A SSVEP asynchronous identification method based on CCA coefficient threshold

Abstract

A SSVEP asynchronous recognition method based on CCA coefficient threshold, firstly collects EEG signals in the occipital area, in the offline training stage, trains the CCA-PSD mixed coefficient decision algorithm, calculates the optimal length of signal calculation and the optimal EEG data acquisition channel combination; in the online classification stage, adopts a dynamic windowing method, uses the trained CCA-PSD mixed coefficient decision algorithm to process the EEG signals in real time, calculates the result coefficient, and then votes for the 5 adjacent windows to obtain the final asynchronous control result. The classification accuracy of the invention is high, the error rate of the gaze state and the idle state is effectively reduced, and it has excellent asynchronous classification performance.

Description

Technical Field

[0001] The present invention relates to the field of Steady State Visual Evoked Potential (SSVEP) and brain-computer interface, and is an SSVEP asynchronous recognition method combining Power Spectral Density (PSD) and Canonical Correlation Analysis (CCA) coefficient thresholds. Background Art

[0002] As a system that detects neural activity in the brain and converts it into output, the brain-computer interface (BCI) can directly detect a person's actions and thoughts from the brain and transmit the information directly to the machine. The brain-computer interface system is mainly composed of EEG signal acquisition, signal processing and decoding (preprocessing, feature extraction, classification and recognition), command output and other units. The acquisition device first collects the EEG signals of the subjects in the brain area, amplifies, filters, and converts the signals into analog-to-digital signals, and transmits them to the processing system; the processing system first preprocesses the signals, filters out the noise and artifacts contained in the EEG signals, and then extracts the required signal features for classification; finally, the classified signals are converted into control command outputs, thereby achieving the purpose of human-computer interaction.

[0003] Steady-state Visual Evoked Potential (SSVEP) is mainly distributed in the occipital region of the brain. It is an EEG signal generated when the subject is subjected to continuous visual stimulation of a fixed frequency from the outside world. The SSVEP paradigm has become one of the most popular brain-computer interface paradigms due to its excellent electroencephalographic noise, relative immunity to eye artifacts, and the fact that the subject does not need to undergo prior training.

[0004] The control methods of brain-computer interface systems are mainly divided into synchronous control and asynchronous control. Most of the current BCI systems are synchronous control. Synchronous control requires sending control signals within a specified interval and cannot freely complete the control of the brain-computer interface system. Asynchronous control can send control signals at any time, which is more in line with people's usage habits. In asynchronous control, the user's state is usually divided into idle and staring states. In asynchronous control, if a state recognition error occurs, it will cause a greater impact, such as loss of control of the controller. Therefore, how to correctly and quickly identify the idle and staring states is a technical problem that needs to be solved urgently. Summary of the invention

[0005] In order to overcome the shortcomings of the existing technology, the present invention proposes an SSVEP asynchronous recognition method based on CCA coefficient threshold. By first extracting features from the signal and then identifying the gaze state and the idle state based on the extracted features, the classification accuracy of the present invention is high, the error rate of the gaze state and the idle state is effectively reduced, and it has excellent asynchronous classification performance.

[0006] The technical solution adopted by the present invention to solve its technical problem is:

[0007] A SSVEP asynchronous identification method based on CCA coefficient threshold includes the following steps:

[0008] Step 1: Offline training phase. The process is as follows:

[0009] Under the stimulation of target blocks of different frequencies, offline experiments of control task and idle task were carried out respectively, and the signals of occipital area of ​​the brain were collected to obtain two offline data sets of idle state and gaze state.

