Multi-scene fatigue state detection system and method based on electroencephalogram signals and wearable cap body

Abstract

The invention provides a multi-scene fatigue state detection system and method based on an electroencephalogram signal and a wearable cap body, and relates to the technical field of biological signal detection, and the system comprises a processing module which is used for preprocessing an original electroencephalogram signal of a scalp area of a user to obtain a conditioned electroencephalogram signal; performing contact state analysis on the conditioned electroencephalogram signals, and calculating a contact stability index corresponding to each flexible dry electrode to determine screened effective electroencephalogram signals; and the extraction module is used for extracting a first phase lock point, a second phase lock point and a third phase lock point from the electrode channels corresponding to the screened effective electroencephalogram signals, constructing an elliptical space potential field curved surface and calculating an electrode distribution configuration entropy value. According to the invention, detection, comprehensive evaluation and graded alarm of the fatigue state of the user in multiple scenes are realized, the judgment result can be synchronously transmitted, the comprehensiveness and accuracy of fatigue monitoring are improved, and the operation safety of the user is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of biosignal detection technology, and in particular to a multi-scenario fatigue state detection system and method based on electroencephalogram (EEG) signals and a wearable cap. Background Technology

[0002] In scenarios such as long-distance driving, high-intensity work, or sports training, human fatigue is a significant factor leading to decreased efficiency, operational errors, and even safety accidents. Currently, several technical solutions for fatigue detection have been proposed. Some existing fatigue detection methods rely on visual analysis of facial expressions (such as blinking and nodding) or operational behaviors (such as steering wheel grip). However, these methods may be affected by changes in ambient light or obstruction from clothing (such as glasses, masks, and helmet brims), resulting in room for improvement in detection stability. Furthermore, visual characteristics often only show significant changes when fatigue is already substantial, potentially leading to a delay in early warning and insufficient time for intervention.

[0003] Other technologies attempt to use EEG signals for fatigue assessment. EEG signals can directly reflect the state of neural activity in the brain, containing physiological information related to alertness, cognitive load, and fatigue depth. Traditional EEG acquisition devices usually have a large number of electrodes, are relatively cumbersome to wear, require the application of conductive gel, and may not be aesthetically pleasing in everyday work scenarios. Their applicability in scenarios requiring long-term continuous wear or with certain requirements for comfort and concealment still needs further verification. Currently, some consumer-grade EEG devices have improved in terms of portability, but there is still room for exploration in terms of deep integration with wearable carriers required for daily work (such as safety helmets and work caps). Further optimization is still possible in terms of comfort during long-term wear, long-term stability of electrode contact, and continuous availability in multiple scenarios.

[0004] For example, in long-haul freight scenarios, drivers need to wear safety helmets or work hats for extended periods. If traditional multi-electrode EEG acquisition equipment is used, the wearing process may be relatively complicated and the appearance may be quite noticeable, which may affect the driver's acceptance. On the other hand, fatigue monitoring systems based solely on facial vision may have their detection effectiveness affected when driving at night or when the driver is wearing sunglasses or a mask. Summary of the Invention

[0005] This invention provides a multi-scenario fatigue state detection system and method based on electroencephalogram (EEG) signals and a wearable cap, enhancing the adaptability of wearable devices in practical use.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, a multi-scenario fatigue state detection system based on electroencephalogram (EEG) signals and a wearable cap includes: The processing module is used to preprocess the raw EEG signals from the user's scalp area to obtain conditioned EEG signals; and to perform contact state analysis on the conditioned EEG signals to calculate the contact stability index corresponding to each flexible dry electrode in order to determine the effective EEG signals after screening. The extraction module is used to extract the first phase lock point, the second phase lock point, and the third phase lock point from the electrode channels corresponding to the screened valid EEG signals, construct the elliptical space potential field surface, and calculate the electrode distribution configuration entropy value. The fusion module is used to dynamically reconstruct the fusion weights of the selected effective EEG signals based on the electrode distribution configuration entropy value, obtain the reconstructed fusion weights, and perform weighted fusion of the selected effective EEG signals to generate a comprehensive EEG signal sequence. The analysis module is used to perform time-frequency and time-domain waveform analysis on the comprehensive EEG signal sequence, calculate the first fatigue characterization index and determine the characteristics of peripheral behavioral fatigue symptoms; calculate the user's continuous working time from the last rest to the current moment based on the characteristics of peripheral behavioral fatigue symptoms, and obtain a multi-dimensional environmental and working time sequence feature set; The evaluation module is used to input the first fatigue characterization index, peripheral behavioral fatigue symptom characteristics, and multi-dimensional environmental and operational time sequence feature set into the pre-trained evaluation model to obtain the comprehensive fatigue level judgment result and trigger the corresponding graded alarm measures; the comprehensive fatigue level judgment result is transmitted to the external monitoring platform or local monitoring terminal.

[0007] Secondly, a multi-scenario fatigue state detection method based on electroencephalogram (EEG) signals and a wearable cap includes: Step 1: Preprocess the raw EEG signal from the user's scalp area to obtain the conditioned EEG signal; perform contact state analysis on the conditioned EEG signal and calculate the contact stability index corresponding to each flexible dry electrode to determine the effective EEG signal after screening. Step 2: Extract the first phase lock point, the second phase lock point, and the third phase lock point from the electrode channels corresponding to the selected valid EEG signals, construct the elliptical space potential field surface, and calculate the electrode distribution configuration entropy value. Step 3: Based on the electrode distribution configuration entropy value, dynamically reconstruct the fusion weights of the selected effective EEG signals to obtain the reconstructed fusion weights, and then perform weighted fusion of the selected effective EEG signals to generate a comprehensive EEG signal sequence. Step 4: Perform time-frequency and time-domain waveform analysis on the comprehensive EEG signal sequence, calculate the first fatigue characterization index and determine the peripheral behavioral fatigue symptoms; calculate the user's continuous working time from the last rest to the current moment based on the peripheral behavioral fatigue symptoms, and obtain a multi-dimensional environmental and working time sequence feature set; Step 5: Input the first fatigue characterization index, peripheral behavioral fatigue symptom characteristics, and multidimensional environmental and operational time sequence feature set into the pre-trained evaluation model to obtain the comprehensive fatigue level judgment result and trigger the corresponding graded alarm measures; transmit the comprehensive fatigue level judgment result to the external monitoring platform or local monitoring terminal.

[0008] The above-described solution of the present invention has at least the following beneficial effects: Raw EEG signals are acquired using flexible dry electrodes, and the effectiveness of the signals is screened using a contact stability index. This effectively eliminates interference data caused by poor electrode contact, improving the reliability of raw physiological signals and ensuring stable acquisition under different wearing conditions, thus enhancing the adaptability of wearable devices in practical use. An elliptical spatial potential field surface is constructed based on fixed phase lock points, and the electrode distribution configuration entropy value is calculated. This achieves a quantitative description of the spatial distribution characteristics of the electrodes, avoiding the influence of single-channel signal anomalies on the overall detection results. This makes the analysis of EEG signals more closely aligned with the spatial electric field distribution patterns of the head, improving the rationality and accuracy of feature extraction. Signal fusion weights are dynamically reconstructed based on the electrode distribution configuration entropy value, allowing for the self-reconstruction of effective EEG signals. Adaptive weighted fusion can allocate contribution based on the actual acquisition quality of each electrode channel, reducing redundant information and noise interference. It simultaneously combines time-frequency domain fatigue characterization indicators with peripheral behavioral fatigue symptoms for analysis, taking into account both central nervous system physiological signals and external behavioral manifestations. It also incorporates continuous working time and multi-dimensional environmental and working time sequence characteristics, breaking through the limitations of judging fatigue by single EEG features. It adopts a pre-trained evaluation model combined with a cross-attention mechanism to achieve adaptive deep fusion of multi-dimensional features, which can distinguish different fatigue levels and provide graded alarm measures. It can trigger reminders differently according to the degree of fatigue, avoiding excessive alarms from affecting normal work, and providing timely warnings in the case of severe fatigue, ensuring user work safety and physical health. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of a multi-scenario fatigue state detection system based on electroencephalogram (EEG) signals and a wearable cap, provided by an embodiment of the present invention.

