A four-dimensional attention recognition method based on electroencephalogram feature fusion selection

Abstract

The application provides a four-dimensional attention recognition method based on electroencephalogram feature fusion selection, and belongs to the technical field of electroencephalogram signal processing. Attention specifically includes four specific qualities of distribution, breadth, stability and transfer. The application combines electroencephalogram signals and attention qualities, extracts multi-domain electroencephalogram features to construct a fusion vector for four-dimensional attention recognition, and proposes a ReliefF-SBE-L1 three-layer hybrid feature selection model to improve recognition accuracy and calculation efficiency. The application can significantly reduce the calculation complexity while realizing efficient four-dimensional attention state recognition.

Description

Technical Field

[0001] This invention belongs to the field of EEG signal processing technology, especially the field of attention recognition based on EEG signals, specifically a four-dimensional attention recognition method based on EEG feature fusion selection. Background Technology

[0002] Attention refers to a person's ability to direct and concentrate their mental activity on a certain thing. It is a real and one of the most thoroughly studied psychological phenomena. Most studies focus on attention in a narrow sense, using only a single dimension of attention quality to determine an individual's overall attention level. In fact, attention is composed of a multi-dimensional structure and cannot be defined by a single concept. Its connotation should include the development of four specific qualities: 1) the breadth of attention (referring to the number of objects an individual can clearly perceive at the same time); 2) the stability of attention (referring to the duration of attention maintained on a certain activity); 3) the distribution of attention (referring to the ability of an individual to simultaneously direct attention to different objects when engaging in two or more activities); and 4) the shifting of attention (referring to the active shifting of attention from one activity to another according to a certain purpose).

[0003] Attention recognition is often based on subjective scales and questionnaires, but these methods cannot achieve objectivity and accuracy. More and more scholars are focusing on brain science research, and electroencephalogram (EEG) signals have been proven to be one of the most accurate and efficient physiological bases for studying attention. By preprocessing, extracting, and selecting features from EEG signals, and using machine learning classification algorithms, attention levels can be identified and evaluated, thereby accurately grasping attentional states.

[0004] Attention recognition based on EEG signals mainly considers two aspects: recognition accuracy and recognition complexity, which are essentially issues of precision and efficiency. A single feature does not contain enough effective information to achieve good classification and recognition results. Therefore, it is necessary to extract multiple electrode channel parameters of multiple EEG features to obtain higher recognition accuracy. However, excessively high feature dimensionality can lead to significant redundancy in the model, greatly increasing computational complexity and affecting classification and recognition accuracy to some extent. It is also prone to overfitting, resulting in poor generalization performance of the model. Summary of the Invention

[0005] To address the shortcomings of existing research, this invention provides a four-dimensional attention recognition method based on EEG feature fusion selection. In EEG-based attention recognition tasks, a multi-domain feature fusion vector is constructed to improve traditional single-feature recognition methods. The "ReliefF-SBE-L1" three-layer hybrid feature selection model is introduced to overcome the problem of feature redundancy and complex computation. It can achieve efficient recognition results for four-dimensional attention states of allocation, breadth, stability, and transition.

[0006] This invention provides a four-dimensional attention recognition method based on EEG feature fusion selection, comprising the following steps:

[0007] Step 1, Experimental Paradigm Design: Based on the development of attention quality in four dimensions—attention allocation, attention breadth, attention stability, and attention shift—design an experiment to induce four-dimensional attention.

[0008] Step 2, EEG signal acquisition: EEG signals are acquired while inducing four-dimensional attention using an EEG acquisition device;

[0009] Step 3: Signal preprocessing: The raw mixed EEG signals acquired in Step 2 are preprocessed using the EEGLAB toolbox and Matlab software;

[0010] Step 4: Feature extraction and fusion: From the EEG signal processed in Step 3, 16 features are extracted based on time domain, frequency domain, and time-frequency domain comparison. Each feature contains parameters of 17 electrode channels. The extracted feature parameters are fused to construct a multi-domain feature fusion vector.

[0011] Step 5, Feature Selection and Dimensionality Reduction: Use the "ReliefF-SBE-L1" three-layer hybrid feature selection model to perform feature selection and dimensionality reduction on the multi-domain fusion feature vector obtained in Step 4.

