Continuous word spelling method based on steady-state visual evoked potentials

Abstract

The application discloses a continuous word spelling method based on steady-state visual evoked potentials, and mainly solves the problems of low feature extraction efficiency and discontinuous spelling in the prior art.The implementation scheme is as follows: all characters in a stimulation interface are coded by using a joint frequency phase modulation paradigm; a user gazes at the characters in the stimulation interface to induce steady-state visual evoked potentials, and training signals and test signals are obtained; model parameters of a reconstructed template signal are acquired by using the training signals; steady-state visual evoked potential signals of all gaze stages in the test signals are intercepted; feature extraction and identification are performed on the steady-state visual evoked potential signal segments of all gaze stages by using the template signal; and character sequences in the identification results are combined into words to complete continuous word spelling input.The application improves the feature extraction efficiency and identification accuracy of the steady-state visual evoked potential signals, improves the continuity of spelling and the information transmission rate, and can be used for word spelling in a human-computer interaction process.

Description

Technical Field

[0001] This invention belongs to the field of signal processing technology, and specifically relates to a continuous word spelling method based on steady-state visual evoked potentials, which can be used for word spelling input in human-computer interaction. Background Technology

[0002] Existing brain-computer interface technology based on steady-state visual evoked potentials (SSVEP) is widely used in spelling input systems due to its diverse commands, high information transmission rate, and simple configuration. A typical SSVEP-based spelling system includes a visual stimulation interface, a signal acquisition module, a signal processing module, and an information feedback module. The visual stimulation interface contains several target blocks that flash at specific frequencies. Each target block represents a specific character, namely a letter, number, or punctuation mark. When the user focuses on the target block corresponding to the character they want to input, an SSVEP signal with the same or higher harmonic components as the target block's flashing frequency is induced in the occipital region of their head. By acquiring, processing, and recognizing the SSVEP signal, the system can determine the character the user wants to input, thus completing the spelling input.

[0003] To accurately identify user-inputted characters, feature extraction and recognition of SSVEPs are crucial. Commonly used feature extraction algorithms include those based on reference signals and those based on template matching. For example, Zhonglin Lin's 2006 paper "Frequency Recognition Based on Canonical Correlation Analysis for SSVEP-Based BCIs," published in IEEE Transactions on Biological Engineering, proposed a reference signal-based SSVEP recognition method. This method requires no training; it first constructs sine and cosine reference signals corresponding to all stimulus frequencies, then performs correlation analysis between the test signal and all sine and cosine reference signals, using the obtained correlation coefficients as features for SSVEP frequency recognition. The frequency with the highest correlation coefficient is taken as the SSVEP signal frequency. However, this method only extracts frequency domain features and does not fully utilize the time domain and phase information of the signal. Therefore, it has low feature extraction efficiency and inaccurate recognition for SSVEPs within a short time window.

[0004] To further improve the efficiency of SSVEP feature extraction, researchers have applied template matching-based methods. For example, Masaki Nakanishi's 2018 paper, "Enhancing Detection of SSVEPs for a High-Speed ​​BrainSpeller Using Task-Related Component Analysis," published in IEEE Transactions on Biological Engineering, constructs a template signal by averaging training signals, uses a task-related component spatial filter to spatially filter the EEG signal, and then performs correlation analysis between the test signal and the spatially filtered template signal. The resulting correlation coefficient is used as a feature for SSVEP frequency recognition. This method effectively extracts the frequency, time-domain, and phase information of SSVEP signals, significantly improving feature extraction efficiency. However, due to individual user differences and the brain's endogenous attention allocation mechanism, the distance differences between all characters in the spelling system's stimulus interface result in different requirements for gaze shift time. Therefore, this method uses a fixed gaze shift time, increasing the time required for target character selection, leading to a lack of continuity in the spelling system's input process and low information transmission rate.

