SSVEP recognition system and method based on linear frequency modulation visual evoked potential migration

Abstract

The invention discloses an SSVEP (Steady-State Visual Evoked Potential) recognition system and method based on linear frequency modulation visual evoked potential migration. The visual stimulation generation and presentation module generates and presents a visual stimulation signal; the electroencephalogram signal acquisition module is used for acquiring Chirp-VEP and SSVEP electroencephalogram signals; the electroencephalogram signal preprocessing module is used for preprocessing the collected electroencephalogram signals; the Chirp pulse matrix construction module is used for converting a Chirp signal to obtain a Chirp pulse matrix, the Chirp-VEP decomposition module is used for constructing a least square optimization problem to solve common model parameters of Chirp-VEP, and the common model parameters are migrated to SSVEP through the SSVEP identification module to carry out feature extraction and classification. According to the method, the system overhead is reduced, the calibration time is shortened, the training cost is minimized while the optimal performance is kept, and the practicability of the brain-computer interface based on SSVEP is improved.

Description

Technical Field

[0001] This invention belongs to the field of electronic information detection technology, and further relates to the field of intelligent physiological information recognition technology, a system and method for identifying steady-state visual evoked potentials (SSVEPs) based on linear frequency modulation visual evoked potential migration. This invention enables the rapid deployment and application of SSVEP-based BCI systems in fields such as healthcare, entertainment, and engineering. Background Art

[0002] Brain-computer interfaces (BCIs) enable communication between the brain and the external environment by analyzing neural activity. Electroencephalogram (EEG)-based BCIs have attracted attention due to their non-invasiveness and portability. Potential applications include motor rehabilitation, wheelchair navigation, physiological status assessment, robotic arm control, and spelling input. Visual evoked potentials (VEPs) are EEG signals generated in the occipital region of the brain in response to specific visual stimuli. The consistency and stability of these responses make VEPs ideal for designing BCI applications. To implement multi-command systems, the SSVEP paradigm has been adopted in BCIs. SSVEPs are induced by periodic visual flickering stimuli, with joint frequency-phase modulation (JFPM) being the most common modulation method. In SSVEP BCIs, calibration can improve the information transfer rate (ITR), but the calibration process for all stimuli remains time-consuming and labor-intensive. In recent years, cross-stimulus transfer learning methods have been widely adopted in SSVEP-based brain-computer interfaces to reduce the need for calibration time. However, a considerable amount of source domain stimulus data is still required to train transferable model parameters.

[0003] CMWong et al. published a cross-stimulus frequency knowledge transfer method (tlCCA) based on transfer learning in their paper "Transferring Subject-Specific Knowledge Across Stimulus Frequencies in SSVEP-Based BCIs" (IEEE Trans. Autom. Sci. Eng., 2021). The implementation steps of this method are: (1) learning spatial filter parameters from calibration data of the source frequency; (2) using superposition theory to decompose the SSVEP signal into two parts: impulse response and periodic pulse; (3) solving the impulse response and spatial filter, transferring the learned spatial filter and impulse response to the target frequency, and calculating classification features through correlation analysis and CCA to achieve target recognition. The shortcomings of this method are: using SSVEP as the source stimulus, SSVEP usually only contains one frequency component, resulting in low frequency band utilization; in order to train the common model parameters of all stimulus sources, multiple data blocks of multiple source stimuli need to be collected, which can be very time-consuming and has limited feature extraction efficiency. At least 50 seconds of calibration time is required to obtain satisfactory performance, which limits the applicability of this method in real-world applications.

[0004] In their paper "Stimulus-Stimulus Transfer Based on Time-Frequency-Joint Representation in SSVEP-Based BCIs" (IEEE Trans. Biomed. Eng., 2023), Z. Wang et al. published a cross-stimulus transfer learning method based on time-frequency joint representation. The implementation steps of this method are: (1) converting the SSVEP signal from the time domain to a time-frequency joint representation, and synchronizing the signals of different stimuli through the transformation; (2) establishing an SSVEP signal model; (3) using a multi-channel adaptive Fourier decomposition method with different phases to adaptively decompose the SSVEP signals of different source stimuli simultaneously; (4) selecting the common components across stimuli based on energy relationships, decomposition levels, and coefficient distributions; (5) using the selected decomposition coefficients to construct an SSVEP template for the target stimulus; and (6) combining spatial filters for SSVEP recognition. The shortcomings of this method are: the basis function-based method faces the challenge of selecting appropriate basis functions, and relies on the matching pursuit algorithm to solve the common model parameters, resulting in high computational overhead. Summary of the Invention

[0005] The purpose of the present invention is to address the deficiencies of the above-mentioned prior art and propose an SSVEP recognition system and method based on linear frequency modulation visual evoked potential migration to solve the following three problems: First, the existing method uses SSVEP as the source stimulus, and a single SSVEP signal contains only one frequency component, resulting in low frequency band utilization. Second, the basis function-based method faces the challenge of selecting appropriate basis functions and relies on the matching pursuit algorithm to solve the common model parameters, resulting in high computational overhead. Third, in order to train the common impulse response and common spatial filter of all stimulus sources, the existing cross-stimulus migration SSVEP recognition method needs to collect multiple data blocks of multiple source stimuli, which is very time-consuming and has limited feature extraction efficiency. It takes a long time to calibrate to achieve satisfactory performance, which is not conducive to user experience and the overall usability of the system.

