Individual specific channel selection method for SSVEP brain-computer interface system

Abstract

The invention relates to an individual specific channel selection method for an SSVEP brain-computer interface system. The method comprises the following steps: firstly, quantitatively table-verifying the importance of each channel based on task-related spatial filtering; secondly, through channel sparse learning based on spatial distance constraint, the importance difference between the SSVEP task related channel and the unrelated channel is maximized, and the importance of the unrelated channel tends to 0; and finally, based on the significant difference of the importance degrees between the channels, adaptively selecting a proper individual specific channel. According to the method, the SSVEP sensitive channel can be accurately captured, individual specificity knowledge is fully utilized, and the individual specificity SSVEP-BCI performance is further improved.

Description

Technical Field

[0001] The present invention relates to the technical fields of biomedical engineering and brain-computer interface, and particularly relates to an individual-specific channel selection method for an SSVEP brain-computer interface system. Background Art

[0002] As a new, promising, and subversive human-machine interaction technology, a brain-computer interface (BCI) can directly convert brain nerve activities into artificial outputs, thereby realizing direct interaction between the brain and external devices. In a brain-computer interface system, neural or brain decoding is a crucial link. Selecting appropriate electroencephalogram (EEG) signal channels helps extract neural behavior characteristics, accurately understand brain intentions, and thus improve the decoding accuracy.

[0003] Steady-state visual evoked potentials (SSVEP) are one of the most widely used non-invasive brain-computer interface paradigms. Related research shows that the main generation area of SSVEP is in the V1 area of the occipital region of the brain, and signals from other brain regions also contribute to improving the decoding accuracy of SSVEP EEG signals. Currently, the channels used for SSVEP brain decoding are empirical fixed channels or generalized channels. For example, a common channel set is 9 channels: Pz, PO3 / 5 / z / 6 / 4, O1 / z / 2. Since SSVEP signals have strong individual specificity, each subject has an optimal individual-specific channel set. Fixed channels are not conducive to further improving the performance of SSVEP-BCI and developing individual-specific SSVEP-BCI.

[0004] Existing SSVEP channel selection methods mainly take the decoding accuracy as the optimization goal, and obtain the channel set corresponding to the highest accuracy as the optimal channel by traversing all possible channel combinations (X. Chen, Y. Wang, S. Gao, T. P. Jung, and X. Gao, "Filterbank canonical correlation analysis for implementing a high-speed SSVEP-based brain-computer interface," J Neural Eng, vol. 12, no. 4, p. 046008, Aug 2015, doi: 10.1088 / 1741-2560 / 12 / 4 / 046008.). Such methods not only require collecting a large amount of training data but also are an inefficient method, seriously affecting the BCI experience effect and lacking generality and specificity. Therefore, it is necessary to design a new SSVEP channel selection method based on the characteristics of SSVEP itself. Summary of the Invention

[0005] To overcome the above-mentioned disadvantages of the prior art, the purpose of the present invention is to provide an individual-specific channel selection method for an SSVEP brain-computer interface system, which can accurately capture SSVEP-sensitive channels, and then make full use of individual-specific knowledge to further improve the performance of individual-specific SSVEP-BCI.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0007] An individual-specific channel selection method for an SSVEP brain-computer interface system first quantifies the importance of each channel based on a task-related spatial filtering quantization table; then, through channel sparse learning based on spatial distance constraints, it maximizes the importance difference between SSVEP task-related channels and non-related channels and makes the importance of non-related channels tend to 0; finally, based on the significant difference in the importance degree between channels, it adaptively selects appropriate individual-specific channels.