[0005] To achieve continuous, flexible, and high-speed brain-computer interface (BCI) control technology, researchers have conducted extensive studies on SSVEP-based BCI control methods and decoding algorithms. For example, patent application CN111487988 A discloses a "brain-controlled drone method based on steady-state visual evoked potential (SSVEP) brain-computer interface." This method uses a sliding time window strategy based on task-related component analysis and a linear discriminant model to extract features and classify EEG signals that increase in a fixed step size, achieving continuous, flexible, and stable control of the drone. Patent application CN114371784 A discloses a "brain-computer interface decoding method for steady-state visual evoked potentials." This method uses periodic cyclic shifting to expand the original data or construct a template signal, and then uses the expanded data or template signal to decode the SSVEP. In both of these methods, because a fixed-step sliding window is used, for application scenarios with non-fixed-step displacement, the original SSVEP features in the constructed expanded data or template signal will be broken, leading to reduced feature extraction efficiency or even unusable features. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the existing technology and propose a continuous word spelling method based on steady-state visual evoked potentials, which is applicable to both fixed-time character spelling and non-fixed-time continuous word spelling, thereby improving the feature extraction efficiency and recognition accuracy of SSVEP.

[0007] To achieve the above objectives, the technical solution of the present invention includes the following steps:

[0008] (1) The 40 target blocks in the stimulation interface flash according to the joint frequency phase modulation paradigm. Each target represents a different character, including 26 English letters, 10 Arabic numerals and 4 common symbols.

[0009] (2) The user gazes at the flashing target block on the stimulus interface to induce steady-state visual evoked potential (SSVEP) signals. The SSVEP signals from c channels are collected and preprocessed to obtain the training signal X. train and test signal X test Where c≥9;

[0010] (3) Obtain the model parameters for reconstructing the SSVEP template signal:

[0011] (3a) Based on the linear superposition theory generated by SSVEP, construct a framework for the training signal X. train Least squares problem with reconstructed SSVEP template signals:

[0012]

[0013] in w represents the average of the multiple trial training signals and is then spatially filtered. i For spatial filters, r represents the average training signal from multiple trials; i Τ H i The reconstructed SSVEP template signal, r i H represents the user's response to different visual stimuli. i This indicates the frequency and phase information of visual stimuli;

[0014] (3b) Solve the least squares problem in (3a) to obtain the impulse response used to reconstruct the SSVEP template signal: r = {r1, r2, ..., r i ,…,r 40}, and spatial filters: w={w1,w2,…,w i ,…,w 40}, where r i w represents the impulse response at the i-th stimulus frequency. i This represents the spatial filter at the i-th frequency;

[0015] (4) Test signal segmentation processing:

[0016] (4a) Using the spatial filter w obtained in (3b) i For the test signal X test Perform spatial filtering to obtain the spatially filtered signal.

[0017] (4b) Test signal after spatial filtering Perform a continuous wavelet transform to obtain the wavelet coefficients W, and then normalize them to obtain the normalized wavelet coefficients.

[0018] (4c) Based on the amplitude differences of the normalized wavelet coefficients, slice the spatially filtered test signal, i.e., extract the test signal X. test SSVEP signal segments during n fixation phases: X = {X1, X2, ..., X...} j ,…,X n Simultaneously record the start times of n SSVEP signal segments X: t={t1,t2,…,t j ,…,t n}, where X j t represents the SSVEP signal induced during the j-th fixation phase. j Indicates the start time of the j-th fixation phase;

[0019] (5) Recognizing words spelled consecutively:

[0020] (5a) Let the j-th segment SSVEP signal X j , j≤n, and let j=1;

[0021] (5b) Based on the SSVEP signal X and the start time t during the fixation phase obtained in (4c), calculate the SSVEP signal X of the j-th segment. j Phase: Then, construct a periodic pulse based on the phase φ: Where φ i j This indicates that the frequency of the i-th stimulus is at t j Phase of time, The phase is represented by φ j The periodic pulse of the i-th stimulation frequency, i≤40;

[0022] (5c) For the SSVEP signal X j Perform the i-th feature extraction and let i = 1;

[0023] (5d) Using the impulse response r obtained in (3b) iand the periodic pulse obtained in (5b) SSVEP signal X during reconstruction and fixation phase j Synchronous template signal:

[0024] (5e) Construct the reference signal Y at the i-th frequency i ;

[0025] (5f) Using template signals and reference signal Y i For SSVEP signal X j Perform correlation analysis to obtain the fusion correlation coefficient.

