Electroencephalogram attention level monitoring method based on double-domain deep supervision feature fusion network

Abstract

The invention discloses an electroencephalogram attention level monitoring method based on a double-domain deep supervision feature fusion network, and the method comprises the steps: collecting an EEG (electroencephalogram) signal and an EOG (eye movement) signal, and forming an attention level database; a dual-domain deep supervision feature fusion network is constructed, a data processing module in the network analyzes time sequence data and frequency spectrum data of EEG signals, a PERCLOS value is calculated based on the EOG signals to serve as an attention level quantitative index, and a time sequence data set and a frequency spectrum data set are constructed; the time attention module extracts EEG signal time domain features and outputs a main function loss value; the frequency spectrum residual module extracts EEG signal frequency spectrum features and outputs a main function loss value; the deep supervision module receives and processes output values of the slave TransformerEncoder and the first residual block, provides two auxiliary loss values, integrates all the main loss values and the auxiliary loss values, and enhances the capturing capability of the network for feature details; and the feature fusion module monitors and evaluates the attention level of the pilot through a linear regression algorithm.

Description

Technical Field

[0001] The present invention belongs to the technical field of attention level calculation and prediction, and relates to a method for monitoring EEG attention level based on a dual-domain deep supervision feature fusion network. Background Art

[0002] With the rapid development of the aviation industry, the complexity of flight missions has increased significantly, and the challenges facing flight safety have also intensified. Human factors are the main cause of flight accidents, among which improper allocation of pilot attention due to accumulated fatigue and decreased vigilance is an important cause of accidents. Therefore, it is crucial to effectively monitor the pilot's attention level and promptly detect and correct potential attention deviations to reduce human errors and improve flight safety. At present, physiological signal monitoring technologies such as EEG and EOG provide a scientific basis for the quantitative evaluation of attention status by capturing changes in pilots' physiological responses during flight missions.

[0003] However, existing attention monitoring methods are generally limited to the one-sided analysis of a single physiological signal source and a single feature. This single-dimensional analysis method not only limits the comprehensive and in-depth understanding of physiological responses, but also directly affects the accuracy and reliability of attention assessment results due to the lack of key information. Summary of the invention

[0004] Purpose of the invention: In order to solve the problem of lack of objective and accurate monitoring means in the existing pilot attention level monitoring, a method for monitoring EEG attention level based on a dual-domain deep supervision feature fusion network is provided.

[0005] Technical solution: To achieve the above purpose, the present invention provides a method for monitoring EEG attention level based on a dual-domain deep supervision feature fusion network, comprising the following steps:

[0006] S1: Through high-precision EEG and EOG equipment, the physiological reactions of pilots during mission execution are captured in real time, and EEG and EOG data are collected to form an attention level database;

[0007] S2: preprocessing the acquired EEG data;

[0008] S3: Align the EOG data and the preprocessed EEG data according to the timestamp, select 4s as the time window, segment the continuous EOG data, and calculate the PERCLOS value in each time window as a quantitative indicator of attention level;

[0009] S4: Construct a dual-domain deep supervision feature fusion network, including a data processing module, a temporal attention module, a spectral residual module, a deep supervision module, and a feature fusion module;

[0010] S5: Processing the EEG signal time series data and the EEG signal spectrum data respectively through the data processing module to obtain the time series data and the spectrum data;

[0011] S6: construct a time series data set and a spectrum data set according to the PERCLOS value obtained in step S3 and the time series data and spectrum data obtained in step S5; each time series data and spectrum data segment is aligned and labeled with the corresponding PERCLOS value to form a data-label pair required for machine learning, and the PERCLOS value is used as a label and input into the regression network model to guide the learning and training of the model;

[0012] S7: For the acquired time series data set, the multi-head attention mechanism of the time attention module is used to capture the key information in the EEG signal time series data, and the main function loss is output. temporal and the time domain feature F temporal ;

[0013] S8: For the acquired spectrum data set, the spectrum residual module deeply mines the high-order attention-related information in the spectrum data by stacking multiple residual blocks, and outputs the main function loss loss spectral And the spectrum feature F spectral ;

[0014] S9: Input the output of the multi-head attention mechanism in step S7 and the output of the first residual block obtained in step A8 into the deep supervision module respectively, and generate auxiliary loss loss through two deep supervision sub-modules. aux1 and loss aux2 , combined with the main function loss loss temporal 、loss spectral , integrate all main loss values ​​and auxiliary loss values;

[0015] S10: According to the time domain feature F temporal And the spectrum feature F spectral ,The pilot’s attention level is predicted by linear regression algorithm.

