A multi-dimensional psychological state real-time evaluation method based on multi-brain region electroencephalogram signals

Abstract

The application discloses a kind of multi-dimensional psychological state real-time evaluation method based on multiple brain regions electroencephalogram signal, it is related to intelligent psychological state evaluation technical field, including, the present application is by the functional connection modeling of multiple brain regions electroencephalogram signal, adaptive weighted fusion of linear feature and multi-scale nonlinear complexity feature is combined, it is realized in time, space and frequency multidimension, to the collaborative characterization of electroencephalogram information, and can be according to brain region activation degree dynamic adjustment the contribution weight of different features, realizes the continuous, real-time and individualized quantitative evaluation of concentration, peace, fatigue and stress and so on multi-dimensional psychological state, and individual baseline calibration and nonlinear mapping mechanism are combined, effectively reduce the influence of individual difference and noise interference on result, to improve the accuracy, stability of psychological state identification.

Description

Technical Field

[0001] This invention relates to the field of intelligent psychological state assessment technology, and in particular to a multidimensional real-time psychological state assessment method based on multi-brain region electroencephalogram (EEG) signals. Background Technology

[0002] With increasing social competition, sub-optimal mental health (such as inattention, anxiety, and mental fatigue) is becoming increasingly common, leading to a surge in demand for objective monitoring and assessment of mental health. Traditional psychological assessments primarily rely on subjective questionnaires (such as scales) or clinical interviews with professional physicians, which have limitations such as high subjectivity, time-consuming processes, and difficulty in real-time continuous monitoring.

[0003] Electroencephalography (EEG) signals, as a non-invasive detection technology for recording the electrophysiological activity of brain neurons, can directly reflect the functional state of the brain and are considered an objective physiological indicator for assessing psychological state.

[0004] In recent years, psychological state assessment based on electroencephalogram (EEG) signals has become a research hotspot due to its high temporal resolution and direct correlation with brain activity.

[0005] However, current technologies still lack a unified real-time assessment framework for multidimensional psychological states. Most current solutions are still based on single states, single-class features, or fixed discrimination logic, making it difficult to simultaneously take into account the collaborative activities of multiple brain regions, multi-frequency difference mechanisms, and individual dynamic changes. In particular, for different psychological dimensions such as focus, calmness, fatigue, and stress, the formation mechanism often involves the joint action of multiple brain regions at different frequency bands. If judgment is made based solely on local power characteristics or static thresholds, it is often difficult to accurately reflect the real-time changes in psychological states. Summary of the Invention

[0006] This invention provides a method for real-time assessment of multidimensional mental states based on multi-brain region electroencephalogram (EEG) signals, which includes: collecting raw voltage sequence data and preprocessing it, and extracting features from the preprocessed data, including linear features and nonlinear features;

[0007] Acquire multi-window EEG voltage sequences, calculate phase lock values ​​between leads in each frequency band, construct an average functional connectivity matrix, and calculate the hub index of each brain region based on the core frequency band weights of a preset psychological dimension.

[0008] The weights of the linear feature vectors generated by the hub index are used, and the linear and nonlinear feature vectors are adaptively weighted and fused based on the dynamic activation threshold and soft threshold mechanism to obtain the global feature vector. This global feature vector is then input into the three-stage transformation function to output the primary state value of each psychological dimension.

[0009] The initial state values ​​are baseline-calibrated to obtain the final state values, and the psychological state is evaluated based on the final state values.

[0010] As a preferred embodiment of the multidimensional real-time assessment method for mental states based on multi-brain region electroencephalogram (EEG) signals described in this invention, wherein: the collection of raw voltage sequence data includes:

[0011] The central processing unit connects to a portable multi-channel EEG acquisition device via Bluetooth Low Energy technology. The device uses the Protobuf protocol to encode and encapsulate the raw EEG voltage sequence, while the receiving end uses the same protocol definition file to decode the data packets and arrange them into the raw voltage sequence in chronological order.

[0012] The original voltage sequence includes prefrontal / frontal lobe lead signal sequence, central zone lead signal, parietal lobe lead signal, occipital lobe lead signal, and temporal lobe lead signal.