[0026] (5g) Determine if i≤40 is true: if yes, let i=i+1 and return (5d); otherwise, execute (5h).

[0027] (5h) Take all fusion correlation coefficients: The maximum value ρ max and ρ max In ρ j The index in the code is used as the SSVEP signal segment X. j The target character index;

[0028] (5i) Determine whether j≤n is true:

[0029] If so, let j = j + 1 and return to (5b).

[0030] Otherwise, combine the characters corresponding to the target character indices of all SSVEP signal segments X into words to complete the spelling input of the words.

[0031] Compared with the prior art, the present invention has the following advantages:

[0032] 1. This invention solves the problems of lack of continuity in the input process and fixed gaze shift time in existing spelling systems based on steady-state visual evoked potentials by designing a continuous word spelling method. This allows users to spell words continuously and adaptively adjust the gaze shift time according to the distance between target characters, thereby improving the continuity of the spelling process and the information transmission rate of the spelling system.

[0033] 2. Because the present invention reconstructs the synchronization template signal, compared with existing feature extraction algorithms that can only be applied to SSVEP with a fixed initial phase, the present invention can construct a synchronized template signal for SSVEP signals with any initial phase during continuous spelling, providing a more efficient feature extraction method for continuous spelling systems and improving the recognition accuracy of SSVEP signals. Attached Figure Description

[0034] Figure 1 This is a flowchart illustrating the overall implementation of the present invention;

[0035] Figure 2 This is a sub-flowchart for obtaining training and testing signals in this invention;

[0036] Figure 3 This is a sub-flowchart for recognizing consecutively spelled words in this invention; Detailed Implementation

[0037] The specific embodiments and effects of the present invention will be further described in detail below with reference to the accompanying drawings:

[0038] Reference Figure 1 The implementation steps for this example are as follows:

[0039] Step 1. Design of the steady-state visual evoked potential (SSVEP) stimulation interface.

[0040] 1.1) Using English letters, Arabic numerals and common symbols as target blocks, 40 target blocks are arranged in a 5×8 matrix to form a stimulation interface for inducing SSVEP signals. This stimulation interface contains 26 English letters, 10 Arabic numerals and 4 common symbols, and is displayed on an LCD screen.

[0041] 1.2) Assign different frequencies f and phases φ to the target block:

[0042] f = {f1, f2, ..., f i ,…,f 40},

[0043] φ = {φ1, φ2, ..., φ i ,…,φ 40},

[0044] Where f i φ represents the coding frequency of the i-th target. i Indicates the coding frequency f i The phases are represented as follows:

[0045] f i =f0+(i-1)×Δf,

[0046] φ i =φ0+(i-1)×Δφ,

[0047] In the formula, f0 represents the initial frequency of 8Hz, Δf is 0.2Hz, φ0 represents the initial phase of 0π, and Δφ is 0.5π.

[0048] 1.2) Encode all target blocks according to the joint frequency-phase modulation paradigm to obtain the stimulus source signals of the target blocks:

[0049]

[0050] Where s(f) i ,φ i ,nFrame) represents the stimulus source signal of the i-th target block obtained by encoding, nFrame represents the frame index of the display, and RefreshRate represents the refresh rate of the display;

[0051] 1.3) Set the stimulus source signal s(f) i ,φ i The value range of (nFrame) is [0,1], which controls the brightness change of the target block in each frame of the display, where 0 represents the darkest and 1 represents the brightest, and is used to control the target block according to the frequency f. i and phase φ i Flashing.

[0052] Step 2. Obtain training and test signals.

[0053] Reference Figure 2 The specific implementation of this step is as follows:

[0054] 2.1) All character blocks in the stimulation interface start flashing during the spelling phase. The user focuses on a specific character block for 5 seconds during the flashing time. After the user finishes focusing, the stimulation interface stops flashing and a 0.5-second time is set for the user to shift their gaze. After 0.5 seconds, the user enters the next spelling phase. The above process is repeated to complete the visual stimulation of all 40 character blocks.