[0016] S11: Input the time series dataset and spectrum dataset into the DBPNet model, output the error with the true PERCLOS value, and evaluate the performance of the model through the mean squared error (MSE), mean absolute error (MAE) and correlation coefficient (CORR) indicators.

[0017] Furthermore, the preprocessing of the EEG data in step S2 includes re-referencing, downsampling, and band-pass filtering, specifically:

[0018] A1: Re-reference the EEG data using the formula:

[0019]

[0020] Where N represents the total number of electrodes, EEG i (t) represents the signal of the i-th electrode;

[0021] A2: Downsample the EEG data using the formula:

[0022] x′[n]=x[Mn]

[0023] Where x′[n] is the downsampled sequence, x[Mn] is one sample taken for every M samples, and M is the downsampling factor;

[0024] A3: Perform Butterworth bandpass filtering on the EEG data. The formula is:

[0025]

[0026] Among them, ω l and ω h are the normalized high and low frequency cutoff frequencies, respectively, and Q is the quality factor.

[0027] Furthermore, the calculation method of the quantitative index of the attention level in step S3 is:

[0028] B1: Calculate the PERCLOS value in each time window. The formula is:

[0029]

[0030] Among them, CLOS represents the eye closure time, blink represents the blink time, fixation represents the fixation time, and saccade represents the saccade time;

[0031] B2: Get attention level quantitative labels Labels = {PERCLOS1, PERCLOS2, ..., PERCLOS n}.

[0032] Furthermore, in the data processing module of step S5, the EEG signal time series data is processed by co-spatial pattern analysis, and the EEG signal spectrum data is processed by multi-band DE feature extraction and 2D topological map mapping technology. The specific processing flow is as follows:

[0033] C1: Analyzing EEG time series data {X1,X2,...,X N}, calculate the covariance matrix for each sample:

[0034]

[0035] Compute the correlation covariance matrix:

[0036]

[0037] Among them, y i is the target variable value associated with the i-th sample;

[0038] Find the projection matrix that maximizes the objective function:

[0039]

[0040] Where R is the average of all sample covariance matrices;

[0041] Generate regression task features:

[0042] Z i =W T X i ;

[0043] C2: Decompose the EEG signal into five bands, namely delta band, theta band, alpha band, beta band and gamma band;

[0044] C3: For each frequency band, the time domain signal is converted to a frequency domain signal by fast Fourier transform (FFT):

[0045]

[0046] Where n is the time index and N is the number of samples;

[0047] Calculate PSD:

[0048]

[0049] Calculate the DE value using the normalized PSD:

[0050] DE=-Σ f∈band p[f]log(p[f])

[0051] Where f is the band.

[0052] Furthermore, the temporal attention module of step S7 is composed of a TransformerEncoder, a convolutional layer, an average pooling layer and a fully connected layer, and the processing flow is specifically as follows:

[0053] D1: Capturing EEG time series data features through Transformer Encoder:

[0054] EncoderOutput=TransformerEncoder(EEG temporal );

[0055] D2: Temporal attention module outputs temporal features F temporal :

[0056]

[0057] Where W is the weight of the fully connected layer, and b is the bias of the fully connected layer;

[0058] D3: The loss value of the temporal attention module output is calculated through the Mean Squared Error (MAE) loss function:

[0059]

[0060] Among them, y i is the true value, is the predicted value and N is the total number of samples.

[0061] Furthermore, the spectrum residual module of step S8 is composed of a convolution layer, three multi-band fusion residual blocks and a pooling layer. The specific processing flow of the spectrum residual module is as follows:

[0062] E1: Capturing EEG spectrum data features through multi-band fusion residual blocks:

[0063] ResBlockOutput = σ(F(x)+x)

[0064] Among them, x is the input of the residual block, F(x) is the internal operation of the residual block, and σ is the activation function;

[0065] E2: Spectral residual module outputs spectral features F spectral :

[0066] F spectral =Avg(ResBlock(ResBlock(ResBlock(Conv(x)))))

[0067] E3: Calculate the loss value of the spectral residual module output through the MAE loss function:

[0068]

[0069] Among them, y i is the true value, is the predicted value and N is the total number of samples.

