An orthogonal integration of emotion, fatigue, and subjective will, and a method for assessing mental state.

Abstract

This application relates to the field of data processing technology, and discloses an orthogonal fusion model of emotion, fatigue, and subjective will, as well as a method for assessing mental state. For the problem of emotion recognition, this application proposes a spatiotemporal entropy feature extraction method, which has good adaptability to small sample EEG data; for the problem of low recognition rate in fatigue state recognition, it proposes a power spectrum normalization feature extraction method for brain functional connectivity, which shows higher accuracy in fatigue state recognition; for the problem of feature fusion of emotion, fatigue, and work will, it proposes a mental state assessment model based on the Taguchi orthogonal method that fuses subjective and objective indicators. Experimental results show that the spatiotemporal entropy feature emotion recognition, the power spectrum recognition fatigue state recognition of brain functional connectivity, and the orthogonal fusion mental state assessment model of subjective and objective indicators proposed in this invention can complete the task of recognizing emotional state, fatigue state, and mental state.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to an orthogonal fusion of emotion, fatigue, and subjective will, and a method for assessing mental state. Background Technology

[0002] Factors affecting safe production include equipment factors, human factors, and environmental factors, among which human factors are the core and most important. Typically, mining workers spend long hours in the monotonous and dimly lit underground working environment, which can easily lead to irritability, depression, and fatigue. These negative emotions increase the rate of operational errors among underground personnel, resulting in coal mine safety accidents.

[0003] Poor mental state among miners, leading to negligence, improper operation, fatigue, lack of concentration, and low mood, are all human factors contributing to safety accidents. Currently, research on miners' mental state is limited, and preventing negative mental states in miners relies solely on managerial supervision, which not only wastes human resources but also negatively impacts miners' work morale. Therefore, research on human mental state is crucial for safe production in mining. Furthermore, research on mental state has significant application value in environments such as transportation and production workshops.

[0004] Patel P et al. (Brain informatics, 2021) used entropy as a metric for feature extraction to describe the complexity and uncertainty of EEG signals. Entropy is a concept in information theory used to measure the randomness and irregularity of signals. By calculating the entropy value of EEG signals, features related to emotional states can be extracted. Disadvantages of sample entropy include: sensitivity to noise in spatiotemporal data, which can significantly impact the results; significant dependence on data length and sampling frequency, requiring adjustments based on specific data and potentially introducing uncertainty; and high computational complexity when processing high-dimensional spatiotemporal data.

[0005] Wang H et al. (IEEE Transactions on Cognitive and Developmental Systems, 2020) collected electroencephalogram (EEG) data from drivers and used phase synchronization analysis to study the functional connectivity between different brain regions. The researchers constructed a functional connectivity matrix from phase synchronization values ​​to describe the connection strength and patterns between different brain regions. They then used machine learning algorithms to classify the functional connectivity matrix to determine driver fatigue. Chaudhuri A et al. (IEEE Transactions on Intelligent Transportation Systems, 2019) used 19-channel EEG data from 12 groups of test subjects and employed the sLORETA algorithm to infer the spatial distribution of active brain regions by analyzing EEG or MEG signals. It is based on the inverse problem, estimating source activity based on the potential distribution measured on the scalp. sLORETA uses regularization methods to handle the inverse problem to reduce estimation uncertainty. Combined with an SVM algorithm, a final accuracy of 86.00% was achieved.

[0006] Shang Y et al. (Electronics, 2022) used an emotion recognition and fatigue detection model to establish a model based on extracted features. They employed time series fusion, and state determination and decision-making were based on the fused time series, using appropriate algorithms or rules to determine the driver's emotion and fatigue state.

[0007] However, existing technologies suffer from problems such as incomplete single features in emotion state recognition and low recognition rate in fatigue state recognition, and the decision fusion of the two tasks of emotion state and fatigue state needs further improvement. Summary of the Invention

[0008] In view of the above-mentioned shortcomings of the existing technology, the purpose of the present invention is to provide a method for assessing mental state by orthogonal integration of emotion, fatigue and subjective will.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] An orthogonal integration of emotion, fatigue, and subjective will as a method for assessing mental state, comprising the following steps:

[0011] S1. Collect EEG data, preprocess the raw EEG data, and then use independent component analysis to remove hidden noise components to obtain processed EEG data.

[0012] S2. Based on the defined spatiotemporal entropy, feature extraction is performed on the emotional EEG data, and a Bi-LSTM emotion state recognition model is built to realize emotion recognition based on EEG spatiotemporal entropy features.

[0013] S3. Extract features from the fatigue EEG dataset, apply the regional relationships of brain functional connectivity to the regional relationships of multi-channel power spectrum, normalize the multi-channel power spectrum density according to the principle of brain functional connectivity, and classify fatigue states using support vector machine.

[0014] S4. Establish Taguchi orthogonal matrices for valence, arousal, and fatigue status as objective evaluation indicators of mental state to obtain objective mental state indicators;

[0015] S5. The mental state of the subjects was identified using a two-dimensional mental state assessment method that combines subjective and objective mental indicators.

[0016] Furthermore, in S1, the preprocessing of the collected EEG data includes: removing noise, power frequency interference and electrocardiogram biological signals, using an FIR filter to filter the EEG data in the range of 0.1Hz to 40Hz, setting the global EEG data average value as a reference point, and normalizing the data.

[0017] Furthermore, in S2, the spatiotemporal entropy specifically includes the following steps:

[0018] Step 1: Map the spatial electrode positions to a matrix

[0019] Step 2: Construct an m-dimensional vector matrix

[0020] Step 3: Calculate the standard deviation

[0021] Calculate the standard deviation σt of each channel at time t.

[0022]

[0023] In the formula, t represents time, and x i is the brain potential value of channel i, and I is the number of samples for channel i; This represents the average value of I channel values ​​at time t;

[0024] Step 4: Calculate the distance between vectors

[0025] Calculate the distance between sequence X(i) and all other sequences X(j) in step 2. The distance is calculated by subtracting the sequences at equal positions and taking the absolute value. The value with the largest absolute value is taken as the distance d[X(i),X(j)].