[0010] The power spectrum density of the two offline data sets is calculated for data of different lengths at the fundamental frequency and different harmonic frequencies, and a power spectrum density analysis threshold for dividing the idle state and the staring state is obtained;

[0011] The two offline data sets are used to calculate CCA coefficient thresholds for data of different lengths, respectively, to obtain CCA coefficient thresholds for dividing the idle state and the staring state;

[0012] The above two offline data sets are calculated and analyzed to obtain the optimal signal partition window length and optimal processing frequency, as well as the optimal channel area CCA coefficient threshold and power spectrum density weight coefficient;

[0013] Step 2: Online classification stage. The process is as follows:

[0014] The EEG signals of the best channel area of ​​the brain are collected in real time, the window data is filtered, and then the power spectrum density and CCA coefficient threshold are calculated to obtain the classification result; if the result is a gaze state, it is recorded as +1; the idle state is recorded as -1; a judgment is made every 5 windows, and the 5 classification results are added together. If it is greater than 0, it is a gaze state, and if it is less than 0, it is an idle state.

[0015] Further, the process of step one is as follows:

[0016] 1.1) Collecting brain occipital area signals, including the following steps: placing measuring electrodes at 9 positions, namely Oz, O1, PO3, P1, POz, Pz, P2, PO4, and O2, and displaying 6 stimulus sources with different flashing frequencies, namely 8 Hz, 9 Hz, 10 Hz, 11 Hz, and 12 Hz, on the interface;

[0017] 1.2) Data preprocessing: bandpass filtering and notching are performed on the EEG signals to divide the data into gaze state data and idle state data. The gaze of the subjects with both eyes fixed on the stimulus target is recorded as the gaze state, and the gaze away from the stimulus target is recorded as the idle state;

[0018] 1.3) Window the data. The length of each experimental data is 5s. Divide the data into 0.5s-4s in length with a step size of 0.1s. Window the data with a length of 5s.

[0019] 1.4) Process and calculate the data. The specific steps are as follows:

[0020] A. Use the CCA algorithm to process the data in 1.3). The obtained feature matrix corresponds to the number of stimulation signals. The N-dimensional stimulation vector is expressed as: F = [R1, R2, R3, R4, R5, R6]. The CCA coefficient is calculated as:

[0021]

[0022] Per: CCA coefficient;

[0023] Rmax: the maximum eigenvalue in the F feature matrix;

[0024] Rmax: the second eigenvalue in the F eigenmatrix;

[0025] B. Calculate the power spectrum of the data in 1.3) at the fundamental frequency and different harmonic frequencies (8-13Hz, 8-26Hz, 8-39Hz, 8-52Hz, 16-26Hz, 16-39Hz, 16-52Hz, 24-39Hz, 24-52Hz, 32-52Hz);

[0026] 1.5) Substitute the results from 1.4) into the calculation to find the EEG acquisition channel, window size, optimal power spectrum selection frequency band and threshold Pow that produces the best results. Thre , CCA coefficient threshold P Thre , and weight coefficient W = [w1, w2].

[0027] Furthermore, the process of step 2 is as follows:

[0028] 2.1) Collect EEG data of the best channel combination online, and the subjects choose the time and duration of staring at the target block by themselves;

[0029] 2.2) performing bandpass filtering and notching on the signal in 2.1);

[0030] 2.3) According to the optimal window size obtained by offline training, the data is windowed with a step size of 0.1s, and the data is recorded as Data = [D1, D2, D3, D4, D5];

[0031] 2.4) Calculate the CCA coefficient and power spectrum of the data in 2.3) and compare them with the corresponding threshold. After comparison, the result matrix is ​​reduced in dimension using the weight coefficient to obtain the final result coefficient R all =[R1,R2,R3,R4,R5]:

[0032]

[0033] R: result coefficient;

[0034] Comb: CCA-PSD mixing coefficient;

[0035] Comb = [w1, w2] · [P RESULT , Pow RESULT ]

[0036] P RESULT : The result coefficient of the CCA coefficient threshold,

[0037] Pow RESULT : The result coefficient of the power spectrum threshold, Pow is the power spectrum calculation result in the optimal frequency band;

[0038] 2.5) Accumulate the values ​​of the result coefficients in 2.4). If the result is greater than 0, it is considered that the signal is in the control state, and Rmax = max F is output as the result of classification detection; if the result is less than 1, it is considered that it is still in the idle state, move forward one window, and continue to identify and vote on the signal.