[0006] The idea for achieving the purpose of the present invention is as follows: the present invention proposes to use a linear frequency modulation chirp signal with a frequency variation range of 8.0-15.0 Hz as the source stimulus, and utilizes the wide spectrum characteristics of the chirp-VEP signal frequency that varies linearly with time. A single signal can cover the frequencies of all target SSVEPs. Compared with traditional single-frequency SSVEP signals, it has higher information density and better stimulus differentiation ability, thereby solving the problem of low frequency band utilization caused by a single SSVEP signal containing only one frequency component in the prior art. In response to the high cost of learning impulse responses corresponding to different instantaneous frequencies in the prior art, the present invention learns a common impulse response for Chirp-VEPs of different instantaneous frequencies, transforms the Chirp signal to obtain a Chirp impulse matrix, and the first dimension of the transformed Chirp impulse matrix is ​​determined by the lowest frequency of the Chirp-VEP. The first dimension of the Chirp impulse matrix at different instantaneous frequencies has the same length to ensure convolution operation with the common impulse response, while the second dimension represents the information of the correct Chirp coding sequence. The present invention is based on the linear superposition theory and constructs a least squares optimization problem between a spatially filtered Chirp-VEP and a reconstructed Chirp-VEP. The spatially filtered Chirp-VEP is obtained by averaging the electroencephalogram (EEG) signals of each Chirp-VEP trial and then performing spatial filtering. The reconstructed Chirp-VEP is obtained by convolving the common impulse response of the Chirp-VEP with the Chirp impulse matrix. The alternating least squares method is selected and the Tikhonov regularization method is added to the optimization process of the least squares problem. The common impulse response and common spatial filter of the Chirp-VEP are solved by maximizing the correlation coefficient between the reconstructed Chirp-VEP and the spatially filtered Chirp-VEP. This avoids complex basis function selection and matching pursuit algorithms, greatly reduces computational complexity, and solves the problem of high computational overhead in the prior art. The present invention migrates the common impulse response and common spatial filter of Chirp-VEP into SSVEP, uses the common impulse response of Chirp-VEP to convolve with a periodic impulse matrix to reconstruct the SSVEP template signal, uses the common spatial filter of Chirp-VEP to spatially filter the single-trial SSVEP signal, and calculates the Pearson correlation coefficient between the reconstructed SSVEP template signal and the single-trial SSVEP signal after spatial filtering as the feature of the SSVEP signal. This eliminates the need to use SSVEP data of different frequencies for training and calibration, and solves the problem of the existing method that it is time-consuming to collect multiple data blocks of multiple source stimuli to train the common impulse response and the common spatial filter.The present invention takes the canonical correlation coefficient as the second feature, fuses it with the Pearson correlation coefficient to construct a feature vector, and adds filter bank analysis to fully extract the fundamental and harmonic components in the Chirp-VEP and SSVEP signals. The weighted sum of the eigenvalues ​​obtained in each subband is used as the final fused eigenvalue to identify SSVEP, thereby solving the problems of limited feature extraction efficiency in the existing technology, a long calibration time required to achieve satisfactory performance, and a disadvantageous effect on user experience and overall system usability.

[0007] The technical solutions for achieving the purpose of the present invention are as follows:

[0008] The SSVEP recognition system of the present invention includes a visual stimulus generation and presentation module, an EEG signal acquisition module, an EEG signal preprocessing module, a Chirp pulse matrix construction module, a Chirp-VEP decomposition module and an SSVEP recognition module; wherein,

[0009] The visual stimulus generation and presentation module includes a visual stimulus generation unit and a visual stimulus presentation unit, wherein the visual stimulus generation unit is used to generate a linear frequency modulation chirp signal with a frequency variation range of 8.0-15.0 Hz as a source stimulus and a sine wave modulation signal as a target stimulus for inducing SSVEP; the visual stimulus presentation unit is used to control the stimulus presentation timing and stimulus interface layout through a stimulus program, and present the visual stimulus using an LCD display;

[0010] The EEG signal acquisition module uses an EEG system with at least 9 channels to respectively acquire the subject's linear frequency modulation visual evoked potential Chirp-VEP and SSVEP EEG signals;

[0011] The EEG signal preprocessing module is used to preprocess the collected EEG signals;

[0012] The Chirp pulse matrix construction module is used to transform the Chirp signal to obtain the Chirp pulse matrix;

[0013] The Chirp-VEP decomposition module is used to construct a least squares optimization problem between the spatially filtered Chirp-VEP and the reconstructed Chirp-VEP, and to obtain the common impulse response and common spatial filter of the Chirp-VEP;

[0014] The SSVEP recognition module is used to migrate the common impulse response and common spatial filter of Chirp-VEP into SSVEP, calculate the fusion eigenvalue of each candidate target by combining correlation analysis and filter bank analysis, and select the target corresponding to the maximum fusion eigenvalue as the recognition result.

[0015] The specific steps of the SSVEP identification method of the present invention are as follows:

[0016] Step 1: The visual stimulus generation and presentation module selects a linear frequency modulation chirp signal with a frequency range of 8.0-15.0 Hz as the source stimulus and a sinusoidal wave modulation signal as the target stimulus to induce SSVEP, and collects the subject's linear frequency modulation visual evoked potential chirp-VEP and SSVEP EEG signals respectively;

[0017] Step 2: The EEG signal acquisition module uses an EEG system with at least 9 channels to collect the subject's linear frequency modulation visual evoked potential (Chirp-VEP) and SSVEP EEG signals, and then the EEG signal preprocessing module preprocesses the collected EEG signals;

[0018] Step 3: The Chirp pulse matrix construction module transforms the Chirp signal to obtain the Chirp pulse matrix;

[0019] Step 4: The Chirp-VEP decomposition module constructs a least squares optimization problem between the spatially filtered Chirp-VEP and the reconstructed Chirp-VEP to obtain the common impulse response and common spatial filter of the Chirp-VEP.

[0020] In step 5, the SSVEP recognition module migrates the common impulse response and common spatial filter of Chirp-VEP into SSVEP; combines correlation analysis and filter bank analysis to calculate the fusion eigenvalue of each candidate target, and selects the target corresponding to the maximum fusion eigenvalue as the recognition result.

[0021] Furthermore, the Chirp signal is as follows:

[0022]

[0023] Among them, S chirp represents the Chirp signal, f0 represents the initial frequency of the Chirp signal, c represents the rate of change of frequency over time, φ0 represents the initial phase, t represents the time from the start of the stimulus to the current moment, sin represents the sine operation, and π represents pi.

[0024] Furthermore, the sinusoidal wave modulation signal is as follows:

[0025]

[0026] Among them, S ssvep represents the sinusoidal wave modulation signal, f k represents the kth stimulation frequency of the sinusoidal modulation signal, φ k represents the kth stimulation phase.

[0027] Furthermore, the steps of preprocessing the collected EEG signals are as follows:

[0028] The first step is to intercept the EEG signals of each trial according to the trigger mark from the collected EEG signals to achieve precise synchronization;

[0029] The second step is to cut off the EEG signals corresponding to the visual delay period of the subjects in Chirp-VEP and SSVEP respectively to avoid the delay of the visual pathway;

[0030] In the third step, a notch filter is used to remove power line interference, and the Chirp-VEP and SSVEP are band-pass filtered to maintain the fundamental and harmonic components. The filter uses zero phase to avoid phase distortion.

[0031] Furthermore, the Chirp pulse matrix is ​​obtained by the following formula:

[0032]

[0033] Where h'(·) represents the Chirp pulse matrix obtained by transformation, i' represents the row index in the Chirp pulse matrix, j' represents the column index in the Chirp pulse matrix, h(·) represents the original pulse matrix obtained by time discretization of the Chirp signal, t(j') represents the sampling time point corresponding to the element of the j'th column, and f represents the reference frequency, which is set to the lowest value of the SSVEP frequency.

[0034] Furthermore, the Chirp-VEP of the spatial filtering is as follows:

[0035]

[0036] in, represents the spatially filtered Chirp-VEP, w c represents the common spatial filter to be optimized, Indicates that the dimension obtained by averaging the EEG signals of each Chirp-VEP trial is N e ×N p The matrix, N e Indicates the number of electrodes, N p Indicates the number of sampling points.