[0010] Figure 2 This is a flowchart illustrating a multi-scenario fatigue state detection method based on electroencephalogram (EEG) signals and a wearable cap, as provided in an embodiment of the present invention. Detailed Implementation

[0011] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0012] like Figure 1 As shown, an embodiment of the present invention proposes a multi-scenario fatigue state detection system based on electroencephalogram (EEG) signals and a wearable cap, comprising: The processing module is used to preprocess the raw EEG signals from the user's scalp area to obtain conditioned EEG signals; and to perform contact state analysis on the conditioned EEG signals to calculate the contact stability index corresponding to each flexible dry electrode in order to determine the effective EEG signals after screening. The extraction module is used to extract the first phase lock point, the second phase lock point, and the third phase lock point from the electrode channels corresponding to the screened valid EEG signals, construct the elliptical space potential field surface, and calculate the electrode distribution configuration entropy value. The fusion module is used to dynamically reconstruct the fusion weights of the selected effective EEG signals based on the electrode distribution configuration entropy value, obtain the reconstructed fusion weights, and perform weighted fusion of the selected effective EEG signals to generate a comprehensive EEG signal sequence. The analysis module is used to perform time-frequency and time-domain waveform analysis on the comprehensive EEG signal sequence, calculate the first fatigue characterization index and determine the characteristics of peripheral behavioral fatigue symptoms; calculate the user's continuous working time from the last rest to the current moment based on the characteristics of peripheral behavioral fatigue symptoms, and obtain a multi-dimensional environmental and working time sequence feature set; The evaluation module is used to input the first fatigue characterization index, peripheral behavioral fatigue symptom characteristics, and multi-dimensional environmental and operational time sequence feature set into the pre-trained evaluation model to obtain the comprehensive fatigue level judgment result and trigger the corresponding graded alarm measures; the comprehensive fatigue level judgment result is transmitted to the external monitoring platform or local monitoring terminal.

[0013] In this embodiment of the invention, flexible dry electrodes are used to acquire raw EEG signals, and the effectiveness of the signals is screened using a contact stability index. This effectively eliminates interference data caused by poor electrode contact, improves the reliability of raw physiological signals, adapts to stable acquisition under different wearing conditions, and enhances the adaptability of wearable devices in practical use. An elliptical spatial potential field surface is constructed based on fixed phase lock points, and the electrode distribution configuration entropy value is calculated. This achieves a quantitative description of the spatial distribution characteristics of the electrodes, avoiding the influence of single-channel signal abnormalities on the overall detection results. This makes the analysis of EEG signals more consistent with the spatial electric field distribution law of the head, improving the rationality and accuracy of feature extraction. The signal fusion weights are dynamically reconstructed based on the electrode distribution configuration entropy value to effectively analyze EEG signals. The system employs adaptive weighted fusion to allocate contribution based on the actual acquisition quality of each electrode channel, reducing redundant information and noise interference. It simultaneously combines time-frequency domain fatigue characterization indicators with peripheral behavioral fatigue symptoms for analysis, taking into account both central nervous system physiological signals and external behavioral manifestations. Furthermore, it incorporates continuous work duration and multi-dimensional environmental and work sequence characteristics, overcoming the limitations of judging fatigue based on single EEG features. A pre-trained evaluation model combined with a cross-attention mechanism achieves adaptive deep fusion of multi-dimensional features, distinguishing different fatigue levels and providing tiered alarm measures. Alerts can be triggered differently based on fatigue levels, avoiding excessive alarms that could disrupt normal work while providing timely warnings in cases of severe fatigue, ensuring user safety and health.

[0014] In a preferred embodiment of the present invention, step 100a involves preprocessing the raw EEG signal from the user's scalp region to obtain a conditioned EEG signal. The raw EEG signal from the user's scalp region is acquired using flexible dry electrodes integrated into the wearable cap at the center of the forehead, the left temporal region, and the right temporal region. Specifically, after the user wears the wearable cap with integrated flexible dry electrodes, the system continuously acquires the raw EEG signal corresponding to the user's scalp region using the flexible dry electrodes located inside the cap at the center of the forehead, the left temporal region, and the right temporal region. After acquisition, the system sequentially performs preprocessing operations on the acquired raw EEG signal. First, the weak raw EEG signal is linearly amplified using an instrumentation amplification circuit, increasing the raw signal intensity. The signal amplitude at the microvolt level is amplified to the 0.5V to 5V range, and then a notch filter circuit is used to filter out power frequency interference introduced by the power grid environment, blocking the interference components corresponding to the power grid frequency from entering the subsequent processing stage. Baseline drift is eliminated by a baseline correction algorithm. Specifically, a moving average fitting method is used to calculate the slow shift trend of the signal. First, the sliding window length is set to 100ms. Using this window as a unit, the amplified and notch-filtered EEG signal is traversed point by point. The average value of all signal sampling points in each sliding window is calculated. The average values ​​of all windows are connected sequentially to form a smooth curve. This smooth curve is the slow shift trend of the signal. Then, the difference between the original conditioning signal and this shift trend is calculated to eliminate the baseline rise or fall caused by slow changes in scalp potential and electrode polarization shift. By adaptively filtering out irrelevant noise, signal components above 30Hz are identified as high-frequency noise. This upper limit of frequency is set according to the effective frequency band distribution of EEG signals. EEG characteristics are mainly concentrated in the range below 30Hz. Components above this frequency are mostly motion artifacts and invalid interference caused by environmental electromagnetic radiation. After filtering and suppressing the above high-frequency components, the environmental interference mixed in the signal and the inherent interference generated by the device itself are removed step by step, and finally the conditioned EEG signal is obtained.

[0015] Step 100b involves analyzing the contact state of the conditioned EEG signals and calculating the contact stability index for each flexible dry electrode to determine the effective EEG signals after screening. This includes: analyzing the contact state of the conditioned EEG signals according to the electrode channels corresponding to each flexible dry electrode; continuously collecting the contact impedance value sequence of each electrode channel within a preset time window; calculating the mean and variance of the contact impedance value sequence; and calculating the contact stability index of each electrode channel based on the mean and variance. Specifically, the system performs independent contact state analysis for each flexible dry electrode channel corresponding to each conditioned EEG signal. The system uses a continuous time window of 10 minutes for data acquisition. Within this time window, the system synchronously collects the contact impedance value sequence generated by each electrode channel during signal acquisition. The smaller the impedance value in the sequence, the tighter the flexible dry electrode adheres to the scalp, and the smoother the signal acquisition pathway. If the impedance value is consistently high or fluctuates significantly, it directly indicates that the electrode is loose, misaligned, or has poor contact. The mean value of the contact impedance value sequence for each electrode channel is calculated. This result characterizes the overall level of contact impedance of the electrode channel within the detection period. A higher mean value indicates a lower overall electrode fit. After calculating the mean value, the variance of the contact impedance value sequence is calculated. The variance characterizes the dispersion of contact impedance over time. A larger variance indicates that the electrode contact state is more severely affected by head movement and wearing displacement, resulting in poorer stability. After obtaining the mean and variance of the contact impedance value sequence for each electrode channel, the system calculates the contact stability index corresponding to that electrode channel by the ratio of the mean to the variance. A higher index value indicates a more stable and reliable contact state between the corresponding flexible dry electrode and the user's scalp.

[0016] Step 101b involves comparing the contact stability index of each electrode channel with a preset threshold, retaining the conditioned EEG signals corresponding to electrode channels whose contact stability index is not lower than the preset threshold, and using these as the selected effective EEG signals. Simultaneously, a correspondence table between electrode channel identifiers and contact stability indices is established, recording the electrode channel identifier and corresponding contact stability index of each retained electrode channel. Specifically, after obtaining the contact stability index of each electrode channel, the contact stability index of each channel is compared one by one with a preset judgment threshold of 10000. This threshold is determined based on the typical contact impedance fluctuation range of the flexible dry electrode and the signal-to-noise ratio requirements for effective EEG signal acquisition, and can distinguish between effective signal channels and contact failure channels. Only the conditioned EEG signals corresponding to electrode channels with a contact stability index greater than or equal to the preset threshold are retained, and these signals are identified as the selected effective EEG signals for subsequent fatigue analysis. At the same time, a correspondence table between electrode channel identifiers and contact stability indices is established, recording and storing the unique identifier information of all retained effective electrode channels and the corresponding contact stability index values ​​for each effective channel.

[0017] This embodiment acquires EEG signals using flexible dry electrodes in the midline of the forehead and bilateral temporal regions. These electrodes conform to the physiological structure of the head, ensuring comfortable wear while stably acquiring central nervous system-related physiological signals, thus adapting to various wearable scenarios. Standardized preprocessing of the raw EEG signals effectively eliminates various interference factors, improving signal quality. The contact stability index is calculated based on the mean and variance of the contact impedance sequence, objectively quantifying the actual contact quality of each flexible dry electrode and avoiding invalid signals caused by poor electrode contact due to wearing misalignment or head movement. Threshold filtering retains valid signals, and a corresponding relationship table is established to eliminate abnormal channel data, improving the overall accuracy of fatigue detection results.

