Method for identifying fatigue state through EEG signal

Abstract

The invention discloses a method for identifying a fatigue state through an EEG signal, and the method comprises the steps: collecting the EEG signal, and extracting a plurality of feature values from the EEG signal, and enabling the feature values to serve as a construction sample; a shallow neural network is constructed, each characteristic value of the sample is operated by an input layer, a hidden layer and an output layer of the shallow neural network, and a prediction result is output by the output layer; and carrying out normalization processing on a prediction result, wherein a normalization processing result is a result for judging whether fatigue exists or not. According to the method for identifying the fatigue state through the EEG signal, a plurality of feature values are comprehensively considered, and the interaction and complementarity among the feature values are utilized to construct a more comprehensive and fine fatigue evaluation model. The model not only can more accurately reflect fatigue characteristics in the EEG signal, but also can better adapt to differences among different individuals, so that the practicability and effectiveness of fatigue monitoring are improved.

Description

Technical Field

[0001] The present invention relates to the field of brain wave analysis and brain-computer interface technology, and in particular to a method for identifying fatigue status through EEG signals. Background Art

[0002] Using brain waves to study fatigue is a cutting-edge and important technology. EEG is a record of potential changes caused by the electrical activity of brain nerve cells, which reflects the functional state of the brain. In recent years, with the continuous development of neuroscience and electronic technology, brain wave analysis technology has gradually been applied to many fields, including the assessment of fatigue status.

[0003] Fatigue is a physiological and psychological state that occurs after long hours of work, study or physical activity, manifested as inattention, slow reaction, depression, etc. For certain groups of people, such as athletes, drivers, night shift workers, etc., fatigue may lead to serious consequences, such as sports injuries, traffic accidents, etc. Therefore, accurate and real-time assessment of fatigue status is of great significance for preventing related risks and improving work efficiency.

[0004] Brainwave analysis technology provides a new means for studying fatigue states. Different brainwave frequency components are closely related to different physiological and psychological states. For example, alpha waves are usually associated with relaxation and rest, while beta waves are associated with active and excited states. When the human body is in a state of fatigue, the frequency and amplitude of brainwaves tend to undergo specific changes. Therefore, by analyzing these changes, the fatigue state of the human body can be inferred.

[0005] The external manifestation of EEG is the voltage difference between any electrodes, that is, EEG is a time series signal. The time domain scale feature can directly reflect the information contained in the time series EEG signal. In the study of fatigue EEG, it is very important to accurately use the eigenvalue to judge the fatigue state. At present, the eigenvalues ​​of EEG fatigue signals are generally: mean, variance, standard deviation, amplitude, Hjorth parameter (three time domain feature sets describing EEG single channel), power spectrum entropy, Shannon entropy, etc. However, using only a single eigenvalue as a criterion for judging fatigue is too one-sided and limited, because EEG signals (EEG) are a complex bioelectric activity, which contains multi-level and multi-dimensional information. A single eigenvalue can often only capture the characteristics of one aspect of the signal, while ignoring other information that may be equally important or complementary. For example, although the mean can reflect the overall level of the signal, it is difficult to reflect the volatility and dynamic changes of the signal; although the variance and standard deviation can measure the discreteness of the signal, they may not be able to accurately capture subtle differences when facing complex and changeable fatigue states. In addition, different subjects are affected by various factors such as individual differences or skin contact resistance. If a constant threshold is used to distinguish them, the accuracy of the judgment results is difficult to guarantee. Summary of the invention

[0006] Purpose of the invention: The purpose of the present invention is to address the deficiencies of the prior art and provide a method for identifying fatigue status through EEG signals, comprehensively considering multiple eigenvalues ​​of EEG signals and utilizing the interaction and complementarity between them to construct a more comprehensive and sophisticated fatigue assessment model, thereby improving the accuracy and reliability of fatigue status judgment.

