-band right P n Sub-periods are averaged to obtain Copy multiple connections Synthesize the steady-state visual evoked potential template signal of the required length Collect unknown steady-state visual evoked potentials from subjects Using the N fb The unknown steady-state visual evoked potential K is preprocessed by a filter bank to obtain N fb The unknown steady-state visual evoked potential K after preprocessing of sub-bands (m) ; By integrating spatial filters and steady-state visual evoked potential template signal Calculate N fb Correlation coefficient of sub-bands N fb indivual The weighted summation is used to obtain the unknown steady-state visual evoked potential K and the original steady-state visual evoked potential X n The correlation coefficient ρ between n ; The unknown steady-state visual evoked potential K and frequency are f1,…f i ,…f n ,,…f Nf All original steady-state visual evoked potentials [X1,…X i ,…,X n ,...,X Nf ] between the correlation coefficient [ρ1,…ρ i ,…,ρ n ,…,ρ Nf ], the i-th correlation coefficient ρ i The maximum, then the visual stimulation frequency of the unknown steady-state visual evoked potential K is f iThe present invention solves the technical problem that the recognition performance drops significantly when the training data is small when using the TRCA algorithm, especially when there is only a single trial training data, and achieves the technical effect of improving the recognition accuracy of unknown visual stimuli.

[0082] Example 2

[0083] Figure 7 is a schematic diagram of the structure of a low-training-cost EEG feature decoding device for steady-state visual evoked potentials, provided in Example 2 of this application. This EEG feature decoding device can execute the low-training-cost EEG feature decoding method for steady-state visual evoked potentials provided in Example 1 of this application, and has the corresponding functional modules and effects of the execution method.

[0084] A low-training-cost electroencephalogram (EEG) feature decoding device for steady-state visual evoked potentials, comprising:

[0085] The pre-processing module 610 is configured to present N f The frequencies are f1,…f i ,…f n ,,…f Nf The original steady-state visual evoked potential signals of the subjects looking at these visual stimuli are collected as training data. The subject's gaze frequency is f n The original steady-state visual evoked potential of the subject's multi-electroencephalogram channels collected when the visual stimulation is Among them, N c is the number of EEG channels, N s is the number of sampling points, N t is the number of trials; N fb The filter bank is used to filter the original steady-state visual evoked potential X n Perform preprocessing to obtain N fb Preprocessed steady-state visual evoked potential Where m = 1, 2, ..., N fb ;

[0086] The sub-cycle division module 620 is configured to divide the pre-processed steady-state visual evoked potential into Divided into P n Sub-cycle

[0087] The integrated spatial filter acquisition module 630 is configured according to Get the spatial filter of the mth sub-band N f The spatial filter of the mth sub-band of the frequency Integrate to obtain the integrated spatial filter of the mth sub-band

[0088] The template signal acquisition module 640 is configured to P n Sub-periods are averaged to obtain Copy multiple connections Synthesize the steady-state visual evoked potential template signal of the required length

[0089] The unknown steady-state visual evoked potential acquisition module 650 is configured to acquire the unknown steady-state visual evoked potential of the subject. Using the N fb The unknown steady-state visual evoked potential K is preprocessed by a filter bank to obtain N fb The unknown steady-state visual evoked potential K after preprocessing of sub-bands (m) , where m = 1, 2, ..., N fb ;

[0090] The correlation coefficient calculation module 660 is configured to calculate the correlation coefficient by integrating the spatial filter. and steady-state visual evoked potential template signal Calculate N fb Correlation coefficient of sub-bands N fb indivual The weighted summation is used to obtain the unknown steady-state visual evoked potential K and the original steady-state visual evoked potential X n The correlation coefficient ρ between n ;

[0091] The visual stimulation frequency acquisition module 670 is configured to obtain the unknown steady-state visual evoked potential K and the frequencies f1, ...f i ,…f n ,,…f Nf All original steady-state visual evoked potentials [X1,…X i ,…,X n ,...,X Nf ] between the correlation coefficient [ρ1,…ρ i ,…,ρ n ,…,ρ Nf ], the i-th correlation coefficient ρ i The maximum, then the visual stimulation frequency of the unknown steady-state visual evoked potential K is f i .

