A fatigue driving recognition method based on multi-channel weighted multi-scale permutation entropy

Abstract

The present application relates to the field of intelligent transportation, and proposes a kind of fatigue driving recognition method based on multi-channel weighted multi-scale permutation entropy, uses multi-channel weighted multi-scale permutation entropy as electroencephalogram feature, give consideration to time and space information, solve the deficiency of existing research method when extracting electroencephalogram space-time characteristics.First, the multi-scale permutation entropy value of each channel is calculated, then the brain function network is constructed to the preprocessed electroencephalogram signal using the Pearson correlation coefficient, the brain function network obtained by this method can better reflect the spatial relationship and functional characteristics of electroencephalogram signal.Next, according to the close degree of connection between nodes of brain function network, the importance of node is sorted using PageRank algorithm, the weight of each channel is obtained, then the weight of each channel and multi-scale permutation entropy are combined, and multi-channel weighted multi-scale permutation entropy is obtained.Finally, based on the combination feature of multi-channel weighted multi-scale permutation entropy, electroencephalogram signal is recognized for fatigue driving.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation, and in particular to a fatigue driving recognition method based on multi-channel weighted multi-scale permutation entropy. Background Technology

[0002] Frequent traffic accidents in recent years have drawn widespread public attention. Studies have shown that fatigued driving has become a direct cause of major traffic accidents, resulting not only in significant casualties but also numerous negative social impacts. Therefore, researching the mechanisms and physiological characteristics of driver fatigue, and designing effective methods for identifying fatigued driving, is of paramount importance for preventing fatigued driving and improving driving safety. Current methods for detecting driver fatigue primarily rely on the detection of the driver's physiological and behavioral characteristics, with electroencephalogram (EEG) signals being considered an effective objective means of assessing mental state.

[0003] Identifying fatigued driving states based on electroencephalogram (EEG) signals has become a research hotspot in recent years. This involves feature optimization of EEG signal sample data, followed by modeling to identify fatigued driving states. To improve the recognition rate of fatigued driving states, feature optimization is necessary. In recent years, researchers both domestically and internationally have conducted extensive research on EEG-based feature optimization and fatigued driving identification, with some researchers successfully identifying fatigue states. Currently, commonly used methods for extracting EEG features mainly include wavelet transform, sample entropy, power spectral density, and brain network analysis methods. Some of these methods focus on the time-frequency domain features of EEG signals, while others study the spatial features of EEG signals.

[0004] However, EEG signals are complex physiological signals with spatially discrete and non-stationary characteristics, making feature selection difficult. Analyzing only from the time-frequency domain or spatial connectivity will result in the loss of some effective information, affecting the final recognition accuracy. Therefore, further research is needed to extract features that reflect the spatiotemporal information of EEG. Summary of the Invention

[0005] To address the above issues, this invention proposes a fatigue driving identification method based on multi-channel weighted multi-scale permutation entropy. This method primarily targets the non-stationary and spatially discrete characteristics of EEG signals, selecting features that represent both time-frequency and spatial information. This improves the recognition rate between fatigue and wakefulness states, providing a better approach for EEG-based research on fatigue driving.

[0006] The present invention discloses a fatigue driving recognition method based on multi-channel weighted multi-scale permutation entropy, the technical solution of which includes the following steps:

[0007] Step 1: Preprocess the raw EEG data, use Pearson correlation coefficient to calculate the correlation between channels, and construct a brain functional network;

[0008] Step 2: Based on the tightness of the connections between brain functional network nodes, the PageRank algorithm is used to rank the importance of the nodes and obtain the weight of each node, which is the weight of each channel. The multi-scale permutation entropy of the EEG signal of each channel is calculated, and then the multi-scale permutation entropy of each channel and the node weight are weighted and combined. The resulting multi-channel weighted multi-scale permutation entropy is used as a feature to construct a feature dataset as a feature dataset representing fatigue or wakefulness.

[0009] Step 3: Use the feature set or a subset thereof as the training dataset, select a classification algorithm for training, and obtain a classifier;

[0010] Step 4: Use a classification algorithm to process the unknown feature dataset to obtain the recognition results for the two states of fatigue and wakefulness.

[0011] In step 2, a feature reflecting driving state is obtained through a weighted combination operation of multi-scale permutation entropy and node weights: multi-channel weighted multi-scale permutation entropy. This feature takes into account both temporal and spatial information of EEG signals, making up for the shortcomings of studying only from a single perspective of time or space.