[0007] In order to achieve the above object, the present invention provides a method for identifying fatigue status through EEG signals, comprising the following steps:

[0008] S1, collect EEG signals and extract multiple feature values ​​from them for constructing samples;

[0009] S2, construct a shallow neural network, each characteristic value of the sample undergoes the operation of the input layer, hidden layer and output layer of the shallow neural network, and the output layer outputs the prediction result;

[0010] S3. Perform normalization processing on the prediction results, and the normalization processing result is the determination result of whether fatigue occurs.

[0011] A further preferred technical solution of the present invention is that the step S1 of collecting EEG signals and extracting a plurality of characteristic values ​​therefrom for constructing samples includes:

[0012] S11, collecting EEG signals;

[0013] S12, calculating Hjorth parameter, power spectrum entropy and Shannon entropy according to the collected EEG signal;

[0014] S13. Arrange the Hjorth parameter, power spectrum entropy, Shannon entropy and EEG signal value in order to form a sample.

[0015] Preferably, in step S11, a 10-20 international standard lead system is used to collect EEG signals.

[0016] Preferably, in step S12, the method for calculating the Hjorth parameter according to the collected EEG signal is:

[0017] Hjorth parameters include three main parameters: activity A, mobility M and complexity C;

[0018] The calculation formula of activity A is:

[0019]

[0020] The calculation formula of mobility M is:

[0021]

[0022] The calculation formula of complexity C is:

[0023]

[0024] In each formula, N is the number of sampling points, s(n) is the sampling value, μ , Represents the signal average value, s 7 (n) represents the first-order derivative of s(n), and var represents the variance.

[0025] Preferably, in step S12, the method for calculating the power spectrum entropy according to the collected EEG signal is:

[0026] For an uncertain system, the random variable X(n) is used to represent the characteristics of its state. The value of X(n) is {X(n)|n=1,2,…,N}(N≥1); {P(n)|n=1,2,…,N} is the probability of taking the value corresponding to each state characteristic of the system, and 0≤P(n)≤1,(n=1,2,…,N), and

[0027] Assume that the discrete Fourier transform of a signal is X(ω n ), angular frequency f s is the sampling frequency, N is the number of sampling points, n=1,2,…,N; power spectral density Will Normalized according to the total spectral power, the power spectral density distribution function is obtained The power spectrum entropy calculation formula is:

[0028]

[0029] Preferably, in step S12, the method for calculating the Shannon entropy according to the collected EEG signal is:

[0030] The Shannon entropy calculation formula is:

[0031]

[0032] Among them, p(x U ) is the probability of different information appearing.

[0033] Preferably, the shallow neural network constructed in step S2 includes an input layer, a hidden layer and an output layer; the n eigenvalues ​​X of the sample n After the operation of the input layer, hidden layer and output layer, the output layer outputs the prediction result;

[0034] The mathematical model of the shallow neural network is calculated as follows:

[0035] Z [ / ] =W [ / ] X [ / ] +b [ / ] ;

[0036] Among them, Z is the output result of the neuron, n is the number of neural network layers, W and b are the weight information of this neuron, X [ / ] The input feature value of this neuron;

[0037] It is the output result after Z passes through the activation function, expressed as:

[0038]

[0039] Represents the activation function, which is selected according to the network structure.

[0040] Preferably, in step S3, the prediction result is normalized, and the normalization algorithm calculation formula is:

[0041]

[0042] The output layer outputs the prediction result Z [ / ] ≥0.55 is considered as fatigue state, Z [ / ] <0.55 was considered as non-fatigue state.