[0037] Furthermore, the reconstructed Chirp-VEP is as follows:

[0038]

[0039] in, represents the reconstructed Chirp-VEP signal, r cRepresents the common impulse response to be optimized, including the common response characteristics at different instantaneous frequencies, H ' c It represents the Chirp pulse matrix obtained after the transformation in step 3, and the superscript T represents the transpose operation symbol.

[0040] Furthermore, the least squares optimization problem is as follows:

[0041]

[0042] in, Represents finding w that minimizes the objective function value c and r c The operation of ||·|| 2 Represents an operation that computes the square of the Frobenius norm.

[0043] Furthermore, the steps of obtaining the common impulse response and common spatial filter of Chirp-VEP are as follows:

[0044] The first step is to randomly initialize a common space filter and satisfy them

[0045] In the second step, use the common space filter of the current iteration Update the common impulse response:

[0046]

[0047] in, represents the updated common impulse response, I represents the identity matrix, λ represents the regularization factor, (·) -1 Indicates the matrix pseudo-inverse operation;

[0048] The third step is to use Updated public space filters:

[0049]

[0050] in, represents the updated common spatial filter;

[0051] The fourth step is to calculate the objective function value after the m+1th round of iterative update:

[0052]

[0053] Among them, J (m+1) Represents the objective function value after the m+1th round of iterative update;

[0054] The fifth step is to determine whether the objective function value after the current iteration meets the convergence condition. If so, the current iteration As the common impulse response of Chirp-VEP, As the common spatial filter of Chirp-VEP; otherwise, return to the second step to continue iteration; the convergence condition is that |J (m+1) -J (m) |<ε or m+1≥M; where ε represents the convergence threshold and M represents the maximum number of iterations.

[0055] Furthermore, the steps of migrating the common impulse response and common spatial filter of Chirp-VEP to SSVEP are as follows:

[0056] The first step is to reconstruct each SSVEP template signal:

[0057]

[0058] in, represents the reconstructed k-th SSVEP template signal, r ' c represents the common impulse response of Chirp-VEP obtained in step 4, represents the periodic pulse matrix of the kth SSVEP stimulation frequency, k = 1, 2, …, K, K represents the number of SSVEP stimulation frequencies;

[0059] The second step is to perform spatial filtering on the single-trial SSVEP signal:

[0060] x ss =(w ' c ) T X ss

[0061] Among them, x ss represents the single-trial SSVEP signal after spatial filtering, w ' c represents the common spatial filter of Chirp-VEP obtained in step 4, X ss Represents the single-trial SSVEP signal matrix, the matrix dimension is N e ×N q , N q Indicates the number of sampling points.

[0062] Furthermore, the steps of the correlation analysis are as follows:

[0063] The first step is to calculate the Pearson correlation coefficient between the reconstructed SSVEP template signal and the single-trial SSVEP signal after spatial filtering as the first feature of the SSVEP signal:

[0064]

[0065] Among them, ρ k,1 represents the first eigenvalue of the k-th target, corr(·) represents the Pearson correlation coefficient calculation function;

[0066] The second step is to construct the reference signal matrix for each stimulation frequency:

[0067]

[0068] Among them, Y k represents the reference signal matrix of the kth stimulation frequency, N h Indicates the number of harmonics, n represents the time sampling point sequence, f s Indicates the sampling frequency, N p represents the pth sampling point, cos represents the cosine operation;

[0069] The third step is to calculate the canonical correlation coefficient as the second feature:

[0070] ρ k,2 =CCA(x ss ,Y k )

[0071] Among them, ρ k,2 represents the second eigenvalue of the k-th target, and CCA(·) represents the canonical correlation analysis calculation function;

[0072] The fourth step is to combine the two features obtained above by weighted summation:

[0073]

[0074] in, represents the eigenvalue of the kth target, sign(·) represents the sign function, and m represents the feature index.

[0075] Furthermore, the filter bank analysis refers to obtaining the fusion feature value of each target through the following weighted summation:

[0076]

[0077] in, represents the fusion feature value of the kth target, N fb Indicates the number of filter subbands, w(n) represents the weight coefficient of the nth subband, represents the eigenvalue of the kth target in the nth subband.

[0078] Compared with the prior art, the present invention has the following advantages:

[0079] First, since the visual stimulus generation and presentation module in the system of the present invention selects a Chirp signal with a frequency variation range of 8.0-15.0 Hz as the source stimulus, and utilizes the wide spectrum characteristics of the Chirp-VEP signal, a single signal can cover the frequencies of all target SSVEPs, overcoming the defect in the prior art that SSVEP is used as the source stimulus, which usually only contains one frequency component and leads to low frequency band utilization. Compared with the traditional single-frequency SSVEP signal, the present invention has higher information density and better stimulus discrimination ability, and significantly improves the frequency band utilization.

[0080] Second, since the method of the present invention constructs a least squares optimization problem between the spatially filtered Chirp-VEP and the reconstructed Chirp-VEP, the alternating least squares method is selected and combined with the Tikhonov regularization method. The common impulse response and common spatial filter of the Chirp-VEP are solved by maximizing the correlation coefficient between the reconstructed Chirp-VEP and the spatially filtered Chirp-VEP, thereby avoiding complex basis function selection and matching pursuit algorithm, overcoming the challenges of selecting appropriate basis functions faced by the basis function-based methods in the prior art and the shortcomings of high computational overhead caused by relying on the matching pursuit algorithm. Therefore, the present invention greatly reduces the computational complexity and system overhead, and is more suitable for real-time BCI application requirements.

[0081] Third, since the method of the present invention only uses a single Chirp-VEP signal to train the common impulse response and the common spatial filter, and at the same time fuses the two features of the Pearson correlation coefficient and the canonical correlation coefficient, and combines the filter bank analysis to perform weighted summation on the multi-band features, it can extract features more accurately and efficiently, overcoming the shortcomings of the prior art that it is time-consuming to collect multiple source domain stimulation data to train the common model parameters, and the feature extraction efficiency is limited, and a long calibration time is required to achieve satisfactory performance. The present invention greatly reduces the training cost, significantly shortens the calibration time to only 4.5 seconds, and can achieve classification accuracy and information transmission rate comparable to the current most advanced methods with the minimum calibration time, greatly improving the user experience and the actual usability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 is a block diagram of the system of the present invention;

[0083] Figure 2 is a flow chart of the method of the present invention;

[0084] Figure 3 The original Chirp pulse and the transformed Chirp pulse and the transformation method thereof are flowcharts of the present invention;

[0085] Figure 4This is a schematic diagram of Chirp-VEP decomposition of the present invention;

[0086] Figure 5 This is a comparison chart of the average classification accuracy and information transmission rate of the method of the present invention and the prior art FBCCA, tlCCA, and CIR under different time windows in the offline simulation experiment of the present invention;

[0087] Figure 6 This is a comparison chart of the average classification accuracy and information transmission rate of the method of the present invention and the prior art TRCA under different time windows in the offline simulation experiment of the present invention;

[0088] Figure 7 This is a comparison chart of the learning curves of the method of the present invention and the prior art under different calibration times of tlCCA and CIR in the offline simulation experiment of the present invention. DETAILED DESCRIPTION

[0089] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0090] Reference Figure 1 The system of the present invention is further described.