[0026] Step 5: Statistical analysis of the relationship between distance and threshold

[0027] When the distance d[X(i),X(j)] at time t is less than τ, the time difference is calculated. t The number of points and the ratio of this number to the total distance N-m+1:

[0028]

[0029] In the formula, N is the number of channels, τ t It is an empirical threshold, τ t =0.2σ t ;

[0030] Approximate entropy-based methods:

[0031] Take the logarithm, then calculate the average of all the values:

[0032]

[0033] The approximate entropy of spacetime is obtained as follows:

[0034] StApEn(m,τ t )=φ m (τ t )-φ m+1 (τ t (6)

[0035] φ in the above formula m (r t ) is taken from equation (3), and after steps 1 to 5, a new spatiotemporal entropy sequence L is formed:

[0036] F 11 ={StApEn(m,τ1),StApEn(m,τ2),…,StApEn(m,τ 40 )} (7)

[0037] In the above formula, M represents the point in time when time is at its maximum;

[0038] The Bi-LSTM model was selected for classification of the new spatiotemporal entropy sequence F11.

[0039] Further, in step 1, the 32 conductive electrode location distribution map is mapped to a 9×9 matrix.

[0040] Further, in step 2, we take m = 2; construct two types of vectors X(i) with dimensions m = 2 and m + 1 = 3 respectively; use Fz, Cz, Pz, Oz as intermediate axes as the ends of the vectors; when m + 1 = 3, the first two values ​​of the vector matrix are a pair of symmetrical positions, and finally find that there are 28 vectors when m = 2 and 14 vectors when m + 1 = 3.

[0041] Furthermore, in step 4, the mathematical expressions for the distance at different values ​​of m are as follows:

[0042] When m = 2:

[0043] d[X(i),X(j)]=max[|X(i)-X(j)|,|X(i+1)-X(j+1)|] (2)

[0044] When m+1 = 3:

[0045] d[X(i),X(j)]=max[|X(i)-X(j)|,|X(i+1)-X(j+1)|,|X(i+2)-X(j+2)|] (3)

[0046] Further, step S3, specifically, is a method for jointly identifying fatigue states based on brain functional connectivity and power spectrum, including:

[0047] The correlation between all channels of the EEG was calculated using the Pearson correlation coefficient method to obtain a correlation matrix. The absolute value of the correlation matrix was then taken.

[0048]

[0049] In the above formula, |r nu | represents the Pearson correlation coefficient in the nth row and uth column of the multi-channel correlation matrix, where N is the total number of electrode channels. It is the channel correlation matrix after taking the absolute value;

[0050] The average periodogram method is used to divide the observed signal data of length N into K segments, each segment of length M, and finally calculate the average power spectral density of the K segments. The density function of the average periodogram method is expressed as follows:

[0051]

[0052] In the formula, x m k (m) represents the k-th data point in the m-th data segment; ω represents the angular frequency; m is the dimension of the oriented quantity;

[0053] The channel normalization process based on the brain functional connectivity correlation matrix can be expressed mathematically as follows:

[0054]

[0055] In the formula, W(n) represents the weight of channel n;

[0056] The single-channel power spectral density of the multi-channel composite is obtained by multiplying the weights corresponding to each channel and summing the results. This can be expressed mathematically as:

[0057]

[0058] In the above formula, psd* (ω) is the power spectral density after multi-channel normalization;

[0059] The preprocessed data is filtered for different frequency ranges of rhythm, and the power spectral density of the four frequency bands is obtained by combining the principle of normalized composite power spectral density:

[0060]

[0061] Select (P) α +P θ ) / (P α +P β () index as a fatigue detection index;

[0062] The fatigue EEG dataset was segmented into multiple segments, and then the two types of datasets were spliced ​​together to form a new dataset. The classification of indicators based on fatigue determination can be classified as a binary classification problem, and support vector machines were chosen to solve the classification problem.

[0063] Furthermore, step S4 specifically includes:

[0064] The mathematical relationship between objective indicators and three important factors: valence, arousal, and fatigue state:

[0065]

[0066] In equation (21), i,j∈{1,2,3},k∈{1,2}; k i ∈{3,4.25,8.25} represents the coefficients corresponding to the three valence levels, k j ∈{3.75,4.75,7} represents the coefficients corresponding to the three levels of arousal, k k ∈{6.27,8.75} represents the coefficients corresponding to the two fatigue levels; P vi P is the valence-to-accuracy ratio at level i. Aj P represents the accuracy corresponding to the arousal level at the j-th level. Pk It is the accuracy corresponding to the fatigue state at the k-th level.

[0067] Furthermore, in step S5, the mathematical equation for the mental state assessment is expressed as:

[0068] S = S1 + S2 + b s (twenty one)

[0069] S2 = 0.425E S +0.25F S +0.325W S (twenty two)

[0070] In the formula, S is the overall mental state evaluation index, S1 is the objective index, S2 is the subjective index, and bs This represents the perturbation error of the evaluation indicators, which is normally set to 0. When necessary, the evaluation indicators can be fine-tuned based on the actual situation. Es is the emotional state score, Fs is the fatigue state score, and Ws is the intensity of willingness to do this job.

[0071] Based on the valence and arousal levels of the second level, as well as fatigue and subjective indicators, the threshold should be set as follows:

[0072] τ=21.5+b S (twenty three)

[0073] In the above formula, 21.5 = 4.25 + 4.75 + (6.25 + 8.75) / 2 + 5;

[0074] Whether a person's mental state is satisfactory can be determined by comparing mental state evaluation indicators with thresholds.

[0075]

[0076] If the mental state index is greater than or equal to the threshold τ, it can be determined that the relevant staff member is in good mental condition and capable of performing the relevant work; conversely, if the mental state index is less than the threshold, it can be determined that the relevant staff member is in poor mental condition and unsuitable for work. Compared with the prior art, the present invention has the following beneficial effects:

[0077] (1) This application employs an EEG spatiotemporal entropy feature extraction algorithm and an emotion recognition method, which can better handle noisy spatiotemporal data and has good adaptability to data sparsity and incompleteness. It is not affected by data length and sampling frequency, and can also obtain good results for shorter data sequences. When processing high-dimensional spatiotemporal data, the computational complexity is relatively low. Moreover, the results of this application show that spatiotemporal approximate entropy as a feature has better accuracy. Therefore, the technical solution of this application selects spatiotemporal approximate entropy as a feature.