[0043] Beneficial effects: The method for identifying fatigue status through EEG signals provided by the present invention comprehensively considers multiple characteristic values ​​and utilizes the interaction and complementarity between them to construct a more comprehensive and sophisticated fatigue assessment model. Such a model can not only more accurately reflect the fatigue characteristics in EEG signals, but also better adapt to the differences between different individuals, thereby improving the practicality and effectiveness of fatigue monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flow chart of the method for identifying fatigue state through EEG signal of the present invention;

[0045] Figure 2 It is a structural diagram of the shallow neural network of the present invention;

[0046] Figure 3 Comparison of Hjorth activity parameters in different states of subjects 1-4;

[0047] Figure 4 Comparison of Hjorth mobility parameters in different states for subjects 1-4;

[0048] Figure 5 Comparison of Hjorth complexity parameters in different states for subjects 1-4;

[0049] Figure 6 Comparison of power spectrum entropy parameters in different states for subjects 1-4;

[0050] Figure 7 Comparison of Shannon entropy parameters under different conditions for subjects 1-4;

[0051] Figure 8 Comparison of fatigue state classification accuracy after neural network. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0053] Combine the following Figure 1-Figure 8 The present invention describes a method for identifying fatigue status through EEG signals.

[0054] Embodiment 1: This embodiment provides a method for identifying fatigue status through EEG signals, such as Figure 1 As shown, the following steps are included:

[0055] S1, collect EEG signals and extract multiple feature values ​​from them for constructing samples;

[0056] S2, construct a shallow neural network, each characteristic value of the sample undergoes the operation of the input layer, hidden layer and output layer of the shallow neural network, and the output layer outputs the prediction result;

[0057] S3. Perform normalization processing on the prediction results, and the normalization processing result is the determination result of whether fatigue occurs.

[0058] In terms of specific steps, step S1 collects EEG signals and extracts multiple feature values ​​from them for constructing samples, including:

[0059] S11, collecting EEG signals;

[0060] In step S11, the 10-20 international standard lead system is used to collect EEG signals. That is, the 32 electrodes of the SynAmps2 system are arranged according to the international 10-20 system for collecting EEG data, of which: 30 electrodes record EEG signals and 2 electrodes are reference electrodes. The 30 recording electrodes are FP1, FP2, F7, F3, Fz, F4, F8, FT7, FC3, FCz, FC4, FT8, T3, C3, Cz, C4, T4, TP7, CP3, CPz, CP4, TP8, T5, P3, Pz, P4, T6, O1, Oz and O2, and the 2 reference electrodes are A1 and A2. The data sampling rate is 500Hz.

[0061] S12, calculating Hjorth parameter, power spectrum entropy and Shannon entropy according to the collected EEG signal;

[0062] The method to calculate the Hjorth parameter is:

[0063] Hjorth parameters include three main parameters: activity A, mobility M and complexity C;

[0064] The calculation formula of activity A is:

[0065]

[0066] The calculation formula of mobility M is:

[0067]

[0068] The calculation formula of complexity C is:

[0069]

[0070] In each formula, N is the number of sampling points, s(n) is the sampling value, μ , Represents the signal average value, s 7 (n) represents the first-order derivative of s(n), and var represents the variance.

[0071] The method for calculating power spectrum entropy is:

[0072] For an uncertain system, the random variable X(n) is used to represent the characteristics of its state. The value of X(n) is {X(n)|n=1,2,…,N}(N≥1); {P(n)|n=1,2,…,N} is the probability of taking the value corresponding to each state characteristic of the system, and 0≤P(n)≤1,(n=1,2,…,N), and

[0073] Assume that the discrete Fourier transform of a signal is X(ω n ), angular frequency f s is the sampling frequency, N is the number of sampling points, n=1,2,…,N; power spectral density Will Normalized according to the total spectral power, the power spectral density distribution function is obtained The power spectrum entropy calculation formula is:

[0074]

[0075] The method to calculate Shannon entropy is:

[0076] The Shannon entropy calculation formula is:

[0077]

[0078] Among them, p(x U ) is the probability of different information appearing.

[0079] S13, arranging the Hjorth parameter, power spectrum entropy, Shannon entropy and EEG signal value in order to form a sample. The sample composition is shown in Table 1.