[0091] The SSVEP recognition system of the present invention includes a visual stimulus generation and presentation module, an EEG signal acquisition module, an EEG signal preprocessing module, a Chirp pulse matrix construction module, a Chirp-VEP decomposition module and an SSVEP recognition module.

[0092] The visual stimulus generation and presentation module includes a visual stimulus generation unit and a visual stimulus presentation unit, wherein the visual stimulus generation unit is used to generate a linear frequency modulation chirp signal with a frequency variation range of 8.0-15.0 Hz as a source stimulus, and encodes a sine wave modulation signal with a modulation frequency range of 8.0 Hz to 15.8 Hz, a frequency interval of 0.2 Hz, and a phase interval of 0.35π through the JFPM method as a target stimulus for inducing SSVEP; the visual stimulus presentation unit is used to develop a stimulus program using the Psychophysics Toolbox Ver.3 toolbox on the MATLAB platform to control the stimulus presentation timing and stimulus interface layout, and use an LCD display to present the visual stimulus.

[0093] The EEG signal acquisition module acquires EEG signals after the subject receives visual stimulation. In the embodiment of the present invention, a g.USBamp device from g.tec is used, nine electrodes on the occipital region are selected, and Chirp-VEP and SSVEP EEG signals of the subject are respectively acquired at a sampling frequency of 256 Hz.

[0094] The EEG signal preprocessing module preprocesses the collected EEG signals.

[0095] The Chirp pulse matrix construction module transforms the Chirp signal to obtain the Chirp pulse matrix.

[0096] The Chirp-VEP decomposition module constructs a least squares optimization problem between the spatially filtered Chirp-VEP and the reconstructed Chirp-VEP to obtain the common impulse response and common spatial filter of the Chirp-VEP.

[0097] The SSVEP recognition module migrates the common impulse response and common spatial filter of Chirp-VEP into SSVEP, calculates the fusion eigenvalue of each candidate target by combining correlation analysis and filter bank analysis, and selects the target corresponding to the maximum fusion eigenvalue as the recognition result.

[0098] Reference Figure 2 The implementation steps of the SSVEP identification method of the present invention are further described.

[0099] Step 1: Design a visual stimulation interface.

[0100] A 23.8-inch LCD monitor (HP Z24nfG2, 5 millisecond response time) with a resolution of 1920 × 1080 pixels and a refresh rate of 60 Hz was used. The stimulation program was developed using the Psychophysics Toolbox Ver. 3 in MATLAB to achieve precise control of the timing of visual stimulation.

[0101] When constructing the Chirp-VEP stimulation interface, a 120×120 pixel square stimulus block is displayed in the center of the screen, and the character "A" is displayed inside the stimulus block to maintain attention. The brightness modulation sequence is generated according to the Chirp signal formula:

[0102]

[0103] The initial frequency f0 = 8.0 Hz, the modulation rate c = 1.75 Hz / s to ensure that the frequency range of 8.0–15.0 Hz is covered within 4 seconds, and the initial phase φ0 = 0π. Stimulus temporal sampling is performed at the display refresh rate of 60 Hz, and the luminance value corresponding to each frame is calculated and converted into the display RGB color value.

[0104] To construct the SSVEP stimulation interface, 40 stimulus targets were created in a 5×8 matrix layout. Each stimulus block was 120×120 pixels in size and contained 26 English letters, 10 numbers, and 4 symbols. Targets were spaced 130 pixels horizontally and 80 pixels vertically. Multiple targets were presented in parallel through synchronized display refresh. All stimulus blocks were coded and modulated using the JFPM method. An independent luminance modulation sequence was generated for each target according to the SSVEP stimulus formula:

[0105]

[0106] Among them, f k represents the kth stimulation frequency of the sinusoidal modulation signal. The frequency of the 40 targets increases from 8.0 Hz to 15.8 Hz at 0.2 Hz intervals. k represents the kth stimulation phase, which is distributed at intervals of 0.35π starting from 0π.

[0107] Step 2: Collect EEG signals.

[0108] The embodiment of the present invention uses the g.USBamp device of g.tec company to collect EEG signals, and sets the sampling frequency to 256Hz. According to the international 10-20 system, 9 recording electrodes are positioned: 1-Pz, 2-O1, 3-O2, 4-POz, 5-PO8, 6-Oz, 7-PO7, 8-PO3, 9-PO4. Special conductive glue is used to improve the conductivity between the electrodes and the subject's scalp. The impedance of all electrodes is kept below 10kΩ. The ground electrode is placed on the FPz electrode, and the reference electrode is placed on the subject's earlobe. In order to keep the stimulation program synchronized with the EEG acquisition, the trigger mark is sent from the stimulation program to the EEG acquisition device via the parallel port of the computer.

[0109] Step 3: Preprocess the EEG signal.

[0110] The EEG signals of each trial were intercepted from the collected EEG signals according to the trigger markers to achieve precise synchronization. For Chirp-VEP, 4 seconds of data were extracted, corresponding to 10 trials, and for SSVEP, 3 seconds of data were extracted, corresponding to 40 trials in each data block. Taking into account visual delay, the first 0.115 seconds of data were truncated for Chirp-VEP, and the first 0.135 seconds of data were truncated for SSVEP. Digital filtering was applied: first, a 50Hz notch filter was used to remove power line interference, and then a 7-90Hz bandpass filter was used to maintain the fundamental and harmonic components. The filter used zero phase to avoid phase distortion and ensure the integrity of the time domain characteristics of the visual evoked potential.

[0111] Step 4: Construct the Chirp pulse matrix.

