A method for evaluating mental fatigue based on K-TRCA

Abstract

The present invention discloses a method for evaluating mental fatigue based on K‑TRCA, including the following steps: preprocessing stage: filling the vacant data segments with "0"; filtering the original EEG data to remove electrooculographic artifacts; obtaining pupil diameter data sample sets and EEG data sample sets; calibration stage: calculating the P80 values ​​of all pupil data samples to obtain the initial clustering center; re-clustering the pupil data sample sets based on the K‑means algorithm to obtain the EEG data sample sets. Training stage: training the samples using the TRCA algorithm to calculate the spatial filter group and the average value of the training samples; testing stage: filtering the test samples and the sample average values, and calculating the Pearson correlation coefficient to determine the category of the test samples. The present invention calibrates the eye movement data by combining the P80 threshold determination method with the K‑means clustering algorithm, thereby ensuring the objectivity and accuracy of fatigue category division, improving the fatigue assessment accuracy in the RSVP weak target recognition task, and enhancing robustness and generalization capabilities.

Description

Technical Field

[0001] The present invention relates to the field of mental fatigue assessment, and in particular to a mental fatigue assessment method based on K-TRCA, which is a method for identifying mental fatigue induced by a rapid serial visual presentation (RSVP) weak target detection task based on a K-means clustering method combined with an Improved Task-relevant Component Analysis (TRCA) algorithm. Background Art

[0002] Whether it is autonomous driving in daily life or in rapid response fields such as modern high-tech warfare, the system is required to have a fast reaction time and a long detection distance for small targets, that is, to achieve RSVP small target rapid detection. It is difficult to achieve accurate recognition by relying solely on computer vision processing. Combining the advantages of machine intelligence and human brain intelligence and human-machine hybrid strategies can improve the perception of small targets. However, the human brain is prone to fatigue in high-intensity and long-term small target detection tasks. If it is not evaluated and intervened in time, it will lead to serious consequences. Therefore, it is of great significance to quickly and accurately evaluate the fatigue state of the human brain and implement early warning in time to prevent man-made accidents caused by operator fatigue.

[0003] In recent years, the evaluation methods of mental fatigue have become diversified, mainly based on subjective scales, physiological indicators, psychological indicators and biochemical indicators. The fusion of multimodal indicators to achieve fatigue evaluation is the main direction of future development. However, in terms of mental fatigue, the fusion method of multimodal data often adopts the decision-level fusion method. The defect of this method is that the final recognition accuracy is overly dependent on the recognition accuracy of a single modality data. At present, there is an urgent need for a method of re-identification by fusing multimodal data at the feature layer to effectively improve the fatigue evaluation level in the RSVP weak target recognition task and ensure the stability of the results.

[0004] EEG is considered to be one of the most reliable indicators for judging fatigue, but it has the following shortcomings: First, the calibration of fatigue data is often based on a subjective fatigue scale, which is affected by personal subjective judgment, resulting in errors in the calibration process, which in turn affects the classification results; second, most studies are based on frequency domain analysis of changes in the θ, δ, and α bands of EEG to evaluate fatigue. There is a lack of research on the relationship between the various channels of EEG at a deeper level, and there is no analysis from the perspective of the relevant components of EEG and fatigue, which hinders the further development and application of EEG as a fatigue assessment indicator.

[0005] Eye movement indicators including blinking frequency, pupil diameter, eyelid closure and other parameters are often used to evaluate fatigue. Among them, the P80 standard based on the PERCLOS (percentage of eyelid closure over the pupil over time) algorithm is recognized to have the best correlation with fatigue. However, it is difficult to accurately divide the fixed threshold of P80 according to different tasks and different people, which also limits the application of this method in multiple fields. Summary of the invention

[0006] The purpose of the present invention is to overcome the shortcomings of the above existing technologies and provide a mental fatigue assessment method based on K-TRCA. The eye movement data is calibrated based on the P80 threshold judgment method combined with the K-means clustering algorithm to ensure the objectivity and accuracy of the category division. The maximum fatigue-related component is extracted based on the TRCA algorithm for training and classification recognition, and fatigue is analyzed from the deep-level relationship between the channels of EEG data, thereby improving the fatigue assessment accuracy in the RSVP weak target recognition task and enhancing the robustness and generalization ability.

