Driving fatigue state classification and evaluation method based on leg semg

Abstract

The present invention belongs to the technical field of fatigue classification. Disclosed are a driving fatigue state classification and evaluation method based on a leg sEMG. The method comprises: analyzing a driving fatigue cause; on the basis of the driving fatigue cause, collecting a leg surface electromyogram signal, and establishing a leg surface electromyogram signal data set; segmenting and processing the leg surface electromyogram signal data set; and using an improved SVM classification model to perform classification prediction on a driving fatigue state. In the present invention, by using the method, a leg surface electromyogram signal is collected, processed and analyzed, an SVM classification model is constructed, and a whale optimization algorithm is used to optimize the SVM classification model, thereby improving the effect of performing classification on the basis of leg surface electromyogram signals, and thus improving the classification precision.

Description

A classification and evaluation method for driving fatigue status based on leg sEMG Technical Field

[0001] The present invention relates to the technical field of fatigue classification, and in particular to a method for classifying and evaluating driving fatigue status based on leg sEMG. Background Art

[0002] Traffic accidents are a significant cause of casualties and property damage, and driver fatigue is a primary contributing factor. Long driving hours, traffic congestion, and complex road conditions can lead to increased leg muscle fatigue, slowing drivers' reaction times and increasing the risk of accidents.

[0003] In recent years, domestic and foreign scholars have carried out a lot of research on the problem of driving fatigue, mainly analyzing the problem of driving fatigue through subjective fatigue detection methods and objective fatigue detection methods. Objective fatigue detection methods include vehicle driving status detection methods, driver behavior characteristics detection methods and driver physiological characteristics detection methods.

[0004] Through the analysis of the above methods, the shortcomings of the research are reflected in the following aspects:

[0005] (1) The results of subjective testing methods for driving fatigue are affected by factors such as individual differences, psychological state, and social expectations of drivers, resulting in subjective bias and inaccuracy. Compared with objective testing methods, subjective testing methods are less efficient and require drivers to conduct self-assessments before, during, or after driving, which greatly affects the driver's attention and reaction time, increasing driving risks.

[0006] (2) The accuracy of vehicle driving state detection methods is affected by factors such as vehicle type, road conditions, traffic conditions, and weather conditions, and may result in false positives or false negatives. In addition, vehicle driving state detection methods have poor adaptability and cannot be applied to different driving scenarios and driving habits.

[0007] (3) Driver behavior feature detection methods are also affected by factors such as individual differences, emotional changes, and environmental interference, and may result in misidentification or missed recognition. Furthermore, driver behavior feature detection methods are complex and require the collection, processing, analysis, and recognition of a large amount of behavioral feature data, consuming significant computing resources and time, which reduces the real-time performance and efficiency of the system.

[0008] (4) Among the detection methods based on the driver's physiological characteristics, although EEG signals are the gold standard for detecting physiological fatigue and can directly reflect the driver's fatigue level from EEG signals, with a strong correlation, EEG wearable devices greatly affect the driver's normal driving operation; although ECG signals can clearly reflect the changes in the driver's stress level when detecting driver fatigue, the difference between mild fatigue and moderate fatigue is small, and only when the driver is severely fatigued will there be a significant change. Electromyographic signals can meet the requirements of real-time and continuous detection of the driver's fatigue state. Electromyographic signal acquisition equipment is non-invasive and does not affect the driver's normal driving operation. Sensors can also be placed in the car cabin, which is more practical. Summary of the Invention

[0009] The purpose of the present invention is to provide a driving fatigue status classification and evaluation method based on leg sEMG. By processing and analyzing the surface electromyography signals of the legs, an SVM classification model is constructed, and the SVM classification model is optimized using the whale optimization algorithm, thereby improving the classification effect based on the surface electromyography signals of the legs and improving the classification accuracy.

[0010] To achieve the above objectives, the present invention provides a method for classifying and evaluating driving fatigue status based on leg sEMG, comprising the following steps:

[0011] S1. Construct a human-machine-environment system model based on the graphical model method, analyze the human-machine-environment system model, and determine the causes of driving fatigue;

[0012] S2. Based on the analyzed driving fatigue causes, driving fatigue data is collected, including leg surface electromyography signals, self-assessed fatigue status, and fatigue status label files, to establish a leg surface electromyography signal dataset.