[0055] 2.2) Repeat step (2.1) 10 times, and collect the EEG signals of c channels of the user's visual cortex during each acquisition process. Perform preprocessing on the EEG signals of each channel in sequence, including bandpass filtering of [5,40] Hz, power frequency filtering of 50 Hz, and eye movement artifact removal. Extract all the preprocessed EEG signals according to the visual stimulus time and use them as training signal X. train Where c≥9, in this embodiment c=9, which are the 9 channels defined by the 10-20 international standard: {PO3,PO4,PO5,PO6,PO7,PO8,O1,O2,O...} z};

[0056] 2.3) Stimulate all character blocks in the interface to start flashing again. The user continuously looks at multiple characters that make up the word according to the input intention, and adaptively adjusts the eye transfer time between adjacent characters during the spelling process, stimulating the interface to flash continuously until the user has looked at all characters.

[0057] 2.4) Repeat step (2.3) a total of 5 times, and during each acquisition process, acquire the EEG signals of c channels of the user's visual cortex. Perform preprocessing on the EEG signals of each channel in sequence, including bandpass filtering of [5,40] Hz, power frequency filtering of 50 Hz, and eye movement artifact removal. Use all preprocessed signals as the test signal X. test Where c≥9, in this embodiment c=9, which are the 9 channels defined by the 10-20 international standard: {PO3,PO4,PO5,PO6,PO7,PO8,O1,O2,O...} z}

[0058] Step 3. Obtain the model parameters for reconstructing the SSVEP template signal.

[0059] 3.1) Based on the linear superposition theory generated by SSVEP, construct a model for the training signal X. train Least squares problem with reconstructed SSVEP template signals:

[0060]

[0061] in w represents the average of the multiple trial training signals and is then spatially filtered. i For spatial filters, r represents the average training signal from multiple trials; i Τ H i The reconstructed SSVEP template signal, r i H represents the user's response to different visual stimuli. i This indicates the frequency and phase information of visual stimuli;

[0062] 3.2) Solve the least squares problem in step (3.1) to obtain the impulse response used to reconstruct the SSVEP template signal: r = {r1, r2, ..., r i ,…,r 40}, and spatial filters: w={w1,w2,…,w i ,…,w 40}, where r i w represents the impulse response at the i-th stimulus frequency. i This represents the spatial filter at the i-th frequency.

[0063] Step 4. Test signal segmentation processing.

[0064] 4.1) Use the spatial filter w obtained in step (3.2) i For the test signal X test Spatial filtering is performed to obtain the spatially filtered signal X:

[0065]

[0066] in Let represent the spatial filter with the i-th frequency and the l-th channel, with a total of c channels; This represents the k-th sampling point of the l-th channel of the test signal, with a total of s sampling points for each channel;

[0067] 4.2) Spatially filtered test signal Perform a continuous wavelet transform to obtain the wavelet coefficients W, and then normalize them to obtain the normalized wavelet coefficients. They are represented as follows:

[0068]

[0069]

[0070] Where a represents the wavelet transform scale, b represents the offset; W mean W represents the mean of the wavelet coefficients W. std This represents the variance of the wavelet coefficients W;

[0071] 4.3) Based on the amplitude differences of the normalized wavelet coefficients, the spatially filtered test signal is sliced, i.e., the test signal X is extracted. test SSVEP signal segments during n fixation phases: X = {X1, X2, ..., X...} j ,…,X n Simultaneously record the start times of n SSVEP signal segments X: t={t1,t2,…,t j ,…,t n}, where X j t represents the SSVEP signal induced during the j-th fixation phase. j This indicates the start time of the j-th gaze phase.

[0072] Step 5. Recognize words spelled consecutively.