[0008] An individual-specific channel selection method for an SSVEP brain-computer interface system includes the following steps:

[0009] 1) SSVEP training data collection and preprocessing;

[0010] 2) Initialization of channel sparse learning parameters; using the preprocessed SSVEP training data, calculate the autocovariance matrix Q of the SSVEP signal as shown in Equation (1):

[0011]

[0012] In the formula: N - the number of SSVEP-BCI targets in the training data, N t — the number of each target sample in the training data, N s — the sample length, that is, the number of sampling points, — the i-th training sample of the n-th target;

[0013] Calculate the cross-covariance matrix S of the SSVEP signal as shown in Equation (2):

[0014]

[0015] Based on the EEG cap parameters, calculate the Euclidean space distance D from each electrode channel to channel O as shown in Equation (3):

[0016]

[0017] In the formula: (x i , y i , z i) — Coordinates of the i-th channel; x i — X-axis coordinate; y i — Y-axis coordinate; z i — Z-axis coordinate; Similarly, (x Oz , y Oz , z Oz ) are the Euclidean space coordinates of channel Oz;

[0018] 3) Establish a channel sparse learning model based on spatial distance constraints and solve it;

[0019] 4) Individual-specific channel selection.

[0020] The specific process of step 1) SSVEP training data acquisition is as follows: Arrange the EEG cap acquisition electrodes at the positions of the channels to be optimized, and collect SSVEP training data according to the SSVEP-BCI process; The SSVEP training data is the training dataset for the SSVEP decoding algorithm or the training data for obtaining individual-specific channels; The preprocessing in step 1) includes: performing detrending, power frequency notch filtering, band-pass filtering, and normalization preprocessing operations on the collected SSVEP training data.

[0021] The channel sparse learning model based on spatial distance constraints in step 3) is established by combining the task-related component analysis spatial filter solution model, L1 regularization, and spatial distance constraints, and its mathematical expression is shown in Equation (4):

[0022]

[0023] In the formula: w e — The SSVEP task-related spatial filter to be solved; α1 — L1 regularization parameter; ⊙ — Hadamard product; J — Optimization objective;

[0024] The solution of the channel sparse learning model based on spatial distance constraints in step 3) is carried out through iterative optimization, and the derivative of Equation (4) is Equation (5):

[0025]

[0026] In the formula: J1(w e ) — The derivative of w e , and the derivative of the i-th channel is Equation (6):

[0027]

[0028] In the formula: v — Infinitesimal; By optimizing and solving Equation (4), the optimized SSVEP task-related spatial filter w e is obtained.

[0029] Step 4) Use the absolute value of the obtained SSVEP task-related spatial filter to characterize the importance of channels, and then adaptively select individual-specific channels based on the significant differences in importance between channels, as shown in Equation (7):

[0030]

[0031] In the formula: Cs—the set of individual-specific channels; abs()—the operation of taking the absolute value; max()—the operation of taking the maximum value; β—the significant difference coefficient, set in advance or optimized; N c —the number of channels; by sequentially obtaining and comparing the significant differences between each channel and the most important channel, adaptively select the channels with smaller differences from the most important channel, and then obtain the final individual-specific channels Cs for subsequent SSVEP decoding.

[0032] The described method for selecting individual-specific channels for an SSVEP brain-computer interface system is used to optimize the channels to cover all channels in the occipital region of the brain, or customize the channels to be optimized.

[0033] The described method for selecting individual-specific channels for an SSVEP brain-computer interface system is used to optimize the target number N of the training data of the channels to be greater than or equal to 1.

[0034] The described method for selecting individual-specific channels for an SSVEP brain-computer interface system is used to optimize the number of training data samples N of a single target of the channels t to be greater than or equal to 2, and the number of training data samples between different targets does not have to be the same.

[0035] The described method for selecting individual-specific channels for an SSVEP brain-computer interface system is used to optimize the sum of the number of training data samples of all targets of the channels. The larger the sum, the more reliable the obtained individual-specific channels.

[0036] The described method for selecting individual-specific channels for an SSVEP brain-computer interface system is used to optimize the length of the training data signal and the decoding signal of the channels, which do not have to be the same. The longer the training data signal, the more reliable the obtained individual-specific channels.