[0080] Table 1 Sample structure composition

[0081] 0 1 2 3 4 5 …… 105 A M C <![CDATA[H M ]]> S s(0) …… s(100)

[0082] Furthermore, the shallow neural network constructed in step S2 is as follows Figure 2 As shown, it includes input layer, hidden layer and output layer; the n eigenvalues ​​X of the sample n After the operation of the input layer, hidden layer and output layer, the output layer outputs the prediction result;

[0083] The mathematical model of the shallow neural network is calculated as follows:

[0084] Z [ / ] =W [ / ] X [ / ] +b [ / ] ;

[0085] Among them, Z is the output result of the neuron, n is the number of neural network layers, W and b are the weight information of this neuron, X [ / ] The input feature value of this neuron;

[0086] It is the output result after Z passes through the activation function, expressed as:

[0087]

[0088] Represents the activation function, which is selected according to the network structure.

[0089] In neural network training, the choice of r is crucial. A higher learning rate can speed up the training, but it may cause instability in the training process, causing the parameters to oscillate near the optimal solution. Although a lower learning rate is stable, it will significantly extend the training time. In order to balance training speed and stability, this embodiment adopts a strategy of dynamically adjusting the learning rate. That is, a larger learning rate is used in the early stage of training to quickly approach the optimal solution, and then as the training proceeds, the learning rate is gradually reduced to make the training process more stable and accurate when approaching the optimal solution. This method helps to improve the training effect of the model while ensuring training efficiency.

[0090] Therefore, as a preferred method, the learning rate r is set to a dynamic learning rate, and the calculation formula of the learning rate r is:

[0091]

[0092] Among them, Num is the current training number of the neural network, and r0 is the initial learning rate.

[0093] In view of the individual differences and time variability of bioelectric signals, in order to enhance the generalization ability of the model, multiple samples need to be collected across individuals. However, the high collection cost limits the sample size, so regularization technology is particularly important. In this embodiment, the HuberLoss regularization algorithm is used to add the cost function. The cost function calculation formula after adding regularization is:

[0094]

[0095] Furthermore, the sample consisting of 106 values ​​is sent into the neural network to obtain the calculation result. In step S3, the prediction result is normalized. The calculation formula of the normalization algorithm is:

[0096]

[0097] The output layer outputs the prediction result Z [ / ] ≥0.55 is considered as fatigue state, Z [ / ] <0.55 was considered as non-fatigue state.

[0098] In order to verify the accuracy and reliability of the recognition results of the present invention, a verification experiment was carried out.

[0099] The experimental data comes from a public simulated driving data set, which records the EEG signals of 32 male subjects in four different age groups in different driving states in a simulated driving environment. Two subjects were randomly selected from each age group, and 8 people formed a group, for a total of four groups. The activity parameters, mobility parameters, complexity parameters, power spectrum entropy parameters and Shannon entropy parameters extracted from the EEG signals of one subject randomly selected from the age group of 30-35 in each group were compared. Figure 3-Figure 7 shown.

[0100] The three eigenvalues ​​of Hjorth parameter, power spectrum entropy and Shannon entropy are used to perform feature analysis on the data separately, and the classification accuracy is statistically analyzed; then the recognition method proposed by the present invention is used for feature analysis, and the accuracy obtained after the fusion of the eigenvalues ​​is statistically analyzed. The statistical results are shown in Table 2:

[0101] Table 2 Accuracy of fatigue state judgment using three eigenvalues ​​of Hjorth parameter, power spectrum entropy, Shannon entropy and fused eigenvalues

[0102] Hjorth parameter Power Spectral Entropy Shannon Entropy After fusion Group 1 73.36％ 69.95％ 83.48％ 93.33％ Group 2 74.53％ 72.36％ 83.36％ 90.79％ Group 3 78.84％ 68.98％ 82.94％ 96.83％ Group 4 75.68％ 69.37％ 84.68％ 92.28％ average 75.60％ 70.17％ 83.62％ 93.31％

[0103] The comparison results are as follows Figure 8 As shown, the results show that the accuracy of fatigue state identification by the present invention is greatly improved compared with the accuracy of fatigue state judgment by single eigenvalue analysis.