[0070] Furthermore, the two deep supervision sub-modules in the deep supervision module of step S9 are respectively a time domain deep supervision sub-module and a frequency domain deep supervision sub-module. The time domain deep supervision sub-module consists of a convolutional layer and two fully connected layers, and the frequency domain deep supervision sub-module consists of two convolutional layers, an average pooling layer and a fully connected layer.

[0071] Furthermore, the processing flow of the deep supervision module in step S9 is as follows:

[0072] F1: Directly receive the TransformerEncoder output value EncoderOutput through the time domain deep supervision submodule and calculate the auxiliary loss loss aux1 :

[0073]

[0074] F2: Directly receive the first residual block output value ResBlockOutput through the frequency domain deep supervision submodule and calculate the auxiliary loss loss aux2 :

[0075]

[0076] F3: Integrate the main function loss value and the auxiliary loss value to get the total loss value:

[0077] L total =loss temporal +loss spectral +λ1loss aux1 +λ2loss aux2 .

[0078] Furthermore, in step S10, the time domain feature F is obtained temporal and frequency domain feature F spectral , the features are fused by connection, and the fused features are expressed as; F = [F temporal ,F spectral ].

[0079] The method of the present invention is based on simulating an aircraft cockpit scene, designs flight mission scenes with different difficulty gradients, induces changes in the pilot's attention level, and synchronously collects high-precision EEG and EOG signals to form an attention level database; constructs a dual-domain deep supervision feature fusion network, the data processing module in the network analyzes the EEG signal time series data and spectrum data, and calculates the PERCLOS value based on the EOG signal as a quantitative indicator of the attention level, and constructs a time series data set and a spectrum data set; the time attention module extracts the time domain features of the EEG signal and outputs the main function loss value; the spectrum residual module extracts the spectrum features of the EEG signal and outputs the main function loss value; the deep supervision module receives and processes the output values ​​from the TransformerEncoder and the first residual block, provides two auxiliary loss values, integrates all the main loss values ​​and auxiliary loss values, and enhances the network's ability to capture feature details; the feature fusion module monitors and evaluates the pilot's attention level through a linear regression algorithm.

[0080] In order to realize accurate monitoring and evaluation of pilots' attention level, the present invention proposes a method for monitoring EEG attention level based on dual-domain deep supervision feature fusion network. The method collects EEG and eye movement signals to form a multimodal attention level database; constructs a time attention module and a spectrum residual module to extract the time domain and frequency domain features of EEG data respectively, and calculates the output total loss value; the deep supervision module further extracts local feature information and outputs auxiliary loss value; the feature fusion module fuses time domain features and frequency domain features through a linear regression algorithm; the model is optimized based on the total loss value and auxiliary loss value, and finally the monitoring and evaluation of pilots' attention level is realized.

[0081] Beneficial effects: Compared with the prior art, the method proposed in the present invention is based on the multimodal data input of EEG and EOG, uses a dual-domain deep supervision feature fusion network, focuses on global temporal features through the time attention module, and obtains more information from the frequency domain features through the spectrum residual module, thereby improving the model's ability to capture details and the richness of feature expression. The deep supervision module enhances the model's feature learning and generalization capabilities through the combined optimization of the main loss and the auxiliary loss, enabling the model to effectively adapt to data changes in complex task environments, and improves the accuracy of the model's prediction of attention levels, making it suitable for attention monitoring and evaluation in complex operating environments such as pilots and drivers. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 is a flow chart of the method of the present invention;

[0083] Figure 2 This is a diagram of the network architecture based on dual-domain deep supervision feature fusion proposed in the present invention;

[0084] Figure 3is the evaluation index value of different models under different test conditions in the embodiment;

[0085] Figure 4 are the evaluation index values ​​of different models under within-subject conditions in the embodiments. DETAILED DESCRIPTION

[0086] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0087] The present invention provides a method for monitoring EEG attention level based on a dual-domain deep supervision feature fusion network. Figure 1 As shown, it includes the following steps:

[0088] S1: Based on the aircraft cockpit simulation platform, different flight mission scenarios are designed. The scenarios are arranged in increasing difficulty to induce changes in pilots' attention under different stress levels. High-precision EEG and EOG equipment are used to capture the pilots' physiological reactions in real time during the mission, collect EEG and EOG data, and form an attention level database.