[0078] (2) This application employs an EEG power spectrum normalization feature extraction algorithm based on channel correlation matrix and a fatigue state identification method. The method, which combines brain functional connectivity and power spectrum to identify fatigue states, fully utilizes multiple features of EEG signals, exhibiting high accuracy and reliability, and wide applicability. It provides a new and effective approach for determining fatigue states. Compared to DCPM and brain connectivity fatigue identification methods, this method combines the advantages of both methodologies, demonstrating better performance and higher accuracy.

[0079] (3) This application employs an orthogonal fusion model of objective and subjective indicators and a mental state assessment method. The orthogonal fusion method based on the Taguchi orthogonal method fully utilizes multidimensional information, improving the accuracy and reliability of decisions by optimizing decision weights and reducing decision errors. Experimental results show that the mental state assessment method based on the fusion of emotional and fatigue states proposed in this invention can complete the task of identifying the emotional and fatigue states of miners. The method based on subjective and objective indicators is feasible for assessing the mental state of miners. It has certain reference value for miners' underground work and safe production. Attached Figure Description

[0080] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0081] Figure 1 This is a block diagram of the system structure of this application;

[0082] Figure 2 This is a technical roadmap of the present invention;

[0083] Figure 3 A framework diagram for an orthogonal fusion model of emotion, fatigue, and subjective will, and a method for assessing mental state;

[0084] Figure 4 This is a 32-channel timing diagram during excitation;

[0085] Figure 5 The timing diagram for the 32-channel signal when the signal is lost;

[0086] Figure 6 Two spatiotemporal entropy time series diagrams for excited emotions;

[0087] Figure 7 Two spatiotemporal entropy time series diagrams for excited emotions;

[0088] Figure 8 The accuracy rates of three methods for determining fatigue and normal states;

[0089] Figure 9 Mapping the spatial locations of the 32 conductive electrodes in the DEAP dataset to a matrix;

[0090] Figure 10 The model structure for Bi-LSTM;

[0091] Figure 11 This is a trend graph showing the relationship between experimental factors and experimental indicators. Detailed Implementation

[0092] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention. These all fall within the scope of protection of the present invention.

[0093] Example 1

[0094] like Figure 1 and Figure 2 As shown, an orthogonal fusion of emotion, fatigue, and subjective will, and a method for assessing mental state, includes the following steps:

[0095] S1. Acquire EEG data and remove noise, power line interference, ECG, and other biological signals. The raw EEG data is filtered from 0.1Hz to 40Hz using an FIR filter, with the global EEG average value set as a reference point for data normalization. Even after this processing, the EEG signal may still contain hidden noise components, and deep-seated noise is difficult to remove through observation alone. Therefore, Independent Component Analysis (ICA) is used to decompose the signal into independent components, and then EEG topographic maps of each sub-component are drawn. These EEG topographic maps allow for visual observation and targeted removal of noise sub-components in the EEG data. The noise-based ICA method is used to remove hidden noise components, resulting in processed EEG data.

[0096] S2. Feature extraction is performed on the emotion EEG dataset based on spatiotemporal entropy, and a Bi-LSTM emotion state recognition model is built for emotion recognition. The Bi-LSTM model, which can effectively process time-series signals, is selected to classify the transformed spatiotemporal entropy sequence.

[0097] S3. Feature extraction is performed on the fatigue EEG dataset. Regional relationships of brain functional connectivity are applied to the regional relationships of multi-channel power spectra. The multi-channel power spectral density is normalized using the principles of brain functional connectivity, thus achieving a combination of brain functional connectivity and power spectral analysis. These indicators are then used in the fatigue assessment model to identify fatigued mental states. Finally, support vector machines are used for classification.

[0098] S4. Establish Taguchi orthogonal matrices for valence, arousal, and fatigue status as objective evaluation indicators of mental state to obtain objective mental state indicators;

[0099] S5. The mental state of the subjects was identified using a two-dimensional mental state assessment method that combines subjective and objective mental indicators.

[0100] When collecting EEG data, conductive gel is needed to improve signal quality and ensure that there is no interference from power frequency signals in the room.

[0101] Spatiotemporal entropy characteristics with time and space domains

[0102] The spatiotemporal entropy calculation involves five steps, as shown below. Step 5 introduces two methods: spatiotemporal approximate entropy and spatiotemporal sample entropy, for comparison in subsequent experiments.

[0103] Step 1: Map the spatial electrode positions to a matrix

[0104] Taking a 32-channel EEG dataset as an example, the 32-channel electrode location distribution map is mapped onto a 9×9 matrix. Positions without a corresponding electrode are represented by 0, while those with a corresponding electrode are indicated by the channel name. The channel names in the mapping matrix are then replaced with the names of the corresponding electrodes in the order they appear. Here, t represents time, i.e., a specific moment, and i represents the channel number, such as... Figure 9 As shown.

[0105] Step 2: Construct an m-dimensional vector matrix

[0106] Lower dimensionality results in more vectors; generally, m=2 is optimal. Two types of vectors X(i) are constructed, with dimensions m=2 and m+1=3 respectively. Fz, Cz, Pz, and Oz are used as intermediate axes, serving as the vector terminals. When m+1=3, the first two values ​​of the vector matrix are a pair of symmetrical positions. Finally, it is found that there are 28 vectors when m=2 and 14 vectors when m+1=3.

[0107] Step 3: Calculate the standard deviation

[0108] Calculate the standard deviation σt of each channel at time t.

[0109]

[0110] In the formula, t represents time, and x i is the brain potential value of channel i, and I is the number of samples for channel i. This represents the average value of I channel values ​​at time t.

[0111] Step 4: Calculate the distance between vectors

[0112] Calculate the distance between sequence X(i) and all other sequences X(j) in step 2. The distance is calculated by subtracting the sequences at their positions and taking the absolute value. The largest absolute value is taken as the distance d[X(i),X(j)]. The mathematical expressions for the distance for different values ​​of m are shown below.

[0113] When m = 2:

[0114] d[X(i),X(j)]=max[|X(i)-X(j)|,|X(i+1)-X(j+1)|] (2)

[0115] When m+1 = 3:

[0116] d[X(i),X(j)]=max[|X(i)-X(j)|,|X(i+1)-X(j+1)|,|X(i+2)-X(j+2)|] (3)

[0117] Step 5: Statistical analysis of the relationship between distance and threshold

[0118] When the distance d[X(i),X(j)] at time t is less than τ, the time difference is calculated. t The number of points and the ratio of this number to the total distance N-m+1:

[0119]

[0120] In the formula, N is the number of channels, τ t It is an empirical threshold, τ t =0.2σ t ;

[0121] ① Methods based on approximate entropy:

[0122] Take the logarithm of equation (5) and then calculate the average of all the equations.

