Operating personnel fatigue state detection and early warning method, device and equipment

Abstract

The invention discloses a worker fatigue state detection and early warning method, device and equipment, and relates to the technical field of intelligent safety monitoring. Electroencephalogram signals are obtained through a safety helmet, and an electroencephalogram unbalance index is established through mirror symmetry analysis; analyzing an environmental noise spectrum to identify a silent frequency band, and mapping a fatigue characteristic frequency to generate an intervention window parameter; selecting an optimal carrier frequency point based on the intervention window parameter, and modulating and fusing the carrier signal by using the electroencephalogram imbalance index to generate a composite intervention signal; recognizing fatigue precursor characteristics by monitoring a coupling relationship between head micro-actions and electroencephalogram signals, and determining the optimal release time of intervention signals; playing an intervention signal in a noise silence frequency band, continuously monitoring the change of the electroencephalogram unbalance index, and generating a fatigue relieving curve to evaluate an intervention effect; and establishing a recovery sensitivity curve to determine the optimal interval duration, and analyzing and outputting a grading early warning instruction through the sensibilization effect to control vibration warning, thereby realizing real-time detection, active intervention and grading early warning of the fatigue state.

Description

Technical Field

[0001] This invention relates to the field of intelligent safety monitoring technology, and in particular to a method, device, and equipment for detecting and warning of worker fatigue. Background Technology

[0002] In high-risk work environments such as mines, construction, and transportation, worker fatigue directly impacts production and personal safety. Fatigue-induced distraction, slowed reaction time, and misjudgment are significant contributing factors to safety accidents. Traditional fatigue detection methods rely primarily on subjective assessment, physiological parameter monitoring, or behavioral characteristic analysis. However, these methods often suffer from poor real-time performance, low accuracy, and susceptibility to external interference, making them unsuitable for real-time safety monitoring in high-risk work environments.

[0003] While existing EEG fatigue detection technologies can objectively assess fatigue levels by monitoring brain activity, most remain at a single detection function, lacking effective intervention mechanisms and personalized early warning strategies. Furthermore, current technologies often overlook the dynamic process of fatigue development and individual differences, failing to achieve accurate fatigue prediction and timely intervention responses. In addition, key technical issues such as effectively suppressing electromagnetic interference in complex industrial environments, implementing effective interventions in noisy environments, and establishing closed-loop fatigue management systems remain largely unresolved. Summary of the Invention

[0004] This invention provides a method, device, and equipment for detecting and warning of worker fatigue. It aims to quantify the degree of fatigue through the electroencephalogram (EEG) imbalance index, determine the optimal carrier parameters of the intervention signal based on environmental noise spectrum analysis, identify fatigue precursor features by combining head micro-movements with EEG signal coupling analysis, achieve active fatigue relief using composite intervention signals, optimize the intervention strategy through sensitization effect evaluation and recovery sensitivity analysis, and finally form a personalized graded early warning output, providing a comprehensive, intelligent, and real-time fatigue management solution for safety monitoring in high-risk work environments.

[0005] The first aspect of this invention provides a method for detecting and warning of worker fatigue, comprising the following steps: The brainwave signals of the workers are acquired, and the mirror symmetry analysis of the brainwave signals is performed to determine the activity difference between the left and right hemispheres. Based on the activity difference, a brainwave imbalance index is generated. The dominant frequency distribution is extracted by performing spectral analysis on the EEG signal, noise signals in the work environment are collected and their spectral gaps are identified to obtain noise silence frequency bands, and the dominant frequency distribution is mapped to the noise silence frequency bands to generate intervention window parameters. The optimal carrier frequency is selected based on the intervention window parameters, a pure carrier signal is generated based on the optimal carrier frequency, and the pure carrier signal is modulated and fused by the EEG imbalance index to generate a composite intervention signal. The coupling relationship between the operator's head micro-movements and the electroencephalogram (EEG) signals is monitored to identify fatigue precursor features, and the timing of releasing the composite intervention signal is determined based on the fatigue precursor features. According to the release timing, the composite intervention signal is played in the noise-silent frequency band, and the change of the brain electrical imbalance index during the playback process is monitored to generate a fatigue relief curve. The slope of the fatigue relief curve is analyzed to generate an intervention effect score. The real-time change of the EEG imbalance index is monitored after the composite intervention signal stops playing to establish a recovery sensitivity curve. The moment when the brain's demand for intervention is strongest in the recovery sensitivity curve is identified as the optimal interval. The optimal interval duration is used to control the release rhythm of the composite intervention signal to form an interval playback sequence. The changes in the intensity of the EEG response under the interval playback sequence are monitored to generate sensitization effect data. Based on the current fatigue level, the sensitization effect data and the intervention effect score, a graded early warning instruction is generated.

[0006] A second aspect of the present invention provides a device for detecting and warning of worker fatigue, comprising: The EEG acquisition module is used to acquire the EEG signals of the operator, perform mirror symmetry analysis on the EEG signals to determine the activity difference between the left and right hemispheres, and generate an EEG imbalance index based on the activity difference. The spectrum analysis module is used to perform spectrum analysis on the EEG signal to extract the dominant frequency distribution, collect noise signals in the work environment and identify their spectral gaps to obtain noise silence frequency bands, and map the dominant frequency distribution to the noise silence frequency bands to generate intervention window parameters. The signal generation module is used to select the optimal carrier frequency point based on the intervention window parameters, generate a pure carrier signal based on the optimal carrier frequency point, and modulate and fuse the pure carrier signal through the EEG imbalance index to generate a composite intervention signal. The fatigue recognition module is used to monitor the coupling relationship between the operator's head micro-movements and the electroencephalogram (EEG) signals to identify fatigue precursor features and determine the timing of releasing the composite intervention signal based on the fatigue precursor features. An intervention playback module is used to play the composite intervention signal in the noise-silent frequency band according to the release timing, and monitor the changes in the EEG imbalance index during the playback process to generate a fatigue relief curve. The effect evaluation module is used to perform slope analysis on the fatigue relief curve to generate an intervention effect score, monitor the real-time changes of the EEG imbalance index after the composite intervention signal stops playing to establish a recovery sensitivity curve, and identify the moment in the recovery sensitivity curve when the brain's demand for intervention is strongest to determine the optimal interval duration. The early warning control module is used to control the release rhythm of the composite intervention signal to form an interval playback sequence using the optimal interval duration, monitor the changes in EEG response intensity under the interval playback sequence to generate sensitization effect data, and generate graded early warning instructions based on the current fatigue level, the sensitization effect data and the intervention effect score.

[0007] A third aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of a method for detecting and warning of worker fatigue disclosed in the first aspect.

[0008] The beneficial effects of this invention are reflected in the following points: 1. By analyzing the difference in activity between the left and right hemispheres, an EEG imbalance index is established. This index integrates multiple dimensions such as instantaneous state, cumulative effect, development trend, and individual baseline, overcoming the one-sidedness of traditional single-indicator assessment. Combined with the analysis of the coupling relationship between head micro-movements and EEG signals, it can identify rhythmic changes and abnormal breakage characteristics in the fatigue development process, and detect subtle changes in cognitive function imbalance in the early stage of fatigue accumulation, realizing the transformation from passive detection to active prediction. 2. By identifying the spectral gaps in environmental noise, fatigue-related EEG characteristic frequencies are mapped to noise-silent frequency bands, ensuring that intervention signals are not masked by environmental noise. Time-reversal technology is used to generate fatigue-resistant waveforms, so that the temporal characteristics of the intervention signal form an antagonistic relationship with the fatigue development process. This active intervention method based on physiological mechanisms has a more direct fatigue relief effect than traditional audio-visual reminders, realizing a functional improvement from simple warning to active regulation. 3. By analyzing the recovery sensitivity curve, the moment when the brain's need for intervention is most intense is identified, and the optimal intervention interval is dynamically determined, avoiding the adaptation problem of fixed-period intervention. The sensitization effect of repeated interventions was discovered and utilized, leading to a sustained enhancement of the effectiveness of subsequent interventions. A tiered early warning system was generated based on current fatigue levels, intervention effectiveness scores, and sensitization status, with multimodal alert output ensuring the effectiveness of the warnings.