[0089] S2: Preprocess the acquired EEG data. The preprocessing steps include re-referencing, downsampling, and bandpass filtering, which aim to improve the quality of EEG data and reduce noise interference. The specific calculation formula is:

[0090] A1: Re-reference: By reselecting the reference electrode, the noise caused by improper selection of the reference electrode position is reduced, and the EEG data is re-referenced. The formula is:

[0091]

[0092] Where N represents the total number of electrodes, EEG i (t) represents the signal of the i-th electrode;

[0093] A2: Downsampling reduces the amount of calculation and data storage requirements by lowering the sampling rate, increasing the effective information content of the signal, and downsampling the EEG data. The formula is:

[0094] x′[n]=x[Mn]

[0095] Where x′[n] is the downsampled sequence, x[Mn] is one sample taken for every M samples, and M is the downsampling factor;

[0096] A3: Bandpass filtering can effectively remove irrelevant frequency components and retain the EEG frequency bands related to attention level. The Butterworth bandpass filter is applied to EEG data. The formula is:

[0097]

[0098] Among them, ω l and ω h are the normalized high and low frequency cutoff frequencies, respectively, and Q is the quality factor.

[0099] S3: Align the EOG data and the preprocessed EEG data according to the timestamp, select 4s as the time window, segment the continuous EOG data, and calculate the PERCLOS value in each time window as a quantitative indicator of attention level;

[0100] The quantitative index of attention level is calculated as:

[0101] B1: Calculate the PERCLOS value in each time window. The formula is:

[0102]

[0103] Among them, CLOS represents the eye closure time, blink represents the blink time, fixation represents the fixation time, and saccade represents the saccade time;

[0104] B2: Get attention level quantitative labels Labels = {PERCLOS1, PERCLOS2, ..., PERCLOS n}.

[0105] S4: Construct a dual-domain deep supervision feature fusion network, such as Figure 2 The dual-domain deep supervision feature fusion network shown includes a data processing module, a time attention module, a spectrum residual module, a deep supervision module and a feature fusion module; refer to Figure 2 Execute steps S5 to S10;

[0106] S5: Processing the EEG signal time series data and the EEG signal spectrum data respectively through the data processing module to obtain the time series data and the spectrum data;

[0107] In the present invention, the EEG signal time series data is processed by co-spatial pattern analysis, and the EEG signal spectrum data is processed by multi-band DE feature extraction and 2D topological map mapping technology. The specific processing flow is as follows:

[0108] C1: Analyzing EEG time series data {X1,X2,...,X N}, calculate the covariance matrix for each sample:

[0109]

[0110] Compute the correlation covariance matrix:

[0111]

[0112] Among them, y i is the target variable value associated with the i-th sample;

[0113] Find the projection matrix that maximizes the objective function:

[0114]

[0115] Where R is the average of all sample covariance matrices;

[0116] Generate regression task features:

[0117] Z i =W T X i ;

[0118] C2: Decompose the EEG signal into five bands, namely delta band, theta band, alpha band, beta band and gamma band;

[0119] C3: For each frequency band, the time domain signal is converted to a frequency domain signal by fast Fourier transform (FFT):

[0120]

[0121] Where n is the time index and N is the number of samples;

[0122] Calculate the power spectral density PSD:

[0123]

[0124] Calculate the DE value using the normalized PSD:

[0125] DE=-∑ f∈band p[f]log(p[f])

[0126] Where f is the band.

[0127] S6: According to the PERCLOS value obtained in step S3 and the time series data and spectrum data obtained in step S5, a time series data set and a spectrum data set are constructed. Each piece of time series data and spectrum data is aligned and labeled with the corresponding PERCLOS value to form the data-label pair required for machine learning. The PERCLOS value is used as a label and input into the regression network model to guide the learning and training of the model.