[0013] As a preferred embodiment of the multidimensional real-time assessment method for mental states based on multi-brain region electroencephalogram (EEG) signals described in this invention, the preprocessing includes:

[0014] The original voltage sequence is cleaned, including frequency domain filtering and time domain outlier removal, to obtain the cleaned voltage sequence, which is then stored in the data buffer pool.

[0015] The frequency domain filtering refers to notch filtering and bandpass filtering;

[0016] The time-domain outlier cleaning refers to grouping the frequency-domain filtered voltage sequence using a sliding window, and calculating the first and third quartiles of the data within each group of voltage sequences.

[0017] The interquartile range is calculated based on the first and third quartiles to set the upper and lower limits of outliers. Data points that exceed the upper and lower limits of outliers are marked as outliers and replaced with linear interpolation.

[0018] As a preferred embodiment of the multidimensional real-time assessment method for mental states based on multi-brain region electroencephalogram (EEG) signals described in this invention, the step of extracting features from the preprocessed data includes:

[0019] Within the sliding window, the Hanning window function is used to window the signal sequence of each brain region lead, and a fast Fourier transform is performed after windowing to obtain the spectrum;

[0020] Within four standard EEG frequency bands, the spectrum is integrated to obtain the power spectral density, and the power spectral density within the four standard EEG frequency bands is used as the linear feature vector of the window.

[0021] The signal sequences of each brain region lead within the sliding window are subjected to multi-scale coarse-graining processing. The sample entropy of each coarse-grained sequence is calculated, and the complexity index and scale slope difference are fitted from the sample entropy as the nonlinear feature vector of the window.

[0022] As a preferred embodiment of the multidimensional real-time assessment method for mental states based on multi-brain region EEG signals described in this invention, the steps of acquiring multi-window EEG voltage sequences, calculating phase lock values ​​between leads of each frequency band, constructing an average functional connectivity matrix, and calculating the hub index of each brain region based on the core frequency band weights of a preset mental dimension include:

[0023] Select the voltage sequence of the current sliding window and the previous M windows from the data buffer pool. Within the sliding window, calculate the phase lock value of any two leads for each data segment and each frequency band. For all data segments within the current sliding window, calculate the average functional connection matrix of each frequency band.

[0024] Based on the average functional connectivity matrix, the brain region-level connectivity strength is calculated.

[0025] The sum of the connectivity strength of each brain region and the connectivity strength of other brain regions in the average connectivity matrix of the brain region is calculated as the weighted degree centrality of each brain region.

[0026] Based on neuroscience consensus, a core frequency band and weight are assigned to each psychological dimension;

[0027] The psychological dimensions include focus, calmness, fatigue, and stress;

[0028] The weighted degree centrality of each brain region across all its core frequency bands is weighted and summed according to the frequency band weights to obtain the pivot index of each brain region for each psychological dimension.

[0029] As a preferred embodiment of the multidimensional real-time assessment method for mental states based on multi-brain region EEG signals described in this invention, wherein: the weights of the linear feature vector generated by the pivot index are adaptively weighted and fused with the linear and nonlinear feature vectors based on a dynamic activation threshold and soft threshold mechanism to obtain a global feature vector, which is then input into a three-stage transformation function to output the primary state value of each mental dimension, including:

[0030] For each psychological dimension, the pivot index of all brain regions is subjected to max-min normalization and used as the linear feature weights.

[0031] An activation threshold is set. The nonlinear features of a brain region are given significant weight only when the pivot index of the brain region is greater than the activation threshold. When the pivot index of the brain region is lower than the activation threshold, a soft threshold mechanism is used, and its weight is smoothly decayed to close to 0 according to the Sigmoid curve.

[0032] The significant weight refers to the difference between the brain region hub index and the activation threshold, and is normalized.

[0033] For the linear feature vectors of each psychological dimension and each brain region, a weighted fusion is performed using linear feature weights to obtain a global linear feature vector.

[0034] Similarly, nonlinear feature vectors are weighted and fused using nonlinear feature weights to obtain global nonlinear feature vectors.