[0037] Compared with the prior art, the beneficial effects of the present invention are:

[0038] Based on the SSVEP task-related features and the SSVEP generation mechanism, the present invention uses spatial distance to constrain the channel sparse learning, so as to obtain more important channels that can better represent the SSVEP task; in addition, the present invention is an efficient channel selection method, which can quickly obtain individual-specific channels with only a small amount of training data, thereby improving the performance of the individual-specific SSVEP-BCI system; finally, the individual-specific channel selection method of the present invention is independent of the decoding method and can be combined with existing SSVEP decoding methods to further improve the decoding accuracy. Brief Description of the Drawings

[0039] Figure 1 is a flowchart of an embodiment of the present invention.

[0040] Figure 2 is a result diagram of optimizing channels in an embodiment of the present invention, where (a) is a schematic diagram of using traditional fixed channels, and (b) is a statistical chart of individual-specific channels after optimizing 64 channels of 35 subjects by the method of this embodiment.

[0041] Figure 3 is an effect diagram for improving the performance of the SSVEP-BCI in an embodiment of the present invention. Detailed Embodiment

[0042] The technical solution of the present invention will be further described and described in detail below in conjunction with the embodiments and the drawings. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0043] Refer to Figure 1 , an individual-specific channel selection method for an SSVEP brain-computer interface system, including: first quantifying the importance of each channel based on task-related spatial filtering; then through channel sparse learning based on spatial distance constraints, maximizing the importance difference between SSVEP task-related channels and non-related channels, and making the importance of non-related channels tend to 0; finally, based on the significant difference in the importance degree between channels, adaptively select appropriate individual-specific channels for subsequent SSVEP decoding.

[0044] An individual-specific channel selection method for an SSVEP brain-computer interface system includes the following steps:

[0045] 1) SSVEP training data collection and preprocessing;

[0046] The specific process of SSVEP training data acquisition is as follows: For a specific SSVEP task, the EEG cap acquisition electrodes are arranged at the positions of the channels to be optimized, and according to the conventional process of SSVEP-BCI, SSVEP training data is acquired; the SSVEP training data can be the training dataset of the SSVEP decoding algorithm or the training data for obtaining individual-specific channels; in this embodiment, a 40-target SSVEP-BCI based on optical flicker with a stimulation frequency of 8 - 15.8 Hz and an interval of 0.2 Hz encoding is adopted. The electrode channels are set according to the 10 / 20 electrode system, and the signals of 64 channels of the whole brain are recorded; 6 samples are collected for each target, the sampling frequency is 1000 Hz, and the sample duration is 5 s.

[0047] The preprocessing includes: According to the specific task and the characteristics of SSVEP, conventional preprocessing operations such as detrending, power frequency notch filtering, band-pass filtering, and normalization are performed on the acquired SSVEP training data to remove interference components and retain the useful components of the signal; in this embodiment, the detrending, 50 Hz power frequency notch filtering, 8 - 90 Hz band-pass filtering, and Z-Score normalization preprocessing operations are sequentially performed on each channel of each SSVEP training data sample.

[0048] 2) Initialization of channel sparse learning parameters, specifically: Using the preprocessed SSVEP training data, calculate the autocovariance matrix Q of the SSVEP signal, as shown in Equation (1):

[0049]

[0050] In the formula: N—the number of SSVEP-BCI targets in the training data, in this embodiment N = 40; N t —the number of each target sample in the training data, in this embodiment N t = 5; N s —the sample length, that is, the number of sampling points, in this embodiment N s = 5000; —the i-th training sample of the n-th target;

[0051] Calculate the cross-covariance matrix s of the SSVEP signal, as shown in Equation (2):

[0052]

[0053] Based on the EEG cap parameters, calculate the Euclidean space distance D from each electrode channel to channel Oz, as shown in Equation (3):

[0054]