[0039] The beneficial effects of the present invention are mainly manifested in: high classification accuracy, effective reduction of the error rate of the staring state and the idle state, and excellent asynchronous classification performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a flow chart of a SSVEP asynchronous identification method based on CCA coefficient threshold. DETAILED DESCRIPTION

[0041] The present invention will be further described below in conjunction with the accompanying drawings.

[0042] Reference Figure 1 , a SSVEP asynchronous identification method based on CCA coefficient threshold, comprising the following steps:

[0043] Step 1: Offline training phase. The process is as follows:

[0044] Under the stimulation of different frequency target blocks (such as 8, 9, 10, 11, 12, and 13 Hz), offline experiments of control tasks and idle tasks were carried out respectively, and the EEG signal data of 9 channels, such as Oz, O1, PO3, P1, POz, Pz, P2, PO4, and O2, were collected from the occipital area of ​​the brain to obtain two offline data sets of idle state and gaze state.

[0045] The power spectral density of the two offline data sets is calculated for data of different lengths at the fundamental frequency and different harmonic frequencies (such as 8-13 Hz, 8-26 Hz, 8-39 Hz, 8-52 Hz, 16-26 Hz, 16-39 Hz, 16-52 Hz, 24-39 Hz, 24-52 Hz, and 32-52 Hz) to obtain the power spectral density analysis threshold for dividing the idle state and the gaze state;

[0046] The two offline data sets are used to calculate CCA coefficient thresholds for data of different lengths, respectively, to obtain CCA coefficient thresholds for dividing the idle state and the staring state;

[0047] The above two offline data sets are calculated and analyzed to obtain the optimal signal partition window length and optimal processing frequency, as well as the optimal channel area (such as single channel Oz, multi-channel 9-channel mixed data) CCA coefficient threshold and power spectral density weight coefficient;

[0048] Step 2: Online classification stage. The process is as follows:

[0049] The EEG signal of the best channel area of ​​the brain is collected in real time, and the window data (step length is 0.1s) is filtered, and then the power spectrum density and CCA coefficient threshold are calculated to obtain the classification result. If the result is a gaze state, it is recorded as +1; the idle state is recorded as -1. A judgment is made every 5 windows, and the 5 classification results are added. If it is greater than 0, it is a gaze state, and if it is less than 0, it is an idle state.

[0050] In the solution of this embodiment, after collecting the EEG data of the offline test, the occipital area data is processed and trained to obtain the corresponding thresholds and weight coefficients, which are put into the online classification algorithm.

[0051] Among them, the offline training phase is divided into 5 steps:

[0052] 1.1) Collecting brain occipital area signals, including the following specific steps: placing measurement electrodes at 9 positions, namely Oz, O1, PO3, P1, POz, Pz, P2, PO4, and O2, and displaying 6 stimulation sources with different flashing frequencies, namely 8 Hz, 9 Hz, 10 Hz, 11 Hz, and 12 Hz, on the interface;

[0053] 1.2) Data preprocessing: bandpass filtering and notching are performed on the EEG signals to divide the data into gaze state data and idle state data. In this experiment, the subjects' gaze on the stimulus target is recorded as the gaze state, and the gaze away from the stimulus target is recorded as the idle state;

[0054] 1.3) Window the data. The length of each experimental data is 5s. Divide the data into 0.5s-4s (with a step size of 0.1s). Window the data with a length of 5s. For example, divide the data into 3s windows, and then divide the data into 0-3s, 0.1-3.1s, 0.2-3.2s...4.1-5s, for a total of 21 data.