[0073] Reference Figure 3 The specific implementation of this step is as follows:

[0074] 5.1) Let X be the SSVEP signal of the j-th segment. j , j≤n, and let j=1;

[0075] 5.2) Based on the SSVEP signal X and start time t obtained in step (4.3) during the fixation phase, calculate the SSVEP signal X of the j-th segment. j Phase: Then, construct a periodic pulse based on the phase φ: Where φ ij This indicates that the frequency of the i-th stimulus is at t j Phase of time, The phase is represented by φ j The periodic pulses of the i-th stimulation frequency are represented as follows:

[0076]

[0077]

[0078] In the formula f i t represents the frequency of the i-th stimulus. j φ represents the start time of the j-th fixation phase. i Indicates the stimulus frequency f i The initial phase, h(t) j ) indicates that at t j The pulse signal at time t, there are a total of L pulse signals, i≤40;

[0079] 5.3) For signal X j Perform the i-th feature extraction and let i = 1;

[0080] 5.4) Use the impulse response r obtained in step (3.2) i and the periodic pulse obtained in step (5.2) SSVEP signal X during reconstruction and fixation phase j Synchronous template signal

[0081]

[0082] Where, r i T Indicates r i transpose;

[0083] 5.5) Construct the reference signal Y at the i-th frequency i :

[0084]

[0085] Where N represents the Nth harmonic, and there are a total of N harmonics. h Subharmonic, f i Let t represent the i-th frequency and t represent time. fs represents the sampling frequency, k represents the kth sampling point, and there are a total of s sampling points;

[0086] 5.6) Using template signals and reference signal Y i For SSVEP signal X j Perform a correlation analysis, as follows:

[0087] 5.6.1) Calculate the SSVEP signal X j With template signal Pearson correlation coefficient:

[0088]

[0089] Where cor represents Pearson correlation analysis, w i This represents the spatial filter at the i-th frequency;

[0090] 5.6.2) Calculate the SSVEP signal X j With reference signal Y i Canonical correlation coefficient:

[0091]

[0092] Where CCA represents canonical correlation analysis, Y i This represents the reference signal at the i-th frequency;

[0093] 5.6.3) on and The fusion correlation coefficient is obtained by weighted summation.

[0094]

[0095] Where sign() represents the sign function;

[0096] 5.7) Determine if i≤40 is true:

[0097] If so, let i = i + 1 and return to step (5.4);

[0098] Otherwise, proceed to step (5.8);

[0099] 5.8) Take all fusion correlation coefficients: The maximum value ρ max and ρ max In ρ j The index in the code is used as the j-th segment X of the SSVEP signal. j The target character index;

[0100] 5.9) Determine if j≤n is true:

[0101] If so, let j = j + 1 and return to step (5.2).

[0102] Otherwise, combine the characters corresponding to the target character indices of all SSVEP signal segments X into words to complete the spelling input of the words.

[0103] The effectiveness of this invention can be further illustrated by the accuracy of spelling and the information transmission rate:

[0104] Experiment 1: Offline experiments were conducted using spelling methods of the present invention and existing technologies, each employing a spelling device capable of producing 12 characters. The fixation time for each target character was set to 2 seconds, and the gaze shift time to 0.5 seconds. The accuracy rates of the two methods under different time windows were statistically analyzed, and the results are shown in Table 1.

[0105] Table 1

[0106]

[0107] As can be seen from Table 1, the accuracy of the present invention is higher than that of the prior art within a time window of 0.5s to 1.2s, especially within a short time window, where it is significantly higher than that of the prior art. This indicates that the feature extraction method of the present invention can improve the feature extraction efficiency and recognition accuracy of SSVEP within a short time window.

[0108] Experiment 2: Offline experiments were conducted using spelling methods of the present invention and existing technologies, each employing a spelling device capable of producing 12 characters. The fixation time for each target character was set to 2 seconds, and the gaze shift time to 0.5 seconds. The information transmission rate of the two methods under different time windows was statistically analyzed, and the results are shown in Table 2.