[0009] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0010] The accompanying drawings illustrate specific examples of the technical solutions described in this invention and, together with the detailed embodiments, form part of the specification, serving to explain the technical solutions, principles, and effects of this invention.

[0011] Unless otherwise specified, the same reference numerals in different figures represent the same or similar technical features, and different reference numerals may be used to represent the same or similar technical features.

[0012] Figure 1 This is a flowchart illustrating a method for detecting and warning of worker fatigue according to the present invention.

[0013] Figure 2 This is a structural block diagram of a worker fatigue detection and early warning device according to the present invention.

[0014] Figure 3 This is a schematic diagram of the structure of a computer device according to the present invention. Detailed Implementation

[0015] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0016] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0017] The technical solutions of the embodiments of this application will be described below.

[0018] like Figure 1 As shown, this embodiment of the invention provides a method for detecting and warning of worker fatigue, including the following steps S110-S170: Step S110: Obtain the EEG signal of the operator, perform mirror symmetry analysis on the EEG signal to determine the activity difference between the left and right hemispheres, and generate an EEG imbalance index based on the activity difference.

[0019] Specifically, the multi-point EEG acquisition device is integrated into the lining of the safety helmet, employing a flexible electrode array structure to achieve close contact with the worker's scalp. The electrode layout follows international standard electrode placement systems, with multiple pairs of acquisition electrodes symmetrically positioned in the left and right hemispheres, including the frontal, central, parietal, and occipital regions, ensuring coverage of key fatigue-related brain regions. A reference electrode is positioned along the midline of the head, and a ground electrode is placed in the center of the forehead, forming a stable differential acquisition system. The signal acquisition circuit uses a high-input-impedance bioelectric amplifier with strong common-mode rejection capabilities, effectively suppressing electromagnetic interference in industrial environments. The sampling frequency is set to a standard sampling rate that meets the requirements of EEG signal spectrum analysis, ensuring the capture of fatigue-related characteristic frequency bands. Signal preprocessing includes bandpass filtering, power frequency notch filtering, and baseline drift removal to extract EEG signals suitable for fatigue analysis.

[0020] Mirror symmetry analysis of EEG signals was performed to determine the activity difference between the left and right hemispheres. Signal grouping divided the symmetrical electrode signals according to their location in the left and right hemispheres, forming left-brain and right-brain signal groups. Temporal activity calculation involved sliding window analysis for each channel signal, with window length and overlap set to ensure temporal continuity. The root mean square value of the signal within the window served as the instantaneous activity index for that channel, reflecting the intensity level of EEG activity. Frequency domain activity was calculated by extracting the power of each characteristic frequency band using short-time Fourier transform, including the delta, theta, alpha, and beta wave bands closely related to fatigue. Left-brain activity was obtained by spatially weighting the temporal and frequency domain characteristics of each channel in the left-brain group. Right-brain activity was calculated using the same method to determine the overall activity of the right-brain group signals. Spatial weights were determined based on the correlation between electrode location and fatigue-sensitive brain regions, with higher weights assigned to the prefrontal cortex and lower weights to the occipital cortex. The activity difference is calculated using the normalized difference formula, ΔA=(A_left-A_right) / (A_left+A_right), to eliminate the influence of the individual's baseline activity level.

[0021] The EEG imbalance index is generated based on activity difference. The instantaneous imbalance component I_inst maps the difference ΔA to a standard interval using a sigmoid function; a larger difference indicates more severe instantaneous imbalance, reflecting the degree of imbalance at the current moment. The cumulative imbalance component I_cum is calculated by integrating the difference over time, considering the absolute levels of activity in both hemispheres. When A_left or A_right remains consistently high, the cumulative weight is increased. The integration window covers the fatigue accumulation process over the past 5 minutes, reflecting the persistent effect of the imbalance state. The trend component I_trend is calculated by combining the rate of change of the difference and frequency domain power change characteristics; an upward trend indicates increasing fatigue, while a downward trend indicates easing fatigue, predicting the direction of fatigue development. Individual baseline calibration establishes an individual characteristic baseline by collecting awake and resting state data before the task begins, eliminating the influence of individual differences. The relative imbalance index R quantifies the degree of deviation of the current state from the individual baseline; a larger deviation indicates a higher risk of fatigue. The time-period correction factor T_factor is adjusted according to physiological rhythms. A correction coefficient of 0.8 is used for night shifts, a standard coefficient of 1.0 for day shifts, and a transition coefficient of 0.9 for shift handover. The comprehensive imbalance index is calculated through multi-component weighted fusion: BEI = 100 × (0.4 × I_inst + 0.3 × I_cum + 0.2 × I_trend + 0.1 × R) × T_factor. The weight of each component is determined based on its contribution to fatigue prediction. The instantaneous component has the highest weight, reflecting real-time performance, followed by the cumulative component, reflecting persistence. The trend component and relative indicators provide auxiliary information. The index grading maps BEI values ​​to fatigue levels: 0-30 for normal, 30-50 for mild fatigue, 50-70 for moderate fatigue, and 70-100 for severe fatigue. For example, when miners operate coal mining machines for extended periods, they need to continuously identify coal seam interfaces and monitor equipment status. These cognitive tasks primarily activate the right hemisphere of the brain. As the underground working time increases, the activity level of the right hemisphere remains higher than that of the left hemisphere, the difference in activity level gradually increases, and the accumulated imbalance component rises rapidly. When the comprehensive imbalance index exceeds 50, the system issues a moderate fatigue warning.

[0022] Step S120: Perform spectral analysis on the EEG signal to extract the dominant frequency distribution, collect noise signals in the work environment and identify their spectral gaps to obtain the noise silence frequency band, and map the dominant frequency distribution to the noise silence frequency band to generate intervention window parameters.

[0023] Specifically, spectral analysis was performed on the EEG signal to extract the dominant frequency distribution. A 2-second time window was used to segment the continuous EEG signal, with 50% overlap between windows to ensure the continuity of spectral evolution. A Hamming window was chosen to reduce spectral leakage and improve frequency resolution. Fast Fourier Transform (FFT) was used to convert the time-domain signal to the frequency domain, with 1024 FFT points, corresponding to a frequency resolution of approximately 0.5 Hz. Power spectral density was calculated using the periodogram method, performing modulo-square operations on the complex spectrum at each frequency point to obtain the power value. Multi-channel spectral fusion was achieved by weighted averaging of the power spectra of all electrode channels, with the weights dynamically adjusted based on the signal quality of each channel. Fatigue characteristic frequency band identification focused on frequency ranges related to fatigue states, including the theta wave enhancement band (4-8 Hz), the alpha wave suppression band (8-12 Hz), and the beta wave decline band (13-30 Hz). Dominant frequency extraction was achieved by identifying the peak frequencies within each characteristic band; the peak detection algorithm required the peak power to exceed 1.5 times the average power of neighboring frequencies. When the imbalance index BEI exceeds 50, the abnormal enhancement components of the theta band and the suppression features of the alpha band are extracted first.