[0018] In a preferred embodiment of the present invention, a first phase lock point, a second phase lock point, and a third phase lock point are extracted from the electrode channels corresponding to the screened valid EEG signals, an elliptical spatial potential field surface is constructed, and the electrode distribution configuration entropy value is calculated, including: Step 200a: Extract the first phase lock point, the second phase lock point, and the third phase lock point from the electrode channels corresponding to the screened valid EEG signals. The first phase lock point is located at the geometric center of the flexible dry electrode in the center of the forehead inside the wearable cap; the second phase lock point is located at the geometric center of the flexible dry electrode in the left temporal region inside the wearable cap; and the third phase lock point is located at the geometric center of the flexible dry electrode in the right temporal region inside the wearable cap. Specifically, after the system completes the screening of the conditioned EEG signals and determines the screened valid EEG signals, it extracts the positions of three fixed phase lock points for each electrode channel to which the screened valid EEG signals belong. The first phase lock point is set at the geometric center of the flexible dry electrode installed in the center of the forehead inside the wearable cap; the second phase lock point is set at the geometric center of the flexible dry electrode installed in the left temporal region inside the wearable cap; and the third phase lock point is set at the geometric center of the flexible dry electrode installed in the right temporal region inside the wearable cap. The spatial coordinates of the three phase locking points on the cap are pre-fixed by the structural design of the wearable cap and will not shift due to changes in the tightness of the user's wear, head movement, or electrode contact status.

[0019] Step 200b, the construction of the elliptical spatial potential field surface and the calculation of the electrode distribution configuration entropy value, includes: when the contact stability index of the electrode channel corresponding to any one of the first phase lock point, the second phase lock point, and the third phase lock point is detected to be lower than a preset dynamic calibration threshold, the physical position of the first phase lock point on the wearable cap is used as the spatial reference origin. The selection of the spatial reference origin depends only on the physical structure of the wearable cap and does not depend on the contact stability index of the electrode channel corresponding to the corresponding phase lock point; the physical position of the first phase lock point on the wearable cap is used as the origin, the line connecting the first phase lock point to the second phase lock point is used as the first direction vector, and the line connecting the first phase lock point to the third phase lock point is used as the second direction vector; wherein, the geometric center positions of the first phase lock point, the second phase lock point, and the third phase lock point are fixed by the physical structure of the wearable cap, specifically including: the system continuously connects the electrode channels corresponding to the first phase lock point, the second phase lock point, and the third phase lock point. The system monitors the contact stability index in real time. When the contact stability index of the electrode channel corresponding to any phase lock point is lower than the preset dynamic calibration threshold of 8000, the system automatically initiates the construction process of the elliptical spatial potential field surface. This threshold is lower than the judgment threshold used in the previous signal screening. This can identify electrode channels that are close to failure but have not yet completely failed, and can also avoid the spatial reconstruction process being triggered by slight fluctuations, ensuring that the timing of the spatial potential field construction is reasonable and stable. When constructing the spatial reference, the system uses the inherent physical position of the first phase lock point on the wearable cap as the spatial reference origin. The selection of this spatial reference origin is determined only by the physical structure and size of the wearable cap itself and is completely unaffected by the contact stability of the electrode channel to which the corresponding phase lock point belongs. Using this spatial reference origin as the starting point, the spatial line connecting the first phase lock point to the second phase lock point is defined as the first direction vector, and the spatial line connecting the first phase lock point to the third phase lock point is defined as the second direction vector. The two direction vectors together constitute the basic reference system for describing the spatial distribution of EEG signals in the head.

[0020] Step 201b: Taking the starting point of the first direction vector as the reference endpoint and the ending point of the first direction vector as the first reference endpoint, extract the line segment expression of the first direction vector and convert the line segment expression of the first direction vector into a parameterized line equation to obtain a first parameterized line equation; taking the starting point of the second direction vector as the reference endpoint and the ending point of the second direction vector as the second reference endpoint, extract the line segment expression of the second direction vector and convert the line segment expression of the second direction vector into a parameterized line equation to obtain a second parameterized line equation, specifically including: taking the starting endpoint of the first direction vector as a unified reference endpoint and the ending endpoint of the first direction vector as the first reference endpoint; assuming the coordinates of the reference endpoint in the preset spatial coordinate system are (x0, y0, z0) and the coordinates of the first reference endpoint are (x1, y1, z1), then the line segment coordinate expression of the first direction vector is: = = Then, through parameter substitution and coordinate transformation, the line segment expression is transformed into a parameterized line equation, resulting in the first parameterized line equation: Where t is a continuously changing parameter, and still using the same reference endpoint (x0, y0, z0) as the starting point, the ending endpoint of the second direction vector is taken as the second reference endpoint. Let the coordinates of the second reference endpoint be (x2, y2, z2). The line segment coordinate expression of the second direction vector is extracted in the same way as follows: = = This is converted into the corresponding parametric line equation, resulting in the second parametric line equation: .

[0021] Step 201b1: Based on the first parameterized straight line equation and the second parameterized straight line equation, calculate the tilt angle of the first direction vector relative to the preset reference coordinate system and the tilt angle of the second direction vector relative to the preset reference coordinate system to obtain the first tilt angle and the second tilt angle. Specifically, this includes: using a pre-established head space reference coordinate system as a reference, calculating the tilt angle based on the spatial orientation of the first parameterized straight line equation and the second parameterized straight line equation; using the horizontal reference axis of the coordinate system as the X-axis, the tilt angle of the direction vector is determined by the angle between the projection of the vector in the XOY plane and the X-axis. The calculation formula is: the tangent of the tilt angle = the component of the direction vector in the Y-axis direction ÷ the component in the X-axis direction. The specific angle value is obtained by calculating the arctangent function, thus obtaining the first tilt angle and the second tilt angle. The larger the difference between the two tilt angles, the higher the degree of orientation deviation between the first and second direction vectors in the head space; the smaller the difference, the closer the spatial orientation of the two vectors. This numerical difference directly quantifies the spatial orientation deviation of the two direction vectors.

[0022] Step 201b2: Based on the difference between the first tilt angle and the second tilt angle, determine whether the first direction vector and the second direction vector are collinear. When the difference between the first tilt angle and the second tilt angle is within a preset collinearity threshold range, take the intersection of the perpendicular bisector of the first direction vector and the perpendicular bisector of the second direction vector as the center point of the elliptical spatial potential field surface, and use the average length of the first direction vector and the length of the second direction vector as the reference radius to construct a circular spatial potential field surface. Specifically, this includes: calculating the difference between the first tilt angle and the second tilt angle, and comparing the tilt angle difference with a preset collinearity threshold of 5°. Since there is a fixed geometric relationship between the electrode layout of the forehead and the bilateral temporal regions, an angle difference within 5° can be considered as approximately consistent vector directions, and will not cause significant deviation in the calculation of the potential field center. When the tilt angle difference is within a preset collinearity threshold range, determine the first direction vector determined in step 200b. When the first and second direction vectors are approximately collinear, the system calculates the perpendicular bisectors of the first and second direction vectors respectively. First, based on the coordinate expression of the first direction vector segment obtained in step 201b, the midpoint coordinates of the segment are obtained, and the slope of the line perpendicular to the first direction vector is calculated. Combining the midpoint coordinates and the vertical slope, the equation of the perpendicular bisector of the first direction vector is obtained. Using the same method, based on the coordinate expression of the second direction vector segment obtained in step 201b, the midpoint coordinates and the corresponding vertical slope of the second direction vector are obtained, and the equation of the perpendicular bisector of the second direction vector is obtained. The two perpendicular bisector equations are solved simultaneously, and the coordinates of the spatial intersection point are determined as the center point of the elliptical spatial potential field surface. Then, the spatial lengths of the first and second direction vectors are calculated respectively, and the arithmetic mean of the two length values ​​is taken. This average value is used as the reference radius to construct the circular spatial potential field surface.

[0023] Step 201b3: When the difference between the first tilt angle and the second tilt angle exceeds a preset collinearity threshold range, the endpoint of the first direction vector is used as the first projection point, and the shortest distance from the first projection point to the line corresponding to the second parameterized line equation is calculated to obtain the first projection distance; the endpoint of the second direction vector is used as the second projection point, and the shortest distance from the second projection point to the line corresponding to the first parameterized line equation is calculated to obtain the second projection distance. Specifically, when the difference between the obtained first tilt angle and the second tilt angle exceeds a preset collinearity threshold range, it is determined that the two direction vectors are not collinear. The system uses the terminating endpoint of the first direction vector as the first projection point, and uses the distance formula from a spatial point to a line to calculate the shortest vertical distance from the projection point to the second parameterized line. The calculation result is the first projection distance. At the same time, the terminating endpoint of the second direction vector is used as the second projection point, and the same calculation formula is used to obtain the shortest vertical distance from the second projection point to the spatial line corresponding to the first parameterized line equation, which is the second projection distance.