[0012] Step 6: SVM Recognition: Import the multi-domain feature fusion vector obtained in Step 5 into the SVM classifier to achieve four-dimensional attention recognition.

[0013] Furthermore, in step 2, the requirements for experimental subjects, experimental environment, and equipment during EEG signal acquisition are as follows:

[0014] (1) Experimental subjects and environment: Selected subjects were graduate or undergraduate students aged 19-25 years with normal vision or corrected vision and no history of major diseases. The experiment was conducted in a quiet and undisturbed environment.

[0015] (2) Experimental equipment: Select a multi-channel wet electrode EEG acquisition instrument (17 channels), along with conductive gel, syringe and 19 electrode caps.

[0016] Furthermore, step 3 includes the following steps:

[0017] Step 3.1, EEGLAB toolbox processing;

[0018] Electrode localization was performed using the EEGLAB toolbox to remove 50Hz power frequency interference and bandpass filtering from 1-40Hz. After reducing the sampling rate to 256Hz, bad segments were removed. The signal was then rereferenced using the T3 and T4 bilateral mastoid electrodes and subjected to independent component analysis (ICA) to remove artifacts such as electrooculography (EOG), electromyography (EMG), and electrocardiography (ECG).

[0019] Step 3.2: Wavelet thresholding for noise reduction;

[0020] Wavelet thresholding denoising includes wavelet decomposition, thresholding, and wavelet reconstruction; db4 wavelet is selected for denoising, and a fixed threshold calculation formula is used. Where N is the signal length.

[0021] Step 3.3: Sample set generation;

[0022] Samples are generated from the signals processed in step 3.2. A time window with a duration of 5 seconds and a step size of 2.5 seconds is selected. That is, the EEG signal is segmented into segments with a 5-second sliding window and an overlap rate of 50%, thus obtaining a sample set.

[0023] Furthermore, step 4 includes the following steps:

[0024] Step 4.1: Obtain temporal features.

[0025] After analysis and comparison, this invention selected the following eight time-domain features, including statistical and nonlinear characteristics:

[0026] (1) Standard deviation (Std), which is the arithmetic square root of the variance:

[0027] x(i) is an N-point signal sequence. Let F be the mean of the signal sequence and F be the standard deviation.

[0028] (2) Hjorth activity parameter (HA) refers to the variance of the EEG signal, i.e., the signal power, and represents the width:

[0029] σ0 is the standard deviation of the signal, and HA is the Hjorth activity parameter.

[0030] (3) Hjorth mobility parameter (HM), used to estimate the average frequency of the EEG signal:

[0031] σ1 is the standard deviation of the first derivative of the signal, and HM is the Hjorth mobility parameter.

[0032] (4) The Hjorth complexity parameter (HC) refers to the mobility of the first derivative relative to the EEG itself:

[0033] σ² is the standard deviation of the second derivative of the signal, and HC is the Hjorth complexity parameter.

[0034] (5) Lempel-Ziv complexity (LZC) refers to the complexity of data directly without data decomposition or transformation, used to characterize the corresponding time series:

[0035] C N Let be the time complexity, N be the number of loops, and C be the normalized Lempel-Ziv time complexity.

[0036] (6) Higuchi fractal dimension (HFD) can be used to describe the complexity and self-similarity of signals;

[0037] A higher Higuchi fractal dimension indicates stronger self-similarity and greater signal complexity. This invention uses the Higuchi algorithm to calculate the fractal dimension, dividing the original time series into k segments and calculating the average curve length L(k) of each segment, based on ln... L(k) with ln 1 / k A straight line is obtained by fitting the linear relationship between them, and its slope is the Higuchi fractal dimension D.

[0038] (7) Permutation entropy (Pec) is a nonlinear method used to detect time series complexity (randomness) or dynamic abrupt changes, and can quantitatively assess the random noise contained in the signal.

[0039] The smaller the permutation entropy value, the simpler and more regular the signal; the larger the value, the more complex and random the signal. The basic principle and algorithm flow are shown in Table 1 below:

[0040] Table 1. Principle of the Permutation Entropy Algorithm

[0041]

[0042]

[0043] (8) Sample entropy, an improvement on approximate entropy, is used to describe the probability of a time series generating new patterns when its complexity and dimensionality change. Higher complexity and greater probability result in a larger sample entropy value. Its algorithm is shown in Table 2 below:

[0044] Table 2 Sample Entropy Algorithm Principle

[0045]

[0046]

[0047] This invention selects parameter values ​​of m=2 and r=0.15σ to calculate the sample entropy characteristics of 17 electrode channels.