[0055] 1.4) Process and calculate the data. The specific steps are as follows:

[0056] A. Use the CCA algorithm to process the data in 1.3). The obtained feature matrix corresponds to the number of stimulation signals. The N-dimensional stimulation vector is expressed as: F = [R1, R2, R3, R4, R5, R6]. The CCA coefficient is calculated as:

[0057]

[0058] Per: CCA coefficient;

[0059] Rmax: the maximum eigenvalue in the F feature matrix;

[0060] Rmax: the second eigenvalue in the F eigenmatrix;

[0061] B. Calculate the power spectrum of the data in 1.3) at the fundamental frequency and different harmonic frequencies (8-13Hz, 8-26Hz, 8-39Hz, 8-52Hz, 16-26Hz, 16-39Hz, 16-52Hz, 24-39Hz, 24-52Hz, 32-52Hz);

[0062] 1.5) Substitute the results from 1.4) into the calculation to find the EEG acquisition channel, window size, optimal power spectrum selection frequency band and threshold Pow that produces the best results. Thre , CCA coefficient threshold P Thre , and weight coefficient W = [w1, w2].

[0063] Online classification and output are divided into the following five steps:

[0064] 2.1) Collect EEG data of the best channel combination online, and the subjects choose the time and duration of staring at the target block by themselves (must be greater than 0.5s);

[0065] 2.2) performing bandpass filtering and notching on the signal in 2.1);

[0066] 2.3) According to the optimal window size obtained by offline training, the data is windowed with a step size of 0.1s, and the data is recorded as Data = [D1, D2, D3, D4, D5];

[0067] 2.4) Calculate the CCA coefficient and power spectrum of the data in 2.3) and compare them with the corresponding threshold. After comparison, the result matrix is ​​reduced in dimension using the weight coefficient to obtain the final result coefficient R all =[R1,R2,R3,R4,R5]:

[0068]

[0069] R: result coefficient;

[0070] Comb: CCA-PSD mixing coefficient;

[0071] Comb = [w1, w2] · [P RESULT , Pow RESULT ]

[0072] P RESULT : The result coefficient of the CCA coefficient threshold,

[0073] Pow RESULT : The result coefficient of the power spectrum threshold, Pow is the power spectrum calculation result in the optimal frequency band.

[0074] 2.5) Accumulate the values ​​of the result coefficients in 2.4). If the result is greater than 0, it is considered that the signal is in the control state, and Rmax = max F is output as the result of classification detection; if the result is less than 1, it is considered that it is still in the idle state, move forward one window, and continue to identify and vote on the signal.

[0075] The SSVEP asynchronous classification method of the fusion CCA coefficient threshold and PSD disclosed in the present invention, namely the CCA-PSD hybrid coefficient decision algorithm, uses dynamic windowing and dynamic channel combination to perform threshold processing on the CCA coefficient and PSD in the classification process of EEG signals, performs weighted processing, and finally judges the state of the subject by voting. This method effectively reduces the error rate of the gaze state and the idle state, can judge the state of the subject in a short time, and has a more superior asynchronous classification performance.

[0076] The contents described in the embodiments of this specification are merely enumerations of implementation forms of the inventive concept and are for illustrative purposes only. The protection scope of the present invention should not be considered to be limited to the specific forms described in this embodiment, and the protection scope of the present invention also extends to equivalent technical means that can be thought of by ordinary technicians in this field based on the inventive concept.