[0109] Table 2

[0110]

[0111] As can be seen from Table 2, the information transmission rate of the present invention is higher than that of the prior art within a time window of 0.5s to 1.2s. The highest information transmission rate of the prior art reaches its maximum at 1.1s, while the highest information transmission rate of the present invention reaches its maximum at 0.8s. Furthermore, the highest information transmission rate of the present invention is higher than that of the prior art, indicating that the method of the present invention can achieve a higher information transmission rate in a shorter time.

[0112] Experiment 3 involved online experiments using spelling methods from both the present invention and existing technologies, employing spellers capable of writing 12 characters. For both methods, the time window length representing the highest information transmission rate from offline experiments was used as the target fixation time. The gaze shift time was adaptively adjusted by the user based on the distance between the target characters. The highest information transmission rates of these two methods were statistically analyzed, and the results are shown in Table 3.

[0113] Table 3

[0114] The highest information transmission rate of existing technology The highest information transmission rate of this invention Offline Experiments 116.27 bits / min 128.98 bits / min Online Experiment 170.49 bits / min 196.41 bits / min

[0115] As shown in Table 3, in the online experiment, the information transmission rate of both the existing technology and the present invention is significantly higher than that in the offline experiment. Furthermore, the highest information transmission rate of the present invention is higher than that of the existing technology. This indicates that the continuous spelling method proposed in this invention can improve online spelling efficiency, increase information transmission rate, and is more suitable for spelling input in human-computer interaction.

Claims

1. A method for continuous word spelling based on steady-state visual evoked potentials, characterized in that, Includes the following steps: (1) The 40 target blocks in the stimulation interface flash according to the joint frequency phase modulation paradigm. Each target represents a different character, including 26 English letters, 10 Arabic numerals and 4 common symbols. (2) The user gazes at the flashing target block on the stimulus interface, inducing a steady-state visual evoked potential (SSVEP) signal, which has a significant impact on the user's visual perception. The SSVEP signals from each channel are acquired and preprocessed to obtain the training signal. and test signal ,in ; (3) Obtain the model parameters for reconstructing the SSVEP template signal: (3a) Based on the linear superposition theory generated by SSVEP, construct a model for the training signal. Least squares problem with reconstructed SSVEP template signals: ， in This represents the average training signal from multiple trials, after which spatial filtering is performed. For spatial filters, This represents the average training signal from multiple trials; This represents the reconstructed SSVEP template signal. H represents the user's response to different visual stimuli. i This indicates the frequency and phase information of visual stimuli; (3b) Solve the least squares problem in (3a) to obtain the impulse response used to reconstruct the SSVEP template signal: And spatial filters: ,in Indicates the first Impulse response at each stimulus frequency, Indicates the first A spatial filter with a frequency; (4) Test signal segmentation processing: (4a) Using the spatial filter obtained in (3b) For test signals Perform spatial filtering to obtain the spatially filtered signal. ; (4b) Spatially filtered test signal Perform continuous wavelet transform to obtain wavelet coefficients. The wavelet coefficients are then normalized to obtain normalized wavelet coefficients. ; (4c) Based on the amplitude differences of the normalized wavelet coefficients, the spatially filtered test signal is sliced, i.e., the test signal is extracted. middle SSVEP signal segments during the fixation phase: Simultaneously record Start time of each SSVEP signal segment X: ,in Indicates the first SSVEP signals induced during the fixation phase Indicates the first The start time of each fixation phase; (5) Recognize words spelled consecutively: (5a) Let the first SSVEP signal segment , and order ; (5b) SSVEP signal during the fixation phase obtained in (4c) and start time Calculate the first SSVEP signal Phase: Then based on the phase Constructing periodic pulses: ,in Indicates the first The stimulation frequency is Phase of time, Indicates phase as The A periodic pulse with a stimulation frequency, ; (5c) For SSVEP signal Conduct the first Secondary feature extraction, and let ; (5d) Using the impulse response obtained in (3b) and the periodic pulse obtained in (5b) SSVEP signal during reconstruction and fixation phase Synchronous template signal: ; (5e) Construct the first Reference signal at each frequency ; (5f) Using template signals and reference signal For SSVEP signal Perform correlation analysis to obtain the fusion correlation coefficient. The implementation steps include: (5f1) Calculate the SSVEP signal With template signal Pearson correlation coefficient: , in This indicates a Pearson correlation analysis. Indicates the first A spatial filter with a frequency; (5f2) Calculate the SSVEP signal With reference signal Canonical correlation coefficient: ， in This indicates canonical correlation analysis. Indicates the first A reference signal at a frequency; (5f3) and The fusion correlation coefficient is obtained by weighted summation. : , in Represents a symbolic function; (5g) judgment Is it true? If so, then let If it returns (5d), otherwise, execute (5h); (5h) Take all fusion correlation coefficients: maximum value and will exist The index in the code is used as the SSVEP signal segment. The target character index; (5i) Judgment Is it true or false? If so, then let Return (5b). Otherwise, all SSVEP signal segments The target character index corresponds to the characters that form a word, and the spelling input of the word is completed.