[0024] Noise-silent frequency bands were identified through spectral analysis using environmental microphone data acquisition. Environmental noise was collected via an external microphone mounted on a safety helmet, with sampling parameters consistent with those used for EEG signals to ensure comparability of the spectral analysis. Noise signal preprocessing included DC component removal and normalization to adjust noise signals of varying intensities to a uniform analytical scale. Spectral analysis employed the same FFT parameters as the EEG signals to maintain consistent frequency resolution. A stable spectral envelope was obtained through long-term averaging of the noise power spectrum, with an averaging duration of at least 30 seconds to cover various transient noises. Spectral envelope extraction used a polynomial fitting method, with the fitting order adaptively adjusted based on noise complexity. Silent frequency band identification was achieved by setting a power threshold; a frequency band whose average power was below 20% of the overall average power was marked as a potential silent frequency band. Continuity checks ensured that silent frequency bands had sufficient bandwidth, with a minimum bandwidth requirement of at least 2Hz to accommodate modulated signals. Frequency band stability was assessed by calculating the power variance over different time periods; frequency bands with variances less than a threshold were considered stable silent frequency bands. Noise quiescent frequency bands, including center frequency, bandwidth, noise floor level, and stability indicators, are determined through noise spectrum analysis.

[0025] In some embodiments, the step of mapping the dominant frequency distribution to the noise-silent frequency band to generate intervention window parameters includes: extracting peak frequency components from the dominant frequency distribution; finding an embedding position within the noise-silent frequency band that matches the peak frequency components; determining a frequency offset based on the embedding position and the dominant frequency distribution; and generating intervention window parameters based on the frequency offset.

[0026] Peak frequency components are extracted from power spectral density data based on the dominant frequency distribution through local maximum search. A sliding window method is employed, with a window width of 1 Hz and a step size of 0.1 Hz, searching for the maximum power within each window. The search process begins at the starting frequency of the spectrum, calculates the power values ​​of all frequency points within the current window, and records the maximum power value and its corresponding frequency. The peak determination criterion requires that the power of a candidate peak exceeds 1.5 times the average power of its neighboring frequencies. Neighboring frequencies are defined as a 0.5 Hz range before and after the peak. The average power within this range is used as a reference benchmark to exclude spurious peaks in flat spectra. In the case of multiple peaks, they are sorted by power magnitude, and the top three most significant peaks are retained as the main frequency components. Peak feature extraction includes peak frequency f_peak, peak power P_peak, and peak width W_peak, where f_peak is accurate to 0.01 Hz. The peak width is obtained through -3 dB bandwidth measurement, i.e., the frequency range decreasing by 3 dB from the peak power, reflecting the concentration of frequency components. Peak extraction prioritizes peak components in fatigue characteristic frequency bands such as theta wave enhancement (4-8Hz) and alpha wave suppression (8-12Hz). When abnormal peaks are detected in the theta wave band, they are taken as the primary intervention target.

[0027] The embedding location is determined through an optimized search based on the spectral characteristics of the silent frequency band and the intervention requirements of the peak frequency. The search range is limited to the identified silent frequency band, and all possible embedding locations are traversed by a sliding search window with a search step of 0.5 Hz. The matching degree evaluation function is M(f) = w1 × N(f) + w2 × B(f) + w3 × S(f), where N(f) is the reciprocal of the noise floor level, i.e., 1 / noise floor power; the lower the noise floor, the larger the value of N(f); B(f) is the available bandwidth, defined as the continuous frequency range without interference before and after the frequency point; S(f) is the frequency band stability, obtained by calculating the reciprocal of the power variance of the frequency point within the time window. The weighting coefficients w1, w2, and w3 correspond to the importance of noise floor, bandwidth, and stability, respectively, with typical values ​​of 0.5, 0.3, and 0.2, satisfying w1 + w2 + w3 = 1.0. The optimal location is determined by maximizing the matching degree function M(f). The M value is calculated by traversing all candidate locations, and the frequency point with the largest M value is selected as the embedding location. Bandwidth matching requires that the available bandwidth at the embedding location be no less than twice the bandwidth of the peak frequency component, i.e., B(f_embed) ≥ 2 × W_peak, to reserve space for frequency modulation. The center frequency selection considers avoiding power frequency harmonics, checking whether candidate frequencies are integer multiples of 50Hz; if so, skip those frequencies and prioritize frequencies that are not integer multiples of 50Hz.

[0028] The frequency offset is determined by calculating the difference between the embedding position frequency and the peak frequency. The basic offset Δf = f_embed - f_peak, where f_embed is the embedding position frequency and f_peak is the peak frequency; the offset can be positive or negative. When multiple peak frequencies exist, their respective offsets Δf1, Δf2, and Δf3 are calculated. The offset accuracy is controlled within 0.1Hz, achieved through a phase accumulator in digital frequency synthesis technology. The phase accumulator has at least 32 bits to ensure a frequency resolution of 0.01Hz. A relative offset preservation mechanism ensures that the original frequency interval relationship is maintained after mapping multiple frequency components, i.e., (f_peak2 - f_peak1) = (f_embed2 - f_embed1), keeping the relative relationship between frequency components unchanged. The time-varying characteristic of the offset is achieved by establishing an offset sequence Δf(t), calculating the current offset every 5 seconds to form a time series. Offset boundary constraints ensure that the mapped frequency does not exceed the effective range of the silent band. An upper limit f_embed_max and a lower limit f_embed_min are set. When f_peak + Δf exceeds the boundaries, Δf is automatically adjusted to ensure the mapped frequency falls within the effective range. Offset smoothing employs a first-order low-pass filter to avoid frequency jumps caused by abrupt offset changes.

[0029] Intervention window parameters are generated based on frequency offset through parameter configuration. The window center frequency f_center = f_peak + Δf is set as the mapped target frequency, with an accuracy of 0.1Hz. The window bandwidth B_window is determined comprehensively based on the original peak width W_peak and the available bandwidth B_available in the silent band, calculated as B_window = min(2 × W_peak, 0.8 × B_available), which must cover the mapped frequency components while reserving sufficient space for carrier modulation. The bandwidth setting follows these rules: when W_peak < 1Hz, B_window = 2Hz to ensure minimum bandwidth; when W_peak is between 1-2Hz, B_window = 2 × W_peak; when W_peak > 2Hz, B_window = W_peak + 2Hz, while not exceeding 80% of the available bandwidth. The frequency range of the window boundary [f_center - B_window / 2, f_center + B_window / 2] defines the selectable range of carrier frequencies. The window shape parameter adopts a raised cosine characteristic, with a roll-off factor α set to 0.35, achieving a balance between spectral efficiency and adjacent channel interference suppression. The frequency resolution parameter specifies the accuracy requirement for carrier frequency selection, with a typical value of 0.1Hz, consistent with the offset accuracy. The guard interval parameter reserves a 0.5Hz frequency interval at each window boundary to prevent signal power leakage outside the window from affecting adjacent frequency bands. When the EEG signal exhibits an abnormal theta wave peak at 6.5Hz, it is mapped to the 48Hz silent band, generating intervention window parameters with a center frequency of 48Hz and a bandwidth of 3Hz, with a window range of [46.5Hz, 49.5Hz], providing precise frequency localization for fatigue intervention signal generation.

[0030] Step S130: Select the optimal carrier frequency point based on the intervention window parameters, generate a pure carrier signal based on the optimal carrier frequency point, and modulate and fuse the pure carrier signal through the EEG imbalance index to generate a composite intervention signal.

