Method for distinguishing asynchronous control state of SSVEP signal

Abstract

A method for distinguishing an asynchronous control state of an SSVEP (Steady-State Visual Evoked Potential) signal comprises the following steps: averaging preprocessed signals based on multiple channels, and carrying out phase-space reconstruction; constructing a bubble recurrence plot for the reconstructed multi-dimensional signal based on bubble sorting; extracting four quantitative characteristics of recursion rate, certainty, average diagonal length and Shannon entropy from the recursion plot; and finally, the control state is judged through a support vector machine. According to the method, based on nonlinear dynamics, the accuracy of distinguishing the asynchronous control state of the SSVEP signal is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of brain-computer interface asynchronous SSVEP systems, and in particular to a method for distinguishing the asynchronous control state of SSVEP signals. Background Art

[0002] Steady-state visual evoked potential (SSVEP) signals have potential in many application fields. In brain-computer interfaces, SSVEP signals can be used to achieve high-speed and precise neural feedback interaction, especially in tasks that require rapid decision-making and response. The SSVEP-based BCI system is divided into synchronous SSVEP system and asynchronous SSVEP system. Unlike the synchronous SSVEP system, the asynchronous SSVEP system is a more free and flexible method. Users can independently select the target of attention without relying on stimulation of a specific frequency, which makes the system more natural and provides a better user experience.

[0003] One of the core issues restricting the asynchronous SSVEP-BCI system is the identification of the control state (IC) and the idle state (NC) in the asynchronous system. The traditional method for distinguishing the asynchronous control state of the SSVEP signal is mainly based on the characteristics of the SSVEP signal itself, that is, from three aspects: spatial filter selection, feature extraction, and classifier selection. This implementation method does not rely on other physiological signals and can meet the requirements of ease of use and application scenarios. However, the SSVEP signal features selected in the current study do not meet the requirements of specificity and generality, and are mostly linear features. Therefore, the system accuracy is not satisfactory, and it is necessary to design new signal features and combine existing algorithms to improve the system accuracy. Summary of the invention

[0004] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide a method for distinguishing the asynchronous control state of SSVEP signals, based on nonlinear dynamics, constructing a bubble recursive graph, extracting quantitative features of the recursive graph, and improving the accuracy of distinguishing the asynchronous control state of SSVEP signals.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for distinguishing the asynchronous control state of an SSVEP signal comprises: reconstructing the phase space of a preprocessed SSVEP signal based on multi-channel averaging; constructing a bubble recursion graph based on bubble sorting for the reconstructed multidimensional signal; extracting four quantitative features of the recursion graph, namely, recursion rate, certainty, average diagonal length, and Shannon entropy; and finally distinguishing the control state through a support vector machine.

[0007] A method for determining an asynchronous control state of an SSVEP signal comprises the following steps:

[0008] Step 1) SSVEP signal acquisition: multi-channel SSVEP signals near the subject's occipital area are collected, and then amplified, filtered, and converted into digital analogs;

[0009] Step 2) signal preprocessing;

[0010] Step 3) constructing the SSVEP signal bubbling recursive graph model;

[0011] Step 4) quantitative feature extraction;

[0012] Step 5) training the support vector machine;

[0013] Step 6) Use the support vector machine model to complete the control state discrimination of the test SSVEP signal.

[0014] In the step 1), six measuring electrodes are arranged based on the occipital area of ​​the subject's head: OZ, O1, O2, PO3, POZ, and PO4, a reference electrode FPZ is arranged on the subject's unilateral earlobe, and a ground electrode GND is arranged on the subject's forehead, and EEG signals are collected by using a get.c EEG acquisition cap.

[0015] In the step 2), the collected signal is firstly subjected to bandpass filtering, then the channel signal is averaged, and finally the obtained signal is subjected to detrending and smoothing processing.

[0016] In the step 3), the phase space of the one-dimensional signal obtained in the step 2) is firstly reconstructed, and a binary bubble recursive graph is constructed by bubble sorting for the reconstructed multi-dimensional signal.

[0017] In the step 4), the bubble recursive graph model obtained in the step 3) is used to extract four quantitative features of the recursive graph, namely, recursive rate, certainty, average diagonal length, and Shannon entropy.

[0018] In the step 5), for each stimulation frequency, a set of training data is selected, and the recursive graph quantization features in step 4) are obtained based on the training data, and a linear binary classification support vector machine is trained.

[0019] In the step 6), the support vector machine model obtained in the step 5) is used to determine the control state of the test SSVEP signal.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: in view of the problems of low accuracy and high cost in the prior recognition methods for the asynchronous control state of SSVEP signals, the present invention proposes a method for distinguishing the control state of asynchronous SSVEP signals based on a bubble recursive graph from the perspective of nonlinear dynamics, which provides a new method and new idea for improving the distinction of the control state of asynchronous SSVEP signals, and has the following advantages:

[0021] (1) Since the method used in the present invention does not require the assistance of other physiological signals, the cost of use is reduced, which provides prospects for the application of asynchronous SSVEP systems.