[0112] According to the linear change characteristics of Chirp-VEP frequency, the original Chirp pulse H is obtained by time discretization of Chirp signal. c The spacing of the chirp pulses is gradually reduced and the length is gradually shortened to adapt to the convolution of impulse responses of different instantaneous frequencies. Taking into account the characteristics of the common impulse response, the present invention designs a transformed Chirp pulse H' c ,like Figure 3As shown in (a), the first dimension has the same length to ensure the convolution operation with the common impulse response. The specific transformation method is as follows Figure 3 (b) is shown. The length of the first dimension of the transformed Chirp pulse is determined by the lowest frequency of the Chirp-VEP, which is set to 8 Hz in the embodiment of the present invention to include more user-personalized feature information. During the transformation process, h(i, j) represents the i-th column and j-th row of the original Chirp pulse, and h'(i', j') represents the i'th column and j'th row of the transformed Chirp pulse. The transformation formula is:

[0113]

[0114] The reference frequency f is set to 8.0 Hz, the lowest value of the Chirp-VEP frequency, to determine the length of the common impulse response; the linear modulation rate c is 1.75 Hz / s; the time index t(j') is the sampling time point corresponding to the j'th column; and the initial frequency f0 is set to 8.0 Hz.

[0115] Step 5: Train the Chirp-VEP common model parameters.

[0116] The Chirp-VEP signals of each trial are averaged to obtain a dimension of N e ×N p Matrix N e Indicates the number of electrodes. In the embodiment of the present invention, the number of electrodes is 9. N p Indicates the number of sampling points.

[0117] Use the common spatial filter w c Perform spatial filtering on the multi-channel Chirp-VEP signal to extract the Chirp-VEP components:

[0118] Based on the linear superposition theory, the decomposition diagram of Chirp-VEP is as follows Figure 4 As shown, through the common impulse response r c The Chirp pulse matrix H' obtained by transformation in step 4 c Convolutional reconstruction Chirp-VEP:

[0119] In order to simultaneously solve the common spatial filter and impulse response, a least squares optimization problem between the spatially filtered Chirp-VEP and the reconstructed Chirp-VEP is constructed:

[0120]

[0121] in, Represents finding w that minimizes the objective function valuec and r c The operation of ||·|| 2 Represents an operation that computes the square of the Frobenius norm.

[0122] Use the alternating least squares method to solve the optimization problem. The steps are as follows:

[0123] The first step is to randomly initialize a common space filter and satisfy them

[0124] In the second step, use the common space filter of the current iteration Update the common impulse response: in, represents the updated common impulse response, I represents the identity matrix, λ represents the regularization factor, which is set to 0.2 in the embodiment of the present invention, (·) -1 Represents the matrix pseudo-inverse operation.

[0125] The third step is to use Updated public space filters: in, represents the updated common spatial filter.

[0126] The fourth step is to calculate the objective function value after the m+1th round of iterative update:

[0127] The fifth step is to determine whether the objective function value after the current iteration meets the convergence condition. If so, the current iteration As the common impulse response of Chirp-VEP, As the common spatial filter of Chirp-VEP; otherwise, return to the second step to continue iteration; the convergence condition is that |J (m+1) -J (m) |<ε or m+1≥M; wherein ε represents the convergence threshold, M represents the maximum number of iterations, and in the embodiment of the present invention, ε is set to 0.001 and M is set to 200.

[0128] Step 6: Migrate the common model parameters to SSVEP for feature extraction and classification.

[0129] Step 6.1, use the common impulse response r of the Chirp-VEP obtained from step 5 ' c Reconstruct the SSVEP template signal. For the kth SSVEP stimulation frequency, reconstruct the corresponding template signal through the convolution operation: in, represents the reconstructed k-th SSVEP template signal, The periodic pulse matrix representing the k-th SSVEP stimulation frequency, k=1, 2, ..., K, where K represents the number of SSVEP stimulation frequencies and is set to 40 in the embodiment of the present invention.

[0130] Use the common spatial filter w' of the Chirp-VEP obtained in step 5 c Perform spatial filtering on the single-trial SSVEP signal: Among them, x ss represents the single-trial SSVEP signal after spatial filtering, X ss Represents the single-trial SSVEP signal matrix, whose dimension is the product of the number of electrodes and the number of sampling points.

[0131] Step 6.2: Calculate the Pearson correlation coefficient between the reconstructed SSVEP template signal and the spatially filtered single-trial SSVEP signal as the first feature of the SSVEP signal:

[0132] Construct the reference signal matrix for each stimulation frequency:

[0133]

[0134] Among them, Y k represents the reference signal matrix of the kth stimulation frequency, N h Indicates the number of harmonics. In the embodiment of the present invention, N h Set to 5.

[0135] Calculate the canonical correlation coefficient as the second feature: ρ k,2 =CCA(x ss ,Y k ), where ρ k,2 represents the second eigenvalue of the k-th target, and CCA(·) represents the canonical correlation analysis calculation function.

[0136] The two features obtained above are fused by weighted summation:

[0137]

[0138] in, represents the eigenvalue of the kth target, sign(·) represents the sign function, and m represents the feature index.

[0139] In step 6.3, to fully extract the fundamental and harmonic components, a filter bank analysis is performed, with five subbands set, and the nth filter subband range is [7×n, 90] Hz. The filter bank uses a zero-phase Chebyshev Type I IIR filter, implemented using the filtfilt() function in MATLAB. The final fused eigenvalue is obtained through weighted summation:

[0140]

[0141] Among them, the number of subbands N fb Set to 5, w(n) = n -1.25 +0.25 represents the weight of each sub-band, Represents the eigenvalue of the nth subband.

[0142] Compare the fused eigenvalues ​​of all 40 candidate targets and select the target with the largest eigenvalue as the recognition result. Display the recognized target character in the designated area above the stimulation interface, completing the command output. This process can be performed continuously in online spelling mode.

[0143] The effects of the present invention are further described below in conjunction with simulation experiments:

[0144] 1. Simulation experiment conditions:

[0145] The hardware platform for the simulation experiment of the present invention is: g.USBamp EEG system from g.tec, with a sampling frequency of 256 Hz, nine occipital electrodes (1-Pz, 2-PO3, 3-PO4, 4-PO7, 5-PO8, 6-POz, 7-O1, 8-PO2, 9-Oz), and a 23.8-inch LCD monitor (HP Z24nfG2, response time: 5 ms).

[0146] The software platform for the simulation experiment of the present invention is: Windows 10 operating system and Psychophysics Toolbox Ver.3 (PTB-3) in MATLAB platform.

[0147] The input signals used in the simulation experiments were chirp-VEP and SSVEP signals. The chirp-VEP signal had an initial frequency of 8.0 Hz, a modulation rate of 1.75 Hz / s, and a visual stimulation duration of 4 seconds. The SSVEP signal contained 40 stimuli and used joint frequency phase modulation (JFPM) with a frequency range of 8.0 Hz to 15.8 Hz, a frequency interval of 0.2 Hz, a phase interval of 0.35π, and a visual stimulation duration of 3 seconds.

[0148] 2. Simulation content and results analysis:

[0149] There are three simulation experiments of the present invention.