[0024] Step 201b4: Determine the first and second boundary control points of the elliptical spatial potential field surface based on the first and second projection distances, and use these two boundary control points as the two foci of the ellipse. Use the distance between the first and second boundary control points as the focal length of the ellipse, and the sum of the lengths of the first and second direction vectors as the length of the major axis of the ellipse. This process constructs the elliptical spatial potential field surface. Specifically, this includes: based on the first and second projection distances calculated in step 201b3, and considering their relative positions in the head space reference coordinate system, performing spatial coordinate mapping to determine the first and second boundary control points of the elliptical spatial potential field surface; using these two boundary control points as the two foci of the ellipse, and the straight-line distance between the two foci in the spatial reference coordinate system as the focal length of the corresponding ellipse; summing the spatial lengths of the first and second direction vectors calculated in step 201b, and using the total sum as the length of the major axis of the ellipse. Based on the determined focal positions, focal length, and major axis length of the ellipse, and combined with the standard geometric relationships of the ellipse, a complete construction of the elliptical spatial potential field surface is completed. This elliptical spatial potential field surface uses the natural physiological contour of the midline of the forehead to the bilateral temporal regions of the human head as a construction reference. The geometric shape of the elliptical surface accurately fits the curvature and spatial extension of the scalp surface in this region, ensuring that the surface shape is consistent with the actual physiological structure of the head. During the potential simulation process, the system collects the potential values ​​of EEG signals from flexible dry electrodes at different spatial locations. The system maps the potential values ​​of each electrode to matching coordinate points on an elliptical potential field surface, corresponding to their actual physical positions on the wearable cap. For the distribution of EEG potential intensity, the system assigns values ​​to points on the elliptical surface based on the mapped potential values ​​of each electrode. Higher potential values ​​correspond to higher potential levels at the corresponding points, and vice versa. This difference in potential levels at different points fully simulates the distribution of potential intensity across the forehead and bilateral temporal scalp areas. Regarding the spatial diffusion trend of EEG potential, the system uses the surface points corresponding to three phase-locking points as the core diffusion source. Following the arc extension direction of the elliptical surface, it calculates the gradient attenuation of potential values ​​from the core points to the surrounding areas. The farther the surface area is from the core point, the greater the attenuation of the potential value, thus recreating the natural trend of EEG potential gradually diffusing and attenuating from the central nervous system to the surrounding scalp areas.For the overall potential field morphology, the system integrates the geometric contour of the elliptical surface, the potential strength assignment at each point, and the potential gradient diffusion law, transforming the discrete electrode potential data into a continuous and smooth spatial surface morphology. The undulation of the surface directly corresponds to the high and low changes of the EEG potential, and the boundary range of the surface corresponds to the actual area of ​​action of the EEG potential on the scalp surface. Finally, it completely simulates the overall distribution morphology of the EEG potential generated by the brain tissue in the forehead and bilateral temporal regions on the scalp surface.

[0025] Step 202b involves using the line connecting the first phase lock point and the second phase lock point as the first equipotential segmentation reference line, and the line connecting the first phase lock point and the third phase lock point as the second equipotential segmentation reference line. The elliptical spatial potential field surface is then equipotentially segmented along the directions of the first and second equipotential segmentation reference lines to form multiple solid angle unit cells. Specifically, this includes: defining the spatial connection between the first and second phase lock points determined in step 200a as the first equipotential segmentation reference line; simultaneously, the first phase lock point and... The spatial connection between the third phase lock points is clearly defined as the second equipotential segmentation baseline. Both baselines closely conform to the surface orientation of the elliptical spatial potential field surface, and the only intersection of the two baselines is the first phase lock point, thus forming a cross-segmentation baseline system centered on the first phase lock point. After determining the two baselines, the intersection of the two equipotential segmentation baselines (i.e., the first phase lock point) is used as the segmentation origin. The baselines extend uniformly towards the edge region of the elliptical spatial potential field surface along their respective extension directions, performing a uniform equipotential segmentation operation on the elliptical spatial potential field surface. During the segmentation process, the segmentation angle and segmentation spacing are controlled, with the segmentation angle set at 30° and the segmentation spacing at 5mm to avoid situations where there are no electrodes or the electrodes are too densely packed within a unit. By controlling the above segmentation angle and segmentation spacing, it is ensured that the angle interval of each segmentation remains consistent and the segmentation range is completely equal. Ultimately, the elliptical spatial potential field surface is divided into multiple solid corner unit cells of uniform spatial size, uniform angle distribution, and similar potential distribution characteristics.

[0026] Step 203b: When the number of flexible dry electrodes integrated within the wearable cap reaches the preset spatial distribution analysis threshold, the number of flexible dry electrodes corresponding to the electrode channels contained in each solid angle unit cell and retained after screening, as well as the coefficient of variation of the contact stability index of each corresponding flexible dry electrode, are counted, and the electrode distribution configuration entropy value is calculated. Specifically, when the total number of flexible dry electrodes integrated inside the wearable cap reaches the preset spatial distribution analysis threshold of 8, data statistics are performed on each solid angle unit cell. If the number is less than this, the electrode spatial distribution is too sparse, and the statistical results are not representative. When the number reaches 8 or more, the uniformity of the electrode layout on the head surface can be fully reflected. During the statistical process, the system counts one by one. The number of flexible dry electrodes corresponding to the effective electrode channels contained within each unit after screening is calculated. Simultaneously, for the contact stability index of each flexible dry electrode within the unit, the standard deviation of the index is first calculated, and then the standard deviation is divided by the mean of the contact stability index to obtain the corresponding coefficient of variation. Using the probability distribution of the number of electrodes within each solid angle unit cell and the coefficient of variation of contact stability as calculation parameters, the entropy value of the electrode distribution configuration is obtained through the entropy value calculation formula. The larger the entropy value, the more uneven the distribution of flexible dry electrodes in the head space and the higher the dispersion of contact stability of each electrode. The smaller the entropy value, the more regular the electrode spatial distribution and the better the consistency of the contact state. This achieves a quantitative characterization of the electrode spatial distribution and contact stability.

[0027] This embodiment constructs a spatial vector and electric potential field surface by fixing three phase lock points at physical locations, which can avoid the failure of the spatial reference due to poor electrode contact and ensure the stability of EEG signal spatial analysis. It adopts tilt angle judgment and dynamic projection calculation to adaptively construct elliptical or circular electric potential field surfaces, which can adapt to the spatial characteristics of the head under different wearing postures. By forming solid angle unit cells through equipotential segmentation and statistically analyzing electrode distribution, the spatial layout characteristics of electrodes can be quantified. Combined with the contact stability variation coefficient to calculate the configuration entropy, it can objectively reflect the spatial reliability of signal acquisition and effectively improve the accuracy of fatigue feature extraction and anti-interference ability.

[0028] In a preferred embodiment of the present invention, the fusion weights of the selected effective EEG signals are dynamically reconstructed based on the electrode distribution configuration entropy value to obtain the reconstructed fusion weights. The selected effective EEG signals are then weighted and fused to generate a comprehensive EEG signal sequence, including: Step 300: Using the electrode distribution configuration entropy value as a weight adjustment factor, perform a nonlinear mapping transformation on the contact stability index of each electrode channel in the correspondence table between electrode channel identifiers and contact stability indices to obtain the initial fusion weight for each electrode channel. Specifically, this includes: using the electrode distribution configuration entropy value calculated in step 203b as a weight adjustment factor, performing a nonlinear mapping transformation on the contact stability index corresponding to each electrode channel in the correspondence table between electrode channel identifiers and contact stability indices established in step 101b, to achieve the initial allocation of signal fusion weights. The transformation adopts an exponential nonlinear mapping method, i.e. = In the formula, This represents the initial fusion weight for the current electrode channel. It is a natural constant. This is the contact stability index corresponding to the electrode channel. The entropy value of the electrode distribution configuration can be used to enhance the adjustment effect of the electrode distribution configuration entropy on the weights through this exponential mapping relationship. This allows electrode channels with more regular electrode distribution in the head space and higher contact stability index to be given higher initial fusion weights, thereby completing the calculation of the initial fusion weights of all electrode channels.

[0029] Step 301: Based on the number of electrode channels contained in each solid corner unit cell, normalize the initial fusion weights of all electrode channels within the same solid corner unit cell to obtain the reconstructed fusion weights of each electrode channel. Specifically, this includes: based on the divided solid corner unit cells, counting the number of effective electrode channels contained in each cell, and normalizing the initial fusion weights of all electrode channels belonging to the same solid corner unit cell to unify the weight scale within the same spatial cell and eliminate dimensional differences. That is, the normalized reconstructed fusion weight of the electrode channel = the initial fusion weight of the channel ÷ the sum of the initial fusion weights of all electrode channels within the solid corner unit cell. After normalization, the weights of each electrode channel have a unified comparison benchmark and a reasonable distribution ratio within the same spatial cell, and finally, the reconstructed fusion weights of all electrode channels are obtained.