[0123]

[0124] The approximate entropy of spacetime is obtained as follows:

[0125] StApEn(m,τ t )=φ m (τ t )-φ m+1 (τ t (6)

[0126] φ in the above formula m (r t (6) is the formula. After steps 1 to 4 and step 5, a new spatiotemporal entropy sequence L is formed.

[0127] F 11 ={StApEn(m,τ1),StApEn(m,τ2),…,StApEn(m,τ 40 )} (7)

[0128] In the above formula, M represents the point in time at which the time is at its maximum.

[0129] ②Methods based on sample entropy:

[0130] Calculate all φ is calculated by direct averaging. m (r t ).

[0131]

[0132] In the above formula, r t =0.2*Std t N represents the sequence length, and equation (3) is a spatial vector, so in equation (9) N represents the number of channels; m is the dimension taken in step 2.

[0133] Calculate the spatiotemporal sample entropy:

[0134]

[0135] φ in the above formula m (r t (9) is the formula. After steps 1 to 4 and step 5, a new time-space entropy sequence L is formed.

[0136] L={SpSaEn(m,r1),SpSaEn(m,r2),...,SpSaEn(m,r M )} (10)

[0137] In the above formula, M represents the point in time at which the time is at its maximum.

[0138] At this point, both forms of the spatiotemporal entropy sequence can be calculated. Each point in the spatiotemporal entropy sequence contains spatial feature information, and the entire sequence contains temporal information. Transforming the original multi-channel sequence into a single-channel spatiotemporal entropy sequence not only includes dual-domain features but also shortens the training time for the effective sequence after multi-channel normalization. The practicality of the two spatiotemporal entropies is evaluated, and the optimal spatiotemporal entropy is selected as the feature extraction for emotion recognition. A Bi-LSTM model, which can effectively process temporal signals, is used to classify the transformed spatiotemporal entropy sequence.

[0139] like Figure 10 The diagram shows the Bi-LSTM model structure, which consists of two LSTM layers: a forward layer and a backward layer. Using a two-layer LSTM unidirectional structure allows for better extraction of EEG temporal features. Bi-LSTM can acquire information from both previous and subsequent time steps. With sufficient data, Bi-LSTM performs better than a unidirectional LSTM in classification. The forward LSTM calculates the output matrix h1 from the input of time step 1 to time step t of sequence L, while the backward LSTM calculates the output matrix h2 from the input of time step 1 to time step t of sequence L. The outputs of the forward and backward LSTMs are concatenated to obtain the final output. This can be expressed mathematically as:

[0140] O t =σ(W1h1+W2h2) (11)

[0141] In the above formula, O t σ represents the final output, W1 represents the weights of the forward LSTM output segment, and W2 represents the weights of the backward LSTM output segment.

[0142] A method for identifying fatigue states based on a combination of brain functional connectivity and power spectrum analysis:

[0143] The channels acquired during EEG signal acquisition cover various regions of the brain. By performing functional connectivity analysis between these channels, the goal of understanding the functional connectivity of brain regions can be achieved. The Pearson correlation coefficient measures the degree of correlation between two variables. The Pearson correlation coefficient ranges from -1 to 1; a value closer to 1 indicates a stronger positive correlation, while a value closer to -1 indicates a stronger negative correlation. Assume that after preprocessing, there are two channels X = {x1, x2, ..., x...}. n}, and Y = {y1, y2, ..., y n Then, the correlation coefficient between the X and Y channels can be expressed using the Pearson correlation coefficient as follows:

[0144]

[0145] In the formula, The mean of the data in channel X. Let be the mean of the data for channel Y, and r be the Pearson correlation coefficient.

[0146] The correlation between all EEG channels was calculated using the Pearson correlation coefficient method, resulting in a correlation matrix. Since the Pearson correlation coefficient can show both positive and negative correlations, the absolute values ​​of the correlation matrix need to be taken.

[0147]

[0148] In the above formula, r ij This represents the Pearson correlation coefficient between the i-th and j-th channels, where n represents the total number of electrode channels in the dataset. This represents the matrix after taking the absolute value.

[0149] When a person is fatigued, brain activity decreases, and the frequency and energy of brain electrical signals also decrease accordingly. Power spectral density reflects the power distribution of the signal in different frequency domains. Using the average periodogram method, the observed signal data of length N is divided into K segments, each of length M, and finally the average power spectral density of the K segments is calculated. The density function of the average periodogram method is expressed as follows:

[0150]

[0151] In the formula, x m k (m) represents the k-th data point in the m-th data segment; ω represents the angular frequency; m is the dimension of the oriented quantity;

[0152] Power spectral density analysis (PSD) and functional correlation modeling (FRM) are two commonly used methods for characterizing electroencephalograms (EEGs). These methods are not independent and may have influencing relationships. Therefore, the correlation matrix of nodes in brain functional connectivity methods is introduced into PSD. Combining the brain functional correlation matrices under normal and fatigue states, this correlation matrix is ​​incorporated into the multi-channel normalization process of PSD. First, the correlation matrix is ​​normalized to obtain the channel weight matrix. Then, the weights corresponding to each channel are multiplied and summed to obtain a multi-channel composite PSD. This normalization process highlights the important channels related to fatigue states and reduces the influence of irrelevant channels, resulting in clearer results and more accurate PSD identification of fatigue states. The channel normalization process based on the brain functional connectivity correlation matrix is ​​mathematically expressed as:

[0153]

[0154] In the formula, W c (n) represents the weight of channel n, N C Indicates the number of channels. This represents the value in the i-th row and j-th column of the correlation matrix.

[0155] The single-channel power spectral density of the multi-channel composite is obtained by multiplying the weights corresponding to each channel and summing the results. This can be expressed mathematically as:

[0156]

[0157] In the above formula, psd * (ω) is the power spectral density after multi-channel normalization;

[0158] The preprocessed data is filtered for different frequency ranges of rhythms. The power spectral density of four frequency bands is obtained by combining the principle of normalized composite power spectral density. The power spectral density of an EEG rhythm represents the power per unit frequency, and the power spectral value of each rhythm segment of the EEG can be obtained by integrating the power spectral density within a specified frequency band.

