A method for decoding EEG signals based on orthogonal experiments

Abstract

The present invention discloses an EEG signal decoding method based on an orthogonal experiment. The method steps adopted by the present invention are: (1) using a motor imagery dataset as the EEG signal to be analyzed; (2) extracting and designing EEG signal parameters; (3) designing an orthogonal experiment based on the EEG signal parameters; (4) generating optimized parameters based on the orthogonal experiment; (5) extracting features from the EEG signal parameters selected by the orthogonal experiment; and (5) classifying the extracted features to achieve EEG signal decoding. The method of the present invention can overcome the problems of the prior art, such as high computational complexity, inaccurate preprocessing, and reliance on neurophysiological cognition.

Description

Technical Field

[0001] The present invention relates to the fields of brain-computer interface, computer and information processing technology, and in particular to an electroencephalogram (EEG) signal decoding method based on orthogonal experiments. Background Art

[0002] A brain-computer interface (BCI) is a system that enables communication between human and computer signals. It includes software and hardware that recognizes and manipulates human signals to control computers and various communication devices. A complete BCI system consists of four components: preprocessing, feature extraction, signal classification, and final control. BCIs are used in robotic wheelchair control, robotic arm control, stroke rehabilitation, and computer game control. EEG signals have been widely analyzed and applied in the BCI field due to their non-invasiveness, ease of acquisition, and high temporal resolution.

[0003] In most existing literature, researchers focus on feature extraction algorithms and classification methods for brain signal decoding to improve the efficiency of EEG signal decoding. The EEG feature extraction process includes frequency filtering, window selection, feature extractor, and feature selection. After preprocessing, the EEG signal is fed into one or more types of feature extraction algorithms to extract features in the time, frequency, and spatial domains. In BCI systems, band power features and time domain features represent EEG signals. Band power features represent the average power of the EEG signal in a given frequency band over a time window, while time domain features are a combination of EEG signals from all channels. In motor imagery BCIs, the most commonly used or referenced feature extraction techniques are short-term Fourier transform (STFT), autoregressive model (AR), wavelet transform (WT), principal component analysis (PCA), and cospatial patterns (CSP). In BCIs, the spatial filtering used by the CSP algorithm aims to compute features whose variance is optimal for distinguishing between two types of EEG measurements. The performance of this spatial filtering depends on the EEG frequency band. Wei W et al. wrote an article titled “Probabilistic Common Spatial Patterns for Multichannel EEG Analysis” in IEEE Transactions on Pattern Analysis & Machine Intelligence, proposing a probabilistic CSP method to address the overfitting problem of CSP and developing a spatiotemporal modeling framework based on EEG feature decomposition based on maximum a posteriori estimation.

[0004] In motor imagery brain-computer interfaces (BCIs), features extracted through various feature extraction techniques are converted into different motor imagery tasks, such as hand movements, foot movements, and tongue movements, through classification algorithms. Currently, algorithms used for EEG signal classification include linear classifiers, neural networks, nonlinear classifiers, and deep neural networks. In their paper, "Decoding of Intuitive Visual Motion Imagery Using Convolutional Neural Network under 3D-BCI Training Environment," BH Kwon et al. developed a 3D BCI training platform, used a functional connectivity method for visual motion imagery decoding, and proposed a convolutional neural network architecture for EEG signal classification.

[0005] However, there are great differences in the EEG signals of different subjects. For each subject, it is very necessary to conduct preliminary design and estimation of EEG signal decoding parameters in the preprocessing stage, which can reduce the accuracy and scientificity of subsequent EEG decoding and improve the portability of the brain-computer interface system on different subjects. Summary of the Invention

[0006] In response to the actual needs of EEG signal processing and the shortcomings of the existing technology, the purpose of the present invention is to provide an EEG signal decoding method based on orthogonal experiments, which has low computational complexity and simple method. It optimizes the parameters of EEG signal decoding in the preprocessing stage, which can effectively improve the effect of subsequent decoding.

[0007] The technical solution of the present invention is specifically described as follows.

[0008] The present invention provides an electroencephalogram (EEG) signal decoding method based on an orthogonal experiment, comprising the following steps:

[0009] (1) Using self-collected motor imagery datasets as EEG signals to be analyzed;

[0010] (2) Extraction and generation of EEG signal parameters; EEG signal parameters include EEG signal channel area, frequency band and motor imagery time;

[0011] (3) Design of EEG signal parameters based on orthogonal experiments;

[0012] (4) Feature extraction of EEG signal parameters generated by orthogonal experiments;

[0013] (5) Classify the extracted features and complete the decoding of EEG signals.