[0048] Step 4.2: Obtain frequency domain features.

[0049] Frequency domain feature extraction requires separating delta waves, theta waves, alpha waves, and beta waves from the original signal for analysis. Since EEG signals are non-stationary, this invention uses wavelet packet decomposition to analyze the signal, considering both low-frequency and high-frequency components, and extracts the following frequency domain features:

[0050] (1) Energy and ratio;

[0051] This invention calculates the sum of squares of the wavelet packet coefficients in different frequency bands after decomposition to obtain the energy of the corresponding frequency bands of δ wave, θ wave, α wave, and β wave. Then, it calculates the energy ratios of the corresponding frequency bands of θ wave and α wave, θ wave and β wave, and α wave and β wave. The formula for calculating the frequency band energy is shown below:

[0052]

[0053] Where i represents the corresponding frequency band, n represents the number of coefficients, and D i,j E represents the coefficient of the j-th wavelet packet in frequency band i. i This refers to frequency band energy.

[0054] (2) Differential entropy (DE);

[0055] Differential entropy is Shannon information entropy Generalization for continuous variables:

[0056]

[0057] Here, p(x) represents the probability density function of continuous information, and [a,b] represents the interval of information values. For a given length, it approximately follows a Gaussian distribution. The differential entropy of the EEG signal is:

[0058]

[0059] The differential entropy is equal to the logarithm of its energy spectrum in a specific frequency band. For ease of calculation, the power spectrum can be used to approximate the differential entropy. The specific formula for the differential entropy is:

[0060]

[0061] PSD is the power spectral density of the signal, which can be obtained using the periodogram method or the autocorrelation method.

[0062] Step 4.3: Multi-domain feature vector fusion;

[0063] To achieve better recognition results, this invention fuses the extracted features to construct a multi-domain fusion feature vector, resulting in a 272-dimensional (16 features * 17 channels) feature vector.

[0064] Furthermore, in step 5, the “ReliefF-SBE-L1” three-layer hybrid feature selection model includes three layers of selection: First, based on the Filter method, the ReliefF algorithm is used for the first layer of channel filtering, calculating the weight of each channel under a single feature according to the correlation between the channel parameters and the category label, and eliminating redundant channels; Second, based on the Wrapper method, the Backward Search SBE algorithm is used for the second layer of feature selection, with an embedded SVM classifier continuously removing features to select the feature subset with the best performance index; Finally, based on the Embedded method, Lasso regression, i.e., L1 regularization, is used for the third layer of global optimization to further eliminate redundant optimization models.

[0065] Beneficial effects: The four-dimensional attention recognition method based on EEG feature fusion selection proposed in this invention improves recognition accuracy and computational efficiency; by extracting multi-domain EEG features to construct fusion vectors for four-dimensional attention recognition, it improves the traditional single-feature recognition method; by proposing the "ReliefF-SBE-L1" three-layer hybrid feature selection model, it significantly reduces computational complexity while achieving efficient four-dimensional attention state recognition. Attached Figure Description

[0066] Figure 1 This is a framework diagram of the four-dimensional attention recognition method based on EEG feature fusion selection proposed in this invention;

[0067] Figure 2 This invention presents a four-dimensional attention-inducing experimental paradigm.

[0068] Figure 3 This is a schematic diagram of the electrode channels of the EEG acquisition device of the present invention;

[0069] Figure 4 This is a schematic diagram of the EEG signal preprocessing process of the present invention;

[0070] Figure 5 This is a schematic diagram of the processing of the EEGLAB toolbox of the present invention;

[0071] Figure 6 This is a comparison of the signals before and after wavelet threshold denoising in this invention;

[0072] Figure 7 This is a schematic diagram illustrating the generation of sample data according to the present invention;

[0073] Figure 8 This is a schematic diagram illustrating the wavelet packet decomposition method used in extracting frequency domain features according to the present invention;

[0074] Figure 9 This is a schematic diagram of the multi-domain fusion feature vector combination constructed by the present invention;

[0075] Figure 10This is a schematic diagram of the "ReliefF-SBE-L1" three-layer hybrid feature selection model proposed in this invention;

[0076] Figure 11 This is a schematic diagram of the four-dimensional attention recognition using the SVM classifier in this invention. Detailed Implementation

[0077] To make the objectives, technical solutions, and advantages of this invention clearer, the embodiments of this invention will be further described in detail below with reference to the accompanying drawings.