[0007] In order to achieve the above object, the technical solution of the present invention is:

[0008] A method for evaluating mental fatigue based on K-TRCA, comprising the following steps:

[0009] Based on the RSVP small target detection task, the eye movement data and EEG data of the subjects are collected simultaneously. Both types of data contain synchronization labels when the target image appears. The data processing is divided into four stages:

[0010] (1) Preprocessing stage

[0011] (1-1) Fill the missing data segments of the pupil diameter data of the left and right eyes with "0";

[0012] (1-2) Filter the raw EEG data and remove eye artifacts;

[0013] (1-3) Starting from the time tag when the first target image appears, intercept the pupil diameter data and EEG data with a unit time length of t0, until the time tags when all target images appear are traversed, and save the intercepted data segment samples as pupil data sample sets and EEG data sample sets respectively;

[0014] (2) Calibration stage

[0015] (2-1) Calculate the P80 value of each sample in the pupil data sample set, and classify the results into three categories: awake, tired, and sleepy according to the internationally recognized P80 threshold standard;

[0016] (2-2) Re-clustering all pupil data samples by K-means algorithm, and representing the EEG data sample sets corresponding to the newly clustered pupil diameter data samples as X1, X2, and X3, where X1 is the awake sample, X2 is the fatigue sample, and X3 is the sleepy sample, for training;

[0017] (3) Training phase

[0018] (3-1) Using the TRCA algorithm to perform fatigue-related component analysis on X1, X2, and X3, a spatial filter group W capable of extracting three types of fatigue-related components is calculated;

[0019] (3-2) Calculate the superposition average of all training samples for each fatigue category

[0020] (4) Testing phase

[0021] (4-1) Use the spatial filter bank W to simultaneously filter the test sample Y∈R Nc×Nt and Filter and calculate the Pearson correlation coefficient r k，k＝1,2,3 ; The calculation formula is as follows:

[0022]

[0023] Where k = 1, 2, 3, R represents a real number set, Nc represents the number of channels, Nt represents the number of sampling points, ρ(*,*) represents the calculation of the Pearson correlation coefficient, Y represents the test sample, T represents the transpose operation of the matrix, and W represents the spatial filter bank. express The transpose of the kth in ;

[0024] (4-2) Compare the sizes of r1, r2, and r3. If r1 is the largest, it is determined that the test sample Y represents an awake state. If r2 is the largest, it is determined that the test sample Y represents a fatigued state. If r3 is the largest, it is determined that the test sample Y represents a sleepy state.

[0025] Preferably, the P80 numerical calculation method in step (2-1) is specifically implemented as follows:

[0026] By calculating the percentage of the time t1 in which the pupil area is covered for more than 80% per unit time to the time t0, the degree of brain fatigue can be preliminarily assessed. Assuming that a certain value in the pupil diameter sample set is d, and the average pupil diameter of the subjects in the resting state with eyes open is d, the screening condition of the time t1 in which the pupil area is covered for more than 80% can be expressed as:

[0027]

[0028] Thus, the numerical calculation formula of P80 can be expressed as:

[0029]

[0030] Preferably, the P80 threshold determination criterion in the step (2-1) is specifically as follows:

[0031] When P80 ≤ 0.075, it is in a waking state;

[0032] When 0.075 < P80 ≤ 0.15, it is in a fatigued state;

[0033] When P80 > 0.15, it is in a drowsy state.

[0034] Preferably, the specific implementation of the K-means algorithm in (2-2) is as follows:

[0035] Initialization of centers: Assume that the pupil diameter sample sets of three types of fatigue states can be expressed as C j , j = 1, 2, 3. Randomly select a set of data from each type of pupil diameter data, and a total of 3 sets of data are used as the initial clustering centers A1, A2, and A3 of the K-means algorithm respectively;

[0036] Calculation of distances: Assume that the pupil diameter data sample set F = {f1, f2, …, f n}}, calculate the distance D i between the i-th pupil diameter data sample f ij and the j-th clustering center. The preferred calculation formula can be expressed as:

[0037] D ij = ||f i - A j ||2, 1 ≤ i ≤ n,, j = 1, 2, 3

[0038] where n represents the number of pupil diameter data samples, i represents a positive integer greater than or equal to 1 and less than or equal to n, f i represents the i-th pupil diameter data sample, and A j represents one of the initial clustering centers A1, A2, and A3;

[0039] Re-grouping: Assign the pupil diameter data sample f i to the class corresponding to the clustering center with the minimum distance until all samples are grouped;

[0040] Calculation of new center points: Recalculate the clustering center B j , j = 1, 2, 3, and replace the original clustering center. The calculation formula of B j is:

[0041]

[0042] Among them, C j represents one of the three types of pupil diameter sample sets of fatigue states, and f represents C j A single sample in

[0043] Iterate until convergence: Repeat the process of "calculating distance", "regrouping" and "calculating new center points" until the current cluster center is the same as the previously calculated cluster center, and then end the loop.

[0044] Preferably, the TRCA algorithm in step (3-1) is specifically implemented as follows:

[0045] The EEG training sample X k，k＝1,2,3 Represented as X k =[x1, x2, …x Nc ], the linear model of EEG fatigue component with Nc leads can be expressed as:

[0046]

[0047] Among them, x m Represents the EEG training sample X k The mth subcomponent in, m is a positive integer greater than or equal to 1 and less than or equal to Nc, t represents the time variable, w m represents the spatial filter matrix;

[0048] After the spatial filter matrix w m After filtering, y(t) contains only fatigue-related components, Q is the covariance matrix between leads, S is the covariance matrix between similar samples, and W k =[w1,w2,...,w Nc ], k = 1, 2, 3 can be solved by solving Q -1 The S feature vector is obtained, and the spatial filter bank of all fatigue categories can be expressed as: W = [W1, W2, W3].

[0049] Compared with the prior art, the present invention has the following obvious outstanding substantive features and remarkable advantages:

[0050] 1. The present invention clusters samples based on the K-means algorithm on the PERCLOS fatigue calibration method, solving the problem that the P80 threshold fluctuates due to individual differences, which ultimately affects fatigue calibration; this method ensures the consistency of features of similar samples, which is conducive to improving the final fatigue degree recognition accuracy;

[0051] 2. The present invention searches for fatigue-related components in EEG signals under different fatigue states based on the TRCA algorithm. Compared with the fatigue judgment method based on the time-frequency domain, the present method searches for the relationship between each lead and the maximum fatigue-related component of the same sample in the spatial domain, and decodes the correlation between fatigue and EEG at a deeper level.

[0052] 3. The present invention combines the K-means algorithm and the TRCA algorithm to perform data fusion at the feature layer to evaluate fatigue. Compared with the method of performing fusion at the data layer, the amount of data calculation is reduced and the calculation efficiency is improved. Compared with performing fusion at the decision layer, the accuracy and robustness of the evaluation are improved.

[0053] 4. The present invention is conducive to improving the efficiency of workers in RSVP-based small target detection tasks, quickly and accurately assessing the fatigue state of the human brain, and implementing early warning in a timely manner, which is of great significance for preventing man-made accidents caused by worker fatigue. It is conducive to promoting the further application of collaborative tasks of human-machine integration. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a block diagram of the overall algorithm implementation principle of the preferred embodiment of the present invention.

[0055] Figure 2 It is a flow chart of the K-means algorithm of the preferred embodiment method of the present invention. DETAILED DESCRIPTION

[0056] The preferred embodiments of the present invention are described in detail as follows in conjunction with the accompanying drawings:

[0057] Embodiment 1:

[0058] A method for evaluating mental fatigue based on K-TRCA, the method comprising:

[0059] See also Figure 1 Based on the RSVP small target detection task, the eye movement data and EEG data of the subjects are collected simultaneously. Both types of data contain synchronization tags when the target image appears. The data processing is divided into four stages:

[0060] (1) Preprocessing stage

[0061] (1-1) Extract the pupil diameter data of the left and right eyes from the collected original eye movement data, and fill the missing data segments with "0";

[0062] (1-2) The collected raw EEG data were subjected to 0-100 Hz bandpass filtering and 50 Hz notch filtering, and the electrooculographic artifacts were removed.