[0030] Step 302: Based on the reconstructed fusion weights, multiply the filtered effective EEG signal of each electrode channel with the corresponding reconstructed fusion weight to obtain the weighted EEG signal of each electrode channel; sum the weighted EEG signals of each electrode channel to obtain a comprehensive EEG signal sequence. Specifically, this includes: according to the reconstructed fusion weights corresponding to each electrode channel, multiplying the filtered effective EEG signal sequence of each electrode channel with the corresponding reconstructed fusion weight at each time point. In the formula, For the first Weighted EEG signals from individual electrode channels The reconstructed fusion weights for this channel. The effective EEG signals after filtering by this channel. This is a time series. After completing the channel-by-channel weighted calculation, the system sums the weighted EEG signals from all electrode channels point-by-point on the same time axis, i.e. In the formula, To synthesize the EEG signal sequence, The total number of all effective electrode channels is used to integrate the effective signal components of electrodes at different spatial locations through weighted summation, suppressing noise and interference from abnormal channels, ultimately obtaining a comprehensive EEG signal sequence that can fully and stably reflect the overall EEG activity characteristics of the head.

[0031] In this embodiment, the electrode distribution configuration entropy is used as a weight adjustment factor for nonlinear mapping, which can dynamically allocate signal weights by combining the rationality of electrode spatial distribution and contact stability, avoiding the interference of single channel abnormalities on the overall signal quality. The weights within the same solid angle unit cell are normalized to balance the signal contribution of different spatial regions, thereby improving the rationality and robustness of signal fusion. By generating a comprehensive EEG signal sequence through weighted multiplication and summation, the effective signal components can be fully preserved and noise and invalid interference can be suppressed.

[0032] In a preferred embodiment of the present invention, time-frequency and time-domain waveform analysis is performed on the comprehensive EEG signal sequence to calculate a first fatigue characterization index and determine peripheral behavioral fatigue symptoms; based on the peripheral behavioral fatigue symptoms, the user's continuous working time from the last rest to the current moment is calculated, and a multi-dimensional environmental and working time sequence feature set is obtained, including: Step 400: Perform time-frequency analysis on the comprehensive EEG signal sequence, extract the power spectral density of the δ band, θ band, α band, and β band from the comprehensive EEG signal sequence, and calculate the first fatigue characterization index based on the power spectral density of the δ band, θ band, α band, and β band. Specifically, this includes: performing time-frequency analysis on the comprehensive EEG signal sequence, firstly decomposing the comprehensive EEG signal sequence using discrete wavelet transform, selecting the Daubechies4 wavelet basis function, and combining the characteristic that the effective frequency distribution range of the EEG signal is 0.5Hz to 30Hz, fixing the number of decomposition layers to 5 layers to ensure that the decomposition results of each layer can completely cover the full frequency range of δ, θ, α, and β. During the decomposition process, the original EEG signal undergoes alternating iterative low-pass and high-pass filtering operations. First, the original time-series signal is simultaneously input into a low-pass filter and a high-pass filter with a cutoff frequency of 15Hz. After low-pass filtering, low-frequency components from 0.5Hz to 15Hz are output; after high-pass filtering, high-frequency components from 15Hz to 30Hz are output. After completing the first level of decomposition, the low-frequency components output from the previous low-pass filtering are again simultaneously input into the next level's low-pass filter and high-pass filter with a cutoff frequency of 7.5Hz, outputting low-frequency components from 0.5Hz to 7.5Hz and high-frequency components from 7.5Hz to 15Hz. This frequency division process is repeated. The process iterates five times, with each round focusing only on the latest low-frequency approximation component. High-frequency detail components are no longer included in subsequent iterations. This decomposes the original time-series signal layer by layer into approximation and detail components in different frequency bands. The approximation components are obtained from the low-pass filter output at each stage, preserving the overall slow trend and low-frequency energy characteristics of the signal, corresponding to the lower frequency band of 0.5Hz to 5Hz. The detail components are obtained from the high-pass filter output at each stage, reflecting the rapid fluctuations and high-frequency characteristic information of the signal, corresponding to the higher frequency band of 5Hz to 30Hz. Through this layer-by-layer separation, the effective differentiation and independent extraction of EEG components at different frequencies are achieved.

[0033] After wavelet decomposition, combined with the periodogram power spectrum estimation method, frequency domain energy statistics and power spectrum calculations are performed on the components obtained from each decomposition layer. First, a discrete Fourier transform is performed on each layer component to obtain a frequency domain complex sequence. Then, the square of the modulus of the complex sequence is taken and the ratio with the signal length is calculated to obtain the power spectrum distribution of the corresponding frequency band. Finally, the power spectral density values ​​corresponding to the δ, θ, α, and β frequency bands in the signal are located and extracted respectively, and the energy quantization results of each frequency band are obtained. After completing the power spectral density extraction of the above four typical EEG frequency bands, the first fatigue characterization index is calculated based on the energy change law related to EEG fatigue. In the formula, It is the primary indicator of fatigue. , , The values ​​represent the power spectral density of the delta, theta, alpha, and beta bands, respectively. This index directly quantifies the user's fatigue state from the perspective of central nervous system electrophysiology. The larger the ratio, the stronger the slow wave activity and the weaker the fast wave activity in the EEG signal, reflecting a decrease in the excitability of the cerebral cortex and a tendency for inhibition of neural activity, corresponding to a higher degree of central nervous system fatigue.

[0034] Step 401 involves performing time-domain waveform analysis on the integrated EEG signal sequence to identify and separate the first type of oculomotor artifact waveform induced by blinking and the second type of oculomotor artifact waveform induced by yawning. Specifically, this includes: conducting continuous time-domain waveform feature analysis on the weighted and fused integrated EEG signal sequence; constructing dual-judgment constraint rules using pre-set waveform amplitude abrupt change thresholds and waveform duration thresholds to identify and effectively separate behavioral artifact waveforms induced by the user's facial movements. The baseline amplitude of the EEG signal is set to 10 μ. V, this value is determined based on the statistical mean amplitude of the electroencephalogram (EEG) signals of the forehead and temporal scalp in a resting state. It can truly reflect the signal baseline level when there is no obvious EEG activity or behavioral interference. The waveform amplitude mutation threshold is set to 3 times the baseline amplitude, i.e., 30μV. The amplitude fluctuation of normal EEG signals is small, usually fluctuating within a small range around the 10μV baseline. However, the amplitude of artifacts produced by behaviors such as blinking and yawning is significantly higher than that of normal EEG signals. Using 3 times the baseline amplitude as the judgment threshold can effectively distinguish between normal EEG fluctuations and behavioral artifacts. It will not miss weak amplitude artifacts due to an excessively high threshold, nor will it introduce noise interference due to an excessively low threshold. In the specific waveform determination process, the amplitude change rate of the signal waveform is monitored in real time based on the amplitude change threshold. Single-pulse waveforms that show a sharp upward trend in amplitude within a very short time and quickly reach the peak value, and then quickly fall back to the signal baseline level are identified. The waveform duration is then used for secondary verification. The waveform duration threshold is divided into two segments. The determination duration range for blinking is set to 0.1 seconds to 0.4 seconds. This value is determined based on the statistical analysis of the duration of normal human physiological blinking. The duration of a single natural blink generally falls within this range. Waveforms that exceed this range can be directly excluded as blink artifacts. When the overall duration of this type of waveform is stably within the range of 0.1 seconds to 0.4 seconds, it is determined to be a type I electrooculography artifact waveform induced by the user's blinking action.

[0035] For waveforms with relatively smooth amplitude changes, no drastic jumps, and an overall wide-amplitude bell-shaped symmetrical undulating contour, a duration threshold is also used for judgment. The duration threshold for the waveform corresponding to yawning is set to 1.5 to 3 seconds. This value is derived from the statistical analysis of the physiological duration of yawning. The duration of yawning is much longer than blinking, and it is mostly concentrated between 1.5 and 3 seconds. Using this as a judgment criterion, the waveform characteristics corresponding to yawning can be accurately matched. When the fluctuation duration of this type of waveform is significantly longer and is stable within the range of 1.5 to 3 seconds, it is judged as a second type of electrooculography artifact waveform induced by the user's yawning. Through the dual constraints of amplitude and duration characteristics, interference from non-target waveforms is effectively eliminated, and waveform misjudgment and missed detection caused by a single judgment condition are avoided. Finally, the distinction between the two types of behavioral artifact waveforms that are highly correlated with mental fatigue is achieved.