[0128] S7: For the acquired time series data set, the multi-head attention mechanism of the time attention module is used to capture the key information in the EEG signal time series data, and the main function loss is output. temporal and the time domain feature F temporal ;

[0129] The temporal attention module consists of TransformerEncoder, convolutional layer, average pooling layer and fully connected layer. The specific processing flow is as follows:

[0130] D1: Capturing EEG time series data features through TransformerEncoder:

[0131] EncoderOutput=TransformerEncoder(EEG temporal )

[0132] D2: Temporal attention module outputs temporal features F temporal :

[0133]

[0134] Where W is the weight of the fully connected layer, and b is the bias of the fully connected layer;

[0135] D3: Calculate the loss value of the temporal attention module output through the MAE loss function:

[0136]

[0137] Among them, y i is the true value, is the predicted value and N is the total number of samples.

[0138] S8: For the acquired spectrum data set, the spectrum residual module deeply mines the high-order attention-related information in the spectrum data by stacking multiple residual blocks, and outputs the main function loss loss spectral And the spectrum feature F spectral ;

[0139] The spectrum residual module consists of a convolutional layer, three multi-band fusion residual blocks and a pooling layer. The specific processing flow of the spectrum residual module is as follows:

[0140] E1: Capturing EEG spectrum data features through multi-band fusion residual blocks:

[0141] ResBlockOutput = σ(F(x)+x)

[0142] Among them, x is the input of the residual block, F(x) is the internal operation of the residual block, and σ is the activation function;

[0143] E2: Spectral residual module outputs spectral features F spectral :

[0144] F spectral =Avg(ResBlock(ResBlock(ResBlock(Conv(x)))));

[0145] E3: Calculate the loss value of the spectral residual module output through the mean absolute error loss function:

[0146]

[0147] Among them, y i is the true value, is the predicted value and N is the total number of samples.

[0148] S9: Input the output of the multi-head attention mechanism in step S7 and the output of the first residual block obtained in step A8 into the deep supervision module respectively, and generate auxiliary loss loss through two deep supervision sub-modules. aux1 and loss aux2 , combined with the main function loss loss temporal 、loss spectral , integrate all main loss values ​​and auxiliary loss values;

[0149] The two deep supervision submodules in the deep supervision module are the time domain deep supervision submodule and the frequency domain deep supervision submodule. The time domain deep supervision submodule consists of a convolutional layer and two fully connected layers, and the frequency domain deep supervision submodule consists of two convolutional layers, an average pooling layer, and a fully connected layer.

[0150] The processing flow of the deep supervision module is:

[0151] F1: Directly receive the Transformer Encoder output value EncoderOutput through the time domain deep supervision submodule and calculate the auxiliary loss loss aux1 :

[0152]

[0153] F2: Directly receive the first residual block output value ResBlockOutput through the frequency domain deep supervision submodule and calculate the auxiliary loss loss aux2 :

[0154]

[0155] F3: Integrate the main function loss value and the auxiliary loss value to get the total loss value L totalBy minimizing the total loss value, the model can improve the sensitivity to local details of EEG data while maintaining the understanding of global information, thereby improving the accuracy and robustness of attention level prediction. The total loss value L is calculated. total :

[0156] L total =loss temporal +loss spectral +λ1loss aux1 +λ2loss aux2

[0157] S10: Obtaining time domain features F temporal and frequency domain feature F spectral , the feature fusion module fuses the features through connection, and the fused features are expressed as; F = [F temporal ,F spectral ], predicting pilots’ attention level through a linear regression algorithm.

[0158] S11: Input the time series dataset and spectrum dataset into the DBPNet model, output the error with the true PERCLOS value, and evaluate the performance of the model using MSE, MAE, and CORR indicators.

[0159] In order to verify the effectiveness and effect of the method of the present invention, a comparative experiment is conducted between the method of the present invention and the existing attention monitoring method in this embodiment, as follows:

[0160] In this embodiment, the prediction performance of different models under different test conditions is evaluated. Table 1 lists the evaluation index values ​​of different models under different test conditions. Figure 3 The index values ​​of each model are intuitively displayed, where (a) is the MAE value of different models, (b) is the RMSE value of different models, and (c) is the CORR value of different models. In this embodiment, all models are repeatedly run five times under the same hyperparameter configuration to obtain the mean and standard deviation of each index. Compared with the optimal values ​​of the prediction performance of the classic EEGNet model and the improved EEGNet model, the MAE and RMSE average values ​​of the DSFNet model are reduced by 0.0482 and 0.0425 respectively, and the CORR average value is increased by 0.6154, which verifies that the DSFNet model provided by the present invention can effectively predict the attention level state under cross-subject conditions.