[0031] Specifically, based on the intervention window parameters (center frequency 48Hz, bandwidth 3Hz), the carrier frequency search range is set to [46.8Hz, 48.8Hz], and fine scanning is performed within this range in 0.1Hz increments. Multi-dimensional indicators are employed, including spectral isolation from environmental noise, frequency spacing from power frequency harmonics, and long-term stability of the frequency point. Isolation is calculated by measuring the frequency difference between the candidate frequency point and the most recent noise peak; higher isolation is better. A power frequency harmonic avoidance mechanism ensures that the carrier frequency point does not fall near integer multiples of 50Hz, with an avoidance range set at ±0.5Hz. Frequency point stability is assessed by continuously monitoring the power fluctuation of the frequency point over a specific time period; smaller fluctuations indicate greater stability. A comprehensive score is achieved by weighted combination of various indicators, with noise isolation having the highest weight, followed by harmonic spacing, and stability having a relatively lower weight. A dynamic adjustment mechanism updates the carrier frequency point based on real-time noise changes, automatically switching to a candidate frequency point when the original frequency point is subjected to new interference. Frequency locking is achieved through phase-locked loop technology to achieve high-precision control, ensuring the accuracy and stability of the carrier frequency. Multidimensional evaluation and optimization selection determined the optimal carrier frequency.

[0032] A pure carrier signal is generated using digital synthesis technology based on the optimal carrier frequency. A lookup table method is employed, with the sine waveform lookup table containing high-density sampling points to ensure phase resolution. The sampling frequency is set to be more than 100 times the carrier frequency, satisfying the Nyquist sampling theorem and allowing sufficient margin. The phase accumulator performs high-precision phase accumulation, with the accumulation step size precisely set according to the ratio of the carrier frequency to the sampling frequency. Amplitude control is implemented using a high-resolution digital-to-analog converter to ensure high fidelity of the carrier signal. The initial phase is set to zero to ensure a deterministic start of the carrier signal. Carrier purity is enhanced through digital filtering using a narrowband bandpass filter with a passband range set to a narrow interval near the carrier frequency, effectively suppressing out-of-band noise. Harmonic suppression is achieved through precise calculation and filtering of the lookup table data, keeping total harmonic distortion to an extremely low level. DC bias is removed through a high-pass filter to ensure the zero-mean characteristic of the carrier signal. A double-buffer mechanism is used for signal buffering to ensure the continuity and real-time performance of the output. Digital synthesis technology produces a pure carrier signal with accurate frequency, stable phase, and extremely low distortion.

[0033] In some embodiments, the step of modulating and fusing the pure carrier signal using the EEG imbalance index to generate a composite intervention signal includes: amplitude modulating the pure carrier signal to generate an anti-fatigue audio sequence; identifying fatigue-sensitive feature signals based on the EEG imbalance index; performing time-reversal processing on the fatigue-sensitive feature signals to form a fatigue-resistant waveform; and embedding the fatigue-resistant waveform into the anti-fatigue audio sequence to complete the generation of the composite intervention signal.

[0034] An anti-fatigue audio sequence is generated by applying a time-varying envelope function to a clean carrier signal. The envelope design employs a dual-frequency modulation mode: the main modulation frequency is set to 0.2Hz to correspond to the delta wave rhythm, and the secondary modulation frequency is set to 0.05Hz to generate ultra-slow fluctuations. The modulation envelope function is A(t) = 1 + m1 × sin(2π × 0.2 × t) + m2 × sin(2π × 0.05 × t), where A(t) is the time-varying envelope amplitude, t is time, m1 = 0.4 is the main modulation depth, and m2 = 0.2 is the secondary modulation depth. A dynamic mapping relationship is established between the modulation depth and the brain electrical imbalance index: m_AM = 0.3 + 0.4 × (BEI / 100), where m_AM is the real-time modulation depth, and BEI is the brain electrical imbalance index; the higher the fatigue level, the deeper the modulation. The modulated signal is obtained by multiplying the envelope by the carrier, maintaining the continuity of the carrier phase. Envelope smoothing uses low-pass filtering to eliminate high-frequency components in the envelope that may cause auditory discomfort. Power normalization ensures consistent average power before and after modulation, preventing signals from being too strong or too weak. Sequence segmentation divides the continuous modulated signal into segments of appropriate length, with smooth transitions between segments to avoid abrupt changes. The sequence buffering mechanism supports real-time adjustment and continuous playback.

[0035] The fatigue-sensitive characteristic signals are identified based on the EEG imbalance index. This is achieved by analyzing the correlation between the BEI change rate and external stimuli, calculating the first derivative of the BEI as the change rate indicator, and marking periods of high correlation (correlation coefficient > 0.7) as sensitive periods. Feature extraction employs a sliding window method with a window length of 30 seconds and a step size of 5 seconds, extracting statistical features of the BEI within the window, including mean, standard deviation, kurtosis, and trends such as slope and inflection point location. Spectral features are obtained by performing a short-time Fourier transform on the BEI sequence, focusing on low-frequency oscillations in the 0.01-0.5Hz range, which reflects the slow fluctuations in fatigue state. Principal component analysis identifies the main patterns of BEI changes; the first three principal components typically explain more than 85% of the variance, extracting the feature vectors that best represent the fatigue response. Pattern matching, through comparison with a fatigue feature database, calculates Euclidean distance to identify individual-specific fatigue response patterns. Dynamic tracking uses an exponentially weighted moving average to continuously update sensitive features, with a learning rate set to 0.2, adapting to different stages of fatigue development. The time-domain waveform of the characteristic signal is reconstructed through inverse transformation, preserving the key dynamic characteristics of the fatigue response.

[0036] For example, the step of performing time-reversal processing on the fatigue-sensitive feature signal to form a fatigue-resistant waveform includes: extracting the time-domain envelope of the fatigue-sensitive feature signal; flipping the time-domain envelope on the time axis to determine the reverse timing sequence; and adjusting the phase of the reverse timing sequence according to the fatigue-sensitive feature signal to generate the fatigue-resistant waveform.

[0037] First, the time-domain envelope of the fatigue-sensitive feature signal is extracted, and the Hilbert transform method is used to achieve this. The instantaneous amplitude A(t) = |z(t)| is obtained by constructing the analytic signal z(t) = x(t) + jH[x(t)], where x(t) is the original signal and H[x(t)] is the Hilbert transform, forming a smooth envelope curve. Envelope smoothing uses a moving average with a window length of 0.5 seconds corresponding to 50 sampling points, removing rapid fluctuations while retaining the main trend. Envelope feature point identification includes peak and valley positions, achieved by detecting the zero-crossing points of the first derivative. The peak interval is typically 2-5 seconds, and these points correspond to critical moments in fatigue development. Envelope morphology analysis identifies rising and falling edge characteristics. A typical fatigue envelope exhibits an asymmetry of rapid rise and slow fall, with a rise-to-fall ratio of approximately 1:3, reflecting the time characteristics of fatigue accumulation and recovery. Next, the time-domain envelope is flipped along the time axis to obtain the reverse time series. Time reversal is achieved by inverting the time index of the envelope sequence, E_rev(t) = E(Tt), where T is the total signal duration, thus reversing the temporal order. After reversal, a rapid rise becomes a rapid fall, and a slow fall becomes a slow rise, creating a time pattern opposite to the original fatigue development. Boundary processing employs a mirror extension method, extending the sequence by 10% (0.1T) at both ends to ensure the continuity of the signal after reversal. Amplitude adjustment makes the reverse envelope 0.8 times the original envelope, E_rev_scaled = 0.8 × E_rev, avoiding overstimulation. Finally, the phase of the reverse timing sequence is adjusted based on the fatigue-sensitive characteristic signal to generate a fatigue countermeasure waveform. Phase adjustment is optimized based on the instantaneous phase information of the fatigue characteristic signal, which is obtained by analyzing the arctangent of the ratio of the imaginary to the real part of the signal. The phase offset is set to π to ensure inversion, φ_counter = φ_original + π. By setting an appropriate phase offset, the countermeasure waveform is ensured to be inversely related to the fatigue waveform, with a correlation coefficient close to -1. Phase transition employs cubic spline interpolation for a smooth transition between feature points, avoiding abrupt phase changes. Carrier reconstruction uses the adjusted envelope and phase information, s_counter(t) = E_rev_scaled(t) × cos(φ_counter(t)), to generate a complete adversarial waveform. Waveform energy normalization ensures the total energy is comparable to the original signal; the amplitude is adjusted by calculating the energy ratio to ensure appropriate signal strength. Time alignment uses a cross-correlation function to find the minimum point, achieving precise anti-alignment and guaranteeing the accuracy of the adversarial effect.