Claims

1. A SSVEP asynchronous identification method based on CCA coefficient threshold, characterized in that: The method comprises the following steps: Step 1: Offline training phase. The process is as follows: Under the stimulation of target blocks of different frequencies, offline experiments of control task and idle task were carried out respectively, and the signals of occipital area of ​​the brain were collected to obtain two offline data sets of idle state and gaze state. The power spectrum density of the two offline data sets is calculated for data of different lengths at the fundamental frequency and different harmonic frequencies, and a power spectrum density analysis threshold for dividing the idle state and the staring state is obtained; The two offline data sets are used to calculate CCA coefficient thresholds for data of different lengths, respectively, to obtain CCA coefficient thresholds for dividing the idle state and the staring state; The above two offline data sets are calculated and analyzed to obtain the optimal signal partition window length and optimal processing frequency, as well as the optimal channel area CCA coefficient threshold and power spectrum density weight coefficient; Step 2: Online classification stage. The process is as follows: Collect EEG signals from the best channel area of ​​the brain in real time, filter the window data, and then calculate the power spectrum density and CCA coefficient threshold to get the classification result; if the result is a gaze state, it is recorded as +1; if the idle state is recorded as -1; make a judgment once every 5 windows, add up the 5 classification results, if it is greater than 0, it is a gaze state, and if it is less than 0, it is an idle state; The process of step one is as follows: 1.1) Collecting brain occipital area signals, including the following steps: placing measuring electrodes at 9 positions, namely Oz, O1, PO3, P1, POz, Pz, P2, PO4, and O2, and displaying 6 stimulus sources with different flashing frequencies, namely 8 Hz, 9 Hz, 10 Hz, 11 Hz, and 12 Hz, on the interface; 1.2) Data preprocessing: bandpass filtering and notching are performed on the EEG signals to divide the data into gaze state data and idle state data. The gaze of the subjects with both eyes fixed on the stimulus target is recorded as the gaze state, and the gaze away from the stimulus target is recorded as the idle state; 1.3) Window the data. The length of each experimental data is 5s. Divide the data into 0.5s-4s in length with a step size of 0.1s. Window the data with a length of 5s. 1.4) Process and calculate the data. The specific steps are as follows: A. Use the CCA algorithm to process the data in 1.3). The obtained feature matrix corresponds to the number of stimulation signals. The N-dimensional stimulation vector is expressed as: F = [R1, R2, R3, R4, R5, R6]. The CCA coefficient is calculated as: Per: CCA coefficient; Rmax: the maximum eigenvalue in the F feature matrix; Rmax: the second eigenvalue in the F eigenmatrix; B. Calculate the power spectrum of the data in 1.3) at the fundamental frequency and different harmonic frequencies (8-13Hz, 8-26Hz, 8-39Hz, 8-52Hz, 16-26Hz, 16-39Hz, 16-52Hz, 24-39Hz, 24-52Hz, 32-52Hz); 1.5) Substitute the results from 1.4) into the calculation to find the EEG acquisition channel, window size, optimal power spectrum selection frequency band and threshold Pow that produces the best results. Thre , CCA coefficient threshold P Thre , and weight coefficient W = [w1, w2]; The process of step 2 is as follows: 2.1) Collect EEG data of the best channel combination online, and the subjects choose the time and duration of staring at the target block by themselves; 2.2) performing bandpass filtering and notching on the signal in 2.1); 2.3) According to the optimal window size obtained by offline training, the data is windowed with a step size of 0.1s, and the data is recorded as Data = [D1, D2, D3, D4, D5]; 2.4) Calculate the CCA coefficient and power spectrum of the data in 2.3) and compare them with the corresponding threshold. After comparison, the result matrix is ​​reduced in dimension using the weight coefficient to obtain the final result coefficient R all =[R1,R2,R3,R4,R5]: R: result coefficient; Comb: CCA-PSD mixing coefficient; Comb=[w1,w2]·[P RESULT ,Pow RESULT ] P RESULT : The result coefficient of the CCA coefficient threshold, Pow RESULT : The result coefficient of the power spectrum threshold, Pow is the power spectrum calculation result in the optimal frequency band; 2.5) Accumulate the values ​​of the result coefficients in 2.4). If the result is greater than 0, it is considered that the signal is in the control state, and Rmax = max F is output as the result of classification detection; if the result is less than 1, it is considered that it is still in the idle state, move forward one window, and continue to identify and vote on the signal.

Images