[0150] Simulation Experiment 1 simulates the SSVEP classification performance of the linear frequency modulation visual evoked potential (SSVEP) recognition method based on transfer of SSVEPs in different time windows. The SSVEP used in Simulation Experiment 1 includes 40 stimulus frequencies ranging from 8.0 Hz to 15.8 Hz, with a frequency interval of 0.2 Hz and a phase interval of 0.35π. JFPM modulation is used, and the visual stimulation duration is 3 seconds. Chirp-VEP uses a single stimulus with an initial frequency of 8.0 Hz, a modulation frequency of 1.75 Hz / s, and a visual stimulation duration of 4 seconds. Seven data blocks of SSVEP signals were collected for each subject, each containing 40 trials corresponding to 40 different stimulus frequencies. One data block of Chirp-VEP signals was collected, containing 10 repeated trials.

[0151] The simulation experiment 1 of the present invention adopts the method of the present invention and four existing technologies to obtain the classification accuracy and information transmission rate under different time windows, and then plots the relationship between the obtained classification accuracy, information transmission rate and time window as shown below: Figure 5 and Figure 6 The curve shown.

[0152] In simulation experiment 1, the four existing technologies used are:

[0153] Prior art 1 refers to the filter bank canonical correlation analysis method (FBCCA) proposed by Lin et al. in “Frequency recognition based on canonical correlation analysis for SSVEP-based BCIs, IEEE Trans. Biomed. Eng., vol. 53, no. 12, pp. 2610–2614, 2006”.

[0154] Prior art 2 refers to the canonical correlation analysis method for cross-stimulus transfer learning, abbreviated as tlCCA, proposed by Wong et al. in their paper "Transferring subject-specific knowledge across stimulus frequencies in SSVEP-based BCIs, IEEE Trans. Autom. Sci. Eng., vol. 18, no. 2, pp. 552–563, 2021".

[0155] Prior art 3 refers to the common impulse response method, abbreviated as CIR, proposed by Xiong et al. in their paper "Cross-stimulus transfer method using common impulse response for fast calibration of SSVEP-based BCIs, IEEE Trans. Instrum. Meas., 2024".

[0156] Prior art 4 refers to the task-related component analysis method (TRCA) proposed by Nakanishi et al. in their paper "Enhancing detection of SSVEPs for a high-speed brain speller using task-related component analysis, IEEE Trans. Biomed. Eng., vol. 65, no. 1, pp. 104–112, 2018".

[0157] Simulation Experiment 2 simulates the information transfer rate of the SSVEP recognition method based on linear frequency modulation visual evoked potential transfer under different calibration times. The calibration time in Simulation Experiment 2 is calculated based on the number of source stimuli, the number of training data blocks, the visual stimulation time, and the gaze shift time.

[0158] The simulation experiment 2 of the present invention adopts the method of the present invention and the prior art 2 and the prior art 3 to obtain the information transmission rate under different calibration times, and then plots the relationship between the obtained information transmission rate and the calibration time as shown below: Figure 7 The prior art 2 and prior art 3 used in simulation experiment 2 are the same as those used in simulation experiment 1.

[0159] Simulation Experiment 3 was an online performance verification simulation of the SSVEP recognition method based on linear frequency modulation visual evoked potential transfer. Simulation Experiment 3 used an online spelling task for verification. All 16 subjects participated in the online experiment, and the visual stimulation duration and gaze shift duration were adjusted based on each subject's individual performance.

[0160] Simulation Experiment 3 of the present invention uses the method of the present invention and prior art 2 and prior art 3 to obtain the classification accuracy and information transmission rate of each of the 16 subjects in the online experiment. In Simulation Experiment 3, prior art 2 and prior art 3 are the same as those used in Simulation Experiment 1.

[0161] The following combination Figure 5-7 The simulation diagram of the present invention is further described.

[0162] Figure 5 The horizontal axis of (a) represents the test time window in seconds, and the vertical axis represents the classification accuracy in percentage. The yellow dashed line represents the relationship between the classification accuracy and the time window obtained using the prior art FBCCA, the red dashed line represents the relationship between the classification accuracy and the time window obtained using the prior art tlCCA, the green dashed line represents the relationship between the classification accuracy and the time window obtained using the prior art CIR, and the blue solid line represents the relationship between the classification accuracy and the time window obtained using the method of the present invention.

[0163] from Figure 5 (a) As can be seen, the classification accuracy achieved by the method of the present invention significantly outperforms the existing FBCCA method across all time windows. Compared with the existing tlCCA method, significant differences only exist in the 0.4-second and 0.6-second time windows. Compared with the existing CIR method, no significant differences are observed across all time windows, demonstrating that the method of the present invention achieves classification performance comparable to existing SSVEP methods.

[0164] Figure 5 The horizontal axis of (b) represents the test time window in seconds, and the vertical axis represents the information transmission rate in bits / min. The meaning of the curve labels is the same as (a). Figure 5 (b) As can be seen, the information transmission rate of the present invention significantly differs from that of the prior art tlCCA only in the 0.4-second and 0.6-second time windows. Across all time windows, the information transmission rate of the present invention significantly outperforms the prior art FBCCA and is comparable to that of the prior art CIR.

[0165] The following combination Figure 6 The simulation diagram of the present invention is further described.

[0166] Figure 6 The horizontal axis of (a) represents the test time window in seconds, and the vertical axis represents the classification accuracy in percentage. The red solid line represents the TRCA method using 6 sets of training data blocks, the red dotted line represents the TRCA method using 2 sets of training data blocks, and the blue solid line represents the method of the present invention, which only requires 1 set of Chirp-VEP data blocks. Figure 6 (a) It can be seen that when TRCA is trained with 6 sets of calibration data, the method of the present invention outperforms TRCA only in the 1.6 second time window. When TRCA is trained with 2 sets of calibration data, the method of the present invention significantly outperforms TRCA after 0.4 seconds.

[0167] Figure 6The horizontal axis of (b) represents the test time window in seconds, and the vertical axis represents the information transmission rate in bits / min. The meaning of the curve labels is the same as (a). Figure 6 (b) It can be seen that in terms of information transfer rate performance, when TRCA is trained with 6 sets of calibration data, its information transfer rate is better than that of the method of the present invention before 0.8 seconds, but after 0.8 seconds, the two methods perform equally well, both achieving a stable information transfer rate of approximately 150 bits / min. When TRCA is trained with only 2 sets of calibration data, its information transfer rate is significantly lower than that of the method of the present invention after 0.4 seconds. It is worth noting that TRCA requires calibration training of multiple data blocks for all 40 SSVEP stimuli, while the method of the present invention only requires a single Chirp-VEP data block to achieve comparable performance, indicating that the method of the present invention has significant advantages in calibration efficiency and practical application value.

[0168] The following combination Figure 7 The simulation diagram of FIG. 1 further describes the advantage of the calibration efficiency of the present invention.