[0012] The key algorithm steps are as follows:

[0013] (1) Node weights Let the importance ranking vector of N nodes be R = (r1, r2, ..., r...). N ) T In a brain functional network, the correlation matrix between N nodes is:

[0014] in, k represents the total number of connections between all nodes. Let R represent the number of connections from the m-th node to the n-th node under the x-th interconnection. x If the result of the x-th iteration is given, then... R x =WR x-1 (2)

[0015] When calculating the node importance ranking vector R, it is assumed that all PR values ​​are... Right now The node importance ranking vector R = (r1, r2, ..., r) can be finally obtained through iterative calculation using formula (2). N ) T ;

[0016] (2) Multi-channel weighted multi-scale permutation entropy For a given time series {x(i), i = 1, 2, ..., n}, its phase space is reconstructed to obtain an m-dimensional matrix X. m ,

[0017] Where t is the delay time, m is the dimension of the embedding space, and N is the number of matrix components, N = n - (m - 1)t, X m The m elements in the i-th row are arranged in ascending order, i.e. The time series is reconstructed to obtain the corresponding sequence: U = (j1, j2, ..., j m The sequence has m! possible permutations. Assume U i The probability of occurrence is p j (U), then the value of PE of order m is

[0018] H(x, m) reflects the degree of regularity of the time series x(i). The smaller H(x, m), the more regular the time series x(i) is, and vice versa.

[0019] Multiscale permutation entropy (MPE) is based on permutation entropy and incorporates the concept of "multiscale," resulting in its expression.

[0020] The node importance weights calculated by the PageRank algorithm are denoted as PR = (pr1, pr2, ..., pr...). n The multi-scale arrangement entropy H(x, m, s) is calculated by formula (5). We denote the multi-scale arrangement entropy value of the i-th channel as mpe. i Then we obtain the MPE values ​​corresponding to the n channels, denoted as MPE = (mpe1, mpe2, ..., mpe...). n Then, the node importance weights PR and the multi-scale permutation entropy of n channels are weighted and combined to obtain the multi-channel weighted multi-scale permutation entropy: MC-WMPE i =mpe i ·pr i (6) Attached Figure Description

[0021] Figure 1 Flowchart for fatigue driving state recognition

[0022] Figure 2 Flowchart for EEG signal feature extraction Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the present invention clearer, the specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0024] Step 1: Preprocess the raw EEG data, use Pearson correlation coefficient to calculate the correlation between channels, and construct a brain functional network;

[0025] The Pearson correlation coefficient between two variables X and Y is defined as the quotient of their covariance and standard deviation.

[0026] Equation (7) defines the population correlation coefficient ρ. By estimating the sample covariance and standard deviation, the Pearson correlation coefficient r can be obtained:

[0027] Step 2: Based on the tightness of connections between nodes in the brain functional network, the PageRank algorithm is used to rank the importance of nodes, obtaining the weight of each node, which is the weight of each channel. Multi-scale permutation entropy is calculated for the EEG signal of each channel, and then the multi-scale permutation entropy of each channel and the node weight are weighted and combined to obtain the multi-channel weighted multi-scale permutation entropy as a feature, constructing a feature dataset as a feature dataset representing fatigue or wakefulness.

[0028] (1) Calculate node weights

[0029] Let the importance ranking vector of N nodes be R = (r1, r2, ..., r...). N ) T In a brain functional network, the correlation matrix between N nodes is:

[0030] in, k represents the total number of connections between all nodes. Let R represent the number of connections where the m-th node points to the n-th node under the x-th mutual connection. x If the result of the x-th iteration is given, then... R x =WR x-1 (10)

[0031] When calculating the node importance ranking vector R, it is assumed that all PR values ​​are... Right now The node importance ranking vector R = (r1, r2, ..., r) can be finally obtained through iterative calculation using formula (10). N ) T ;

[0032] (2) Calculate the entropy of multi-channel weighted multi-scale permutation.

[0033] For a given time series {x(i), i = 1, 2, ..., n}, its phase space is reconstructed to obtain an m-dimensional matrix X. m ,

[0034] Where t is the delay time, m is the dimension of the embedding space, and N is the number of matrix components, N = n - (m - 1)t;

[0035] X m The m elements in the i-th row are arranged in ascending order, i.e. The time series is reconstructed to obtain the corresponding sequence: U = (j1, j2, ..., j m The sequence has m! possible permutations. Assume U i The probability of occurrence is p j (U), then the value of PE of order m is

[0036] H(x, m) reflects the degree of regularity of the time series x(i). The smaller H(x, m), the more regular the time series x(i) is, and vice versa.