[0063] (1-3) Starting from the time tag of the first target image, the pupil diameter data and EEG data with a unit time length of t0 are intercepted backward until the time tags of all target images are traversed, and the intercepted data segment samples are saved as pupil data sample sets and EEG data sample sets respectively.

[0064] (2) Calibration stage

[0065] (2-1) Calculate the P80 value of each sample in the pupil data sample set, and classify the results into three categories: awake, tired, and sleepy according to the internationally recognized P80 threshold standard;

[0066] (2-2) All pupil data samples are re-clustered using the K-means algorithm, and the EEG data sample sets corresponding to the three types of pupil diameter data samples of wakefulness, fatigue, and sleepiness are saved as X1, X2, and X3, respectively, where X1 is the wakefulness sample, X2 is the fatigue sample, and X3 is the sleepiness sample for training.

[0067] (3) Training phase

[0068] (3-1) Using the TRCA algorithm to perform fatigue-related component analysis on the EEG data sample sets X1, X2, and X3, a spatial filter group W capable of extracting three types of fatigue-related components is calculated;

[0069] (3-2) Calculate the superposition average of all training samples for each fatigue category

[0070] (4) Testing phase

[0071] (4-1) Use the spatial filter bank W to simultaneously filter the test sample Y∈R Nc×Nt and Filter and calculate the Pearson correlation coefficient r k，k＝1,2,3 ; The calculation formula is as follows:

[0072]

[0073] Where k = 1, 2, 3, R represents a real number set, Nc represents the number of channels, Nt represents the number of sampling points, ρ(*,*) represents the calculation of the Pearson correlation coefficient, Y represents the test sample, T represents the transpose operation of the matrix, and W represents the spatial filter bank. express The transpose of the kth in ;

[0074] (4-2) Compare the magnitudes of r1, r2, and r3. If r1 is the largest, then it is determined that the test sample Y represents the awake state. If r2 is the largest, then it is determined that the test sample Y represents the fatigued state. If r3 is the largest, then it is determined that the test sample Y represents the drowsy state.

[0075] The method of this embodiment calibrates eye movement data based on the P80 threshold determination method in combination with the K-means clustering algorithm, ensuring the objectivity and accuracy of class division. It extracts the maximum fatigue-related components based on the TRCA algorithm for training and classification recognition, analyzes fatigue from the deep relationship between each channel of electroencephalogram data, thereby improving the fatigue assessment accuracy in the RSVP weak target recognition task and enhancing the robustness and generalization ability.

[0076] Embodiment 2:

[0077] This embodiment is basically the same as Embodiment 1, and the special features are as follows:

[0078] The specific implementation of the P80 numerical calculation method in the step (2-1) is as follows:

[0079] Based on the P80 standard of the PERCLOS method, that is, by calculating the percentage of the time t1 when the pupil area is blocked by more than 80% in the unit time t0, the degree of brain fatigue is preliminarily evaluated; assume that a certain value in the pupil diameter sample set is d, and the average pupil diameter of the subject in the open-eye resting state is Then the screening condition for the time t1 when the pupil area is blocked by more than 80% can be expressed as:

[0080]

[0081] Therefore, the P80 numerical calculation formula can be expressed as:

[0082]

[0083] In this embodiment, the P80 threshold determination standard in the step (2-1) is specifically:

[0084] When P80 ≤ 0.075, awake state;

[0085] When 0.075 < P80 ≤ 0.15, fatigued state;

[0086] When P80 > 0.15, drowsy state.

[0087] In this embodiment, referring to Figure 2 , the specific implementation of the K-means algorithm in (2-2) is as follows:

[0088] Center initialization: Assume that the pupil diameter sample set of 3 types of fatigue states can be expressed as C j, j = 1, 2, 3, randomly select a group of data in each type of pupil diameter data, a total of 3 groups of data as the initial clustering centers A1, A2, and A3 of the K-means algorithm;

[0089] Calculate the distance: Assume that the pupil diameter data sample set F = {f1,f2,…,f n}, calculate the i-th pupil diameter data sample f i The distance D from the jth cluster center ij , the optimal calculation formula can be expressed as:

[0090] D ij =||f i -A j ||2, 1≤i≤n, j=1,2,3

[0091] Where n represents the number of pupil diameter data samples, i represents a positive integer greater than or equal to 1 and less than or equal to n, and f i represents the i-th pupil diameter data sample, A j represents one of the initial cluster centers A1, A2, A3;