[0036] Step 402: Calculate the blink frequency and average duration of a single blink based on the first type of electrooculography (EOG) artifact waveform; calculate the frequency of yawning feature waveforms within a unit time based on the second type of EOG artifact waveform; and use the blink frequency, average duration of a single blink, and frequency of yawning feature waveforms within a unit time as the peripheral behavioral fatigue sign characteristics. Specifically, this includes: performing quantitative statistical calculations based on the identified and separated first type of EOG artifact waveforms, using 60 seconds as a uniform unit time period, counting the total number of occurrences of the first type of EOG artifact waveforms within this period to directly obtain the user's blink frequency; simultaneously extracting the duration data of each blink waveform, and calculating the arithmetic mean of all single blink durations to obtain the user's average single blink duration. For the second type of electrooculography artifact waveform, a 60-second time period was also used to count the frequency of yawning feature waveforms within this period. The calculated blink frequency per unit time, average duration of a single blink, and frequency of yawning feature waveforms per unit time were standardized and combined to form peripheral behavioral fatigue signs, which can help characterize fatigue status from the perspective of the user's external facial behavior.

[0037] This embodiment calculates the first fatigue characterization index based on multi-band power spectral density, which can quantify the degree of fatigue from the central nervous system physiological level and improve the objectivity and sensitivity of fatigue judgment. By separating the artifact waveforms related to blinking and yawning from the comprehensive EEG signal and extracting the corresponding features, peripheral behavioral fatigue characteristics can be obtained without adding additional sensors, reducing the system hardware complexity. By combining physiological characterization with behavioral signs, the fatigue state can be comprehensively reflected from both central and peripheral dimensions, effectively improving the comprehensiveness, accuracy and scenario adaptability of fatigue detection.

[0038] In a preferred embodiment of the present invention, the first fatigue characterization index, peripheral behavioral fatigue symptom features, and multidimensional environmental and operational time-series feature set are input into a pre-trained evaluation model to obtain a comprehensive fatigue level determination result and trigger corresponding graded alarm measures; the comprehensive fatigue level determination result is transmitted to an external monitoring platform or a local monitoring terminal, including: Step 500: Input the first fatigue characterization index, the peripheral behavioral fatigue symptom features, and the multidimensional environment and work time-series feature set into the pre-trained evaluation model. Calculate the interaction weights among the three—the first fatigue characterization index, the peripheral behavioral fatigue symptom features, and the multidimensional environment and work time-series feature set—using the cross-attention mechanism within the evaluation model. Specifically, this includes: inputting the first fatigue characterization index, the peripheral behavioral fatigue symptom features (peripheral behavioral fatigue features), and the multidimensional environment and work time-series feature set into the pre-trained fatigue evaluation model. Calculate the interaction weights among the three types of input features using the built-in cross-attention mechanism. The specific process is as follows: First, complete the construction and pre-training of the fatigue evaluation model. The model construction adopts a deep learning framework, with a convolutional neural network combined with an attention mechanism as the core structure. The input layer receives three types of feature data (the first fatigue characterization index, peripheral behavioral fatigue features, and the multidimensional environment and work time-series feature set), and the hidden layer consists of three fully connected layers. The number of neurons in each layer is 128, 64, and 32 respectively. The ReLU function is used as the activation function, and the Softmax function is used as the output layer. The model calculation results are mapped to the probability distribution of four levels: awake, mild fatigue, moderate fatigue, and severe fatigue. All four probability values ​​are between 0 and 1, and the sum of all probability values ​​equals 1. The higher the probability value, the closer the current user state is to the corresponding fatigue level. During the model training process, 1000 sets of sample data covering different fatigue levels and different environmental scenarios are selected (including the first fatigue representation index, peripheral behavioral fatigue features, multidimensional environmental and work time sequence features, and corresponding fatigue level labels). They are divided into a training set (80%) and a validation set (20%). The training batch size is set to 32, the initial learning rate is 0.001, the Adam optimizer is used, and the number of iterations is set to 50 rounds. The model accuracy is verified every 10 rounds. Training stops when the accuracy of the validation set does not improve for 3 consecutive rounds. Finally, the fatigue assessment model is trained, ensuring that the model can capture the correlation between the three types of features.

[0039] The three types of data—the first fatigue characterization index, peripheral behavioral fatigue features, and multidimensional environmental and work time-series feature sets—are simultaneously input into a pre-trained fatigue assessment model. The model uses a cross-attention mechanism to perform dynamic interaction analysis on the three types of features and calculates the interaction coefficient for each type of feature. First, let the first fatigue characterization index be F, the peripheral behavioral fatigue features be B, and the multidimensional environmental and work time-series feature sets be E. The interaction coefficients for the three types are Wf, Wb, and We, respectively. The formula for calculating the interaction coefficient is W = (standardized feature value) ÷ (sum of standardized values ​​of the three types of features), where the standardized feature value = (actual feature value - minimum feature value) ÷ (maximum feature value - minimum feature value). The interaction weights of the three types of features are obtained through this formula.

[0040] Step 501: Based on the interaction weights, adaptive deep fusion is performed on the first fatigue characterization index, the peripheral behavioral fatigue symptom features, and the multidimensional environmental and work time-series feature set to obtain a comprehensive fatigue level determination result. The comprehensive fatigue level determination result includes alertness level, mild fatigue level, moderate fatigue level, and severe fatigue level. Specifically, this includes: extracting the three types of feature interaction coefficients Wf, Wb, and We, and weighting the first fatigue characterization index, peripheral behavioral fatigue features, and multidimensional environmental and work time-series feature set accordingly. The weighted value of a single feature is calculated as follows: The standardized value of the feature multiplied by the corresponding interaction coefficient. For example, the weighted value of the first fatigue characterization index is F × Wf, the weighted value of peripheral behavioral fatigue features is B × Wb, and the weighted value of the multidimensional environmental and operational time-series feature set is E × We. The three weighted features are then fused together: the comprehensive feature value is calculated as: the weighted value of the first fatigue characterization index + the weighted value of peripheral behavioral fatigue features + the weighted value of the multidimensional environmental and operational time-series feature set. This formula fully integrates the effective information from the three types of features, eliminating the bias in evaluation caused by a single feature. Upon completion, based on the numerical range of the comprehensive feature value and the four-category probability distribution output by the model, the level corresponding to the maximum probability is used as the final judgment result. Specifically, the probability range for the alertness level is 0.7 to 1. When the probability of this level is within this range and is the maximum, it is judged as an alert state, indicating that the user has no obvious fatigue, is in good mental condition, and can carry out normal work. The probability range for the mild fatigue level is 0.5 to 0.7. When the probability of this level is within this range and is the maximum, it is judged as a mild fatigue state, indicating that the user shows slight signs of fatigue and needs appropriate rest and adjustment. The probability range for the moderate fatigue level is 0.3 to 0.5. When the probability of this level is within this range and is the maximum, it is judged as a moderate fatigue state, indicating that the user's fatigue level is relatively obvious and the current work needs to be stopped immediately and rested. The probability range for the severe fatigue level is 0 to 0.3. When the probability of this level is within this range and is the maximum, it is judged as a severe fatigue state, indicating that the user's fatigue level is serious, there is a risk to work safety, and the work needs to be stopped immediately and sufficient rest is required, simultaneously triggering a high-level alarm.

[0041] Step 502: Based on the comprehensive fatigue level determination result, match the corresponding graded alarm strategy from the preset alarm strategy library. The graded alarm strategy includes: when the comprehensive fatigue level determination result is mild fatigue, triggering the vibration motor integrated in the wearable cap to generate a tactile vibration alarm with a first preset frequency and a first preset duration; when the comprehensive fatigue level determination result is moderate fatigue, triggering the bone conduction headphones integrated in the wearable cap to play a preset voice prompt and simultaneously sending a first-level text alarm message to the associated mobile terminal; when the comprehensive fatigue level determination result is severe fatigue, simultaneously triggering the tactile vibration alarm and the bone conduction voice prompt, specifically including: A comprehensive hierarchical alarm strategy library is pre-established, clearly defining the alarm methods, operating parameters, and triggering logic corresponding to different fatigue levels. Mild fatigue corresponds to Level 1 alarms, moderate fatigue to Level 2 alarms, and severe fatigue to Level 3 alarms. Specific parameters for each alarm level are as follows: Level 1 alarms for mild fatigue trigger the vibration motor integrated into the wearable cap, with a vibration frequency of 5Hz, a single duration of 1 second, and repeating every 5 seconds until the user indicates a resting state. Level 2 alarms for moderate fatigue simultaneously trigger the vibration motor and bone conduction headphones, with the vibration motor using the same parameters as the Level 1 alarm. The system continuously plays voice prompts during the first level of fatigue, while simultaneously sending text alarm messages to the mobile terminal linked to the wearable device, indicating that the user is currently experiencing moderate fatigue and needs to rest. The third level alarm corresponds to severe fatigue, simultaneously triggering the vibration motor, bone conduction headphones, and external warning devices. The vibration motor frequency is increased to 8Hz, the duration of each alarm is extended to 2 seconds, and it repeats every 3 seconds. The bone conduction headphones continuously play emergency rest prompts, while simultaneously sending emergency alarm messages to the mobile terminal and activating the warning lights at the work site to emit red warning lights, ensuring that users and on-site management personnel can promptly detect and intervene.