[0078] The four-dimensional attention recognition method based on EEG feature fusion selection provided by this invention has the following framework: Figure 1 As shown, this invention combines EEG signals with four-dimensional attention recognition. Different experiments are designed to precisely induce four-dimensional attention in subjects while simultaneously collecting EEG signals. The collected EEG signals are preprocessed, and multi-domain features are extracted from the preprocessed signals. In the time domain, statistical features such as standard deviation, nonlinear features such as sample entropy, differential entropy, permutation entropy, Hjorth parameter, Lempel-Ziv complexity, and Higuchi fractal dimension are extracted. In the frequency domain, four energy features and three energy ratio features are extracted based on wavelet packet decomposition. The 16 features from the 17 extracted channels are combined into a multi-domain feature fusion vector for classification. A three-layer hybrid feature selection model, "ReliefF-SBE-L1," is proposed to achieve complementary advantages and disadvantages. The ReliefF algorithm is used to filter channels based on filtering, packaging, and embedding methods, while the SBE algorithm is used to select features, and L1 regularization is used for global optimization. The resulting multi-domain feature fusion vector significantly reduces computational complexity while more efficiently recognizing four-dimensional attention states.

[0079] This embodiment provides a four-dimensional attention recognition method based on EEG feature fusion selection. The specific steps are as follows:

[0080] Step 1: Design an experimental paradigm to induce four-dimensional attention;

[0081] Unlike previous general attention-inducing experiments, this invention combines the operational definition of attention quality with four dimensions of attention quality development: attention allocation, attention breadth, attention stability, and attention transfer, and designs an experiment to induce four-dimensional attention.

[0082] The four-dimensional attention-inducing experimental paradigm designed in this embodiment, such as Figure 2As shown, four tests are used to elicit attention allocation, breadth, stability, and shift in four corresponding dimensions. Test 1, Pattern Recognition Test: This test consists of two rings of different sizes with notches. Because the notches are in different directions, many similar but different patterns can be formed. The total number of rings is 15 * 20 = 300. Participants are required to identify and mark two designated patterns. Test 2, Four-Circle Selection Test: This test consists of small squares with different numbers of circles drawn on them. The total number of squares is 26 * 50 = 650. Participants are required to identify and mark the squares with four circles drawn on them. Test 3, Visual Tracking Test: This test consists of multiple curves starting on the left and ending on the right. There are 25 curves in total. Participants are required to use their eyes to track a curve starting from the left and write the starting number of the curve in the square where the curve ends on the right. Test 4, Addition and Subtraction Test: The test consists of natural numbers. Test takers are required to alternate between adding and subtracting numbers and write the result between the two numbers. The total number of results is 22 * ​​12 = 264.

[0083] Step 2: EEG signal acquisition;

[0084] (1) Experimental subjects and environment:

[0085] Because EEG signals vary among different individuals, factors such as age, gender, and medical history can have varying degrees of influence. To ensure the accuracy and validity of the experimental data and final results, careful screening of recruited participants is essential. Participants should be current graduate or undergraduate students, aged 19-25, with normal vision or corrected vision, and no history of serious illness. Since EEG signals are easily affected by external interference, the experiment requires a comfortable, quiet environment with minimal disturbances.

[0086] (2) Experimental equipment: The multi-channel wet electrode EEG acquisition instrument produced by Nanjing Weisi Medical was used, along with conductive gel, syringe and 19 electrode cap.

[0087] The electrode channel distribution diagram of the EEG acquisition device designed in this embodiment is shown below. Figure 3 As shown, the 19-lead electrode distribution conforms to the international 10-20 system standard electrode placement method, specifically including frontal electrodes Fp1 and Fp2, frontal lobe electrodes F3, F4, F7, F8, and Fz, central lobe electrodes C3, C4, and CZ, parietal lobe electrodes P3, P4, and PZ, occipital lobe electrodes O1 and O2, temporal lobe electrodes T3, T4, T5, and T6, and reference electrodes A1 and A2.