[0014] In the present invention, in step (1), the acquisition paradigm of the motor imagery EEG signal is an experimental paradigm based on the TDT electrophysiological workstation and python design, and the EEG signal is represented in matrix form as E={x1(t),x2(t),x3(t),...,x n (t)} T ,,x n (t) represents the EEG signal of the nth channel, t represents the experimental time, and T is x n The transposed representation of (t).

[0015] In the present invention, in step (2), the EEG signal channel area is 128 channels and seven areas, and the EEG channels contained in the seven areas are:

[0016] Area 1: Fp1,Fpz,Fp2,AF10,AF9,AF8,AF7,AF4,AF3,AFz,AFF8h,AFF7h,AFF6h,AFF5h,AFF1h,AFF4h,AFF3h,AFF2h,F10,F9,F8,F7,F6,F5,F4,F3,F2,F1,Fz

[0017] Area 2: FFT7h,FFC5h,FFC3h,FFC1h,FFC2h,FFC4h,FFC6h,FFT8h,FC1,FC2,FC3,FC4,FC5,FC6,FCz

[0018] Area 3: FCC1h, FCC2h, FCC3h, FCC4h, FCC5h, FCC6h, C1, C2, C3, C4, C5, C6, Cz, CCP1h, CCP2h, CCP3h, CCP4h, CCP5h, CCP6h

[0019] Region 4: CP1, CP2, CP3, CP4, CP5, CP6, CPz, CPP1h, CPP2h, CPP3h, CPP4h, CPP5h, CPP6h

[0020] Area 5: FT7, FT8, FT9, FT10, T7, T8, T9, T10, TP7, TP8, TP9, TP10, FTT7h, FTT8hTPP7h, TPP8h, TTP7h, TTP8h

[0021] Area 6: P1, P2, P3, P4, P5, P6, P7, P8, P9, P10, Pz, PPO1h, PPO2h, PPO3h, PPO4h, PPO5h, PPO6h, PPO7h, PPO8h, PO3, PO4, PO7, PO8, PO9, PO10, POz

[0022] Area 7: O1, Oz, O2, I1, I2, Iz;

[0023] The frequency band of EEG signals is between 0.5Hz and 80Hz;

[0024] The motor imagery time was 4 s.

[0025] In the present invention, in step (3), the design method of the EEG signal parameters is as follows:

[0026] Parameter 1, selection of channel region parameters: the design randomly selects 1 to 7 regions, for a total of 127 region parameters;

[0027] Parameter 2, selection of frequency band parameters: the frequency band range of 0.5Hz to 80Hz is divided into 5 frequency band parameters according to the EEG rhythm: delta wave, 0.5-4Hz; theta wave, 4-8Hz; alpha, 8-13Hz, also called mu rhythm; beta, 13-30Hz; gamma band, 30-80Hz;

[0028] Parameter 3, selection of motor imagery time parameters: divide the 4s motor imagery time period into 0.5s intervals, namely 0-0.5, 0.5-1, 1-1.5, 1.5-2, 2-2.5, 2.5-3, 3-3.5, 3.5-4; 1s intervals, namely 0-1, 1-2, 2-3, 3-4; 2s intervals, namely 0-2, 2-4; 4s intervals, namely 0-4, for a total of 4 types and 15 time parameters.

[0029] In the present invention, in step (3), an orthogonal table is designed for the EEG signal parameters, and the orthogonal table is a three-factor multi-level orthogonal table.

[0030] In the present invention, in step (4), based on the EEG signal determined by the orthogonal experiment, a spatial filter is generated using a common space pattern to realize the extraction of EEG features.

[0031] In the present invention, in step (5), the EEG signal features of the orthogonal experiment are classified, and two classifiers, support vector machine SVM and linear discriminant analysis LDA, are used to realize the decoding of motor EEG signals.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] 1. The present invention optimizes the EEG signal decoding parameters through orthogonal experiments. Orthogonal experimental design is another design method for studying multiple factors and multiple levels. It selects some representative points from a comprehensive experiment based on orthogonality for testing. These representative points have the characteristics of "even dispersion, neatness and comparability". Orthogonal experimental design is the main method of fractional factorial design. It is an efficient, fast and economical experimental design method. The famous Japanese statistician Genichi Taguchi listed the level combinations selected by orthogonal experiments in a table, which is called an orthogonal table. When the number of experiments required by the factorial design is too many, a very natural idea is to select some representative level combinations from the level combinations of the factorial design for testing.