[0159]

[0160] During fatigue, slow waves increase while fast waves decrease. Simultaneously, the power of delta (δ) and theta (θ) increases, while the power of alpha (α) and beta (β) decreases. Over time, δ and theta activities stabilize, alpha activity decreases slightly, and β activity decreases significantly. In fatigue assessment, the power ratio exponent of different rhythms is more effective than time-domain characteristics. Many studies on fatigue discriminant indicators exclude δ activity; if δ activity is significant, it indicates a deep sleep state. Because these waves exist at all times, only their power varies, an index related to δ rhythm signals is introduced. P is added to the four commonly used rhythm ratios. θ / (P α +P β ), P θ / P α 、(P δ +P θ ) / (P α +P β ), P δ / P β 、(P δ +P α ) / (P α +P β ), P δ / (P α +P β These 10 indicators were selected for experimental comparison, and the best (P) was chosen based on the experimental results. α +P θ ) / (P α +P β The index is used as the fatigue detection index in this application.

[0161] Normalized EEG data can effectively represent the power spectrum of the entire field in an EEG topography map, and fatigue and mental state discrimination indicators can also effectively reflect the difference between fatigued and non-fatigued states. However, EEG data has temporal characteristics and will change slightly over time. If the data is not segmented, a single discrimination indicator will be difficult to use for assessing fatigue. Therefore, it is necessary to divide the dataset into multiple segments and then concatenate the two types of datasets to form a new dataset. The classification of fatigue-based indicators can be categorized as a binary classification problem, and support vector machines can better solve this problem.

[0162] like Figure 3 The diagram shows a framework for an orthogonal fusion model of emotion, fatigue, and subjective will, as well as a method for assessing mental state.

[0163] To address the challenge of integrating emotional and fatigue states, this paper first addresses the limitations of privately collected EEG data, which often suffers from limited dataset size and lack of data authority. Introducing transfer learning not only solves these problems but also saves significant time and manpower. Second, it proposes using the Taguchi orthogonal array method to predict trends in the relationship between experimental factors and indicators, thus integrating emotional and fatigue states. Third, it presents a two-dimensional mental state assessment method combining subjective and objective mental indicators. Finally, the method is validated using a private dataset based on an individual's emotional and fatigue states.

[0164] Valence, arousal, and fatigue were selected as the three decision factors influencing mental state. Since all three categories achieved over 80% accuracy in emotional state recognition, valence and arousal were chosen as the two factors, each with a level of 3, and the orthogonal array was L6(3). 2 ×2 1 If we use the four-category and nine-category categories of emotion as factors, and combine valence and arousal, there will be too few factors affecting mental state, making it difficult to find factors that significantly influence mental state. According to Taguchi's orthogonal experimental ranking method, the three levels of valence correspond to Lv, Mv, and Hv, denoted by 1, 2, and 3; the three levels of arousal correspond to La, Ma, and Ha, denoted by 1, 2, and 3; and the two levels of fatigue correspond to F (Fatigue) and N (Normal), denoted by 1 and 2.

[0165] To calculate the main effect of each factor in the orthogonal array, it is necessary to calculate the average of the sum of evaluations at each level for each parameter factor, and then find the extreme values ​​of multiple averages to obtain the range. Using L6(3) 2 ×2 1 Taking the valence factor in an orthogonal array as an example, the main effect of valence can be expressed as:

[0166]

[0167]

[0168] In equation (19) above, R V K represents the range of valence. S The sum of evaluations at level s; in the above equation (20), V3A2P1 represents the average of the sum of evaluations at level s; V3A2P1 represents the objective index evaluation corresponding to the valence at level 3, the arousal at level 2, and the fatigue state at level 1, with the evaluation index range being [0,10].

[0169] Plot the levels of each factor on the x-axis and the values ​​below the corresponding levels on the y-axis. Plot a trend graph showing the relationship between experimental factors and experimental indicators, with the vertical axis as the ordinate, such as... Figure 11As shown. Based on the comprehensive comparability of orthogonal arrays, the following conclusions can be drawn from the above calculations and trend charts: the levels of Valance, Arousal, and fatigue are all positively correlated with mental state indicators. Valance has the most significant impact on changes in mental state indicators, with a range of 5.25. Fatigue level determines the level of mental state indicators.

[0170] Based on the principles of Taguchi orthogonal arrays, trend diagrams of experimental relationships, and the above conclusions, a mathematical relationship between objective indicators and three important factors is proposed to address the decision fusion problem.

[0171]

[0172] In equation (21), i,j∈{1,2,3},k∈{1,2}; k i ∈{3,4.25,8.25} represents the coefficients corresponding to the three valence levels, k j ∈{3.75,4.75,7} represents the coefficients corresponding to the three levels of arousal, k k ∈{6.27,8.75} represents the coefficients corresponding to the two fatigue levels; P vi P is the valence-to-accuracy ratio at level i. Aj P represents the accuracy corresponding to the arousal level at the j-th level. Pk It is the accuracy corresponding to the fatigue state at the k-th level.

[0173] This invention designs emotional state recognition and fatigue state determination separately, and uses the Taguchi orthogonal method to fuse emotional and fatigue states, obtaining an objective mathematical expression for mental indicators. While EEG based on physiological states can objectively reflect a driver's emotions and fatigue, and distinguish their mental state, some drivers may experience physical injuries or discomfort, leading to a strong desire to drive or not drive. Violating this desire can result in negative emotions during driving. Therefore, a subjective component should be incorporated into mental state assessment, constructing a mechanism that prioritizes objective indicators and supplements them with subjective ones. The mathematical equation for mental state assessment can be expressed as:

[0174] S = S1 + S2 + b s (twenty one)

[0175] In the above formula, S is the overall mental state evaluation index, S1 is the objective index, S2 is the subjective index, and b s This is the disturbance error of the evaluation index. Under normal circumstances, it is taken as 0. When necessary, the evaluation index is fine-tuned according to the actual situation.