[0035] For each psychological dimension, a three-stage transition function is designed to calculate the initial state value of each psychological dimension. In stage 1, linear features are saturated; in stage 2, nonlinear complexity is modulated by gain; and in stage 3, global activation regulation of cross-scale dynamics is performed.

[0036] As a preferred embodiment of the multidimensional real-time assessment method for mental states based on multi-brain region electroencephalogram (EEG) signals described in this invention, the step of baseline calibration of the primary state values ​​to obtain the final state values ​​includes:

[0037] After the system starts, an evaluation phase is conducted, which automatically monitors the initial state values ​​of the first W windows. When the fluctuation of the state values ​​of K consecutive windows is less than a preset threshold, it is determined that the user is in a stable resting state, and the median of the state values ​​of these K windows is used as the user's personal initial baseline value.

[0038] For each time window in the evaluation phase, calculate its relative score with respect to the individual baseline, and map the relative score to the final output interval [0, 100] through a sigmoid function to obtain the final state value.

[0039] As a preferred embodiment of the multidimensional real-time assessment method for psychological states based on multi-brain region electroencephalogram (EEG) signals described in this invention, the psychological state evaluation based on the final state value includes:

[0040] Based on the statistical characteristics of large sample norm distribution and users' personal historical data, low and high thresholds are set for focus, calmness, fatigue, and stress, respectively. Based on the final state value of the psychological dimension, the user's corresponding psychological state level is determined.

[0041] The beneficial effects of this invention are as follows: By modeling the functional connectivity of EEG signals from multiple brain regions and combining adaptive weighted fusion of spectral linear features and multi-scale nonlinear complexity features, this invention achieves a collaborative characterization of EEG information across multiple dimensions of time, space, and frequency. It can also dynamically adjust the contribution weights of different features according to the degree of brain region activation, enabling continuous, real-time, and individualized quantitative assessment of multidimensional psychological states such as focus, calmness, fatigue, and stress. Furthermore, by combining individual baseline calibration and nonlinear mapping mechanisms, it effectively reduces the impact of individual differences and noise interference on the results, thereby improving the accuracy and stability of psychological state recognition. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart of a multidimensional real-time assessment method for psychological states based on multi-brain region EEG signals, as described in Example 1.

[0044] Figure 2 This is a flowchart of linear and nonlinear feature extraction in Example 1.

[0045] Figure 3 This is a flowchart of the functional connection matrix construction and hub index calculation in Example 1.

[0046] Figure 4 This is a flowchart of adaptive fusion, state transition, and baseline calibration in Example 1. Detailed Implementation

[0047] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0048] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0049] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0050] Example 1, referring to Figures 1 to 4 This is the first embodiment of the present invention, which provides a method for real-time assessment of multidimensional mental states based on multi-brain region electroencephalogram (EEG) signals, including the following steps:

[0051] S1. Collect and preprocess the raw voltage sequence data, and extract features from the preprocessed data, including linear and nonlinear features;

[0052] Preferably, the central processing unit (such as a mobile phone, computer, etc.) connects to a portable multi-channel EEG acquisition device via Bluetooth Low Energy technology. The device uses the Protobuf protocol to encode and encapsulate the raw EEG voltage sequence, and the receiving end uses the same protocol definition file to decode the data packets and arrange them into the raw voltage sequence in chronological order.

[0053] The aforementioned original voltage sequences include prefrontal / frontal lobe lead signal sequences (responsible for higher cognition and attention), central lobe lead signals (sensory-motor), parietal lobe lead signals (spatial processing), occipital lobe lead signals (visual processing), and temporal lobe lead signals (auditory and memory).

[0054] Furthermore, the original voltage sequence is cleaned, including frequency domain filtering and time domain outlier removal, to obtain the cleaned voltage sequence, which is then stored in the data buffer pool.

[0055] The frequency domain filtering refers to notch filtering and bandpass filtering;

[0056] The time-domain outlier cleaning refers to grouping the frequency-domain filtered voltage sequence using a sliding window, and calculating the first and third quartiles of the data within each group of voltage sequences.