[0092] Optionally, the sub-period division module 620 is configured to:

[0093] The original steady-state visual evoked potential is X n Divided into P n sub-period, P n=floor(N s f n / F s ), the number of sampling points in each sub-period is L n =round(F s / f n ), where F s is the sampling frequency, floor is the rounding function, and round is the rounding function; according to the number of sub-period sampling points L n segmentation get

[0094] Optionally, the template signal acquisition module 640 is configured to:

[0095] Copy multiple connections Synthesize the steady-state visual evoked potential template signal of the required length in, for part of The number of sampling points is N s .

[0096] Optionally, the integrated spatial filter acquisition module 630 is configured as follows:

[0097] The calculation is performed according to the following generalized eigenvalue decomposition

[0098] Among them, here is the sum of the cross-period covariance matrices, is the sum of the periodic autocovariance matrices, and express p1 and p2 periods in the.

[0099] Optionally, the correlation coefficient calculation module 660 is configured as follows:

[0100] The correlation coefficient is calculated according to the following formula

[0101] Where ρ represents the calculated two-dimensional correlation coefficient.

[0102] Optionally, the correlation coefficient calculation module 660 is configured as follows:

[0103] The correlation coefficient ρ is obtained according to the following formula n

[0104] Where a(m) is defined as a(m)=m -s The optimal parameters of +β, s and β can be determined by grid search, or empirical values ​​of s = 1.25 and β = 0.25 can be used.

[0105] The technical solution of this embodiment solves the technical problem that when the eTRCA algorithm is used, the recognition performance drops significantly when there is less training data, especially when there is only single-trial training data, the recognition performance is very poor, and achieves the technical effect of improving the accuracy of recognition of unknown visual stimuli.

[0106] Example 3

[0107] Figure 8 is a structural diagram of a low-training-cost brain-computer interface system for steady-state visual evoked potentials provided in Example 3 of the present application. As shown in Figure 8, the system includes a processor 710, a memory 720, an input device 730, and an output device 740; the number of processors 710 in the system can be at least one, and Figure 8 takes one processor 710 as an example; the processor 710, memory 720, input device 730, and output device 740 in the system can be connected via a bus or other means, and Figure 8 takes connection via a bus as an example.

[0108] The memory 720, as a computer-readable storage medium, can be used to store software programs, computer executable programs, and modules, such as the program instructions / modules corresponding to the low-training-cost EEG feature decoding method for steady-state visual evoked potentials in the embodiments of the present application (for example, a preprocessing module 610, a sub-cycle division module 620, an integrated spatial filter acquisition module 630, a template signal acquisition module 640, an unknown steady-state visual evoked potential acquisition module 650, a correlation coefficient calculation module 660, and a visual stimulation frequency acquisition module 670 in a low-training-cost EEG feature decoding device for steady-state visual evoked potentials). The processor 710 executes various functional applications and data processing of the system by running the software programs, instructions, and modules stored in the memory 720, thereby realizing the above-mentioned low-training-cost EEG feature decoding method for steady-state visual evoked potentials.

[0109] The memory 720 may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function; the data storage area may store data created based on the use of the terminal, etc. Furthermore, the memory 720 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 720 may include memory remotely located relative to the processor 710, and these remote memories may be connected to the system via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0110] The input device 730 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the system. The output device 740 may include a display device such as a display screen.

[0111] Example 4

[0112] Embodiment 4 of the present application further provides a computer-readable storage medium containing a computer program. When executed by a processor, the computer program is configured to execute a low-training-cost EEG feature decoding method for steady-state visual evoked potentials, the method comprising:

[0113] Present N on the screen f The frequencies are f1,…f i ,…f n ,,…f Nf The original steady-state visual evoked potential signals of the subjects looking at these visual stimuli are collected as training data. The subject's gaze frequency is f n The original steady-state visual evoked potential of the subject's multi-electroencephalogram channels collected when the visual stimulation is Among them, N c is the number of EEG channels, N s is the number of sampling points, N t is the number of trials; N fb The filter bank is used to filter the original steady-state visual evoked potential X n Perform preprocessing to obtain N fb Preprocessed steady-state visual evoked potential Where m = 1, 2, ..., N fb ;

[0114] Steady-state visual evoked potential Divided into P n Sub-cycle

[0115] according to Get the spatial filter of the mth sub-band N f The spatial filter of the mth sub-band of the frequency Integrate to obtain the integrated spatial filter of the mth sub-band

[0116] right P n Sub-periods are averaged to obtain Copy multiple connections Synthesize the steady-state visual evoked potential template signal of the required length