[0176] Subjective evaluations should be intuitive, allowing participants to clearly assess their own physical condition. They mainly consist of three parts: an emotional state score (E), and a...S (1-10 points) Fatigue status score F S (1-10 points) Intensity of driving intention W S (1-10 points), higher scores indicate better mental state and greater suitability for driving. Based on the range of each factor, subjective scores are further weighted, ensuring that the subjective component accounts for 25% of all state indicators. The final mathematical expression for subjective mental state is:

[0177] S2 = 0.425E S +0.25F S +0.325W S (twenty two)

[0178] In the above formula, 0.425 is the coefficient of influence of both valence and arousal on mental state; the coefficient of influence of fatigue on mental state is 0.25, i.e., R0. P The remaining component of / 10.1-0.425-0.25=0.325 is used as the intention influence coefficient.

[0179] Objective indicators include three factors: valence, arousal, and fatigue. Since these are all important components of mental state, the theoretical range for objective indicators is [0, 30]. The theoretical range for objective component indicators, after weighting by influence coefficients, is [0, 10]. Therefore, the theoretical range for the overall mental state indicator is [0, 40], with lower scores indicating worse mental state and higher scores indicating better mental state. In reality, S... min ≈16, S max ≈33, but the actual index range is [16,33].

[0180] Since the three factors are proportional to the mental indicators, but to varying degrees, the average value cannot be used as the threshold indiscriminately; the median value should be chosen instead. Based on the second level of valence and arousal, as well as fatigue and subjective indicators, the threshold should be set as follows:

[0181] τ=21.5+b S (twenty three)

[0182] In the above formula, 21.5 = 4.25 + 4.75 + (6.25 + 8.75) / 2 + 5.

[0183] Whether a person's mental state is up to standard can be determined by comparing mental state evaluation indicators with thresholds.

[0184]

[0185] If the mental state index is greater than or equal to the threshold τ, it can be determined that the relevant staff member is in good mental condition and is able to perform the relevant work; conversely, if the mental state index is less than the threshold, it can be determined that the relevant staff member is in poor mental condition and is not suitable for work.

[0186] Example 2

[0187] This embodiment uses methods for identifying emotional and fatigue states based on electroencephalogram (EEG) signals to assess the mental state of subjects using both subjective and objective indicators, considering both emotional and fatigue states. Specific content includes:

[0188] (1) A spatiotemporal entropy feature with temporal and spatial domains is proposed.

[0189] In the mining industry, miners are professionals engaged in mining, extraction, and mine production. Due to the unique working environment and high-intensity physical labor in mines, a miner's emotional state has a significant impact on their work efficiency and safety.

[0190] A miner's emotional state can be influenced by a variety of factors, such as work pressure, harsh working conditions, work intensity, and the status of mineral resource extraction. Changes in emotional state can have a positive or negative impact on a miner's work performance, thereby affecting mine production efficiency and work safety.

[0191] To better understand and assess miners' emotional states, researchers can collect physiological signals and behavioral data, such as heart rate, electroencephalogram (EEG), and skin conductance, to reflect changes in their emotional state. Then, by analyzing this data in both the temporal and spatial domains, key emotional characteristics can be extracted. Temporal analysis can study the fluctuations and trends in emotions, while spatial analysis can explore the distribution of emotions across different time periods and locations.

[0192] Based on the spatiotemporal entropy characteristics in both the temporal and spatial domains, the temporal changes and spatial distribution of miners' emotions can be comprehensively considered, thus enabling emotion recognition and assessment. This method provides a more comprehensive understanding of miners' emotional states, offering scientific basis and support for safety management and work efficiency in mining operations.

[0193] Based on the concepts of space and entropy, a two-dimensional matrix can represent electrode locations, thus reflecting the spatial relationships between brain electrodes. An image is a two-dimensional matrix represented by rows and columns; the number of rows multiplied by the number of columns equals the number of pixels. A video, on the other hand, is formed by continuously displaying frames of images. Images record information at a specific moment, while videos record temporal information. If we map the values ​​of all channels of the EEG signal at a given moment into a two-dimensional matrix, and calculate the entropy of the information in that matrix at that moment based on the concept of entropy, then connecting more moments consecutively constitutes a new time series. We can liken spatiotemporal entropy to an image, and the new time series to a video. This transformed sequence not only reflects informational characteristics but also reduces computation time through normalization.

[0194] Two types of sentiment data from the DEAP dataset were used to verify the experimental results. Figure 4 This is a 60-second time series diagram of emotional excitement (valence = 9, arousal = 9), which simultaneously displays signals from 32 channels. Figure 5 This is a 60-second time series diagram of the emotional excitement (valence = 1, arousal = 4) during disappointment. The time series diagram shows that the absolute values ​​of the amplitudes of both excitement and disappointment are very large, falling within the range [-300, 300], and exhibiting vertical symmetry, making it difficult to clearly represent their distinct characteristics. Figure 6 It is Figure 4 The multi-channel time series was converted into spatiotemporal approximate entropy and spatiotemporal sample entropy. It can be seen that the amplitude of the new spatiotemporal entropy waveform is in the range [-1,1], and oscillates significantly around 0.5. The spatiotemporal approximate entropy and spatiotemporal sample entropy results of excitement emotion look somewhat similar. Figure 7 They are respectively to Figure 5 The multi-channel time series was transformed into spatiotemporal approximate entropy and spatiotemporal sample entropy. It can be seen that the amplitude of the new spatiotemporal approximate entropy waveform lies in the interval [-0.4, 4], while the amplitude of the new spatiotemporal sample entropy waveform lies in the interval [-4, 1]. The waveforms of the spatiotemporal entropy oscillate within a certain range. Compared to the original sequence, the amplitude of the spatiotemporal entropy feature sequence is very small. The experimental results fully verify that spatiotemporal entropy has a significant feature extraction effect.

[0195] (2) A method for identifying fatigue state based on brain functional connectivity and power spectrum

[0196] In the mining industry, miners typically work long hours and endure harsh, dangerous environments. Prolonged physical labor and demanding tasks can easily lead to fatigue. Fatigue can cause loss of concentration and slowed reaction time, increasing the risk of accidents, reducing work efficiency, and even endangering lives.

[0197] To improve mine production efficiency and ensure miner safety, researchers have begun to focus on detecting miners' fatigue. Fatigue detection can assess a miner's level of fatigue by monitoring physiological signals and behavioral data, such as heart rate, electroencephalogram (EEG), and eye movements. EEG signals are an important indicator of brain activity and can reveal a miner's cognitive state and attention level, thus providing crucial information for identifying fatigue.