[0022] (2) Since the bubble recursive graph model constructed in the present invention does not rely on a threshold, it provides the possibility of using this method for cross-subject transfer in the future.

[0023] (3) Since the method adopted by the present invention is simple to implement, it is convenient to deploy and use in various SSVEP-based systems. Compared with the deep learning-based method, the present method requires fewer training templates and has a higher accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a flow chart of the principle of distinguishing the asynchronous control state of the SSVEP signal of the present invention.

[0025] Figure 2 Four quantitative features of the SSVEP bubble recursive graph constructed by the present invention, (a) recursive rate, (b) certainty, (c) average diagonal length, and (d) information entropy.

[0026] Figure 3 This is a curve chart of the accuracy of the present invention on 54 subjects. DETAILED DESCRIPTION

[0027] The present invention is described in detail below in conjunction with the embodiments and drawings.

[0028] Reference Figure 1 , a method for determining the asynchronous control state of an SSVEP signal, comprising the following steps:

[0029] Step 1) SSVEP signal acquisition: multi-channel SSVEP signals near the subject's occipital area are collected, and then amplified, filtered and processed by digital-to-analog conversion, specifically;

[0030] 1.1) The screen presents a multi-target light flickering stimulation paradigm for several seconds, thereby stimulating SSVEP responses near the occipital area of ​​the subject's brain;

[0031] 1.2) Collect EEG signals from 6 electrodes near the occipital area of ​​the subject. The names of the 6 electrodes are: OZ, O1, O2, PO3, POZ, PO4. The reference electrode FPZ is placed on one earlobe, and the ground electrode GND is placed on the forehead of the subject.

[0032] 1.3) Complete the collection of EEG signals by using the get.c EEG acquisition cap;

[0033] Step 2) EEG signal preprocessing: The collected EEG signals are band-pass filtered, and the filter passband bandwidth is set to 5-16 Hz (determined according to the actual test template stimulation frequency). Then, the 6-channel EEG signals are averaged to obtain a single-channel signal, and detrending and smoothing are performed;

[0034] Step 3) Construct the SSVEP signal bubble recursive graph model: First, reconstruct the phase space of the single-channel EEG signal obtained in step 2), select the reconstruction dimension as 6 electrodes and the delay as 1, and perform bubble sorting according to the signal dimension to obtain the number of bubble sorting at each time point; bubble sorting is a simple sorting algorithm, and its principle is to gradually "bubble" larger elements to the end of the array by continuously comparing adjacent elements and exchanging positions; the specific steps are as follows:

[0035] 3.1) Starting from the first element of the array, compare two adjacent elements in sequence;

[0036] 3.2) If the previous element is larger than the next element, swap their positions so that the larger element "bubbles" to the back;

[0037] 3.3) Continue to compare the next pair of adjacent elements and repeat the above operation until the last pair of elements;

[0038] 3.4) Repeat the above steps, reducing the number of elements compared by 1 each time, until all elements are arranged in ascending order;

[0039] This process is similar to bubbles rising gradually in water, hence the name bubble sort. After obtaining the number of bubble sorts at each time point, the SSVEP signal recursive graph G is constructed based on this. If two time points are the same, they are considered to have a connection, and the weight is 1, otherwise it is 0, G ij =G ji ,j,i are different time points, and G ii =1;

[0040] Step 4) Quantitative feature extraction: Quantitative feature extraction is performed on the bubble recursive graph constructed in step 3), mainly including four quantitative features:

[0041] 4.1) Recurrence Rate (RR): RR is the ratio of the number of points in the recurrence graph to the total number of points in the distance matrix. A higher RR usually means a stronger periodic embedding process. RR is defined as

[0042] Where: M represents the total number of vertices in the recursive graph, R i，j Indicates whether there is an edge in the recursive graph (1 if yes, 0 if no);

[0043] 4.2) Deterministic (DET): Deterministic refers to the ratio of points on the diagonal structure (composed of two or more points) to the total recurrence points in the recurrence graph; processes with random behavior lead to no or very short diagonals, while deterministic processes lead to longer diagonals and fewer single recurrence points; DET is defined as

[0044] Where: l represents the length of the diagonal in the recursive graph, p l represents the frequency of occurrence of diagonal lines of length l;

[0045] 4.3) Average diagonal length (L): The average diagonal length refers to the ratio of the points on the vertical line (composed of two or more points) to the total recurrence points in the recurrence graph; it represents the randomness of the dynamic system and is used to illustrate the complexity of the recurrence graph; L is defined as

[0046] Where: v represents the length of the vertical structure line of the recursive graph, p v Indicates the frequency of occurrence of vertical structural lines with a length of v;

[0047] 4.4) Shannon Entropy (ENTR): Entropy is used to measure the amount of information contained in a recursive graph. The stronger the periodicity of the recursive graph, the smaller the entropy value. ENTR is defined as Where: p l Represents the frequency of occurrence of diagonal lines of recursive graph length l;

[0048] Step 5) training the support vector machine: select a certain proportion of experimental data as training data, divide the data into control state and idle state based on the sliding window, construct a bubble recursive graph according to step 3), and then extract quantitative features according to step 4) and import them into the support vector machine for training;

[0049] Step 6) Use the support vector machine model to complete the control state discrimination of the test SSVEP signal: construct a bubble recursive graph of the test data according to step 3), then extract quantitative features according to step 4), and import them into the support vector machine to complete the control state discrimination.