[0117] Collect unknown steady-state visual evoked potentials from subjects Using the N fb The unknown steady-state visual evoked potential K is preprocessed by a filter bank to obtain N fb The unknown steady-state visual evoked potential K after preprocessing of sub-bands (m) , where m = 1, 2, ..., N fb ;

[0118] By integrating spatial filters and steady-state visual evoked potential template signal Calculate N fb Correlation coefficient of sub-bands N fb indivual The weighted summation is used to obtain the unknown steady-state visual evoked potential K and the original steady-state visual evoked potential X n The correlation coefficient ρ between n ;

[0119] The unknown steady-state visual evoked potential K and frequency are obtained as f1,…f i ,…f n ,,…f Nf All original steady-state visual evoked potentials [X1,…X i ,…,X n ,...,X Nf ] between the correlation coefficient [ρ1,…ρ i ,…,ρ n ,…,ρ Nf ], the i-th correlation coefficient ρ i The maximum, then the visual stimulation frequency of the unknown steady-state visual evoked potential K is f i .

[0120] Of course, the computer-readable storage medium containing a computer program provided in an embodiment of the present application is not limited to the method operations described above, and can also execute relevant operations in the low-training-cost EEG feature decoding method for steady-state visual evoked potentials provided in any embodiment of the present application.

[0121] It should be noted that the low-training-cost EEG feature decoding method for steady-state visual evoked potentials described in this application can also be applied to signal-to-noise ratio enhancement and signal recognition of steady-state auditory evoked potentials, steady-state somatosensory evoked potentials and other types of signals, and based on this, corresponding brain-computer interface systems can be constructed.

[0122] Through the above description of the implementation methods, those skilled in the art can clearly understand that the present application can be implemented with the help of software and necessary general-purpose hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer's floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including multiple instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the method described in the embodiments of the present application.

[0123] It is worth noting that in the above-mentioned embodiment of the low-training-cost EEG feature decoding device for steady-state visual evoked potentials, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the names of the various functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application.

Claims

1. A low-training-cost EEG feature decoding method for steady-state visual evoked potentials, comprising: Display N on the screen f The frequencies are f1,…f i ,…f n ,,…f Nf The original steady-state visual evoked potential signals of the subjects looking at these visual stimuli are collected as training data. The subject's gaze frequency is f n The original steady-state visual evoked potential of the subject's multi-electroencephalogram channels collected when the visual stimulation is Among them, N c is the number of EEG channels, N s is the number of sampling points, N t is the number of trials; N fb The filter bank is used to filter the original steady-state visual evoked potential X n Preprocessing is performed to obtain N fb Preprocessed steady-state visual evoked potential Where m = 1, 2, ..., N fb ; The pre-treated steady-state visual evoked potential Divided into P n Sub-period according to Get the spatial filter of the mth sub-band To N f The spatial filter of the mth sub-band of the frequency Integrate to obtain the integrated spatial filter of the mth sub-band right P n Sub-periods are averaged and we get Copy multiple connections Synthesize the steady-state visual evoked potential template signal of the required length Collect unknown steady-state visual evoked potentials from subjects Using the N fb The unknown steady-state visual evoked potential K is preprocessed by a filter bank to obtain N fb The unknown steady-state visual evoked potential K after preprocessing in the sub-bands (m) , where m = 1, 2, ..., N fb ; By integrating the spatial filter and the steady-state visual evoked potential template signal Calculate N fb Correlation coefficient of sub-bands To N fb indivual The unknown steady-state visual evoked potential K and the original steady-state visual evoked potential X are obtained by weighted summation. n The correlation coefficient between n ; The unknown steady-state visual evoked potential K and the frequency are obtained as f1,…f i ,…f n ,,…f Nf All original steady-state visual evoked potentials [X1,…X i ,…,X n ,...,X Nf ] between the correlation coefficients [ρ1,…ρ i ,…,ρ n ,…,ρ Nf ], the i-th correlation coefficient ρ i The maximum visual stimulation frequency of the unknown steady-state visual evoked potential K is f i .

2. The low-training-cost EEG feature decoding method for steady-state visual evoked potential according to claim 1, wherein: The pre-treated steady-state visual evoked potential Divided into P n Sub-period include: The original steady-state visual evoked potential is X n Divided into P n sub-period, P n =floor(N s f n / F s ), the number of sampling points in each sub-period is L n =round(F s / f n ), where F s is the sampling frequency, floor is the floor function, and round is the rounding function; according to the number of sub-period sampling points L n segmentation get 3. The low-training-cost EEG feature decoding method for steady-state visual evoked potential according to claim 1, wherein: The synthesis obtains a steady-state visual evoked potential template signal of a desired length include: Synthesize the steady-state visual evoked potential template signal of the required length in, for part of The number of sampling points is N s .