[0088] Step 3: Preprocessing of EEG signals;

[0089] EEG signals are easily affected by environmental noise and physical magnetic fields. In order to obtain clean and effective EEG signals with a high signal-to-noise ratio, it is necessary to preprocess the raw mixed EEG signals based on the EEGLAB toolbox and Matlab software.

[0090] The flowchart of EEG signal preprocessing in this embodiment is shown below. Figure 4 As shown. Preprocessing is based on the EEGLAB toolbox and Matlab software, and consists of three parts: EEGLAB toolbox processing, wavelet thresholding denoising, and sample set generation.

[0091] Step 3.1: The EEGLAB toolbox processing includes using the EEGLAB toolbox electrode positioning, removing 50Hz power frequency interference and bandpass filtering from 1 to 40Hz, reducing the sampling rate to 256Hz and then removing bad segments, re-referencing the signal with the T3 and T4 bilateral mastoid electrodes of the EEG acquisition device, performing independent component analysis (ICA), and removing artifacts from electrooculography (EOG), electromyography (EMG), and electrocardiography (ECG). Figure 5 This is a schematic diagram of the EEGLAB toolbox, showing a comparison of the effects of removing bad segments and artifacts such as electrooculograms through the EEGLAB toolbox series of processing.

[0092] Step 3.2, Wavelet threshold denoising includes wavelet decomposition, thresholding, and wavelet reconstruction. Wavelet filtering can reduce noise while maximizing the authenticity of the original signal. Its algorithm mainly includes three parts: selecting the wavelet basis, threshold, and threshold function. This invention uses the db4 wavelet for denoising and adopts a fixed threshold calculation formula. Where N is the signal length, a hard threshold function is selected to prevent errors caused by feature loss after soft threshold denoising.

[0093] Figure 6 This is a comparison of the signal before and after wavelet threshold denoising. By comparing the time-domain waveforms and signal-to-noise ratios of each channel of the original signal and the signal after wavelet threshold denoising, it can be clearly seen that the reconstructed signal has a higher signal-to-noise ratio while effectively preserving complete information.

[0094] Step 3.3, Sample generation, refers to segmenting the signal obtained after wavelet threshold denoising to obtain a sample set. Figure 7 This is a schematic diagram of sample set generation. Samples are generated from the processed signals. A time window with a duration of 5 seconds and a step size of 2.5 seconds is selected. That is, the EEG signal is segmented into segments with a 5-second sliding window and an overlap rate of 50%, resulting in a sample set with a single sample length of 5 seconds.

[0095] Step 4: Feature Extraction and Fusion: The purpose of feature extraction is to identify features from EEG signals that can effectively reflect different attentional states.

[0096] From the EEG signals processed in step 3, 16 features were extracted based on time domain, frequency domain, and time-frequency domain comparison. Each feature contains parameters from 17 electrode channels. Figure 8 This is a schematic diagram of wavelet packet decomposition used when extracting frequency domain features. Since the required brainwave frequency band is concentrated in 0-32Hz, this embodiment shows the decomposition from the second layer of the wavelet tree. Node (5,0) mainly corresponds to the delta wave, node (5,1) mainly corresponds to the theta wave, node (5,2) mainly corresponds to the alpha wave, and the five components from node (5,3) to node (5,7) mainly correspond to the beta wave. The extracted feature parameters are fused to construct a multi-domain feature fusion vector.

[0097] Figure 9 This diagram illustrates the combination of multi-domain feature fusion vectors. To effectively identify four-dimensional attention, this invention fuses the extracted features to construct a multi-domain feature fusion vector, resulting in a 272-dimensional (17 channels * 16 feature parameters) feature vector F. The feature parameters represented by F1 to F16 in the diagram are, in order: standard deviation (Std), Hjorth activity parameter (HA), Hjorth mobility parameter (HM), Hjorth complexity parameter (HC), Lempel-Ziv complexity (LZC), fractal dimension (HFD), permutation entropy (Pec), sample entropy (SampEn), delta wave energy (E_Delta), theta wave energy (E_Theta), alpha wave energy (E_Alpha), beta wave energy (E_Beta), the ratio of theta wave energy to alpha wave energy (E_Theta / Alpha), the ratio of theta wave energy to beta wave energy (E_Theta / Beta), the ratio of alpha wave energy to beta wave energy (E_Alpha / Beta), and differential entropy (DE). All of these are feature vectors containing 17 electrode channels.