2. According to the method of claim 1, the 40 target blocks in the stimulation interface in step (1) flash according to the joint frequency phase modulation paradigm, as follows: (1a) Assign different frequencies to 40 target blocks and phase : ， ， in Indicates the first The coding frequency of each target Indicates coding frequency The phases are represented as follows: ， ， In the formula This indicates a starting frequency of 8Hz. It is 0.2 , Indicates initial phase 0 , It is 0.5 ; (1b) Encode all target blocks according to the joint frequency-phase modulation paradigm to obtain the stimulus source signal of the target block: ， in The encoded result is the first Stimulus source signals for each target block, This represents the frame index of the stimulus interface. Indicates the refresh rate of the stimulation interface; (1c) Setting the stimulus source signal The range of values ​​is Each element controls the brightness of the target block on the stimulus interface in each frame, where 0 represents the darkest and 1 represents the brightest, thus controlling the target block according to frequency. and phase Flashing.

3. The method according to claim 1, wherein step (2) is implemented as follows: (2a) The user gazes at a specific flashing block during the flashing time. After gazing, a fixed time is set for the user to shift their gaze and wait for the next flashing stimulus to begin, thus completing the visual stimulation of all 40 character blocks. (2b) Repeat (2a) 10 times and collect multi-channel EEG signals from the user's visual cortex during each process. Perform bandpass filtering, power frequency filtering, and eye movement artifact removal preprocessing on the EEG signals of each channel in sequence. Extract all preprocessed EEG signals according to the visual stimulus time and use them as training signals. ; (2c) Stimulate all character blocks in the interface to start flashing again. The user continuously looks at multiple characters that make up the word according to the input intention, and adaptively adjusts the eye transfer time between adjacent characters during the spelling process, stimulating the interface to flash continuously until the user has finished looking at all characters. (2d) Repeat (2c) 5 times, and during each acquisition process, perform bandpass filtering, power frequency filtering, and eye movement artifact removal preprocessing on the EEG signals of the user's visual cortex. The preprocessed signals are used as test signals. .

4. According to the method of claim 1, the spatially filtered signal obtained in step (4a) is represented as follows: ， in Indicates the first The frequency of the first The spatial filter of the channel has a total of One channel; The test signal is represented by the first... The first channel There are [number] sampling points, and each channel has [number] sampling points. One sampling point.

5. The wavelet coefficients obtained in step (4b) according to claim 1. and normalized wavelet coefficients They are represented as follows: ， ， in Indicates the wavelet transform scale. Indicates the offset; Represents wavelet coefficients The mean, Represents wavelet coefficients The variance.

6. The method according to claim 1, wherein the phase calculated in step (5b) and constructed periodic pulses They are represented as follows: ， ， in Indicates the first Stimulation frequency, Indicates the first The start time of each fixation phase Indicates stimulation frequency The initial phase, Indicates in The pulse signal at time t, total A pulse signal.

7. The method according to claim 1, wherein the reference signal constructed in step (5d) is... , means as follows: ， in Indicates the first Second harmonic, total Second harmonic Indicates the first One frequency, Indicates time: , Indicates the sampling frequency. Indicates the first There are 10 sampling points, totaling 100 sampling points. One sampling point.

Images