[0037] Multiscale permutation entropy (MPE) is based on permutation entropy and incorporates the concept of "multiscale," resulting in its expression.

[0038] The node importance weights calculated by the PageRank algorithm are denoted as PR = (pr1, pr2, ..., pr...). n The multi-scale arrangement entropy H(x, m, s) is calculated by formula (13). We denote the multi-scale arrangement entropy value of the i-th channel as mpe. i Then we obtain the MPE values ​​corresponding to the n channels, denoted as MPE = (mpe1, mpe2, ..., mpe...). n Then, the node importance weights PR and the multi-scale permutation entropy of n channels are weighted and combined to obtain the multi-channel weighted multi-scale permutation entropy. MC-WMPE i =mpe i ·pr i (14)

[0039] Step 3: Use the feature set or a subset thereof as the training dataset, select a classification algorithm for training, and obtain a classifier;

[0040] Step 4: Use a classification algorithm to process the unknown feature dataset to obtain the recognition results for the two states of fatigue and wakefulness.

[0041] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A fatigue driving recognition method based on multi-channel weighted multi-scale permutation entropy, characterized in that, Includes the following steps: Step 1: Preprocess the raw EEG data, use Pearson correlation coefficient to calculate the correlation between channels, and construct a brain functional network; Step 2: Based on the tightness of the connections between nodes in the brain functional network, the PageRank algorithm is used to rank the importance of the nodes and obtain the weight of each node, which is the weight of each channel. The multi-scale permutation entropy of the EEG signal of each channel is calculated, and then the multi-scale permutation entropy of each channel and the node weight are weighted and combined to obtain the multi-channel weighted multi-scale permutation entropy as a feature. This feature is used as the feature data representing fatigue or wakefulness, and a feature dataset is constructed. Step 3: Use the feature dataset or a subset thereof as the training dataset, select a classification algorithm for training, and obtain a classifier; Step 4: Use a classification algorithm to process the unknown feature dataset to obtain the recognition results of the two states: fatigue and wakefulness; Step 2 involves obtaining a feature reflecting driving status—multi-channel weighted multi-scale permutation entropy—through a weighted combination operation of multi-scale permutation entropy and node weights. This feature simultaneously considers the temporal and spatial information of EEG signals, overcoming the shortcomings of studying only from a single temporal or spatial perspective. The specific algorithm process is as follows: (1) Node weights Let the importance ranking vector of N nodes be R = (r1, r2, ..., r...). N ) T In a brain functional network, the association matrix between N nodes is as follows: in k represents the total number of connections between all nodes. Let R represent the number of connections from the m-th node to the n-th node under the x-th interconnection. x If the result of the x-th iteration is given, then: R x =WR x-1 (2) When calculating the node importance ranking vector R, it is assumed that all PR values ​​are... Right now The node importance ranking vector R = (r1, r2, ..., r) can be finally obtained through iterative calculation using formula (2). N ) T ; (2) Multi-channel weighted multi-scale permutation entropy For a given time series {x(i), i = 1, 2, ..., n}, its phase space is reconstructed to obtain an m-dimensional matrix X. m , Where t is the delay time, m is the dimension of the embedding space, and N is the number of matrix components, N = n - (m - 1)t, X m The m elements in the i-th row are arranged in ascending order, i.e. The time series is reconstructed to obtain the corresponding sequence: U = (j1, j2, ..., j m The sequence has m! possible permutations. Assume U i The probability of occurrence is p j (U), then the value of PE of order m is: H(x, m) reflects the degree of regularity of the time series x(i). The smaller H(x, m), the more regular the time series x(i) is, and vice versa. Multiscale permutation entropy (MPE) is based on permutation entropy and incorporates the concept of "multiscale," resulting in the following expression: The node importance weights calculated by the PageRank algorithm are denoted as PR = (pr1, pr2, ..., pr...). n The multi-scale arrangement entropy H(x, m, s) is calculated by formula (5), and the multi-scale arrangement entropy value of the i-th channel is denoted as mpe. i Then we obtain the MPE values ​​corresponding to the n channels, denoted as MPE = (mpe1, mpe2, ..., mpe...). n Then, the node importance weights PR and the multi-scale permutation entropy of n channels are weighted and combined to obtain the multi-channel weighted multi-scale permutation entropy: MC-WMPE i =mpe i ·pr i (6) 。

Images