[0013] S3, cutting the data in the leg surface electromyography signal dataset to obtain the starting points of mild fatigue, moderate fatigue, and severe fatigue;

[0014] S4. Remove data noise from the leg surface electromyography signal dataset and extract time-frequency domain features to analyze the fatigue state of the leg surface electromyography signal;

[0015] S5. Use the improved SVM classification model to classify and predict driving fatigue status.

[0016] Preferably, the step S2 of collecting driving fatigue data includes: collecting the surface electromyography signal of the right leg, continuously questioning the driver during the collection process, recording the driver's fatigue status, and marking the fatigue status in the electromyography signal data to obtain the surface electromyography signal of the right leg, the self-evaluated fatigue status and the fatigue status label file.

[0017] Preferably, in step S3, the fatigue state is segmented according to the corresponding relationship between the original right leg surface electromyographic signal and the fatigue state annotated data at the time, and the starting points of mild fatigue, moderate fatigue and severe fatigue are obtained.

[0018] Preferably, in step S4, a bandpass filter is used to initially remove noise from the original right leg surface electromyography signal, and then a VMD algorithm is used to perform variational mode decomposition to remove interference noise from the original right leg surface electromyography signal.

[0019] Preferably, the fatigue state result of the leg surface electromyography signal analyzed in the time-frequency domain in step S4 is consistent with the self-assessed fatigue state.

[0020] Preferably, the time domain features extracted in step S4 include maximum value, minimum value, root mean square value, amplitude, mean, number of sign changes, variance, kurtosis, skewness, sample entropy and absolute mean; the frequency domain features extracted include median frequency and average power frequency.

[0021] Preferably, the steps in step S5 include:

[0022] S51: Constructing an SVM classification model to classify driving fatigue status, the steps including:

[0023] Suppose the feature vector after training sample mapping is , the partition hyperplane formula in the feature space is:

[0024] ;

[0025] Where, represents the normal vector, Represents the displacement term and introduces the penalty factor With slack variables , the formula is:

[0026] ;

[0027] The dual function is:

[0028] ;

[0029] Where, represents the Lagrange multiplier;

[0030] Define the kernel function:

[0031] ;

[0032] Where, represents the feature vector, Indicates the corresponding label, the value is +1 or -1,

[0033] Bringing the kernel function into the dual function yields:

[0034] ;

[0035] After solving, we get the optimal classification function:

[0036] ;

[0037] The optimal classification function is transformed to obtain the optimal classification function of one-to-many SVM, the formula is:

[0038] ;

[0039] S52. Use the whale optimization algorithm to optimize the penalty parameters and kernel functions in the SVM model to enhance the classification and prediction capabilities of the SVM model.

[0040] Therefore, the present invention adopts the above-mentioned driving fatigue state classification evaluation method based on leg sEMG, which has the following beneficial effects:

[0041] The driver fatigue degree is judged based on leg sEMG, which effectively improves the classification accuracy and prediction effect of driving fatigue;

[0042] By processing and analyzing the surface electromyography signals of the legs, an SVM classification model was constructed, and the SVM classification model was optimized using the whale optimization algorithm, which improved the classification effect based on the surface electromyography signals of the legs and improved the classification accuracy.

[0043] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] FIG1 is a flow chart of a method according to an embodiment of the present invention;

[0045] FIG2 is a diagram of a human-machine-environment system model according to an embodiment of the present invention;

[0046] FIG3 is a sEMG data set according to an embodiment of the present invention;

[0047] FIG4 is a variational mode decomposition diagram of the VMD algorithm according to an embodiment of the present invention;

[0048] FIG5 is a time-frequency domain analysis diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0049] Example

[0050] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0051] 1 , a method for classifying and evaluating driving fatigue status based on leg sEMG includes the following steps:

[0052] S1. Based on the graphical model method, a human-machine-environment system model was constructed and analyzed, as shown in Figure 2. It was found that leg muscle fatigue is easily caused by long-term driving, road congestion, and complex road conditions.

[0053] S2. Collect driving fatigue data based on the analyzed causes of driving fatigue. Because conducting driving fatigue experiments on real roads is dangerous and repeatable, data validity cannot be guaranteed. Therefore, a driving simulator is used to conduct driving fatigue experiments. This simulates driving scenarios on congested roads and allows for in-cockpit experiments to collect experimental data.