[0092] Regrouping: Pupil diameter data samples f i Assign samples to the class corresponding to the cluster center with the smallest distance until all samples are grouped;

[0093] Calculate new center points: Recalculate the cluster center B of each class j , j = 1, 2, 3, and replace the original cluster center, B j The calculation formula is:

[0094]

[0095] Among them, C j represents one of the three types of pupil diameter sample sets of fatigue states, and f represents C j A single sample in

[0096] Iterate until convergence: Repeat the process of "calculating distance", "regrouping" and "calculating new center points" until the current cluster center is the same as the previously calculated cluster center, and then end the loop.

[0097] The TRCA algorithm in step (3-1) is specifically implemented as follows:

[0098] The EEG training sample X k，k＝1,2,3 Represented as X k =[x1, x2, …x Nc ], the linear model of EEG fatigue component with Nc leads can be expressed as:

[0099]

[0100] Among them, x m Represents the EEG training sample X k The mth subcomponent in, m is a positive integer greater than or equal to 1 and less than or equal to Nc, t represents the time variable, w m represents the spatial filter matrix;

[0101] After the spatial filter matrix w m After filtering, y(t) contains only fatigue-related components, Q is the covariance matrix between leads, S is the covariance matrix between similar samples, and W k =[w1,w2,...,w Nc ], k = 1, 2, 3 can be solved by solving Q -1 The S feature vector is obtained, and the spatial filter bank of all fatigue categories can be expressed as: W = [W1, W2, W3].

[0102] The method of this embodiment clusters samples based on the K-means algorithm on the PERCLOS fatigue calibration method, and solves the problem that the P80 threshold fluctuates due to individual differences, which ultimately affects fatigue calibration; the method ensures the consistency of the characteristics of the same type of samples, which is conducive to improving the final fatigue recognition accuracy; the method of this embodiment searches for fatigue-related components in EEG signals under different fatigue states based on the TRCA algorithm. Compared with the fatigue judgment method based on the time-frequency domain, the method of this embodiment searches for the relationship between each lead and the maximum fatigue-related component of the same type of samples from the spatial domain, and decodes the correlation between fatigue and EEG at a deeper level; the method of this embodiment combines the K-means algorithm and the TRCA algorithm to perform data fusion at the feature layer to evaluate fatigue. Compared with the method of fusion at the data layer, the amount of data calculation is reduced and the calculation efficiency is improved. Compared with the fusion at the decision layer, the accuracy and robustness of the evaluation are improved; the method of this embodiment is conducive to improving the efficiency of the staff in the RSVP-based weak target detection task, quickly and accurately evaluating the fatigue state of the human brain, and implementing early warning in time, which is of great significance for preventing man-made accidents caused by fatigue of operators. It is conducive to promoting the further application of collaborative tasks involving human-machine integration.

[0103] Embodiment three:

[0104] This embodiment is basically the same as the above embodiment, with the following special features:

[0105] In this embodiment, the TRCA algorithm in step (3-1) is specifically implemented as follows:

[0106] The EEG training samples are expressed as k=1,2,3, Nc represents the number of leads, Nt represents the number of sampling points, Ns represents the number of samples, k represents the number of categories, X 1,2,3 Represents the data of all sampling points with a length of 1 minute, with the label categories of "awake", "fatigue" and "drowsy". k One of the samples is denoted as X k =[x1, x2, …x Nc ], k = 1, 2, 3, the linear model of the EEG fatigue component of this sample can be expressed as:

[0107]

[0108] Among them, x m represents the mth subcomponent in a single EEG training sample, where m is a positive integer greater than or equal to 1 and less than or equal to Nc, t represents the time variable, and w m represents the spatial filter matrix, Nc represents the number of leads;

[0109] After the spatial filter matrix w m , after filtering, y(t) only contains fatigue-related components, W k =[w1,w2,...,w Nc ], k = 1, 2, 3 can be obtained by maximizing the covariance between each sample, X k The hth sample and fatigue-related components in can be expressed as x (h) (t) and y (h) (t), h = 1, 2, ..., Ns. The sum of the covariances between two samples can be expressed as:

[0110]

[0111] In the formula, i1, i2 represent positive integers greater than or equal to 1 and less than or equal to Nc, h1, h2 represent positive integers greater than or equal to 1 and less than or equal to Ns, Represents the EEG training sample X k The i1th subcomponent of the h1th subsample in Represents the EEG training sample X k The i2th subcomponent of the h2th subsample in W k The i1th subspace filter in , Ns represents the number of samples. It can be expressed as:

[0112]

[0113] Then the normalization constraint of y(t) can be described as

[0114]

[0115] According to the Rayleigh-Ritz theory, the optimal weight vector It can be expressed as:

[0116]

[0117] In the formula, Q is the covariance matrix between leads, S is the covariance matrix between similar samples, By solving Q -1 S eigenvector is obtained; the spatial filter bank of all fatigue categories can be expressed as:

[0118] W=[W1,W2,W3]

[0119] In summary, the present invention discloses a method for evaluating mental fatigue based on K-TRCA. The method is: first, fill the missing data segments of the pupil diameter data of the left and right eyes with "0"; filter the original EEG data and remove the electrooculogram artifacts; intercept and generate a pupil diameter data sample set and an EEG data sample set with a time length of t0; secondly, calculate the P80 value of all pupil data samples, and obtain the initial clustering center of the K-means algorithm based on the recognized threshold standard, re-cluster all pupil data samples based on the K-means algorithm, and obtain the corresponding new clustered EEG data sample sets X1, X2, and X3; then, use the TRCA algorithm to train X1, X2, and X3, and calculate the spatial filter group W that can extract fatigue-related components; calculate the superimposed average value of all training samples of each fatigue category Finally, use W to simultaneously train the test samples and Filtering is performed and the Pearson correlation coefficient is calculated. The fatigue category corresponding to the maximum correlation coefficient is the category of the test sample. The present invention calibrates the eye movement data by combining the P80 threshold judgment method with the K-means clustering algorithm to ensure the objectivity and accuracy of the category division. The maximum fatigue-related component is extracted by the TRCA algorithm for training and classification recognition. Fatigue is analyzed from the deep-level relationship between the channels of the EEG data, thereby improving the fatigue assessment accuracy in the RSVP weak target recognition task and enhancing the robustness and generalization ability.

[0120] The above specific implementations are only descriptions of the preferred implementations of the present invention, and do not limit the protection scope of the present invention. Without departing from the design concept and spirit of the present invention, various modifications, substitutions and improvements made by ordinary technicians in this field to the technical solution of the present invention based on the text description and drawings provided by the present invention should all fall within the protection scope of the present invention. The protection scope of the present invention is determined by the claims.

Claims

1. A method for evaluating mental fatigue based on K-TRCA, characterized in that: The operation steps are as follows: (1) Preprocessing stage: (1-1) Fill the data segments with missing pupil diameter data for both the left and right eyes with "0"; (1-2) Filter the original EEG data and remove electrooculogram artifacts; (1-3) Starting from the time tag when the first target image appears, intercept the pupil diameter data and EEG data with a unit time length of t0 backward until all the time tags when the target images appear are traversed, and save the intercepted data segment samples as the pupil data sample set and the EEG data sample set respectively; (2) Calibration stage: (2-1) Calculate the P80 value of each sample in the pupil data sample set, and classify the results into 3 categories: awake, fatigued, and drowsy according to the internationally recognized P80 threshold standard; (2-2) Through the K-means algorithm, re-cluster all the pupil data samples, and represent the EEG data sample sets corresponding to the newly clustered pupil diameter data samples as X1, X2, and X3, where X1 is the awake sample, X2 is the fatigued sample, and X3 is the drowsy sample for training; (3) Training stage: (3-1) For the above-mentioned X1, X2, and X3, use the TRCA algorithm to perform fatigue-related component analysis and calculate the spatial filter bank W that can extract 3 types of fatigue-related components; (3-2) Calculate the superposition average of all training samples for each fatigue category (4) Testing stage: (4-1) Use the spatial filter bank W to simultaneously filter the test sample Y∈R Nc×Nt and Filter and calculate the Pearson correlation coefficient r k，k＝1,2,3 ; The calculation formula is as follows: Where k = 1, 2, 3, R represents a real number set, Nc represents the number of channels, Nt represents the number of sampling points, ρ(*,*) represents the calculation of the Pearson correlation coefficient, Y represents the test sample, T represents the transpose operation of the matrix, and W represents the spatial filter bank. express The transpose of the kth in ; (4-2) Compare the magnitudes of r1, r2, and r3. If r1 is the largest, it is determined that the test sample Y represents the awake state. If r2 is the largest, it is determined that the test sample Y represents the fatigued state. If r3 is the largest, it is determined that the test sample Y represents the drowsy state.