[0038] The fatigue-resistance waveform is embedded into the fatigue-resistance audio sequence to generate the composite intervention signal. The fusion formula is S_composite(t) = w_audio × S_AM(t) + w_counter × s_counter(t), where S_composite(t) is the composite intervention signal, S_AM(t) is the fatigue-resistance audio sequence, s_counter(t) is the fatigue-resistance waveform, and w_audio and w_counter are the corresponding weight coefficients, satisfying w_audio + w_counter = 1 to ensure signal amplitude normalization. The weight allocation is dynamically adjusted according to the fatigue level BEI: w_audio = 0.7 - 0.2 × (BEI / 100), w_counter = 0.3 + 0.2 × (BEI / 100). For mild fatigue (BEI < 50), the audio sequence weight reaches 0.6, and for severe fatigue (BEI > 70), the weight of the resistance waveform increases to 0.44. Time synchronization is achieved by determining the optimal embedding time through cross-correlation analysis. The peak of the adversarial waveform is embedded at the trough of the audio sequence to ensure optimal embedding and maximize the intervention effect. Transition smoothing employs a raised cosine window function before and after the embedding point, with a window length of 200ms and a roll-off factor of 0.3, to avoid auditory abrupt changes and spectral distortion. Power limiting controls the root mean square value of the synthesized signal to within 0.8 and the instantaneous peak value to not exceed 0.95, ensuring signal safety and preventing excessive stimulation from damaging the auditory system. The resulting composite intervention signal maintains both rhythmic induction and active fatigue countermeasures. Spectral analysis verifies the independence and integrity of the two components, achieving a dual intervention effect.

[0039] Step S140: Monitor the coupling relationship between the operator's head micro-movements and EEG signals to identify fatigue precursor characteristics, and determine the timing of releasing composite intervention signals based on fatigue precursor characteristics.

[0040] In some embodiments, the method of monitoring the coupling relationship between the operator's head micro-movements and the electroencephalogram (EEG) signal to identify fatigue precursor features includes: acquiring the head micro-movements to form a micro-movement time series; extracting waveform segments corresponding to the time from the EEG signal based on the micro-movement time series; analyzing the correlation between the micro-movement time series and the waveform segments to determine the coupling strength; and identifying fatigue precursor features based on the coupling strength.

[0041] Micro-movements are collected to obtain a micro-movement time series. Continuous monitoring is achieved through the inertial measurement unit built into the helmet. A three-axis accelerometer measures the linear acceleration of the head in the forward / backward, left / right, and up / down directions, with a range of ±4g and a resolution of 0.001g. A three-axis gyroscope records the angular velocity changes of the head, with a range of ±250° / s, capturing rotation and tilting movements. Data fusion employs a complementary filtering algorithm, combining acceleration and angular velocity information to calculate the head attitude angle. A micro-movement recognition algorithm extracts feature events from the raw signal, including head nodding (pitch angle change exceeding a set threshold), head shaking (yaw angle change exceeding a set threshold), and tilting (roll angle continuous deviation exceeding a set threshold). Time series construction records the occurrence time, type, amplitude, and duration of each micro-movement. Noise filtering uses Kalman filtering to remove interference caused by normal activities such as walking. Movement classification labels the detected micro-movements with their types, forming structured micro-movement time series data. When the S110's EEG imbalance index (BEI) is at a high level, the detection threshold for micro-movements is correspondingly lowered, improving sensitivity to subtle fatigue movements.

[0042] Waveform segments corresponding to specific moments in EEG signals were extracted based on micromotion time series. Time alignment used the moment of micromotion occurrence as the anchor point, extracting 2 seconds of EEG data before and after it. Segment extraction considered physiological delays; the EEG response typically lags behind the action by 50-300 ms, and the extraction window was shifted accordingly. Multi-channel selection focused on electrodes in the central region related to motor control and the fatigue-sensitive prefrontal cortex. Waveform preprocessing included removing baseline drift and power line interference, retaining effective fatigue-related frequency bands. Event-related potential (ERP) extraction improved the signal-to-noise ratio by superimposing and averaging EEG segments corresponding to multiple micromotions of the same type. Time-frequency analysis used short-time Fourier transform to obtain the time-frequency characteristics of each segment. Phase extraction calculated instantaneous phase information using Hilbert transform. Segment annotation included the corresponding micromotion type and parameters, establishing a one-to-one correspondence between EEG segments and micromotions. The EEG segment characteristics during periods of rising BEI were analyzed in conjunction with the changing trend of the EEG imbalance index (BEI).

[0043] The coupling strength is obtained by analyzing the correlation between the micro-motion time series and waveform segments. Correlation analysis employs multiple methods to comprehensively assess the degree of association between the two signals. Time-domain correlation is calculated using the cross-correlation function, with the maximum correlation coefficient serving as the coupling strength indicator. Frequency-domain coherence is calculated, with the average coherence value in the fatigue-related frequency band reflecting the coupling degree. The phase-locked value (PLV) assesses the synchronization strength by calculating the stability of the phase difference between the micro-motion and EEG signals; the expression is that PLV equals the modulus of the average complex exponential value of the phase difference. Multi-scale analysis calculates coupling indices at different time scales, capturing fast and slow coupling patterns. The comprehensive coupling strength index C_total is obtained by weighted combination of time-domain correlation, frequency-domain coherence, and phase-locked value, with a weight of 0.4 for time-domain correlation, 0.3 for frequency-domain coherence, and 0.3 for phase-locked value. When the EEG imbalance index (BEI) exceeds 50, coupling analysis focuses on coherence changes in the low-frequency band (0.5-4 Hz), which is closely related to fatigue states.

[0044] For example, the step of identifying fatigue precursor features based on the coupling strength includes: performing periodic analysis on the coupling strength to generate a coupling rhythm; identifying abnormal breakpoints in the coupling rhythm; extracting features from the abnormal breakpoints to generate a fatigue warning signal; and determining fatigue precursor features based on the fatigue warning signal.

[0045] First, rhythmic features are extracted using multiple methods based on the time series of coupling strength. Autocorrelation function analysis identifies repetitive patterns in coupling strength; a peak in the autocorrelation function at a certain delay indicates the presence of a rhythm with that period. Power spectral density estimation identifies the dominant frequency components of coupling strength variations. Wavelet decomposition extracts periodic components at different time scales; low-frequency scales correspond to slow fatigue accumulation rhythms, while high-frequency scales reflect rapid alertness fluctuations. Rhythm parameters include the principal period, rhythm strength, and phase stability. Under normal conditions, coupling strength exhibits a stable long-period rhythm, while during fatigue development, the rhythm period shortens and the amplitude increases, reflecting changes in the brain's regulatory mechanisms. Next, abnormal breakpoint detection is based on sudden interruptions or significant changes in rhythm continuity. Breakpoints are identified by calculating abrupt changes in local statistics; a breakpoint is marked when the difference between the mean or variance of the preceding and following windows exceeds a threshold. Phase slip detection monitors instantaneous phase jumps; a phase difference exceeding a set threshold indicates rhythm interruption. Amplitude anomalies are identified by sudden drops or disappearances of rhythm amplitude. Rhythm lockout is determined by a sharp drop in the phase-locked value. Duration analysis distinguishes between transient disturbances and true rhythmic fractures; only anomalies lasting more than several cycles are considered valid fractures. Then, early warning indicators are constructed by extracting features from the fracture point. The fracture feature vector includes fracture depth, fracture width, recovery time, and accumulated features before fracture. Pattern classification categorizes fractures into progressive, sudden, and oscillating types. A hazard score is calculated based on fracture features; fractures with greater depth, wider width, and slower recovery are assigned higher hazard scores. Time-series correlation analysis establishes the relationship between the fracture point and other physiological indicators, creating a multimodal early warning model. Warning levels are set based on the hazard score, including attention level, warning level, and critical level. Finally, fatigue warning signals determine fatigue precursor characteristics. The feature vector comprehensively characterizes fatigue precursors, including rhythm parameters, fracture characteristics, warning level, and time label. Feature classification labels warning signals as early precursors, developmental precursors, and critical precursors. An individual feature database records typical precursor patterns for each worker, improving identification accuracy. A dynamic update mechanism updates feature parameters based on new warning events. Reliability assessment verifies the effectiveness of the precursor characteristics through consistency with subsequent fatigue development.