[0057] The interquartile range is calculated based on the first and third quartiles to set the upper and lower limits of outliers. Data points that exceed the upper and lower limits of outliers are marked as outliers and replaced with linear interpolation.

[0058] Furthermore, within the sliding window, the Hanning window function is used to window the signal sequence of each brain region lead to reduce spectral leakage, and a fast Fourier transform is performed after windowing to obtain the spectrum;

[0059] The sliding window length is set to 4 seconds, and the sliding step size is set to 1 second.

[0060] Within four standard EEG frequency bands, the spectrum is integrated to obtain the power spectral density, and the power spectral density within the four standard EEG frequency bands is used as the linear feature vector of the window.

[0061] The four standard EEG bands mentioned above refer to the delta band, theta band, alpha band, and beta band, respectively. The delta band ranges from 0.5 to 4 Hz, the theta band from 4 to 8 Hz, the alpha band from 8 to 13 Hz, and the beta band from 13 to 30 Hz. These data are set with reference to international clinical and research standards.

[0062] The signal sequences of each brain region lead within the sliding window are subjected to multi-scale coarse-graining processing (scale factor is 1-5; specifically, the average of g adjacent data points is taken as a new coarse-grained data point, where g is greater than 2 and less than the scale factor). The sample entropy of each coarse-grained sequence is calculated, and the complexity index and scale slope difference are fitted from the sample entropy as the nonlinear feature vector of the window.

[0063] S2. Obtain multi-window EEG voltage sequences, calculate phase lock values ​​between leads of each frequency band, construct an average functional connectivity matrix, and calculate the hub index of each brain region based on the core frequency band weights of the preset psychological dimension.

[0064] Preferably, the voltage sequence of the current sliding window and the preceding M windows (e.g., a total of 2-5 minutes) are selected from the data buffer pool. Within this sliding window, the phase lock value of any two leads is calculated for each data segment and each frequency band, using the following formula:

[0065] ;

[0066] in, To be in a specific brainwave frequency band Above, the phase lock value between channel p and channel q. Values , This represents the total number of sampling points in the current sliding window. for index, The imaginary unit, To determine the EEG signals of channels p and q at time t within a specified EEG frequency band. The instantaneous phase difference is obtained by preprocessing the raw multi-channel EEG data, performing a Hilbert transform, and then extracting the instantaneous phase difference of each channel. The difference between channel p and channel q is then calculated. ;

[0067] For all data segments within the current sliding window, calculate the average functional connectivity matrix for each frequency band, which serves as the representative network state for that user at the current moment.

[0068] The brain region-level connectivity strength is calculated based on the average functional connectivity matrix, using the following formula:

[0069] ;

[0070] in, For brain regions in the frequency band The average connectivity strength between the brain region B and brain region B and They belong to brain regions respectively And the number of leads in brain region B, The element value in the p-th row and q-th column of the average functional connectivity matrix is ​​the average phase lock value between lead p and lead q. In the same frequency band, all brain region-level connectivity strengths are combined into a brain region-level average connectivity matrix.

[0071] The sum of the connectivity strength of each brain region and the connectivity strength of other brain regions in the average connectivity matrix of the brain region is calculated as the weighted degree centrality of each brain region.

[0072] Based on neuroscience consensus, a core frequency band and weight are assigned to each psychological dimension;

[0073] The psychological dimensions include focus, calmness, fatigue, and stress.

[0074] The above specifies the most relevant frequency band for each psychological dimension. Specifically, the core frequency bands for focus are the theta frequency band (weighted as 0.3) and the beta frequency band (weighted as 0.7), while the alpha frequency band is set as the inhibition frequency band (which is used as the denominator in the ratio calculation to play a reverse adjustment role).

[0075] The core frequency band for calmness is the α band (weight 1, for example);

[0076] The core frequency bands for fatigue assessment are the θ band (weighted as 0.6) and the δ band (weighted as 0.4).

[0077] The core pressure frequency bands are the θ band (weighted as 0.5) and the β band (weighted as 0.5).