[0042] During alarm execution, the system monitors the user's rest feedback signals in real time. If the user does not stop working or report their rest status within 10 minutes of the alarm being triggered, the alarm level will be automatically escalated: mild fatigue alarm will be upgraded to moderate fatigue alarm, moderate fatigue alarm will be upgraded to severe fatigue alarm, and the reminder intensity will continue to increase until the user stops working and completes their rest. If the user stops working in time and actively reports their rest status, the system will gradually reduce the alarm intensity and extend the alarm interval until the alarm stops completely, ensuring the effectiveness and rationality of the alarm. This avoids excessive alarms interfering with normal operations and also promptly reminds users to pay attention to their own fatigue status, reducing safety hazards.

[0043] This embodiment dynamically calculates the interaction weights of three types of features through a cross-attention mechanism, avoiding the one-sided influence of a single feature on fatigue assessment and solving the assessment bias problem caused by relying solely on a single feature, making the fatigue assessment results more objective and comprehensive. By adaptively and deeply fusing the three types of features and combining the interaction weights to complete the comprehensive total score calculation, a clear total score interval is used to correspond to the fatigue level, providing a clear quantitative standard for fatigue judgment and avoiding errors from subjective judgment. Simultaneously, the three core features are organically integrated, preserving the internal fatigue state reflected by EEG signals while incorporating peripheral behavioral characteristics. By incorporating environmental factors, fatigue level assessment becomes more accurate. A tiered alarm strategy is employed, matching corresponding alarm methods to different fatigue levels. Mild fatigue triggers only a slight vibration alarm to avoid interfering with normal user operations. Moderate and severe fatigue progressively escalate alarm methods, from voice prompts to multi-channel simultaneous alarms. This ensures timely reminders to users while adjusting alarm intensity based on fatigue levels, guaranteeing alarm effectiveness while avoiding excessive alarm interference. Furthermore, mobile terminal alarms allow relevant personnel to promptly monitor user fatigue status, further enhancing the practicality and safety of fatigue monitoring.

[0044] like Figure 2 As shown, embodiments of the present invention also provide a multi-scenario fatigue state detection method based on electroencephalogram (EEG) signals and a wearable cap, including: Step 1: Preprocess the raw EEG signal from the user's scalp area to obtain the conditioned EEG signal; perform contact state analysis on the conditioned EEG signal and calculate the contact stability index corresponding to each flexible dry electrode to determine the effective EEG signal after screening. Step 2: Extract the first phase lock point, the second phase lock point, and the third phase lock point from the electrode channels corresponding to the selected valid EEG signals, construct the elliptical space potential field surface, and calculate the electrode distribution configuration entropy value. Step 3: Based on the electrode distribution configuration entropy value, dynamically reconstruct the fusion weights of the selected effective EEG signals to obtain the reconstructed fusion weights, and then perform weighted fusion of the selected effective EEG signals to generate a comprehensive EEG signal sequence. Step 4: Perform time-frequency and time-domain waveform analysis on the comprehensive EEG signal sequence, calculate the first fatigue characterization index and determine the peripheral behavioral fatigue symptoms; calculate the user's continuous working time from the last rest to the current moment based on the peripheral behavioral fatigue symptoms, and obtain a multi-dimensional environmental and working time sequence feature set; Step 5: Input the first fatigue characterization index, peripheral behavioral fatigue symptom characteristics, and multidimensional environmental and operational time sequence feature set into the pre-trained evaluation model to obtain the comprehensive fatigue level judgment result and trigger the corresponding graded alarm measures; transmit the comprehensive fatigue level judgment result to the external monitoring platform or local monitoring terminal.

[0045] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0046] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A multi-scenario fatigue state detection system based on electroencephalogram (EEG) signals and a wearable cap, characterized in that, include: The processing module is used to preprocess the raw EEG signals from the user's scalp area to obtain conditioned EEG signals; Contact status analysis was performed on the treated EEG signals, and the contact stability index corresponding to each flexible dry electrode was calculated to determine the effective EEG signals after screening. The extraction module is used to extract the first phase lock point, the second phase lock point, and the third phase lock point from the electrode channels corresponding to the screened valid EEG signals, construct the elliptical space potential field surface, and calculate the electrode distribution configuration entropy value. The fusion module is used to dynamically reconstruct the fusion weights of the selected effective EEG signals based on the electrode distribution configuration entropy value, obtain the reconstructed fusion weights, and perform weighted fusion of the selected effective EEG signals to generate a comprehensive EEG signal sequence. The analysis module is used to perform time-frequency and time-domain waveform analysis on the comprehensive EEG signal sequence, calculate the first fatigue characterization index, and determine the characteristics of peripheral behavioral fatigue symptoms. Calculate the user's continuous working time from the last rest to the current moment based on the characteristics of peripheral behavioral fatigue symptoms, and obtain a multi-dimensional environmental and work time sequence feature set; The evaluation module is used to input the first fatigue characterization index, peripheral behavioral fatigue symptom features and multi-dimensional environmental and operational time sequence feature set into the pre-trained evaluation model to obtain the comprehensive fatigue level judgment result and trigger the corresponding graded alarm measures. The comprehensive fatigue level assessment results are transmitted to an external monitoring platform or a local monitoring terminal.

2. The multi-scenario fatigue state detection system based on electroencephalogram (EEG) signals and a wearable cap as described in claim 1, characterized in that, The raw EEG signals of the user's scalp region are collected by flexible dry electrodes integrated into the wearable cap, located in the center of the forehead, the left temporal region, and the right temporal region.

3. The multi-scenario fatigue state detection system based on electroencephalogram (EEG) signals and a wearable cap as described in claim 2, characterized in that, Contact state analysis was performed on the conditioned EEG signals to calculate the contact stability index corresponding to each flexible dry electrode, in order to determine the effective EEG signals after screening, including: For the conditioned EEG signal, contact state analysis is performed on the electrode channel corresponding to each flexible dry electrode. The contact impedance value sequence of each electrode channel is continuously collected within a preset time window. The mean and variance of the contact impedance value sequence are calculated, and the contact stability index of each electrode channel is calculated based on the mean and variance. The contact stability index of each electrode channel is compared with a preset threshold. The conditioned EEG signals corresponding to electrode channels with a contact stability index not lower than the preset threshold are retained as valid EEG signals after screening. At the same time, a correspondence table between electrode channel identifiers and contact stability indices is established to record the electrode channel identifier and corresponding contact stability index of each retained electrode channel.

4. The multi-scenario fatigue state detection system based on electroencephalogram (EEG) signals and a wearable cap as described in claim 3, characterized in that, The first phase lock point is located at the geometric center of the flexible dry electrode in the center of the forehead inside the wearable cap, the second phase lock point is located at the geometric center of the flexible dry electrode in the left temporal region inside the wearable cap, and the third phase lock point is located at the geometric center of the flexible dry electrode in the right temporal region inside the wearable cap.

5. The multi-scenario fatigue state detection system based on electroencephalogram (EEG) signals and a wearable cap as described in claim 4, characterized in that, The construction of the elliptical spatial potential field surface and the calculation of the electrode distribution configuration entropy include: When the contact stability index of the electrode channel corresponding to any one of the first, second, and third phase lock points is detected to be lower than a preset dynamic calibration threshold, the physical position of the first phase lock point on the wearable cap is used as the spatial reference origin. The selection of the spatial reference origin depends only on the physical structure of the wearable cap and does not depend on the contact stability index of the electrode channel corresponding to the phase lock point. The physical position of the first phase lock point on the wearable cap is used as the origin, the line connecting the first and second phase lock points is used as the first direction vector, and the line connecting the first and third phase lock points is used as the second direction vector. The geometric center positions of the first, second, and third phase lock points are fixed by the physical structure of the wearable cap. An elliptical spatial electric potential field surface is constructed based on the plane spanned by the first direction vector and the second direction vector; Using the line connecting the first phase lock point and the second phase lock point as the first equipotential division reference line, and the line connecting the first phase lock point and the third phase lock point as the second equipotential division reference line, the elliptical space potential field surface is divided equipotentially along the directions of the first equipotential division reference line and the second equipotential division reference line to form multiple solid angle unit cells. When the number of flexible dry electrodes integrated in the wearable cap reaches the preset spatial distribution analysis threshold, the number of flexible dry electrodes corresponding to the electrode channels contained in each solid angle unit cell and retained after screening, as well as the coefficient of variation of the contact stability index of each corresponding flexible dry electrode, are counted, and the electrode distribution configuration entropy value is calculated.