[0198] This method, based on the combined use of brain functional connectivity and power spectrum analysis to identify fatigue states, integrates the functional connectivity and spectral characteristics of electroencephalogram (EEG) signals. Brain functional connectivity refers to the degree of functional association between different brain regions. Analyzing brain functional connectivity can reveal the brain's information transmission and processing methods, thereby revealing the miner's cognitive state and brain activity. Power spectrum analysis, which examines the frequency characteristics of EEG signals, can reflect the miner's brain activity state and cognitive load.

[0199] By combining brain functional connectivity and power spectrum features, a more comprehensive understanding of miners' brain function and fatigue levels can be achieved, improving the accuracy and stability of fatigue state identification. This method holds promise for providing an effective means of real-time monitoring and early warning of miners' fatigue, offering a scientific basis for ensuring miners' safety and work efficiency.

[0200] First, the EEG data from the fatigue state dataset were preprocessed. The preprocessed EEG data from 12 groups of test subjects were divided into multiple samples, i.e., 300,000 data points per channel were divided into 300 sub-samples of 1,000 data points each. K-fold cross-validation was used, followed by 3-fold cross-validation. The divided training samples were used as inputs to the DCPM fatigue recognition algorithm, the brain functional connectivity estimation method, and the multi-channel power spectral density normalization method based on brain functional connectivity, respectively. Finally, the recognition rates of the four methods were calculated based on the test set.

[0201] For brain functional connectivity experiments, it is necessary to calculate the channel correlation matrix of the preprocessed data, taking the channel correlation matrix of 34 channels per sample at 1000 sampling points. This results in a 2×30 channel correlation matrix for each subject. Finally, the fatigue state recognition rate is obtained by constructing a hyperplane using SVM, as shown below. Figure 8 As shown.

[0202] Figure 8The data compares the recognition rates of three methods—DCPM algorithm, brain functional connectivity, and the brain functional connectivity-power spectral density joint method proposed in this chapter—for fatigue and normal states. It can be seen that the brain functional connectivity method has the lowest accuracy, with an average accuracy of 65.6% for fatigue and 66.5% for normal states. This is because the correlation matrix data is relatively small, resulting in a limited training set. The DCPM algorithm, which uses average samples as a reference and five weak classifiers as templates for matching, is relatively complex and therefore performs better than the brain connectivity method. The DCPM algorithm achieves an average accuracy of 70.5% for fatigue and 72.7% for normal states. The power spectral density method based on brain functional connectivity normalization achieves an average accuracy of 82.1% for fatigue and 83.2% for normal states. This is because brain functional connectivity reflects the connections between different regions; adjusting the importance components of the power spectra of different regions according to regional relationships results in higher accuracy in identifying fatigue using the normalized power spectrum. Figure 8 The results also verified the effectiveness of this method.

[0203] The fatigue identification results show that the multi-channel power spectrum normalization based on brain connectivity achieves 16.5% and 16.7% higher accuracy than the brain functional connectivity-based methods for fatigue and normal states, respectively, and is also more accurate than the DCPM algorithm. This is because this method combines the methodologies of brain functional connectivity and power spectrum, weighting the power spectrum according to regional importance. It effectively combines two traditional fatigue identification methods.

[0204] (3) A decision-making fusion method based on Taguchi orthogonal method and emotion and fatigue state

[0205] The preprocessed EEG signals were classified for two tasks: emotion recognition and fatigue assessment. Based on the theory of transfer learning, an emotion recognition model trained on the DEAP dataset was used to identify emotions in the preprocessed EEG signals, while a fatigue assessment model trained on a fatigue driving dataset was used to identify fatigue states in the preprocessed EEG signals. The test set for the transfer learning model consisted of the first 300 seconds of resting state data, sampled at a frequency of 512Hz. The data was divided into 30 groups of 10 seconds each, with each group containing 34×5120 data points. A three-class classification system (valence and arousal) was used to identify emotions, while fatigue was classified as a binary system (fatigue and non-fatigue states). The accuracy of the specific model and the transfer learning model was compared. The specific model used EEG data from a video-evoked experiment involving 5 participants as the dataset. The Bi-LSTM classification method based on spatiotemporal entropy feature extraction, as described in this application, was used for emotion recognition, while the multi-channel power spectrum normalization fatigue assessment method based on brain connectivity was used for fatigue assessment.

[0206] The accuracy of transfer learning models is slightly lower than that of proprietary models that require training, due to differences in the data. However, the training time for transfer learning models is much shorter than that for proprietary models, making them easier to apply widely, and the accuracy loss is small and within an acceptable range. Therefore, it is feasible to use transfer learning-based models for assessing the mental state of miners. The mental state of five subjects was assessed using mental state assessment indicators. Subjective indicators included emotional state, fatigue state, and work intention, which were subjective scores given by the subjects based on their own assessments during the experiment. Objective indicators were calculated based on the subjects' EEG signals during the first 300 seconds of resting state, calculating the accuracy of different levels of valence, arousal, and fatigue state, and obtaining the objective indicators using the mathematical relationship of objective evaluation indicators.

[0207] Among the five subjects, Subject 1 had the highest mental state index, and Subject 4 had the lowest. Introducing a threshold of 21.5 as the criterion for judging miners' mental state, Subjects 3 and 4 were considered to have poor mental state and were unsuitable for underground coal mining work. Subjects 1, 2, and 5 all had mental state indices greater than the threshold of 21.5, meeting the criteria for normal and safe work as miners. Although Subject 2 had a neutral mood and low work motivation, their thinking was clear, they were not mentally fatigued, and could safely perform production work. Subject 3, although having a high work motivation, was physically fatigued and emotionally inactive, and would react slowly in the event of an accident during safe production work. The experiment demonstrates that the method of combining subjective and objective indicators proposed in this application is feasible for assessing miners' mental state.

[0208] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various modifications or variations within the scope of the claims, which do not affect the essence of the present invention.