[0055] In the formula: (x i , y i , z i)——Euclidean space coordinates of the i-th channel; x i —x-axis coordinate; y i —y-axis coordinate; z i —z-axis coordinate; similarly, (x Oz , y Oz , z Oz ) are the Euclidean space coordinates of channel Oz;

[0056] 3) Establish a channel sparse learning model based on spatial distance constraints and solve it;

[0057] The channel sparse learning model based on spatial distance constraints is established by combining task-related component analysis of spatial filter solution models, L1 regularization, and spatial distance constraints. Its mathematical expression is shown in Equation (4):

[0058]

[0059] In the formula: w e —SSVEP task-related spatial filter to be solved; α1—L1 regularization parameter, set to 0.18 in this embodiment; ⊙—Hadamard product; J—optimization objective;

[0060] The solution of the channel sparse learning model based on spatial distance constraints is carried out through iterative optimization. In this embodiment, the minFunc optimization solution method based on L-BFGS is adopted, the maximum number of iterations is set to 1000, and the error is set to 10 -9 , and the derivative of Equation (4) is Equation (5):

[0061]

[0062] In the formula: J1(w e )—derivative of w e , and the derivative of the i-th channel is Equation (6):

[0063]

[0064] In the formula: ε—infinitesimal, set to 10 -10 in this embodiment; by optimizing and solving Equation (4), the optimized SSVEP task-related spatial filter w e is obtained;

[0065] 4) Individual-specific channel selection;

[0066] Use the absolute value of the obtained SSVEP task-related spatial filter to characterize the importance of channels, and then adaptively select individual-specific channels through the significant difference in importance between channels, as shown in Equation (7):

[0067]

[0068] Where: Cs is the individual-specific channel set; abs() is the absolute value operation; max() is the maximum value operation; β is the significant difference coefficient, which is set in advance or optimized, and is set to 0.05 in this embodiment; N c —the number of channels. In this embodiment, N c = 64; By sequentially obtaining and comparing the significant differences between each channel and the most important channel, channels with smaller differences from the most important channel are adaptively selected, and finally the individual-specific channel Cs is obtained for subsequent SSVEP decoding.

[0069] To verify the feasibility of the method of the present invention, in this embodiment, the SSVEP signals of 35 subjects collected are implemented according to the above steps to verify its effectiveness.

[0070] Refer to Figure 2 , Figure 2 where (a) is a schematic diagram of the traditional fixed channels used, Figure 2 and (b) is a statistical chart of the individual-specific channels of the 64 channels of 35 subjects optimized by the method of this embodiment. The calculation method is the total number of times each channel is selected as an individual-specific channel / the total number of subjects. It can be seen from the figure that there are obvious differences in the specific channels among the subjects, and the method of this embodiment can effectively obtain individual-specific channels.

[0071] Refer to Figure 3 ,and the SSVEP signals are decoded by the task-related component analysis algorithm using the traditional fixed channels and the channels optimized in this embodiment respectively, and the decoding accuracies of 35 subjects at different signal lengths are statistically analyzed. From Figure 3 it can be seen that the method of this embodiment can make full use of individual-specific knowledge and further improve the performance of SSVEP-BCI.

[0072] The above embodiments are only used to illustrate the present invention, rather than to limit the present invention. Those of ordinary skill in the relevant technical fields can also make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, all equivalent technical solutions also belong to the scope of the present invention, and the patent protection scope of the present invention should be defined by the claims.

Claims

1. An individual-specific channel selection method for an SSVEP brain-computer interface system, characterized in that First, characterize the importance of each channel based on the task-related spatial filtering quantization table; then, through channel sparse learning based on spatial distance constraints, maximize the importance difference between the SSVEP task-related channels and the non-related channels, and make the importance of the non-related channels tend to 0; finally, based on the significant difference in the importance degree between channels, adaptively select appropriate individual-specific channels.