6. The multi-scenario fatigue state detection system based on electroencephalogram (EEG) signals and a wearable cap as described in claim 5, characterized in that, Constructing an elliptical spatial electric potential field surface based on the plane spanned by the first direction vector and the second direction vector includes: Using the starting point of the first direction vector as the reference endpoint and the ending point of the first direction vector as the first reference endpoint, the line segment expression of the first direction vector is extracted, and the line segment expression of the first direction vector is converted into a parameterized line equation to obtain a first parameterized line equation; using the starting point of the second direction vector as the reference endpoint and the ending point of the second direction vector as the second reference endpoint, the line segment expression of the second direction vector is extracted, and the line segment expression of the second direction vector is converted into a parameterized line equation to obtain a second parameterized line equation; Based on the first parameterized line equation and the second parameterized line equation, the tilt angle of the first direction vector relative to the preset reference coordinate system and the tilt angle of the second direction vector relative to the preset reference coordinate system are calculated respectively to obtain the first tilt angle and the second tilt angle. Based on the difference between the first tilt angle and the second tilt angle, it is determined whether the first direction vector and the second direction vector are collinear. When the difference between the first tilt angle and the second tilt angle is within a preset collinearity threshold range, the intersection of the perpendicular bisector of the first direction vector and the perpendicular bisector of the second direction vector is taken as the center point of the elliptical spatial potential field surface, and the average of the lengths of the first direction vector and the second direction vector is taken as the reference radius to construct a circular spatial potential field surface. When the difference between the first tilt angle and the second tilt angle exceeds the preset collinearity threshold range, the endpoint of the first direction vector is used as the first projection point, and the shortest distance from the first projection point to the line corresponding to the second parameterized line equation is calculated to obtain the first projection distance; the endpoint of the second direction vector is used as the second projection point, and the shortest distance from the second projection point to the line corresponding to the first parameterized line equation is calculated to obtain the second projection distance. The first and second boundary control points of the elliptical spatial potential field surface are determined based on the first projection distance and the second projection distance. The first and second boundary control points are used as the two foci of the ellipse, the distance between the first and second boundary control points is used as the focal length of the ellipse, and the sum of the lengths of the first and second direction vectors is used as the length of the major axis of the ellipse. The elliptical spatial potential field surface is thus constructed.

7. The multi-scenario fatigue state detection system based on electroencephalogram (EEG) signals and a wearable cap as described in claim 6, characterized in that, Based on the electrode distribution configuration entropy value, the fusion weights of the selected effective EEG signals are dynamically reconstructed to obtain the reconstructed fusion weights. The selected effective EEG signals are then weighted and fused to generate a comprehensive EEG signal sequence, including: Using the entropy value of the electrode distribution configuration as a weight adjustment factor, a nonlinear mapping transformation is performed on the contact stability index of each electrode channel in the correspondence table between electrode channel identifiers and contact stability indices to obtain the initial fusion weight of each electrode channel. Based on the number of electrode channels contained in each solid corner unit cell, the initial fusion weights of all electrode channels in the same solid corner unit cell are normalized to obtain the reconstructed fusion weights of each electrode channel. Based on the reconstructed fusion weights, the filtered effective EEG signal of each electrode channel is multiplied by the reconstructed fusion weight of the corresponding electrode channel to obtain the weighted EEG signal of each electrode channel; the weighted EEG signals of each electrode channel are summed to obtain a comprehensive EEG signal sequence.

8. The multi-scenario fatigue state detection system based on electroencephalogram (EEG) signals and a wearable cap as described in claim 7, characterized in that, Time-frequency and time-domain waveform analysis was performed on the comprehensive EEG signal sequence to calculate the first fatigue characterization index and determine the characteristics of peripheral behavioral fatigue signs. Based on peripheral behavioral fatigue symptoms, the user's continuous working time from the last rest period to the current moment is calculated, and a multi-dimensional environmental and work time-series feature set is obtained, including: Time-frequency analysis is performed on the comprehensive EEG signal sequence to extract the power spectral density of the δ band, θ band, α band, and β band in the comprehensive EEG signal sequence, and a first fatigue characterization index is calculated based on the power spectral density of the δ band, θ band, α band, and β band. Time-domain waveform analysis was performed on the integrated EEG signal sequence to identify and separate the first type of electrooculogram artifact waveform induced by blinking and the second type of electrooculogram artifact waveform induced by yawning from the integrated EEG signal sequence. The blink frequency and average duration of a single blink are calculated based on the waveform of the first type of electrooculography artifact. The frequency of yawning feature waveforms per unit time is calculated based on the second type of electrooculography artifact waveforms. The blink frequency, the average duration of a single blink, and the frequency of yawning feature waveforms per unit time are used as the peripheral behavioral fatigue signs.

9. The multi-scenario fatigue state detection system based on electroencephalogram (EEG) signals and a wearable cap as described in claim 8, characterized in that, The primary fatigue characterization index, peripheral behavioral fatigue symptoms, and multidimensional environmental and operational time-series feature sets are input into a pre-trained evaluation model to obtain a comprehensive fatigue level determination result and trigger corresponding graded alarm measures, including: The first fatigue characterization index, the peripheral behavioral fatigue symptom features, and the multidimensional environment and work time sequence feature set are jointly input into the pre-trained evaluation model. The interaction weights among the first fatigue characterization index, the peripheral behavioral fatigue symptom features, and the multidimensional environment and work time sequence feature set are dynamically calculated through the cross-attention mechanism in the evaluation model. Based on the interaction weights, the first fatigue characterization index, the peripheral behavioral fatigue symptom features, and the multidimensional environmental and work time sequence feature set are adaptively and deeply fused to obtain a comprehensive fatigue level determination result. The comprehensive fatigue level determination result includes a sober level, a mild fatigue level, a moderate fatigue level, and a severe fatigue level. Based on the comprehensive fatigue level assessment result, a graded alarm strategy corresponding to the comprehensive fatigue level assessment result is matched from a preset alarm strategy library. The graded alarm strategy includes: when the comprehensive fatigue level assessment result is mild fatigue level, triggering a vibration motor integrated in the wearable helmet to generate a tactile vibration alarm with a first preset frequency and a first preset duration; when the comprehensive fatigue level assessment result is moderate fatigue level, triggering a bone conduction headset integrated in the wearable helmet to play a preset voice prompt and simultaneously sending a first-level text alarm message to the associated mobile terminal; when the comprehensive fatigue level assessment result is severe fatigue level, triggering both the tactile vibration alarm and the bone conduction voice prompt.

10. A method for detecting fatigue states in multiple scenarios based on electroencephalogram (EEG) signals and a wearable cap, wherein the method implements the system as described in any one of claims 1 to 9, characterized in that, include: Step 1: Preprocess the raw EEG signals from the user's scalp area to obtain conditioned EEG signals; Contact status analysis was performed on the treated EEG signals, and the contact stability index corresponding to each flexible dry electrode was calculated to determine the effective EEG signals after screening. Step 2: Extract the first phase lock point, the second phase lock point, and the third phase lock point from the electrode channels corresponding to the selected valid EEG signals, construct the elliptical space potential field surface, and calculate the electrode distribution configuration entropy value. Step 3: Based on the electrode distribution configuration entropy value, dynamically reconstruct the fusion weights of the selected effective EEG signals to obtain the reconstructed fusion weights, and then perform weighted fusion of the selected effective EEG signals to generate a comprehensive EEG signal sequence. Step 4: Perform time-frequency and time-domain waveform analysis on the comprehensive EEG signal sequence, calculate the first fatigue characterization index, and determine the characteristics of peripheral behavioral fatigue symptoms; Calculate the user's continuous working time from the last rest to the current moment based on the characteristics of peripheral behavioral fatigue symptoms, and obtain a multi-dimensional environmental and work time sequence feature set; Step 5: Input the first fatigue characterization index, peripheral behavioral fatigue symptom features and multidimensional environmental and operational time sequence feature set into the pre-trained evaluation model to obtain the comprehensive fatigue level judgment result and trigger the corresponding graded alarm measures. The comprehensive fatigue level assessment results are transmitted to an external monitoring platform or a local monitoring terminal.

Images