[0078] The weighted degree centrality of each brain region across all its core frequency bands is weighted and summed according to the frequency band weights to obtain the pivot index of each brain region for each psychological dimension.

[0079] S3. The weights of the linear feature vectors generated by the pivot index are used to adaptively weight and fuse the linear and nonlinear feature vectors based on the dynamic activation threshold and soft threshold mechanism to obtain the global feature vector, which is then input into the three-stage transformation function to output the primary state value of each psychological dimension.

[0080] Preferably, the pivot index of all brain regions for each psychological dimension is subjected to max-min normalization and used as the linear feature weight;

[0081] An activation threshold is set (a dynamic threshold based on the statistical distribution, such as the 50%-60% quantile, to avoid excessive sparsity). The nonlinear features of a brain region are given significant weight only when the hub index of the brain region is greater than the activation threshold. When the hub index of the brain region is lower than the activation threshold, a soft thresholding mechanism is used, and its weight is smoothly decayed to close to 0 according to the Sigmoid curve, rather than being directly truncated to 0.

[0082] The significant weight refers to the difference between the brain region hub index and the activation threshold, and is normalized.

[0083] For the linear feature vectors of each psychological dimension and each brain region, a weighted fusion is performed using linear feature weights to obtain a global linear feature vector.

[0084] Similarly, nonlinear feature vectors are weighted and fused using nonlinear feature weights to obtain global nonlinear feature vectors.

[0085] For each psychological dimension, a transition function is designed to calculate the initial state value of each psychological dimension, including three stages;

[0086] Phase 1: Perform saturation processing on linear features;

[0087] Phase 2 involves performing gain modulation with nonlinear complexity.

[0088] Phase 3 involves global activation regulation of cross-scale dynamics;

[0089] Taking focus level as an example, the conversion function is as follows:

[0090] ;

[0091] ;

[0092] ;

[0093] ;

[0094] in, In the time window The initial state value of focus. For conversion functions, , and In the time window The global linear feature vector of focus, the global complexity index, and the global scale slope difference are used to construct the global nonlinear feature vector, where the global complexity index and the global scale slope difference together constitute the global nonlinear feature vector. , and These are the saturation core term function of focus, the complexity gain term function, and the scale dynamics adjustment term function, respectively. The complexity gain intensity coefficient is an adjustable scalar parameter. It is the hyperbolic tangent function. To focus on power ratio, The slope parameter of the saturation curve. , and In the time window Yes, beta band frequency band and The global frequency band power value is obtained by aggregating the power spectral density at the corresponding frequency band. Use small positive numbers to prevent division by zero. For the Sigmoid function, The gain parameter, such as 8, provides a fixed steepness sufficient to produce a clear distinction near the threshold. The Sigmoid function is close to saturation when the input is ±2 (0.88 and 0.12). mean and A difference of ±0.25 units is sufficient to achieve this effect. This is a scaling factor for the Sigmoid function, with values ​​such as 0.25 (corresponding to the typical standard deviation range of the complexity index), used to scale the physical difference to the optimal response region of the Sigmoid function. The complexity threshold required for focus. It is an exponential function. This is a sensitivity parameter for adjusting focus, with values ​​ranging from 3.0 to 5.0, based on engineering heuristics. The mathematical properties of the function (the boundary between the saturation region and the linear region) and the typical fluctuation amplitude of the standardized EEG features (usually 0.2~0.4) were obtained by fitting a large amount of experimental data through trial and error and grid search. The offset parameter for the scale slope;

[0095] The slope of the saturation curve mentioned above is set using a dynamic range matching method. Specifically, the general value for the slope of the saturation curve is set to 2. Data is collected from the user in a "focused" state, and the average power ratio is calculated. Then, the slope is adjusted... ,make Introduce a linear scaling factor ,in For the power ratio under high concentration, then ;

[0096] The complexity threshold is taken as the P40 (40th percentile) of the statistical distribution of the global complexity index over the past 5-10 minutes as a dynamic threshold. This means that when a user's real-time complexity is lower than the complexity corresponding to their lower performance level (bottom 40%) in highly focused tasks, the system considers the complexity insufficient and the gain on focus is weakened.