[0034] 2. The present invention applies orthogonal experimental methods to EEG signal decoding. Based on the experimental parameters of EEG signal acquisition and decoding (motor imagery time, frequency band, and region of interest), orthogonal experiments are performed to optimize EEG decoding parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 Flowchart of the EEG signal decoding method based on orthogonal experiment. DETAILED DESCRIPTION

[0036] The present invention proposes an EEG signal decoding method based on orthogonal experiment, the flow chart is as follows Figure 1 As shown in the figure, it includes two key steps: designing EEG signal parameters based on orthogonal experiments, and extracting and classifying EEG signal features. The specific steps are as follows:

[0037] Step 1: Use the self-collected motor imagery dataset as the EEG signal to be analyzed. The acquisition paradigm of motor imagery EEG signals is based on the experimental paradigm designed by TDT electrophysiological workstation and Python. The EEG signal can be expressed in matrix form as E={x1(t),x2(t),x3(t),...,x n (t)} T , n represents the number of channels for collecting EEG signals, and T is the transposed representation of X;

[0038] Step 2: Extraction and generation of EEG signal parameters. The number of EEG signal channels collected is 128, and the 128 electrodes are divided into 7 areas. The number of EEG signal experiments collected is 200, the single experiment time is 4 seconds, and the frequency range is 0.5-100 Hz.

[0039] Step 3: Design EEG signal parameters based on orthogonal experiments. The parameters for EEG signal decoding are extracted through orthogonal experiments, including three parameters:

[0040] Parameter 1: Selection of EEG signal channel region parameters (1-7); the design randomly selects 1 to 7 regions, with a total of 127 (C17, C27, C37, C47, C57, C67, C77) regional parameters. The EEG channels contained in the seven regions are:

[0041] Area 1: Fp1,Fpz,Fp2,AF10,AF9,AF8,AF7,AF4,AF3,AFz,AFF8h,AFF7h,AFF6h,AFF5h,AFF1h,AFF4h,AFF3h,AFF2h,F10,F9,F8,F7,F6,F5,F4,F3,F2,F1,Fz

[0042] Area 2: FFT7h,FFC5h,FFC3h,FFC1h,FFC2h,FFC4h,FFC6h,FFT8h,FC1,FC2,FC3,FC4,FC5,FC6,FCz

[0043] Area 3: FCC1h, FCC2h, FCC3h, FCC4h, FCC5h, FCC6h, C1, C2, C3, C4, C5, C6, Cz, CCP1h, CCP2h, CCP3h, CCP4h, CCP5h, CCP6h

[0044] Region 4: CP1, CP2, CP3, CP4, CP5, CP6, CPz, CPP1h, CPP2h, CPP3h, CPP4h, CPP5h, CPP6h],

[0045] Area 5: FT7, FT8, FT9, FT10, T7, T8, T9, T10, TP7, TP8, TP9, TP10, FTT7h, FTT8hTPP7h, TPP8h, TTP7h, TTP8h

[0046] Area 6: P1, P2, P3, P4, P5, P6, P7, P8, P9, P10, Pz, PPO1h, PPO2h, PPO3h, PPO4h, PPO5h, PPO6h, PPO7h, PPO8h, PO3, PO4, PO7, PO8, PO9, PO10, POz],

[0047] Area 7: O1, Oz, O2, I1, I2, Iz;

[0048] Parameter 2: The selection of EEG signal frequency band parameters takes into account the full horizontality under the single factor of the orthogonal experiment, and the frequency band selection range is 0.5Hz to 80Hz; the frequency band range of 0.5Hz to 80Hz is further divided into δ waves (0.5-4Hz), θ waves (4-8Hz), α (8-13Hz, also called mu rhythm), β (13-30Hz), and gamma band (30-80Hz) according to the EEG rhythm, totaling 5 frequency band parameters.