[0046] The timing of releasing composite intervention signals is determined based on fatigue precursor characteristics. First, fatigue development stages are identified through temporal pattern recognition of fatigue precursor characteristics. A state inference model is used to identify stages based on the time series of fatigue precursor characteristics, with four stages defined: alertness, mild fatigue, moderate fatigue, and severe fatigue. Feature sequence preprocessing uses a sliding window to extract statistical features, and pattern matching identifies the current fatigue stage. Stage transition detection is achieved by monitoring abrupt changes in feature patterns, and individual difference calibration is performed by updating model parameters online to adapt to the fatigue patterns of different workers. Combined with the real-time value of the Brain Energy Imbalance Index (BEI), a rapid increase in BEI indicates an early entry into the next fatigue stage. Then, the corresponding intervention intensity level is matched according to the fatigue development stage. Intensity levels are defined as four levels: prevention, mild intervention, moderate intervention, and emergency intervention. Prevention uses intermittent low-intensity stimulation; mild intervention increases intensity and frequency; moderate intervention uses a continuous activation mode; and emergency intervention uses the highest intensity combined with other warning measures. The gradual adjustment strategy smoothly transitions intensity during phase transitions to avoid sudden stimulation. Adaptive adjustment dynamically optimizes intensity parameters based on intervention effect feedback. Safety limits ensure that the maximum intervention intensity does not exceed the auditory comfort threshold. Finally, the release timing of the composite intervention signal is determined based on the intervention intensity level and fatigue precursor characteristics, comprehensively considering the urgency of fatigue development and the optimal intervention window. The timing score is calculated by comprehensively considering fatigue urgency, current sensitivity, and the most recent intervention interval. Fatigue urgency considers fatigue level and deterioration rate. Sensitive window identification is achieved by analyzing the fluctuation cycle of coupling intensity; releasing the intervention during the rising coupling intensity period yields the best effect. Minimum interval limits prevent excessively frequent interventions. Predictive release intervenes early when an accelerating trend of fatigue development is detected. Personalized timing optimization adjusts the release strategy by recording historical intervention effects. The emergency priority mechanism releases immediately upon detecting dangerous precursors, regardless of interval limitations. When the EEG imbalance index (BEI) exceeds 70 or a deep break in the coupling rhythm occurs, the emergency intervention signal is immediately triggered.

[0047] Step S150: Play the composite intervention signal in the noise-silent frequency band according to the release timing, and monitor the changes in the EEG imbalance index during the playback to generate a fatigue relief curve.

[0048] Specifically, a composite intervention signal is played through a helmet speaker in a noise-silent frequency band based on the release timing. Playback control is triggered by the T_score value at the release timing; signal output is initiated immediately when the score exceeds a threshold. The speaker employs a dual-mode design of bone conduction and air conduction, with bone conduction handling the primary signal transmission and air conduction providing spatial enhancement. Precise frequency band positioning positions the composite intervention signal within the 45-55Hz noise-silent frequency band determined by S120, ensuring the signal is not masked by ambient noise. Playback power is set according to the noise floor level of the silent frequency band, maintaining a signal-to-noise ratio 10-15dB above the noise floor. Signal preprocessing includes digital-to-analog conversion, power amplification, and impedance matching to ensure high-fidelity output. Playback timing strictly adheres to the intervention signal's time parameters, lasting 8 seconds, including a 3-second fade-in and a 3-second fade-out. Frequency locking ensures carrier frequency accuracy through a phase-locked loop, with deviation controlled within ±0.01Hz. Real-time monitoring detects the speaker output status to ensure complete signal playback. Environmental adaptation fine-tunes the output power based on real-time noise changes to maintain a constant perceived intensity. Playback is synchronized with precise start and end time markers, providing a time baseline for subsequent performance evaluation. Multiple protection mechanisms ensure reliable signal transmission in complex industrial environments, automatically adjusting the output strategy when sudden changes in environmental noise are detected.

[0049] The system monitors changes in the brain electrical imbalance index (BEI) during playback to generate a fatigue relief curve. Real-time monitoring begins 5 seconds before playback, establishing a baseline reference value (BEI_baseline) and maintaining high-frequency sampling throughout the signal playback. The monitoring system is based on the S110 BEI calculation module, with the sampling frequency increased to 10Hz, updating the BEI value every 100ms to capture rapid changes. Time synchronization ensures precise correspondence between BEI sampling and the playback signal, achieving millisecond-level synchronization accuracy through timestamp matching. Multi-channel fusion integrates the imbalance indices of various brain regions, focusing on the changes in the prefrontal and central regions. Noise filtering employs adaptive filtering to remove electromyography (EMG) and electrooculography (EOG) interference, ensuring the accuracy and stability of BEI calculation. The change trajectory exhibits a typical three-segment characteristic: a rapid response segment where the BEI drops rapidly from the baseline, typically reaching its lowest point within 2-3 seconds of playback start; a stable maintenance segment where the BEI fluctuates at a low level, reflecting effective fatigue suppression; and a slow recovery segment where the BEI gradually rises after playback ends, with the recovery speed significantly slower than the decline speed. Trajectory modeling employs piecewise functions: the response segment is fitted with an exponential decay function, the stable segment is represented by a constant value plus small fluctuations, and the recovery segment is modeled using an exponential recovery function. The fatigue relief curve is generated by smoothing and normalizing the original trajectory, and spline interpolation is used at the segment connection points to ensure curve continuity. Key parameters are extracted to comprehensively quantify the fatigue relief effect: relief depth reflects the degree of fatigue reduction and is calculated as the difference between the baseline value and the lowest value; response speed is characterized by the time constant of the descent segment, with a smaller time constant indicating a faster response; stability duration measures the duration the BEI remains low, reflecting the persistence of the intervention effect; recovery characteristics are described by the time constant of the recovery segment, revealing the speed of fatigue rebound. The effectiveness evaluation index comprehensively considers relief depth, response speed, and stability duration; the intervention is considered effective when the relief depth exceeds 30% of the baseline and the stability duration exceeds 5 seconds. Individual difference analysis records typical relief curves for different workers, establishing personalized effectiveness evaluation benchmarks.

[0050] Step S160: Perform slope analysis on the fatigue relief curve to generate an intervention effect score, monitor the real-time changes in the EEG imbalance index after the composite intervention signal stops playing to establish a recovery sensitivity curve, and identify the moment in the recovery sensitivity curve when the brain's demand for intervention is strongest to determine the optimal interval duration.