[0097] The complexity gain strength is set to a classic value of 0.4. This value ensures that the gain contribution of nonlinear complexity features does not exceed 40% of the contribution of core linear features, which conforms to the design principle of "linear dominance and nonlinear assisted". The value range is 0.3 to 0.5.

[0098] The scale slope offset is set using the median positioning method, specifically by using the user's historical data from the past 30 minutes to 2 hours to calculate the median of the distribution as the scale slope offset.

[0099] S4. Baseline calibration is performed on the initial state values ​​to obtain the final state values, and psychological state evaluation is conducted based on the final state values.

[0100] Preferably, after the system starts, an evaluation phase is performed, automatically monitoring the initial state values ​​of the first W windows (e.g., 10). When the state value fluctuation of K consecutive windows (e.g., 3) is less than a preset threshold (this threshold is set based on the empirical value of the resting state characteristics of physiological signals, such as 0.05 ~ 0.10), it is determined that the user is in a stable resting state, and the median of the state values ​​of these K windows is used as the user's personal initial baseline value.

[0101] For each time window in the assessment phase, calculate its relative score relative to the individual baseline, using the formula:

[0102] ;

[0103] ;

[0104] ;

[0105] ;

[0106] in, , , and In the time window The relative scores of focus, calmness, fatigue, and stress. , , and In the time window The initial values ​​of focus, calmness, fatigue, and stress. , , and These are the initial baseline values ​​for focus, calmness, fatigue, and stress, respectively.

[0107] The relative score is mapped to the final output interval [0, 100] using a sigmoid function (such as a modified sigmoid) to obtain the final state value, as shown in the formula:

[0108] ;

[0109] in, In the time window Psychological dimension The final state value, Including focus, calmness, fatigue, and stress. In the time window Psychological dimension The relative score.

[0110] Furthermore, based on the statistical characteristics of large sample norm distribution and users' personal historical data, low and high thresholds for focus, calmness, fatigue and stress are set respectively. Based on the final state value of the psychological dimension, the user's corresponding psychological state level is determined.

[0111] The above psychological state levels are divided into three categories: high, medium, and low, and are judged by low and high thresholds.

[0112] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for real-time assessment of multidimensional mental states based on multi-regional electroencephalogram (EEG) signals, characterized in that, include: Raw voltage sequence data is collected and preprocessed. Feature extraction is performed on the preprocessed data, including linear and nonlinear features. The system acquires multi-window EEG voltage sequences, calculates the phase lock values ​​between leads in each frequency band, constructs an average functional connectivity matrix, and calculates the hub index of each brain region based on the core frequency band weights of the preset psychological dimension. This includes selecting the voltage sequences of the current sliding window and the previous M windows from the data buffer pool, calculating the phase lock values ​​between any two leads for each data segment and each frequency band within the sliding window, and calculating the average functional connectivity matrix for each frequency band for all data segments within the current sliding window. Based on the average functional connectivity matrix, the brain region-level connectivity strength is calculated. In the same frequency band, all brain region-level connectivity strengths are combined into a brain region-level average connectivity matrix; The sum of the connectivity strength of each brain region and the connectivity strength of other brain regions in the average connectivity matrix of the brain region is calculated as the weighted degree centrality of each brain region. Based on neuroscience consensus, a core frequency band and weight are assigned to each psychological dimension; The psychological dimensions include focus, calmness, fatigue, and stress; The weighted degree centrality of each brain region across all its core frequency bands is weighted and summed according to the frequency band weights to obtain the pivot index of each brain region for each psychological dimension. Based on the dynamic activation threshold and soft threshold mechanism, the linear feature vector and the nonlinear feature vector are adaptively weighted and fused to obtain the global feature vector, which is then input into the three-stage transformation function to output the primary state value of each psychological dimension. This includes the max-min normalization of the hub index of all brain regions in each psychological dimension as the linear feature weight. An activation threshold is set. The nonlinear features of a brain region are given significant weight only when the pivot index of the brain region is greater than the activation threshold. When the pivot index of the brain region is lower than the activation threshold, a soft threshold mechanism is used, and its weight is smoothly decayed to close to 0 according to the Sigmoid curve. The significant weight refers to the difference between the brain region hub index and the activation threshold, and is normalized. For the linear feature vectors of each psychological dimension and each brain region, a weighted fusion is performed using linear feature weights to obtain a global linear feature vector. Similarly, nonlinear feature vectors are weighted and fused using nonlinear feature weights to obtain global nonlinear feature vectors. For each psychological dimension, a three-stage transition function is designed to calculate the primary state value of each psychological dimension. In stage 1, linear features are saturated; in stage 2, gain modulation of nonlinear complexity is performed; and in stage 3, global activation regulation of cross-scale dynamics is performed. The initial state values ​​are baseline-calibrated to obtain the final state values, and the psychological state is evaluated based on the final state values.