[0054] Surface electromyography (sEMG) signals were collected from 10 participants while driving in congested traffic, with their leg muscles at varying levels of fatigue. Specifically, signals from the tibialis anterior muscle in the right leg were collected over a 2-3 hour period. During the collection process, the drivers were continuously questioned, their fatigue status recorded, and their fatigue status annotated within the EMG data. This yielded the right leg surface electromyography signals, self-assessed fatigue status, and fatigue status label files, creating the sEMG dataset, as shown in Figure 3.

[0055] S3. Segment the fatigue state according to the corresponding relationship between the original sEMG data and the fatigue state annotated data at different times, and obtain the starting points of mild fatigue, moderate fatigue, and severe fatigue for each experimenter.

[0056] S4. A bandpass filter is used to initially remove the noise from the original right leg surface electromyography signal. Then, the VMD algorithm is used to perform variational mode decomposition to remove the interference noise in the original right leg sEMG data. The variational mode decomposition is shown in FIG4 .

[0057] Time- and frequency-domain features were extracted from de-noised sEMG data to analyze fatigue status in leg surface electromyography. Time-domain features included maximum, minimum, root mean square (RMS), amplitude, mean, number of sign changes, variance, kurtosis, skewness, sample entropy, and absolute mean. Frequency-domain features included median frequency and mean power frequency.

[0058] Maximum The calculation formula is:

[0059] ;

[0060] Where, For sEMG Amplitude of moment.

[0061] Minimum The calculation formula is:

[0062] ;

[0063] The formula for calculating the root mean square value is:

[0064] ;

[0065] Where, Represents the number of experimental sampling points.

[0066] The mean amplitude reflects the average level of muscle activity and is calculated as:

[0067] ;

[0068] Where, is the number of sEMG samples.

[0069] The sign change number reflects the changing nature of muscle electrical activity and is calculated as follows:

[0070] ;

[0071] Where II is the indicator function, if and If the sign of is opposite, the function value is 1, otherwise it is 0.

[0072] Variance is used to measure the degree of dispersion of signal distribution, and the calculation formula is:

[0073] ;

[0074] Where, is the average value of sEMG.

[0075] Kurtosis is used to measure the sharpness of the probability distribution shape. The calculation formula of kurtosis is:

[0076] ;

[0077] Where, is the sample standard deviation.

[0078] Skewness is used to measure the asymmetry of a probability distribution. The calculation formula of skewness is:

[0079] ;

[0080] Sample entropy is used to express the complexity of time series and reflect the irregularity and unpredictability of signals. The calculation formula of sample entropy is:

[0081] ;

[0082] Where, To calculate the distance less than Vector The number of

[0083] The absolute mean value is used to assess the intensity and fatigue of muscle contraction. The absolute mean value is calculated as follows:

[0084] ;

[0085] Where, Represents the number of experimental sampling points.

[0086] The median frequency refers to the median frequency of the sEMG produced during muscle contraction. MF is obtained by analyzing the power spectral density of the surface sEMG and is calculated as follows:

[0087] ;

[0088] Where, represents the power spectrum of surface sEMG, is the frequency of sEMG.

[0089] The average power frequency refers to the average value of the power frequency distribution of sEMG generated during muscle contraction. It is used to assess the strength and fatigue of muscle contraction. The calculation formula is:

[0090] ;

[0091] Where, represents the power spectrum of surface sEMG, is the frequency of sEMG.

[0092] By analyzing the time-frequency domain, we obtained the distribution of eigenvalues ​​after combining the time windows of each time-frequency domain eigenvalue. As shown in Figure 5, the maximum and minimum values ​​in the time-domain features show an upward trend, indicating an increase in the total amount of muscle fiber discharge involved in the tibialis anterior motor unit, and increased fatigue with increasing driving time. The root mean square, variance, amplitude mean, kurtosis, sample entropy, and absolute mean values ​​showed a significant increase in the magnitude of change, indicating increased contraction of the tibialis anterior muscle and increased fatigue. The sign change number and skewness shifted downward as fatigue deepened, reflecting a decrease in muscle activity. The median frequency and mean power frequency in the frequency-domain features showed an overall decrease, indicating an increase in low-frequency components in the electrical signals generated by the tibialis anterior muscle and increased fatigue. These analysis results are consistent with self-assessed fatigue status, demonstrating the effectiveness of sEMG processing and analysis.