[0098] Step 5, Feature Selection and Dimensionality Reduction: Use the "ReliefF-SBE-L1" three-layer hybrid feature selection model to perform feature selection on the multi-domain fusion feature vector obtained in Step 4.

[0099] To reduce redundancy, a three-layer hybrid feature selection model, "ReliefF-SBE-L1", is proposed to perform feature selection on high-dimensional multi-domain feature fusion vectors. This model aims to achieve complementary advantages among different feature selection methods and select the feature subset that best performs for the classifier.

[0100] Figure 10This diagram illustrates the "ReliefF-SBE-L1" three-layer hybrid feature selection model proposed in this invention. The model primarily comprises three selection layers: First, based on the Filter method, the ReliefF algorithm is used for the first layer of channel filtering, calculating the weight of each channel under a single feature according to the correlation between channel parameters and class labels, and eliminating redundant channels. Second, based on the Wrapper method, the Backward Search SBE algorithm is used for the second layer of feature selection, continuously removing features using an embedded SVM classifier to select the feature subset with the optimal performance index. Finally, based on the Embedded method, Lasso regression (i.e., L1 regularization) is used for the third layer of global optimization to further eliminate redundant and optimize the model. The principles of the Backward Search SBE and L1 regularization algorithms are shown in Tables 3 and 4 below, respectively.

[0101] Table 3. SBE Algorithm Principle

[0102]

[0103] Table 4. Flowchart of the L1 Regularization-based Feature Selection Algorithm

[0104]

[0105]

[0106] Step 6: Import the multi-domain feature fusion vector obtained in Step 5 into the SVM classifier to achieve four-dimensional attention recognition.

[0107] Figure 11 This diagram illustrates the use of an SVM classifier for four-dimensional attention recognition. The results clearly show that the recognition method proposed in this invention significantly reduces computational complexity while achieving more efficient four-dimensional attention state recognition.

Claims

1. A four-dimensional attention recognition method based on EEG feature fusion selection, characterized in that, Includes the following steps: Step 1: Based on the development of attention quality in four dimensions—attention allocation, attention breadth, attention stability, and attention shift—design an experiment to induce four-dimensional attention. Step 2: Collect brain signals while inducing four-dimensional attention using an EEG acquisition device; Step 3: Preprocess the raw mixed EEG signals acquired in Step 2 using the EEGLAB toolbox and Matlab software; Step 4: Extract multiple features from the EEG signal processed in Step 3 based on time domain, frequency domain, and time-frequency domain comparison. Each feature contains parameters of multiple electrode channels. The extracted feature parameters are fused to construct a multi-domain feature fusion vector. Step 5: Use the ReliefF-SBE-L1 three-layer hybrid feature selection model to perform feature selection and dimensionality reduction on the multi-domain fusion feature vector obtained in Step 4; Step 6: Import the multi-domain feature fusion vector obtained in Step 5 into the SVM classifier to achieve four-dimensional attention recognition; In step 5, the ReliefF-SBE-L1 three-layer hybrid feature selection model includes three layers of selection: First, based on the Filter method, the ReliefF algorithm is used for the first layer of channel filtering, calculating the weight of each channel under a single feature according to the correlation between the channel parameters and the category label, and eliminating redundant channels; Second, based on the Wrapper method, the backward search SBE algorithm is used for the second layer of feature selection, with an embedded SVM classifier continuously removing features to select the feature subset with the best performance index; Finally, based on the Embedded method, Lasso regression, i.e., L1 regularization, is used for the third layer of global optimization to further eliminate redundant optimization models.

2. The four-dimensional attention recognition method according to claim 1, characterized in that, The requirements for experimental subjects, experimental environment, and equipment during EEG signal acquisition in step 2 are as follows: (1) Experimental subjects and environment: Selected subjects were graduate or undergraduate students aged 19-25 years with normal vision or corrected vision and no history of major diseases; a quiet and undisturbed environment was required; (2) Experimental equipment: Select a wet electrode EEG acquisition instrument with 17 channels.