[0161] Table 1 Comparison of results of different models across subjects

[0162] method MAE RMSE CORR EEGNet 0.1380 0.1736 0.0261 SE_EEGNet 0.1356 0.1705 -0.0035 ECA_EEGNet 0.1344 0.1701 -0.0111 SK_EEGNet 0.1355 0.1703 0.0400 FC_SK_EEGNet 0.1336 0.1674 0.0279 DBPNet 0.0854 0.1249 0.6043

[0163] At the same time, this embodiment selects data from one subject to evaluate the prediction performance of different models under the within-subject condition. Table 2 lists the evaluation index values ​​of different models under the within-subject condition. Figure 4 The values ​​of various model indicators are intuitively displayed, where (a) is the MAE value of different models, (b) is the RMSE value of different models, and (c) is the CORR value of different models. The average MAE and RMSE values ​​of the DSFNet model are 0.0182 and 0.0183 lower than the optimal values ​​of other models, respectively, and the average CORR value is increased by 0.0676.

[0164] Table 2 Comparison of results of different models within subjects

[0165] method MAE RMSE CORR EEGNet 0.1257 0.1562 -0.0086 SE_EEGNet 0.1253 0.1557 -0.0127 ECA_EEGNet 0.1123 0.1430 -0.0145 SK_EEGNet 0.1046 0.1297 -0.0101 FC_SK_EEGNet 0.1029 0.1280 -0.0220 DBPNet 0.0847 0.1097 0.0590

Claims

1. A method for monitoring EEG attention level based on a dual-domain deep supervision feature fusion network, characterized in that: The steps include: S1: Through EEG and EOG equipment, the physiological reactions of pilots during mission execution are captured in real time, and EEG and EOG data are collected to form an attention level database; S2: preprocessing the acquired EEG data; S3: Align the EOG data and the preprocessed EEG data according to the timestamp, select the time window, segment the continuous EOG data, and calculate the PERCLOS value in each time window as a quantitative indicator of attention level; S4: Construct a dual-domain deep supervision feature fusion network, including a data processing module, a temporal attention module, a spectral residual module, a deep supervision module, and a feature fusion module; S5: Processing the EEG signal time series data and the EEG signal spectrum data respectively through the data processing module to obtain the time series data and the spectrum data; S6: constructing a time series data set and a spectrum data set according to the PERCLOS value obtained in step S3 and the time series data and spectrum data obtained in step S5; S7: For the acquired time series data set, the multi-head attention mechanism of the time attention module is used to capture the key information in the EEG signal time series data, and the main function loss is output. temporal and the time domain feature F temporal ; S8: For the acquired spectrum data set, the spectrum residual module deeply mines the high-order attention-related information in the spectrum data by stacking multiple residual blocks, and outputs the main function loss loss spectral And the spectrum feature F spectral ; S9: Input the output of the multi-head attention mechanism in step S7 and the output of the first residual block obtained in step A8 into the deep supervision module respectively, and generate auxiliary loss loss through two deep supervision sub-modules. aux1 and loss aux2 , combined with the main function loss loss temporal 、loss spectral , integrate all main loss values ​​and auxiliary loss values; S10: According to the time domain feature F temporal And the spectrum feature F spectral ,The pilot’s attention level is predicted by linear regression algorithm.

2. According to claim 1, a method for monitoring EEG attention level based on a dual-domain deep supervision feature fusion network is characterized in that: The preprocessing of the EEG data in step S2 includes re-referencing, downsampling, and band-pass filtering, specifically: A1: Re-reference the EEG data using the formula: Where N represents the total number of electrodes, EEG i (t) represents the signal of the i-th electrode; A2: Downsample the EEG data using the formula: x′[n]=x[Mn] Where x′[n] is the downsampled sequence, x[Mn] is one sample taken for every M samples, and M is the downsampling factor; A3: Perform Butterworth bandpass filtering on the EEG data. The formula is: Among them, ω l and ω h are the normalized high and low frequency cutoff frequencies, respectively, and Q is the quality factor.