2. An individual-specific channel selection method for an SSVEP-based brain-computer interface system, characterized in that, It includes the following steps: 1) SSVEP training data acquisition and preprocessing; 2) Initialization of channel sparse learning parameters; Using the preprocessed SSVEP training data, calculate the autocovariance matrix Q of the SSVEP signal, as shown in Equation (1): Where: N—the number of SSVEP-BCI targets in the training data, N t —the number of each target sample in the training data, N s —the sample length, i.e., the number of sampling points —the i-th training sample of the n-th target Calculate the cross-covariance matrix S of the SSVEP signal, as shown in Equation (2): Based on the EEG cap parameters, calculate the Euclidean spatial distance D from each electrode channel to channel Oz, as shown in Equation (3): where: (x i , y i , z i ) — the coordinates of the i-th channel; x i — the x-axis coordinate; y i — the y-axis coordinate; z i — the z-axis coordinate; similarly, (x Oz , y Oz , z Oz ) are the Euclidean space coordinates of the Oz channel; 3) Establish a channel sparse learning model based on spatial distance constraints and solve it; 4) Individual-specific channel selection.

3. The method according to claim 2, wherein: Step 1) The specific SSVEP training data acquisition is: Arrange the EEG cap acquisition electrodes at the positions of the channels to be optimized, and collect the SSVEP training data according to the SSVEP-BCI process; The SSVEP training data is the training data set of the SSVEP decoding algorithm or the training data used to obtain individual-specific channels; Step 1) The preprocessing includes: performing preprocessing operations of detrending, power frequency notch filtering, band-pass filtering, and normalization on the collected SSVEP training data.

4. The method according to claim 2, characterized in that: Step 3) The channel sparse learning model based on spatial distance constraints is established by combining the task-related component analysis spatial filter solution model, L1 regularization, and spatial distance constraints, and its mathematical expression is as shown in Equation (4): where: w e — the spatial filter related to the SSVEP task to be solved; α1 — the L1 regularization parameter; ⊙ — the Hadamard product; J — the optimization objective; Step 3) The solution of the channel sparse learning model based on spatial distance constraints is carried out through iterative optimization, and the derivative of Equation (4) is Equation (5): where: J1(w e ) — the derivative of w e , and the derivative of the i-th channel is given by Equation (6): where: ε—a very small quantity; by optimizing and solving Equation (4), the optimized task space filter w can be obtained e .

5. The method according to claim 2, characterized in that: Step 4) Use the absolute value of the obtained SSVEP task-related spatial filter to characterize the importance degree of the channels, and then adaptively select individual-specific channels through the significant difference in the importance degree between channels, as shown in Equation (7): Where: Cs - individual-specific channel set; abs() - absolute value operation; max() - maximum value operation; β - significance difference coefficient, set in advance or optimized; N c — number of channels; by sequentially obtaining and comparing the significant differences between each channel and the most important channel, adaptively selecting the channels with smaller differences from the most important channel, and then obtaining the final individual-specific channel Cs for subsequent SSVEP decoding.

6. The method according to any one of claims 1-5, characterized in that: The channels used for optimization cover all channels in the occipital region of the brain, or customize the channels to be optimized.

7. The method according to any one of claims 1-5, characterized in that: The target number N of the training data for the channels to be optimized is greater than or equal to 1.

8. The method according to any one of claims 1-5, characterized in that: The number of training data samples N for a single target used to optimize the channel t is greater than or equal to 2, and the number of training data samples between different targets does not have to be the same.

9. The method according to any one of claims 1-5, characterized in that: The larger the sum of the training data sample numbers of all targets for the channels to be optimized, the more reliable the obtained individual-specific channels.

10. The method according to any one of claims 1-5, characterized in that: The length of the training data signal and the length of the decoded signal for the channels to be optimized do not have to be the same. The longer the training data signal length, the more reliable the obtained individual-specific channels.

Images