[0051] Specifically, slope analysis is performed on the fatigue relief curve to obtain an intervention effect score. Slope analysis quantifies the immediate effect and overall performance of the intervention by calculating the rate of change at each stage of the relief curve. The slope of the response segment reflects the speed of fatigue relief, and the instantaneous rate of change is obtained by differentiating the curve. The maximum slope usually appears at the beginning of the intervention, representing the strongest response of the brain to the stimulus. The average slope integrates the entire response process, eliminating the influence of instantaneous fluctuations. Slope persistence is assessed by calculating the duration of high slope; sustained rapid relief is more valuable than a brief, intense reaction. The relief depth weight W_depth = D / (D + 0.5), where D is the relief depth, converting the degree of relief into a scoring factor in the 0-1 range. The greater the depth, the higher the weight, but there is a saturation effect. The response speed weight W_speed adopts an exponential function form; a rapid response receives a higher evaluation. The stability weight W_stable considers the duration of the relief effect; maintaining a low fatigue level for a long time is the ideal state. The overall score S_effect = 100 × (0.4 × W_depth + 0.3 × W_speed + 0.3 × W_stable), where each weight coefficient is assigned according to its importance, quantifying the overall intervention effect. The effect score will be used for subsequent dynamic adjustment of monitoring parameters and optimization of individual intervention strategies.

[0052] A recovery sensitivity curve was established by monitoring real-time changes in the EEG imbalance index after the termination of playback of the composite intervention signal. The moment playback stopped marks the brain's transition from an externally supported state to an autonomic regulatory mode, a process rich in physiological information. Continuous monitoring began at the end of playback and lasted for 10 minutes to capture the complete recovery dynamics. The BEI sampling frequency was maintained at 10 Hz, i.e., updated every 100 milliseconds, forming a high temporal resolution data stream. The monitoring strategy was dynamically adjusted according to the intervention effect score S_effect. For high scores, the monitoring time was appropriately extended to fully utilize the intervention effect, while for low scores, the monitoring density was increased to detect accelerated recovery in a timely manner. The recovery process exhibited complex nonlinear characteristics: initially, there was a plateau period, with the BEI remaining at a low level after the intervention for 30 seconds to 2 minutes; then, a slow recovery phase began, with the BEI gradually returning to the fatigue baseline; a key turning point occurred at a certain moment, with the recovery speed suddenly accelerating, reflecting the reactivation of the brain's endogenous fatigue mechanisms. Sensitivity is defined as the ratio of the recovery rate to the remaining recovery space, S(t) = dBEI / dt / (BEI_target - BEI_current), where dBEI / dt is the time derivative of BEI, BEI_target is the target recovery value, and BEI_current is the current value, quantifying the recovery speed per unit fatigue difference. The sensitivity curve exhibits a characteristic pattern: slow recovery in the early stage corresponds to low sensitivity, typically below 0.1; during the accelerated recovery in the middle stage, sensitivity increases sharply, reaching above 1.0; and in the later stage, recovery slows down and sensitivity declines. This slow-then-fast-then-slow recovery sensitivity curve fully describes the dynamic process of brain fatigue recovery.

[0053] In some embodiments, identifying the moment in the recovery sensitivity curve when the brain's need for intervention is strongest as the optimal interval duration includes: performing slope analysis on the recovery sensitivity curve to generate slope change data; identifying the starting point of accelerated rise from the slope change data; performing sensitivity threshold analysis on the starting point to obtain critical sensitivity features; and extracting the time value corresponding to the starting point based on the critical sensitivity features to determine the optimal interval duration.

[0054] The slope change data is obtained by differentiating the recovery sensitivity curve S(t). The slope is calculated using a numerical differentiation method, where the instantaneous rate of change is obtained by dividing the sensitivity difference between adjacent points by the time interval. The basic formula is k(t) = [S(t+Δt) - S(t)] / Δt, where Δt is the sampling interval of 0.1 seconds. To reduce computational noise, a five-point central difference formula is used: k(t) = [-S(t-2Δt) + 8S(t-Δt) - 8S(t+Δt) + S(t+2Δt)] / (12Δt), combined with a weighted average. The weighting coefficients follow a distribution of [0.1, 0.2, 0.4, 0.2, 0.1]. The slope sequence k_s(t) visually reflects the dynamic trend of sensitivity change; positive values ​​indicate increased demand, negative values ​​indicate decreased demand, and the absolute value reflects the degree of change. The second derivative d²S / dt² is obtained by differentiating the slope sequence again. A positive value indicates accelerated rise, and a negative value indicates deceleration. It is used to identify the acceleration of change and help locate trend turning points. By changing the sign and magnitude of the slope, the recovery process is divided into different stages: the slope is close to zero in the plateau phase, with typical values ​​between -0.01 and 0.01; the slope increases slowly in the gradual change phase, gradually rising from 0.01 to 0.05; the slope rises sharply in the abrupt change phase, increasing from 0.05 to over 0.25 within 20 seconds, an increase of up to 5 times; the slope gradually falls back from its peak in the stable phase, eventually approaching zero.

[0055] Abrupt changes in slope data often correspond to significant physiological state transitions. The starting point of accelerated ascent is identified using a change-point detection algorithm. This algorithm tracks the cumulative degree of slope deviation from the baseline, specifically employing a cumulative sum method: C(t) = C(t-1) + [k(t) - k_baseline], where k_baseline is the median of the slope data for the first 60 seconds. When the cumulative deviation C(t) exceeds a threshold h determined based on historical data, it is marked as a potential acceleration point. The threshold h is typically set as the product of three times the baseline standard deviation and the time window length. To ensure reliability, the upward slope trend must last for at least 10 seconds, meaning that at least 8 out of 10 consecutive sampling points after the detection point must have slopes higher than the baseline, excluding interference from instantaneous fluctuations. Precise location of the starting point uses a backtracking search method, tracing back from the detected acceleration point to find the moment t_accel when the slope first significantly exceeds the baseline level. Significance is defined as k(t) > k_baseline + 2σ_baseline. This moment has clear physiological significance: the brain's fatigue regulation mechanism shifts from a passive recovery state to an active regulation mode, the endogenous recovery mechanism is activated, manifested as changes in neurotransmitter release patterns and increased cortical excitability.

[0056] The sensitivity value at the starting point has special physiological significance, representing the critical state in which the brain shifts from passive waiting to actively seeking intervention. The critical sensitivity S_critical is not the peak of the sensitivity curve, but rather the threshold level that triggers a rapid rise, typically 40%-60% of the peak value. This characteristic reflects the nonlinear response characteristics of the physiological system. Threshold analysis focuses not only on the numerical value itself, but more importantly, on assessing its stability and repeatability under different conditions. Statistical analysis of multiple intervention data from the same individual revealed that critical sensitivity exhibits high individual specificity and temporal stability. The relative threshold level R_critical = S_critical / S_max remains relatively constant under different fatigue levels; this scale invariance facilitates practical applications. The speed of crossing the critical threshold V_cross reflects the urgency of the need; rapid crossing usually implies a narrower optimal intervention window. The over-threshold maintenance time T_above assesses the persistence of the high-sensitivity state; a longer maintenance time provides more ample intervention opportunities. The critical sensitivity characteristics include four core parameters: threshold level S_critical, relative proportion R_critical, crossing speed V_cross, and over-threshold maintenance time T_above.

[0057] The optimal interval duration is determined based on four dimensions of critical sensitivity characteristics. Time calculation begins at the acceleration starting point t_accel, and the precise interval duration is obtained through comprehensive adjustment of multiple parameters. The preparation time T_prep is dynamically set according to the crossing speed V_cross; faster crossings require more preparation time, typically 10-20 seconds. The safety margin T_margin is calculated based on the above-threshold maintenance time T_above; a shorter maintenance time indicates a narrower intervention window, requiring a larger safety margin, usually 10%-20% of t_accel. The critical level adjustment T_adjust considers the influence of the relative proportion R_critical; a lower relative proportion indicates the brain enters the demand state earlier, requiring earlier intervention. The optimal interval duration T_optimal = t_accel - T_prep - T_margin + T_adjust, and is fine-tuned in conjunction with the intervention effect score S_effect; for high effect scores, the interval can be appropriately extended to fully utilize the intervention effect. Boundary constraints ensure 120 seconds ≤ T_optimal ≤ 600 seconds; the lower limit prevents intervention fatigue, and the upper limit avoids missing the opportunity. The verification step checks whether the sensitivity at the selected time point is close to 0.9 × S_critical, ensuring the effectiveness of the intervention timing. By comprehensively utilizing the critical sensitivity characteristics, the optimal interval duration achieves precise intervention rhythm control based on physiological characteristics.