3. The four-dimensional attention recognition method according to claim 1, characterized in that, Step 3 includes the following steps: Step 3.1, EEGLAB Toolbox Processing: Use the EEGLAB toolbox for electrode positioning, remove 50Hz power frequency interference and bandpass filter 1-40Hz, reduce the sampling rate to 256Hz and remove bad segments, rereference the signal with the T3 and T4 bilateral mastoid electrodes of the EEG acquisition device, perform independent component analysis (ICA) to remove artifacts from electrooculography, electromyography and electrocardiography. Step 3.2, Wavelet Threshold Denoising: This includes wavelet decomposition, thresholding, and wavelet reconstruction; db4 wavelet is selected for denoising, and a fixed threshold calculation formula is used. ,in The signal length; Step 3.3, Sample set generation: Generate samples from the signal processed in step 3.

2. Use a 5-second sliding window to segment the signal with an overlap rate of 50% to obtain the sample set.

4. The four-dimensional attention recognition method according to claim 1, characterized in that, Sixteen features were extracted from the EEG signals processed in step 3 based on time domain, frequency domain, and time-frequency domain comparison. Each feature contains parameters of 17 electrode channels. Step 4 includes the following steps: Step 4.1: Obtain 8 time-domain features; (1) Standard deviation: , for point signal sequence, Let F be the mean of the signal sequence and F be the standard deviation. (2) Hjorth activity parameter refers to the variance of the EEG signal, i.e., the signal power, and represents the width: , is the standard deviation of the signal, and HA is the Hjorth activity parameter; (3) Hjorth mobility parameter, used to estimate the average frequency of the EEG signal: , It is the standard deviation of the first derivative of the signal, and HM is the Hjorth mobility parameter; (4) The Hjorth complexity parameter refers to the mobility of the first derivative relative to the EEG itself: , It is the standard deviation of the second derivative of the signal, and HC is the Hjorth complexity parameter; (5) Lempel-Ziv complexity refers to the complexity of data directly without data decomposition or transformation, used to characterize the corresponding time series: , For complexity, Let C be the number of iterations, and C be the normalized Lempel-Ziv complexity. (6) Higuchi fractal dimension, used to describe the complexity and self-similarity of signals; The Higuchi algorithm is used to calculate the fractal dimension of the original time series. Sub-segments and calculate the average curve length of the sub-segments. ,according to and A linear relationship between them is fitted to obtain a straight line, the slope of which is the Higuchi fractal dimension. ; (7) Permutation entropy, a nonlinear method used to detect time series complexity or dynamic abrupt changes, can quantitatively assess the random noise contained in a signal; (8) Sample entropy, used to describe the complexity of a time series and the probability of a new pattern being generated when the dimension changes; Step 4.2: Obtain 8 frequency domain features; (1) Wave energy Wave energy Wave energy Wave energy Waves and Wave energy ratio, Waves and Wave energy ratio, Waves and Wave energy ratio; Separate from the processed EEG signals in step 3 Wave, Wave, Wave, Wavelet packet decomposition is used to analyze the signal, taking into account both low-frequency and high-frequency components. The sum of squares of the wavelet packet coefficients in different frequency bands after decomposition is obtained. Wave, Wave, Wave, The energy of the corresponding frequency band of the wave is then calculated. Waves and Wave, Waves and Wave, Waves and The ratio of wave energy to corresponding frequency band energy is given by the following formula: ； in, Represents the corresponding frequency band. The number of representative coefficients. Indicates frequency band The Wavelet packet coefficients, For frequency band energy; (2) Differential entropy; The differential entropy is equal to the logarithm of its energy spectrum in a specific frequency band. For ease of calculation, the power spectrum can be used to approximate the differential entropy. The specific formula for the differential entropy is: ； in, The power spectral density of the signal can be obtained using the periodogram method or the autocorrelation method. Step 4.3: Multi-domain feature vector fusion; The features extracted in steps 4.1 and 4.2 are fused to construct a multi-domain fusion feature vector, resulting in a 272-dimensional feature vector.

Images