4. The low-training-cost EEG feature decoding method for steady-state visual evoked potential according to claim 1, wherein: The basis Get the spatial filter of the mth sub-band include: The calculation is based on the following generalized eigenvalue decomposition Among them, here is the sum of the cross-period covariance matrices, is the sum of the periodic autocovariance matrices, and express p1 and p2 periods in the.

5. The low-training-cost EEG feature decoding method for steady-state visual evoked potential according to claim 1, wherein: The integrated spatial filter and the steady-state visual evoked potential template signal Calculate N fb Correlation coefficient of sub-bands include: The correlation coefficient is calculated according to the following formula Wherein, ρ represents the calculation of the two-dimensional correlation coefficient.

6. The low-training-cost EEG feature decoding method for steady-state visual evoked potential according to claim 1, wherein: The unknown steady-state visual evoked potential K and the original steady-state visual evoked potential X n The correlation coefficient between n ,include: The correlation coefficient ρ is calculated according to the following formula n Where a(m) is defined as a(m)=m -s +β, s and β can be selected by grid search to determine the optimal parameters, or empirical values ​​s = 1.25, β = 0.25 can be used.

7. A low-training-cost electroencephalogram feature decoding device for steady-state visual evoked potential, comprising: The pre-processing module (610) is configured to present N f The frequencies are f1,…f i ,…f n ,,…f Nf The original steady-state visual evoked potential signals of the subjects looking at these visual stimuli are collected as training data. The subject's gaze frequency is f n The original steady-state visual evoked potential of the subject's multi-electroencephalogram channels collected when the visual stimulation is Among them, N c is the number of EEG channels, N s is the number of sampling points, N t is the number of trials; N fb The filter bank is used to filter the original steady-state visual evoked potential X n Preprocessing is performed to obtain N fb Preprocessed steady-state visual evoked potential Where m = 1, 2, ..., N fb ; The sub-period division module (620) is configured to divide the pre-processed steady-state visual evoked potential into Divided into P n Sub-period The integrated spatial filter acquisition module (630) is configured according to Get the space of the mth sub-band filter To N f The spatial filter of the mth sub-band of the frequency Integrate to obtain the integrated spatial filter of the mth sub-band The template signal acquisition module (640) is configured to P n Sub-periods are averaged and we get Copy multiple connections Synthesize the steady-state visual evoked potential template signal of the required length An unknown steady-state visual evoked potential acquisition module (650) is configured to acquire the unknown steady-state visual evoked potential of a subject. Using the N fb The unknown steady-state visual evoked potential K is preprocessed by a filter bank to obtain N fb The unknown steady-state visual evoked potential K after preprocessing in the sub-bands (m) , where m = 1, 2, ..., N fb ; The correlation coefficient calculation module (660) is configured to calculate the correlation coefficient by integrating the spatial filter and steady-state visual evoked potential template signal Calculate N fb Correlation coefficient of sub-bands To N fb indivual The unknown steady-state visual evoked potential K and the original steady-state visual evoked potential X are obtained by weighted summation. n The correlation coefficient between n ; The visual stimulation frequency acquisition module (670) is configured to obtain the unknown steady-state visual evoked potential K and the frequencies f1, ...f i ,…f n ,,…f Nf All original steady-state visual evoked potentials [X1,…X i ,…,X n ,...,X Nf ] between the correlation coefficients [ρ1,…ρ i ,…,ρ n ,…,ρ Nf ], the i-th correlation coefficient ρ i The maximum visual stimulation frequency of the unknown steady-state visual evoked potential K is f i .

8. A low-training-cost brain-computer interface system for steady-state visual evoked potentials, comprising: at least one processor; a memory configured to store at least one program, When the at least one program is executed by the at least one processor, the at least one processor implements the low-training-cost EEG feature decoding method for steady-state visual evoked potentials as described in any one of claims 1-6.

9. A computer-readable storage medium comprising a computer program, wherein when the computer program is executed by a processor, the method for low-training-cost electroencephalogram feature decoding for steady-state visual evoked potentials as described in any one of claims 1 to 6 is implemented.

Images