[0104] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the present invention itself. Various changes in form and details may be made without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A method for identifying fatigue status through EEG signals, characterized in that: The steps include: S1, collect EEG signals and extract multiple feature values ​​from them for constructing samples; S2, construct a shallow neural network, each characteristic value of the sample undergoes the operation of the input layer, hidden layer and output layer of the shallow neural network, and the output layer outputs the prediction result; S3. Perform normalization processing on the prediction results, and the normalization processing result is the determination result of whether fatigue occurs.

2. The method for identifying fatigue status through EEG signals according to claim 1, characterized in that: Step S1 collects EEG signals and extracts multiple feature values ​​from them for constructing samples, including: S11, collecting EEG signals; S12, calculating Hjorth parameter, power spectrum entropy and Shannon entropy according to the collected EEG signal; S13. Arrange the Hjorth parameter, power spectrum entropy, Shannon entropy and EEG signal value in order to form a sample.

3. The method for identifying fatigue status through EEG signals according to claim 2, characterized in that: In step S11, the EEG signal is collected using the 10-20 international standard lead system.

4. The method for identifying fatigue status through EEG signals according to claim 2, characterized in that: In step S12, the method for calculating the Hjorth parameter according to the collected EEG signal is: Hjorth parameters include three main parameters: activity A, mobility M and complexity C; The calculation formula of activity A is: The calculation formula of mobility M is: The calculation formula of complexity C is: In each formula, N is the number of sampling points, s(n) is the sampling value, μ , Represents the signal average value, s 7 (n) represents the first-order derivative of s(n), and var represents the variance.

5. The method for identifying fatigue status through EEG signals according to claim 2, characterized in that: In step S12, the method for calculating the power spectrum entropy according to the collected EEG signal is: For an uncertain system, the random variable X(n) is used to represent the characteristics of its state. The value of X(n) is {X(n)|n=1,2,…,N}(N≥1); {P(n)|n=1,2,…,N} is the probability of taking the value corresponding to each state characteristic of the system, and 0≤P(n)≤1,(n=1,2,…,N), and Assume that the discrete Fourier transform of a signal is X(ω / ), angular frequency f , is the sampling frequency, N is the number of sampling points, n=1,2,…,N; power spectral density Will Normalized according to the total spectral power, the power spectral density distribution function is obtained The power spectrum entropy calculation formula is:

6. The method for identifying fatigue status through EEG signals according to claim 2, characterized in that: In step S12, the method for calculating the Shannon entropy according to the collected EEG signal is: The Shannon entropy calculation formula is: Among them, p(x U ) is the probability of different information appearing.

7. The method for identifying fatigue status through EEG signals according to claim 1, characterized in that: The shallow neural network constructed in step S2 includes an input layer, a hidden layer and an output layer; the n eigenvalues ​​X of the sample [ / ] After the operation of the input layer, hidden layer and output layer, the output layer outputs the prediction result; The mathematical model of the shallow neural network is calculated as follows: Z [ / ] =W [ / ] X [ / ] +b [ / ] ; Among them, Z is the output result of the neuron, n is the number of neural network layers, W and b are the weight information of this neuron, X [ / ] The input feature value of this neuron; It is the output result after Z passes through the activation function, expressed as: Represents the activation function, which is selected according to the network structure.

8. The method for identifying fatigue status through EEG signals according to claim 7, characterized in that: In step S3, the prediction results are normalized, and the normalization algorithm calculation formula is: The output layer outputs the prediction result Z [ / ] ≥0.55 is considered as fatigue state, Z [ / ] <0.55 was considered as non-fatigue state.

Images