[0093] S5. Use the improved SVM classification model to classify and predict driving fatigue status.

[0094] S51: Constructing an SVM classification model to classify driving fatigue status, the steps including:

[0095] Suppose the feature vector after training sample mapping is , the partition hyperplane formula in the feature space is:

[0096] ;

[0097] Where, represents the normal vector, Represents the displacement term and introduces the penalty factor With slack variables , the formula is:

[0098] ;

[0099] The dual function is:

[0100] ;

[0101] Where, represents the Lagrange multiplier;

[0102] Define the kernel function:

[0103] ;

[0104] Where, represents the feature vector, Indicates the corresponding label, the value is +1 or -1,

[0105] Bringing the kernel function into the dual function yields:

[0106] ;

[0107] After solving, we get the optimal classification function:

[0108] ;

[0109] The optimal classification function is transformed to obtain the optimal classification function of one-to-many SVM, the formula is:

[0110] ;

[0111] Driving fatigue status is divided into three categories: mild, moderate and severe fatigue. When using SVM to classify fatigue status, , its optimal classification function can be expressed as:

[0112] ;

[0113] S52. Use the whale optimization algorithm to optimize the penalty parameters and kernel functions in the SVM model to enhance the classification and prediction capabilities of the SVM model.

[0114] When the whale algorithm seeks the optimal solution, it is divided into the capture stage, the bubble net sealing stage and the exploration stage.

[0115] Roundup phase

[0116] When a whale is hunting, it will quickly move towards its prey after discovering it and determining its location. During the optimization process, this behavior is simulated as an updated search towards the optimal solution (prey) instead of the initial solution (whale). First, the distance between the whale and the prey is calculated:

[0117] ;

[0118] Where, is the distance vector between the current solution vector and the optimal solution, is the solution vector, is the optimal solution vector in the current iteration, is the coefficient vector.

[0119] The calculation formula for the solution vector of the next iteration in the roundup phase is expressed as:

[0120] ;

[0121] Where, is the solution vector for the next iteration, is the coefficient vector.

[0122] Coefficient vector and The calculation formula is:

[0123] ;

[0124] Where, Linear from 2 to is a random vector in [0, 1].

[0125] Bubble net enclosure stage

[0126] After locating the prey and reaching it, the whale will spit out a spiral bubble ring above the prey to encircle it. At this stage, the whale controls the size of the bubble net to encircle the prey.

[0127] The diameter of the bubble net is expressed as:

[0128] ;

[0129] Where, is the diameter vector of the bubble net.

[0130] The solution vector calculation formula for the next iteration of the bubble network enclosure stage is expressed as:

[0131] ;

[0132] Where, is the spiral shape parameter, It is a random number in the range [-1, 1].

[0133] Search phase

[0134] During the hunting process, whales also perform random searches and may not always find the optimal solution. The solution vector for the next iteration in the random search phase can be expressed as:

[0135] ;

[0136] Where, is a randomly selected solution vector in the search space, is the distance vector between the random solution and the current solution.

[0137] The main logic of optimizing the SVM driving fatigue state classification model based on the whale algorithm is to use the global search capability of the whale algorithm to find and optimize the penalty parameters in the SVM. and kernel function parameters The steps include:

[0138] (1) According to the size of the sample data, set the population size to 20, the maximum number of iterations to 100, and the upper limit of the variable and lower limit are [0.1, 100] and [0.01, 0.1].

[0139] (2) Set the fitness function value and calculate it, and select the one with the smallest fitness function value as the best position for the whale.

[0140] (3) , use the hunting phase to solve the next iteration solution vector calculation formula for calculation, if , the calculation is performed according to the solution vector calculation formula of the next iteration of the bubble network enclosure stage.

[0141] (4) Find the optimal solution vector .

[0142] (5) Determine whether the conditions for the maximum number of iterations are met.

[0143] (6) Output the optimal solution vector .

[0144] Refer to Table 1, which shows the classification results of the SVM model.