3. According to claim 1, a method for monitoring EEG attention level based on a dual-domain deep supervision feature fusion network is characterized in that: The calculation method of the quantitative index of the attention level in step S3 is: B1: Calculate the PERCLOS value in each time window. The formula is: Among them, CLOS represents the eye closure time, blink represents the blink time, fixation represents the fixation time, and saccade represents the saccade time; B2: Get attention level quantitative labels Labels = {PERCLOS1, PERCLOS2, ..., PERCLOS n }.

4. According to claim 1, a method for monitoring EEG attention level based on a dual-domain deep supervision feature fusion network is characterized in that: In the data processing module of step S5, the EEG signal time series data is processed by co-spatial pattern analysis, and the EEG signal spectrum data is processed by multi-band differential entropy (DE) feature extraction and 2D topological map mapping technology. The specific processing flow is as follows: C1: Analyzing EEG time series data {X1,X2,...,X N }, calculate the covariance matrix for each sample: Compute the correlation covariance matrix: Among them, y i is the target variable value associated with the i-th sample; Find the projection matrix that maximizes the objective function: Where R is the average of all sample covariance matrices; Generate regression task features: Z i =W T X i ; C2: Decompose the EEG signal into five bands, namely delta band, theta band, alpha band, beta band and gamma band; C3: For each frequency band, the time domain signal is converted to the frequency domain signal by fast Fourier transform: Where n is the time index and N is the number of samples; Calculate Power Spectral Density (PSD): Calculate the DE value using the normalized PSD: DE=-Σ f∈band p[f]log(p[f]) Where f is the band.

5. According to claim 1, a method for monitoring EEG attention level based on a dual-domain deep supervision feature fusion network is characterized in that: The temporal attention module of step S7 is composed of a TransformerEncoder, a convolutional layer, an average pooling layer, and a fully connected layer. The specific processing flow is as follows: D1: Capturing EEG time series data features through TransformerEncoder: EncoderOutput=TransformerEncoder(EEG temporal ); D2: Temporal attention module outputs temporal features F temporal : Where W is the weight of the fully connected layer, and b is the bias of the fully connected layer; D3: The loss value of the temporal attention module output is calculated through the mean absolute error loss function: Among them, y i is the true value, is the predicted value and N is the total number of samples.

6. The method for monitoring EEG attention level based on dual-domain deep supervision feature fusion network according to claim 5 is characterized in that: The spectrum residual module of step S8 is composed of a convolution layer, three multi-band fusion residual blocks and a pooling layer. The specific processing flow of the spectrum residual module is as follows: E1: Capturing EEG spectrum data features through multi-band fusion residual blocks: ResBlockOutput = σ(F(x)+x) Among them, x is the input of the residual block, F(x) is the internal operation of the residual block, and σ is the activation function; E2: Spectral residual module outputs spectral features F spectral : F spectral =Avg(ResBlock(ResBlock(ResBlock(Conv(x))))); E3: Calculate the loss value of the spectral residual module output through the mean absolute error loss function: Among them, y i is the true value, is the predicted value and N is the total number of samples.

7. The method for monitoring EEG attention level based on dual-domain deep supervision feature fusion network according to claim 1 is characterized in that: The two deep supervision submodules in the deep supervision module of step S9 are respectively a time domain deep supervision submodule and a frequency domain deep supervision submodule. The time domain deep supervision submodule consists of a convolutional layer and two fully connected layers, and the frequency domain deep supervision submodule consists of two convolutional layers, an average pooling layer and a fully connected layer.

8. The method for monitoring EEG attention level based on dual-domain deep supervision feature fusion network according to claim 7 is characterized in that: The processing flow of the deep supervision module in step S9 is as follows: F1: Directly receive the TransformerEncoder output value EncoderOutput through the time domain deep supervision submodule and calculate the auxiliary loss loss aux1 : F2: Directly receive the first residual block output value ResBlockOutput through the frequency domain deep supervision submodule and calculate the auxiliary loss loss aux2 : F3: Integrate the main function loss value and the auxiliary loss value to get the total loss value: L total =loss temporal +loss spectral +λ1loss aux1 +λ2loss aux2 。 9. The method for monitoring EEG attention level based on dual-domain deep supervision feature fusion network according to claim 1, characterized in that: In step S10, the time domain feature F is obtained. temporal and frequency domain feature F spectral , the features are fused by connection, and the fused features are expressed as; F = [F temporal ,F spectral ].

Images