[0049] Parameter 3: Selection of EEG signal motor imagery time parameters; the time of a single motor imagery experiment is 4s, and the 4s time period is divided into 0.5s intervals, namely 0-0.5, 0.5-1, 1-1.5, 1.5-2, 2-2.5, 2.5-3, 3-3.5, 3.5-4; 1s intervals, namely 0-1, 1-2, 2-3, 3-4; 2s intervals, namely 0-2, 2-4; and 4s intervals, namely 0-4, for a total of 4 types and 15 time parameters;

[0050] In step 4, we designed the EEG signal decoding parameters based on an orthogonal experimental table. Orthogonal experimental design is a design method for studying multiple factors and multiple levels. It uses orthogonality to select representative points from a comprehensive experiment for testing. Based on this orthogonal experimental design, we designed an orthogonal table for EEG signal parameters with three factors and multiple levels, as shown in Table 1.

[0051] Table 1 Three-factor multi-level orthogonal table of EEG decoding parameters

[0052] Number of regions Time window / s Frequency band / Hz 1 0.5 0.5-4 2 1 4-8 3 2 8-13 4 4 13-30 5 \ 30-80 6 \ \ 7 \ \

[0053] A total of 35 test cases for the orthogonal experiment were generated based on the three-factor multi-level orthogonal table, as shown in Table 2.

[0054] Table 2 35 test cases generated by orthogonal experiment

[0055]

[0056]

[0057] Step 4: Extract features of common spatial patterns from the EEG signals generated by the orthogonal experiment and determined based on 35 test cases;

[0058] Step 5: Classify the EEG signal features of 35 test cases and use two classifiers, support vector machine (SVM) and linear discriminant analysis (LDA), to decode the motor EEG signals.

[0059] In summary, the present invention utilizes the above-mentioned design method and data to design an EEG signal decoding method based on orthogonal experiments.

[0060] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art may make various variations or modifications within the scope of the claims, which do not affect the essence of the present invention.

Claims

1. A method for decoding EEG signals based on orthogonal experiments, characterized in that: The following steps are involved: (1) Using self-collected motor imagery datasets as EEG signals to be analyzed; (2) Extraction and generation of EEG signal parameters; EEG signal parameters include EEG signal channel area, frequency band and motor imagery time; (3) Design of EEG signal parameters based on orthogonal experiments; (4) Feature extraction of EEG signal parameters generated by orthogonal experiments; (5) Classify the extracted features and complete the decoding of EEG signals; In step (3), the design method of EEG signal parameters is as follows: Parameter 1, selection of channel region parameters: the design randomly selects 1 to 7 regions, for a total of 128 region parameters; Parameter 2, selection of frequency band parameters: the frequency band range of 0.5Hz to 80Hz is divided into 5 frequency band parameters according to the EEG rhythm: delta wave, 0.5-4Hz; theta wave, 4-8Hz; alpha, 8-13Hz, also called mu rhythm; beta, 13-30Hz; gamma band, 30-80Hz; Parameter 3, selection of motor imagery time parameters: divide the 4s motor imagery time period into 0.5s intervals, i.e., 0-0.5, 0.5-1, 1-1.5, 1.5-2, 2-2.5, 2.5-3, 3-3.5, 3.5-4; 1s intervals, i.e., 0-1, 1-2, 2-3, 3-4; 2s intervals, i.e., 0-2, 2-4; and 4s intervals, i.e., 0-4, for a total of 4 types and 15 time parameters; In step (3), an orthogonal table is designed for the EEG signal parameters, and the orthogonal table is a three-factor multi-level orthogonal table.

2. The EEG signal decoding method based on orthogonal experiment according to claim 1, characterized in that: In step (1), the acquisition paradigm of motor imagery EEG signals is based on the experimental paradigm designed by TDT electrophysiological workstation and python. The EEG signals are expressed in matrix form as E = {x1(t), x2(t), x3(t), ..., x n (t)} T , x n (t) represents the EEG signal of the nth channel, t represents the experimental time, and T is x n The transposed representation of (t).

3. The EEG signal decoding method based on orthogonal experiment according to claim 1, characterized in that: In step (4), based on the EEG signals determined by the orthogonal experiment, a spatial filter is generated using a common spatial pattern to extract EEG features.

4. The EEG signal decoding method based on orthogonal experiment according to claim 1, characterized in that: In step (5), the EEG signal features of the orthogonal experiment are classified, and two classifiers, support vector machine (SVM) and linear discriminant analysis (LDA), are used to decode the motor EEG signal.

Images