[0145] Table 1 SVM classification results

[0146] ;

[0147] The accuracy of the SVM driving fatigue training set optimized by the whale algorithm increased from 86.9393% to 96.3061%; the accuracy of the test set increased from 82.8936% to 84.3887%; the accuracy of mild fatigue increased from 97.7200% to 98.8800%; the accuracy of moderate fatigue increased from 81.1200% to 88.5900%; the accuracy of severe fatigue increased from 40.5300% to 82.7500%; and the model classification prediction accuracy increased from 86.1301% to 93.9226%.

[0148] Therefore, the present invention adopts the above-mentioned driving fatigue status classification and evaluation method based on leg sEMG, constructs an SVM classification model by processing and analyzing the surface electromyography signals of the legs, and uses the whale optimization algorithm to optimize the SVM classification model, thereby improving the classification effect based on the surface electromyography signals of the legs and improving the classification accuracy.

[0149] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for classifying and evaluating driving fatigue status based on leg sEMG, characterized in that the steps include: S1. Construct a human-machine-environment system model based on the graphical model method, analyze the human-machine-environment system model, and determine the causes of driving fatigue; S2. Based on the analyzed driving fatigue causes, driving fatigue data is collected, including leg surface electromyography signals, self-assessed fatigue status, and fatigue status label files, to establish a leg surface electromyography signal dataset. S3, cutting the data in the leg surface electromyography signal dataset to obtain the starting points of mild fatigue, moderate fatigue, and severe fatigue; S4. Remove data noise from the leg surface electromyography signal dataset and extract time-frequency domain features to analyze the fatigue state of the leg surface electromyography signal; S5. Use the improved SVM classification model to classify and predict driving fatigue status.

2. The method for classifying and evaluating driving fatigue status based on leg sEMG according to claim 1, characterized in that: The step S2 of collecting driving fatigue data includes: collecting the surface electromyography signal of the right leg, continuously questioning the driver during the collection process, recording the driver's fatigue status, and marking the fatigue status in the electromyography signal data to obtain the surface electromyography signal of the right leg, the self-assessed fatigue status and the fatigue status label file.

3. The method for classifying and evaluating driving fatigue status based on leg sEMG according to claim 2, characterized in that: In step S3, the fatigue state is segmented according to the corresponding relationship between the original right leg surface electromyographic signal and the fatigue state annotation data at the time, and the starting points of mild fatigue, moderate fatigue and severe fatigue are obtained.

4. The method for classifying and evaluating driving fatigue status based on leg sEMG according to claim 3, characterized in that: In step S4, a bandpass filter is used to initially remove noise from the original right leg surface electromyography signal, and then a VMD algorithm is used to perform variational mode decomposition to remove interference noise from the original right leg surface electromyography signal.

5. The method for classifying and evaluating driving fatigue status based on leg sEMG according to claim 4, characterized in that: The fatigue state result of the leg surface electromyography signal analyzed in the time-frequency domain in step S4 is consistent with the self-assessed fatigue state.

6. The method for classifying and evaluating driving fatigue status based on leg sEMG according to claim 5, characterized in that: The time domain features extracted in step S4 include maximum value, minimum value, root mean square value, amplitude, mean, number of sign changes, variance, kurtosis, skewness, sample entropy and absolute mean; the frequency domain features extracted include median frequency and average power frequency.

7. The method for classifying and evaluating driving fatigue status based on leg sEMG according to claim 5, characterized in that: The steps in step S5 include: S51: Constructing an SVM classification model to classify driving fatigue status, the steps including: Suppose the feature vector after training sample mapping is , the partition hyperplane formula in the feature space is: ； Where, represents the normal vector, Represents the displacement term and introduces the penalty factor With slack variables , the formula is: ； The dual function is: ； Where, represents the Lagrange multiplier; Define the kernel function: ； Where, represents the feature vector, Indicates the corresponding label, the value is +1 or -1, Bringing the kernel function into the dual function yields: ； After solving, we get the optimal classification function: ； The optimal classification function is transformed to obtain the optimal classification function of one-to-many SVM, the formula is: ； S52. Use the whale optimization algorithm to optimize the penalty parameters and kernel functions in the SVM model to enhance the classification and prediction capabilities of the SVM model.

Images