Claims

1. A method for assessing mental state through an orthogonal fusion of emotion, fatigue, and subjective will, characterized in that, Includes the following steps: S1. Collect EEG data, preprocess the raw EEG data, and then use independent component analysis to remove hidden noise components to obtain processed EEG data. S2. Based on the defined spatiotemporal entropy, feature extraction is performed on the emotional EEG data, and a Bi-LSTM emotion state recognition model is built to realize emotion recognition based on EEG spatiotemporal entropy features. The spatiotemporal entropy specifically includes the following steps: Step 1: Map the spatial electrode positions to a matrix Step 2: Construct an m-dimensional vector matrix Step 3: Calculate the standard deviation Calculate each channel at the 1st Standard deviation σt at time (1) In the formula, t represents time, and x i I is the brain potential value of channel i, and I is the number of samples of channel i; This represents the average value of I channel values ​​at time t; Step 4: Calculate the distance between vectors Calculate the sequence in step 2 With all other sequences The distance is calculated by subtracting the values ​​at equal positions in the sequence and taking the absolute value. The value with the largest absolute value is taken as the distance. The distance; Step 5: Statistical analysis of the relationship between distance and threshold statistics Time and distance Less than τ t The number of and the total distance The ratio: (4) In the formula, N is the number of channels, τ t It is an empirical threshold, τ t =0.2σ t ; Approximate entropy-based methods: Take the logarithm, then calculate the average of all the values: (5) The approximate entropy of spacetime is obtained as follows: (6) The above formula Taking formula (5), after steps 1 to 5, a new spatiotemporal entropy sequence is formed. : F 11 ={StApEn(m,τ1), StApEn(m,τ2), …, StApEn(m,τ 40 )} (7) In the above formula, Indicates the point in time at its maximum; The Bi-LSTM model was selected for classification of the new spatiotemporal entropy sequence F11; S3. Extract features from the fatigue EEG dataset, apply the regional relationships of brain functional connectivity to the regional relationships of multi-channel power spectrum, normalize the multi-channel power spectrum density according to the principle of brain functional connectivity, and classify fatigue states using support vector machine. S4. Establish Taguchi orthogonal matrices for valence, arousal, and fatigue status as objective evaluation indicators of mental state to obtain objective mental state indicators; S5. The mental state of the subjects was identified using a two-dimensional mental state assessment method that combines subjective and objective mental indicators.

2. The method for orthogonal integration of emotion, fatigue, and subjective will, and for assessing mental state according to claim 1, is characterized in that, In S1, the preprocessing of the collected EEG data includes: removing noise, power frequency interference and electrocardiogram biological signals; using an FIR filter to filter the EEG data in the range of 0.1 Hz to 40 Hz; setting the global EEG data average value as a reference point; and normalizing the data.

3. The method for orthogonally integrating emotion, fatigue, and subjective will as described in claim 1, and for assessing mental state, is characterized in that... In step 1, the 32 conductive electrode location distribution map is mapped to a 9×9 matrix.

4. The method for orthogonal integration of emotion, fatigue, and subjective will, and for assessing mental state according to claim 1, is characterized in that, In step 2, take Construct two types of vectors The dimensions are respectively and ; with Fz, Cz, Pz, Oz as intermediate axes, serving as the terminals of the vector; where When the first two values ​​of the vector matrix are a pair of symmetrical positions, the final result is obtained. There are a total of 28 vectors. There are a total of 14 vectors.

5. The method for orthogonal integration of emotion, fatigue, and subjective will, and for assessing mental state according to claim 1, is characterized in that, In step 4, different The mathematical expression for the distance when the value is given is as follows: When m=2: (2) When m+1=3: (3)。 6. The method for orthogonal integration of emotion, fatigue, and subjective will, and for assessing mental state according to claim 1, is characterized in that, Step S3, specifically, is a method for identifying fatigue states based on the joint use of brain functional connectivity and power spectrum, including: The correlation between all channels of the EEG was calculated using the Pearson correlation coefficient method to obtain a correlation matrix. The absolute value of the correlation matrix was then taken. (13) In the above formula, |r nu | represents the Pearson correlation coefficient in the nth row and uth column of the multi-channel correlation matrix, where N is the total number of electrode channels. It is the channel correlation matrix after taking the absolute value; Using the average periodogram method, the length of the observed signal is... The data is divided into K segments, each with a length of M. The average power spectral density of the K segments is then calculated. The density function of the average periodogram method is expressed as follows: (14) In the formula, x m k (m) represents the k-th data in the m-th data segment; Indicates angular frequency; It is the dimension of the quantity being sought; The channel normalization process based on the brain functional connectivity correlation matrix can be expressed mathematically as follows: (15) In the formula, W(n) represents the channel. The weights; The single-channel power spectral density of the multi-channel composite is obtained by multiplying the weights corresponding to each channel and summing the results. This can be expressed mathematically as: (16) In the above formula, It is the power spectral density after multi-channel normalization; The preprocessed data is filtered for different frequency ranges of rhythm, and the power spectral density of the four frequency bands is obtained by combining the principle of normalized composite power spectral density: (17) choose The index serves as a fatigue detection indicator; The fatigue EEG dataset was segmented into multiple segments, and then the two types of datasets were spliced ​​together to form a new dataset. The fatigue-based index classification can be classified as a binary classification problem, and a support vector machine was chosen to solve the classification problem.

7. The method for orthogonal integration of emotion, fatigue, and subjective will, and for assessing mental state according to claim 1, is characterized in that, Step S4 specifically includes: The mathematical relationship between objective indicators and three important factors: valence, arousal, and fatigue state: (20) In equation (21), ; The coefficients represent the three levels of valence. The coefficients represent the three levels of arousal. The coefficients representing the two levels of fatigue; It is the first Level of valence corresponds to accuracy. It is the first Levels of arousal correspond to accuracy. It is the first The level of fatigue corresponds to the accuracy rate.

8. The method for orthogonal integration of emotion, fatigue, and subjective will, and for assessing mental state according to claim 1, is characterized in that, In step S5, the mathematical equation for assessing mental state is expressed as follows: (21) (22) In the formula, It is an overall indicator for evaluating mental state. It is an objective indicator. It is a subjective indicator. This represents the perturbation error of the evaluation indicators, which is normally set to 0. When necessary, the evaluation indicators can be fine-tuned based on the actual situation. Es is the emotional state score, Fs is the fatigue state score, and Ws is the intensity of willingness to do this job. Based on the valence and arousal levels of the second level, as well as fatigue and subjective indicators, the threshold should be set as follows: (23) In the above formula ; By comparing mental state evaluation indicators with thresholds, it can be determined whether the mental state is satisfactory. (24) If the mental state index is greater than or equal to the threshold It can be determined that the relevant staff members are in good mental condition and capable of performing their duties; Conversely, if the threshold for mental state is lower than the threshold, it can be determined that the relevant staff members are in a poor state of mental fatigue and are not suitable for work.

Images