2. The method for real-time assessment of multidimensional mental states based on multi-brain region EEG signals as described in claim 1, characterized in that, The collection of raw voltage sequence data includes: The central processing unit connects to a portable multi-channel EEG acquisition device via Bluetooth Low Energy technology. The device uses the Protobuf protocol to encode and encapsulate the raw EEG voltage sequence, while the receiving end uses the same protocol definition file to decode the data packets and arrange them into the raw voltage sequence in chronological order. The original voltage sequence includes prefrontal / frontal lobe lead signal sequence, central zone lead signal, parietal lobe lead signal, occipital lobe lead signal, and temporal lobe lead signal.

3. The method for real-time assessment of multidimensional psychological states based on multi-brain region EEG signals as described in claim 2, characterized in that, The preprocessing includes: The original voltage sequence is cleaned, including frequency domain filtering and time domain outlier removal, to obtain the cleaned voltage sequence, which is then stored in the data buffer pool. The frequency domain filtering refers to notch filtering and bandpass filtering; The time-domain outlier cleaning refers to grouping the frequency-domain filtered voltage sequence using a sliding window, and calculating the first and third quartiles of the data within each group of voltage sequences. The interquartile range is calculated based on the first and third quartiles to set the upper and lower limits of outliers. Data points that exceed the upper and lower limits of outliers are marked as outliers and replaced with linear interpolation.

4. The method for real-time assessment of multidimensional psychological states based on multi-brain region EEG signals as described in claim 3, characterized in that, The step of extracting features from the preprocessed data includes: Within the sliding window, the Hanning window function is used to window the signal sequence of each brain region lead, and a fast Fourier transform is performed after windowing to obtain the spectrum; Within four standard EEG frequency bands, the spectrum is integrated to obtain the power spectral density, and the power spectral density within the four standard EEG frequency bands is used as the linear feature vector of the window. The signal sequences of each brain region lead within the sliding window are subjected to multi-scale coarse-graining processing. The sample entropy of each coarse-grained sequence is calculated, and the complexity index and scale slope difference are fitted from the sample entropy as the nonlinear feature vector of the window.

5. The method for real-time assessment of multidimensional psychological states based on multi-regional EEG signals as described in claim 4, characterized in that, The baseline calibration of the primary state value to obtain the final state value includes: After the system starts, an evaluation phase is conducted, which automatically monitors the initial state values ​​of the first W windows. When the fluctuation of the state values ​​of K consecutive windows is less than a preset threshold, it is determined that the user is in a stable resting state, and the median of the state values ​​of these K windows is used as the user's personal initial baseline value. For each time window in the evaluation phase, its relative score relative to the individual baseline is calculated. The relative score is then mapped to the final output interval [0, 100] through a sigmoid function to obtain the final state value.

6. The method for real-time assessment of multidimensional mental states based on multi-brain region EEG signals as described in claim 5, characterized in that, The psychological state evaluation based on the final state value includes: Based on the statistical characteristics of large-sample norm distribution and users' personal historical data, low and high thresholds are set for focus, calmness, fatigue, and stress. Based on the final state value of the psychological dimension, the user's corresponding psychological state level is determined.

Images