[0169] Figure 7 The horizontal axis represents the calibration time in seconds, and the vertical axis represents the information transmission rate in bits / min. The red solid line represents the learning curve of the prior art tlCCA at different calibration times, the green solid line represents the learning curve of the prior art CIR at different calibration times, and the blue thick line represents the learning curve of the method of the present invention. Figure 7 It can be seen that the method of the present invention only requires 4.5 seconds of calibration time to achieve stable information transmission rate performance. The existing CIR method requires 18.0 seconds to achieve similar performance to the method of the present invention. The information transmission rate of the existing tlCCA method only begins to improve after 13.5 seconds of calibration, and even after 45.0 seconds of calibration, it still fails to reach the performance level of the method of the present invention.

[0170] The performance results of the five methods are evaluated using the three evaluation indicators of classification accuracy, information transmission rate and calibration time. The classification accuracy P, information transmission rate ITR and calibration time T are calculated using the following formula: cal :

[0171] Classification accuracy

[0172] Among them, N cor represents the number of correctly identified trials, N incor Indicates the number of incorrectly identified trials.

[0173] Information transmission rate

[0174] Where T represents the target selection time (including visual stimulation time and sight shift time), N t represents the number of SSVEP stimulation frequencies, and P represents the classification accuracy.

[0175] Calibration time T cal =N src ×N blk ×(T stim +T shft )

[0176] Among them, N src Indicates the number of source stimuli, N blk Indicates the number of training data blocks, T stim represents the visual stimulation time, T shft Indicates the time of gaze shift.

[0177] The calculation results of the calibration time and information transmission rate of the present invention and various prior art methods in the offline simulation experiment are shown in Table 1:

[0178] Table 1 Analysis of calibration time and information transmission rate of the present invention and various prior art methods in offline simulation

[0179]

[0180]

[0181] As can be seen from Table 1, the information transfer rate of the present invention is 138.71 bits / min, which is comparable to the existing best cross-stimulus transfer SSVEP identification methods tlCCA (138.62 bits / min) and CIR (138.65 bits / min). Although TRCA achieves the highest information transfer rate, its calibration time is as long as 600 seconds. The calibration time of the present invention is only 4.5 seconds, which is a 90% reduction compared to the 50-second calibration time of tlCCA and a 77.5% reduction compared to the 20-second calibration time of CIR, demonstrating that the present invention can achieve excellent SSVEP classification performance in a very short calibration time.

[0182] The practical application effect of the present invention is further explained below in conjunction with online experiments:

[0183] To verify the effectiveness of the proposed method in practical applications, an online spelling task experiment was conducted. Sixteen subjects participated in the online test, and the system adaptively adjusted the visual stimulation duration and gaze shift duration based on each subject's individual performance. The detailed results of the online experiment are shown in Table 2:

[0184] Table 2 Performance comparison between the method of the present invention and the prior art in online experiments

[0185]

[0186]

[0187] The online experimental results in Table 2 show that the proposed method achieved an average classification accuracy of 91.25±5.63% and an average information transfer rate of 156.39±37.11 bits / min. These performance indicators are very similar to those of the existing methods, tlCCA (accuracy 90.63±7.27%, ITR 156.30±44.81 bits / min) and CIR (accuracy 90.94±6.38%, ITR 155.39±36.89 bits / min). Among the 16 subjects, the proposed method achieved good performance in the majority of subjects, and in some subjects (e.g., S4 and S5), it showed slight improvement over the comparison methods. While the three methods achieved comparable final performance, the proposed method required only 4.5 seconds of calibration, compared to 50 seconds for tlCCA and 20 seconds for CIR. This means that in practical brain-computer interface applications, the proposed method can significantly shorten system initialization time, significantly improving user experience and practicality.

[0188] The above simulation experiments show that the method of the present invention can effectively extract common impulse responses and common spatial filters from a single Chirp-VEP signal, and efficiently migrate common model parameters to the SSVEP brain-computer interface system for feature extraction. By utilizing the designed dual-feature fusion strategy and filter bank analysis, accurate and efficient SSVEP signal recognition can be achieved. This solves the problems in the existing technical methods that a single SSVEP signal only contains one frequency component, resulting in low frequency band utilization, multiple source stimulation data are required for training, resulting in long calibration time, and high computational overhead based on the basis function method. It achieves classification performance comparable to the existing optimal cross-stimulus transfer SSVEP recognition method with only 4.5 seconds of calibration time. It is a very efficient and practical SSVEP recognition method.

Claims

1. A SSVEP recognition system based on linear frequency modulation visual evoked potential migration, characterized in that: The system includes a visual stimulus generation and presentation module, an EEG signal acquisition module, an EEG signal preprocessing module, a Chirp pulse matrix construction module, a Chirp-VEP decomposition module and an SSVEP recognition module; wherein, The visual stimulus generation and presentation module includes a visual stimulus generation unit and a visual stimulus presentation unit, wherein the visual stimulus generation unit is used to generate a linear frequency modulation chirp signal with a frequency variation range of 8.0-15.0 Hz as a source stimulus and a sine wave modulation signal as a target stimulus for inducing SSVEP; the visual stimulus presentation unit is used to control the stimulus presentation timing and stimulus interface layout through a stimulus program, and present the visual stimulus using an LCD display; The EEG signal acquisition module uses an EEG system with at least 9 channels to respectively acquire the subject's linear frequency modulation visual evoked potential Chirp-VEP and SSVEP EEG signals; The EEG signal preprocessing module is used to preprocess the collected EEG signals; The Chirp pulse matrix construction module is used to transform the Chirp signal to obtain the Chirp pulse matrix; The Chirp-VEP decomposition module is used to construct a least squares optimization problem between the spatially filtered Chirp-VEP and the reconstructed Chirp-VEP, and to obtain the common impulse response and common spatial filter of the Chirp-VEP; The SSVEP recognition module is used to migrate the common impulse response and common spatial filter of Chirp-VEP into SSVEP, calculate the fusion eigenvalue of each candidate target by combining correlation analysis and filter bank analysis, and select the target corresponding to the maximum fusion eigenvalue as the recognition result.

2. A SSVEP recognition method based on linear frequency modulation visual evoked potential migration according to the SSVEP recognition system of claim 1, characterized in that: The steps of the identification method include the following: Step 1: The visual stimulus generation and presentation module selects a linear frequency modulation chirp signal with a frequency range of 8.0-15.0 Hz as the source stimulus and a sine wave modulation signal as the target stimulus to induce SSVEP; Step 2: The EEG signal acquisition module uses an EEG system with at least 9 channels to collect the subject's linear frequency modulation visual evoked potential (Chirp-VEP) and SSVEP EEG signals, and then the EEG signal preprocessing module preprocesses the collected EEG signals; Step 3: The Chirp pulse matrix construction module transforms the Chirp signal to obtain the Chirp pulse matrix; Step 4: The Chirp-VEP decomposition module constructs a least squares optimization problem between the spatially filtered Chirp-VEP and the reconstructed Chirp-VEP to obtain the common impulse response and common spatial filter of the Chirp-VEP. In step 5, the SSVEP recognition module migrates the common impulse response and common spatial filter of Chirp-VEP into SSVEP; combines correlation analysis and filter bank analysis to calculate the fusion eigenvalue of each candidate target, and selects the target corresponding to the maximum fusion eigenvalue as the recognition result.