[0050] To verify the feasibility of the method of the present invention, this embodiment uses the collected SSVEP signals of 54 subjects, whose age distribution is 24 to 35 years old and all have normal or corrected normal vision. The stimulation paradigm is multi-target light flickering, and the 4 target SSVEP stimuli are designed to flicker at 5.45, 6.67, 8.57 and 12 Hz, and presented in 4 positions on the display (down, right, left and up respectively); the subjects are required to look at the center of a black screen, and then look at the direction where the target stimulus is highlighted in different colors; each SSVEP stimulus is presented for 4 seconds, and the collection interval is 6 seconds. Each target frequency is presented 25 times.

[0051] The 2s before and after each stimulation were selected as idle state data. The length of each trial data segment was 8s. The data was divided by sliding window with a step size of 0.4s and a window length of 2s. 16 control state and idle state data samples were obtained (the ratio of positive and negative samples was 1:1, the first 8 were used as control state samples, and the last 8 were used as idle state samples). Then the above steps were followed to verify the effectiveness of the algorithm.

[0052] Figure 2 (a) is the recursion rate curve of the idle state and the control state for each trial of subject 3, (b) is the deterministic curve of the idle state and the control state for each trial of subject 3, (c) is the average diagonal length curve of the idle state and the control state for each trial of subject 3, and (d) is the Shannon entropy curve of the idle state and the control state for each trial of subject 3. It can be seen from the figure that for each quantitative feature, the values ​​of the subjects in the idle state and the control state have obvious differences.

[0053] The effect of this embodiment for SSVEP asynchronous control state discrimination is as follows: Figure 3 As shown, the training set and the test set are divided into 1:1, and the result is the statistical result of the average accuracy of all trials of all subjects. The results show that this method is used for SSVEP asynchronous control state discrimination, and the average recognition accuracy of 54 subjects is 85.59%.

Claims

1. A method for determining the asynchronous control state of an SSVEP signal, characterized in that: include: The preprocessed signal is averaged over multiple channels to reconstruct the phase space; Constructing a bubble recursive graph based on bubble sort for the reconstructed multi-dimensional signal; Extract four quantitative features of the recursive graph: recursive rate, certainty, average diagonal length, and Shannon entropy; Finally, the control state is determined by support vector machine.

2. A method for determining the asynchronous control state of an SSVEP signal, characterized in that: The following steps are involved: Step 1) SSVEP signal acquisition: multi-channel SSVEP signals near the subject's occipital area are collected, and then amplified, filtered, and processed by digital-to-analog conversion; Step 2) signal preprocessing; Step 3) constructing the SSVEP signal bubbling recursive graph model; Step 4) quantitative feature extraction; Step 5) training the support vector machine; Step 6) Use the support vector machine model to complete the control state discrimination of the test SSVEP signal.

3. The method according to claim 2, characterized in that In the step 1), six measuring electrodes are arranged based on the occipital area of ​​the subject's head: OZ, O1, O2, PO3, POZ, and PO4, a reference electrode FPZ is arranged on the subject's unilateral earlobe, and a ground electrode GND is arranged on the subject's forehead, and EEG signals are collected by using a get.c EEG acquisition cap.

4. The method according to claim 2, characterized in that: In the step 2), the collected signal is firstly subjected to bandpass filtering, then the channel signal is averaged, and finally the obtained signal is subjected to detrending and smoothing processing.

5. The method according to claim 2, characterized in that: In the step 3), the phase space of the one-dimensional signal obtained in the step 2) is firstly reconstructed, and a binary bubble recursive graph is constructed by bubble sorting for the reconstructed multi-dimensional signal.

6. The method according to claim 2, characterized in that In the step 4), the bubble recursive graph model obtained in the step 3) is used to extract four quantitative features of the recursive graph, namely, recursive rate, certainty, average diagonal length, and Shannon entropy.

7. The method according to claim 2, characterized in that In the step 5), for each stimulation frequency, a set of training data is selected, and the recursive graph quantization features in step 4) are obtained based on the training data, and a linear binary classification support vector machine is trained.

8. The method according to claim 2, characterized in that: In the step 6), the support vector machine model obtained in the step 5) is used to determine the control state of the test SSVEP signal.

Images