2. The method for evaluating mental fatigue based on K-TRCA according to claim 1, characterized in that: The specific calculation method of the P80 value in step (2-1) is as follows: Preliminarily evaluate the brain fatigue degree by calculating the percentage of the time t1 when the pupil area is covered by more than 80% in the unit time t0; Assuming that a certain value in the pupil diameter sample set is d, the average pupil diameter of the subjects in the resting state with eyes open is Then the time t1 when the pupil area is covered for more than 80% can be expressed as:

3. The method for evaluating mental fatigue based on K-TRCA according to claim 1, characterized in that: The specific P80 threshold determination standard in step (2-1) is as follows: When P80 ≤ 0.075, awake state; When 0.075 < P80 ≤ 0.15, fatigued state; When P80 > 0.15, drowsy state.

4. The method for evaluating mental fatigue based on K-TRCA according to claim 1, characterized in that: The specific implementation of the K-means algorithm in step (2-2) is as follows: Center initialization: Assume that the pupil diameter sample set of three types of fatigue states can be expressed as C j , j = 1, 2, 3, randomly select a group of data in each type of pupil diameter data, a total of 3 groups of data as the initial clustering centers A1, A2, and A3 of the K-means algorithm; Calculate the distance: Assume that the pupil diameter data sample set F = {f1,f2,…,f n }, calculate the distance between all pupil diameter data samples and the cluster center; Regrouping: Pupil diameter data samples f i Assign samples to the class corresponding to the cluster center with the smallest distance until all samples are grouped; Calculate new center points: Recalculate the cluster center B of each class j , j = 1, 2, 3, and replace the original cluster center; Iterate until convergence: Repeat the processes of "calculating distances", "re-grouping", and "calculating new center points" until the current cluster center is the same as the previously calculated cluster center, and end the loop.

5. The method for evaluating mental fatigue based on K-TRCA according to claim 4, characterized in that: Calculate the distance between all pupil diameter data samples and the cluster center. The i-th pupil diameter data sample f i The distance D from the jth cluster center ij The calculation formula can be expressed as: D ij =||f i -A j ||2,1≤i≤n,,1≤j≤3 Where n represents the number of pupil diameter data samples, i represents a positive integer greater than or equal to 1 and less than or equal to n, j represents a positive integer greater than or equal to 1 and less than or equal to 3, and f i represents the i-th pupil diameter data sample, A j Represents one of the initial cluster centers A1, A2, and A3.

6. The method for evaluating mental fatigue based on K-TRCA according to claim 4, characterized in that: Recalculate the cluster center B of each class j , j = 1, 2, 3, B j The calculation formula is: Among them, C j represents one of the three types of pupil diameter sample sets of fatigue states, and f represents C j A single sample in .

7. The method for evaluating mental fatigue based on K-TRCA according to claim 1, characterized in that: The specific implementation of the TRCA algorithm in step (3-1) is as follows: The EEG training sample X k，k＝1,2,3 Represented as X k =[x1, x2, …x Nc ], the linear model of EEG fatigue component with Nc leads can be expressed as: Among them, x m Represents the EEG training sample X k The mth subcomponent in, m is a positive integer greater than or equal to 1 and less than or equal to Nc, t represents the time variable, w m represents the spatial filter matrix; After the spatial filter matrix w m After filtering, y(t) contains only fatigue-related components, Q is the covariance matrix between leads, S is the covariance matrix between similar samples, and W k =[w1,w2,...,w Nc ], k = 1, 2, 3 can be solved by solving Q -1 The S feature vector is obtained, and the spatial filter bank of all fatigue categories can be expressed as: W = [W1, W2, W3].

Images