3. The SSVEP identification method according to claim 2, characterized in that: The Chirp signal described in step 1 is as follows: Among them, S chirp represents the Chirp signal, f0 represents the initial frequency of the Chirp signal, c represents the rate of change of frequency over time, φ0 represents the initial phase, t represents the time from the start of the stimulus to the current moment, sin represents the sine operation, and π represents pi.

4. The SSVEP identification method according to claim 3, characterized in that: The sine wave modulation signal described in step 1 is as follows: Among them, S ssvep represents the sinusoidal wave modulation signal, f k represents the kth stimulation frequency of the sinusoidal modulation signal, φ k represents the kth stimulation phase.

5. The SSVEP identification method according to claim 4, characterized in that: The steps of the pre-processing described in step 2 are as follows: The first step is to intercept the EEG signals of each trial according to the trigger mark from the collected EEG signals to achieve precise synchronization; The second step is to cut off the EEG signals corresponding to the visual delay period of the subjects in Chirp-VEP and SSVEP respectively to avoid the delay of the visual pathway; In the third step, a notch filter is used to remove power line interference, and the Chirp-VEP and SSVEP are band-pass filtered to maintain the fundamental and harmonic components. The filter uses zero phase to avoid phase distortion.

6. The SSVEP identification method according to claim 5, characterized in that: The Chirp pulse matrix in step 3 is obtained by the following formula: Where h'(·) represents the Chirp pulse matrix obtained by transformation, i' represents the row index in the Chirp pulse matrix, j' represents the column index in the Chirp pulse matrix, h(·) represents the original pulse matrix obtained by time discretization of the Chirp signal, t(j') represents the sampling time point corresponding to the element of the j'th column, and f represents the reference frequency, which is set to the lowest value of the SSVEP frequency.

7. The SSVEP identification method according to claim 6, characterized in that: The Chirp-VEP of the spatial filtering described in step 4 is as follows: in, represents the spatially filtered Chirp-VEP, w c represents the common spatial filter to be optimized, Indicates that the dimension obtained by averaging the EEG signals of each Chirp-VEP trial is N e ×N p The matrix, N e Indicates the number of electrodes, N p Indicates the number of sampling points.

8. The SSVEP identification method according to claim 7, characterized in that: The reconstructed Chirp-VEP described in step 4 is as follows: in, represents the reconstructed Chirp-VEP signal, r c Represents the common impulse response to be optimized, including the common response characteristics at different instantaneous frequencies, H' c It represents the Chirp pulse matrix obtained after the transformation in step 3, and the superscript T represents the transpose operation symbol.

9. The SSVEP identification method according to claim 8, characterized in that: The least squares optimization problem described in step 4 is as follows: in, Represents finding w that minimizes the objective function value c and r c The operation of ||·|| 2 Represents an operation that computes the square of the Frobenius norm.

10. The SSVEP identification method according to claim 9, characterized in that: The steps for finding the common impulse response and common spatial filter of Chirp-VEP described in step 4 are as follows: The first step is to randomly initialize a common space filter and satisfy them In the second step, use the common space filter of the current iteration Update the common impulse response: in, represents the updated common impulse response, I represents the identity matrix, λ represents the regularization factor, (·) -1 Indicates the matrix pseudo-inverse operation; The third step is to use Updated public space filters: in, represents the updated common spatial filter; The fourth step is to calculate the objective function value after the m+1th round of iterative update: Among them, J (m+1) Represents the objective function value after the m+1th round of iterative update; The fifth step is to determine whether the objective function value after the current iteration meets the convergence condition. If so, the current iteration As the common impulse response of Chirp-VEP, As the common spatial filter of Chirp-VEP; otherwise, return to the second step to continue iteration; the convergence condition is that |J (m+1) -J (m) |<ε or m+1≥M; where ε represents the convergence threshold and M represents the maximum number of iterations.

11. The SSVEP identification method according to claim 10, characterized in that: The steps described in step 5 to migrate the common impulse response and common spatial filter of Chirp-VEP to SSVEP are as follows: The first step is to reconstruct each SSVEP template signal: in, Represents the reconstructed k-th SSVEP template signal, r' c represents the common impulse response of Chirp-VEP obtained in step 4, represents the periodic pulse matrix of the kth SSVEP stimulation frequency, k = 1, 2, …, K, K represents the number of SSVEP stimulation frequencies; The second step is to perform spatial filtering on the single-trial SSVEP signal: x ss =(w' c ) T X ss Among them, x ss represents the single-trial SSVEP signal after spatial filtering, w' c represents the common spatial filter of Chirp-VEP obtained in step 4, X ss Represents the single-trial SSVEP signal matrix, the matrix dimension is N e ×N q , N q Indicates the number of sampling points.

12. The SSVEP identification method according to claim 11, characterized in that: The steps of the correlation analysis described in step 5 are as follows: The first step is to calculate the Pearson correlation coefficient between the reconstructed SSVEP template signal and the single-trial SSVEP signal after spatial filtering as the first feature of the SSVEP signal: Among them, ρ k,1 represents the first eigenvalue of the k-th target, corr(·) represents the Pearson correlation coefficient calculation function; The second step is to construct the reference signal matrix for each stimulation frequency: Among them, Y k represents the reference signal matrix of the kth stimulation frequency, N h Indicates the number of harmonics, n represents the time sampling point sequence, f s Indicates the sampling frequency, N p represents the pth sampling point, cos represents the cosine operation; The third step is to calculate the canonical correlation coefficient as the second feature: r k,2 =CCA(x ss ,Y k ) Among them, ρ k,2 represents the second eigenvalue of the k-th target, and CCA(·) represents the canonical correlation analysis calculation function; The fourth step is to combine the two features obtained above by weighted summation: in, represents the eigenvalue of the kth target, sign(·) represents the sign function, and m represents the feature index.

13. The SSVEP identification method according to claim 12, characterized in that: The filter bank analysis described in step 5 is to obtain the fusion feature value of each target through the following weighted summation: in, represents the fusion feature value of the kth target, N fb Indicates the number of filter subbands, w(n) represents the weight coefficient of the nth subband, represents the eigenvalue of the kth target in the nth subband.

Images