[0058] Step S170: The release rhythm of the composite intervention signal is controlled by the optimal interval duration to form an interval playback sequence. The changes in the intensity of the EEG response under the interval playback sequence are monitored to generate sensitization effect data. Based on the current fatigue level, sensitization effect data and intervention effect score, a graded early warning instruction is generated.

[0059] Specifically, the release rhythm of the composite intervention signal is controlled by the optimal interval duration T_optimal to form an interval playback sequence. A precise timing mechanism is employed, triggering the intervention signal playback once every T_optimal duration. The playback sequence design follows a "play-interval-play" cyclical pattern, with each cycle containing 8 seconds of signal playback and T_optimal-8 seconds of silent waiting. Sequence synchronization is based on the first playback moment, establishing a global time coordinate. In addition to the arrival of the time, playback triggering conditions must also meet constraints such as stable environmental noise and no emergency communication interruptions. The dynamic fine-tuning of the interval duration is based on the real-time fatigue state; it can be shortened by 5-10% when BEI rises rapidly and extended by 5-10% when BEI is stable. Sequence integrity is ensured through a buffering mechanism; even if a playback is interrupted, the next playback will proceed according to the original rhythm. A playback counter records the cumulative number of playbacks and the actual interval time for effect evaluation. Sequence modes include standard mode (fixed interval), adaptive mode (dynamically adjusted according to BEI), and emergency mode (interval shortened to the minimum value).

[0060] In some embodiments, the step of monitoring changes in EEG response intensity under the interval playback sequence to generate sensitization effect data includes: monitoring the execution process of the interval playback sequence to obtain a response intensity sequence; performing incremental analysis on the response intensity sequence to identify sensitization inflection points; constructing sensitization quantification indicators based on the sensitization inflection points, the sensitization quantification indicators including sensitization speed, sensitization amplitude, and sensitization stability; and generating sensitization effect data through the sensitization quantification indicators.

[0061] The response intensity sequence is obtained by monitoring the execution process of the interval playback sequence. Monitoring is performed in each playback cycle, from 30 seconds before playback begins to 60 seconds after playback ends. Intensity calculation uses a normalization method: Rn = (BEImax_n - BEImin_n) / BEImax_n × 100%, where BEImax_n is the peak value before the nth playback, and BEImin_n is the trough value after playback. The sequence is constructed by recording the response intensity values ​​of each time in chronological order, forming {R1, R2, ..., Rn}. Data quality control removes responses severely affected by interference, requiring a signal-to-noise ratio greater than 3dB. Response delay compensation considers individual differences, searching for the lowest point within the range of 3-8 seconds. The sequence length contains at least 8 valid responses to ensure the reliability of trend analysis. Time stamping records the precise time of each response for correlation analysis. Response feature extraction includes not only intensity values ​​but also response rate and recovery time.

[0062] An incremental analysis is performed on the input response intensity sequence {R1, R2, ..., Rn} to identify sensitization inflection points. Sequence preprocessing ensures data quality, requiring at least 8 valid response values, and outliers are handled using the median replacement method. A piecewise linear regression method is used, dividing the sequence into different growth stages using a sliding window method. The window size is set to 3-5 data points, with a step size of 1 data point. For each segment, the slope of the linear regression growth rate is calculated, slope k = (n∑xy - ∑x∑y) / (n∑x² - (∑x)²), where x is the index and y is the response intensity. The inflection point detection algorithm finds locations where the slope changes significantly, traversing all possible segmentation points k and calculating the slope ratio before and after the segmentation, r_k = k_after / k_before. When the slope ratio r_k between adjacent segments is greater than 1.5 and the subsequent segment has at least 3 data points, it is marked as a candidate inflection point. The increasing pattern is classified into linear increasing (constant slope), accelerating increasing (rising slope), and saturating increasing (falling slope), with the pattern type determined by second-order differencing. Statistical significance is tested using the F-test to ensure that the identified inflection points are not random fluctuations; the F-statistic is calculated and compared with the critical value. In the case of multiple inflection points, the inflection intensity score S_turn = (r_k-1) × √n_after is calculated for each candidate point, where n_after is the number of data points after the inflection. The main inflection point with the highest score is identified first. Key output parameters include: inflection point position n_turn, slope before the inflection k_before, slope after the inflection k_after, and inflection intensity score S_turn.

[0063] An augmentation quantification index is constructed based on inflection point parameters and complete response sequence data. The quantification index system includes three core dimensions: augmentation speed, augmentation magnitude, and augmentation stability. Each dimension reflects the augmentation characteristics from different perspectives. Augmentation speed is calculated based on the response change after the inflection point: Vsens = (R_(n_turn+3) - R_n_turn) / 3, where R_n_turn is the response intensity at the inflection point, and R_(n_turn+3) is the third response value after the inflection, expressed as a percentage of the response, reflecting the rapid growth capability after the inflection. When there are fewer than three data points after the inflection, actual available data is used for calculation, and the data is marked as insufficient. Augmentation magnitude is calculated by the ratio of the first and last data points in the entire sequence: Asens = R_max / R_1, where R_max is the maximum response intensity of the sequence (usually occurring at the end of the sequence), and R_1 is the initial response intensity. This dimensionless ratio quantifies the overall increase factor from the initial value to the peak value. Sensitization stability is assessed using the coefficient of variation (CVsens) of all data after the transition, CVsens = σ_post / μ_post, where σ_post is the standard deviation of the data after the transition, and μ_post is the mean of the data after the transition. A smaller value indicates a more stable sensitization process. The comprehensive sensitization index is integrated through a weighted product of three dimensions, Isens = Vsens × Asens / (1 + CVsens). The 1 in the denominator is used to avoid division by zero and to adjust for the weighting effect on stability. Index normalization uses a linear transformation I_norm = 100 × (Isens - I_min) / (I_max - I_min), where I_min and I_max are determined based on historical data, mapping the original values ​​to a standard range of 0-100 for easier comparison and trend analysis between different individuals.

[0064] Structured sensitization effect data are generated using quantified sensitization indicators. Sensitization speed is assessed using a three-level grading system based on Vsens values: rapid sensitization (Vsens > 3.0% / repetition) indicates a swift brain response to repetitive stimuli and a significant learning effect; moderate sensitization (2.0-3.0% / repetition) indicates a moderate response and normal adaptation process; slow sensitization (< 2.0% / repetition) indicates a sluggish response and potential fatigue accumulation. Sensitization magnitude is assessed using Asens values: significant sensitization (Asens > 1.8 times) indicates a significant improvement in intervention effect and a strong cumulative effect; moderate sensitization (1.4-1.8 times) indicates a moderate improvement within the normal range; and small sensitization (< 1.4 times) indicates limited improvement and a near-saturation of intervention effect. Sensitization stability assessment was based on CVsens values ​​to determine process quality: high stability (CVsens < 0.15) indicated a smooth sensitization process with good individual response consistency; moderate stability (0.15-0.3) indicated slight but controllable fluctuations; low stability (> 0.3) indicated large fluctuations and unstable responses. Comprehensive sensitization status was determined based on normalized Isens for overall grading: strong sensitization (Isens ≥ 60) corresponded to a "positive sensitization" state, indicating continuous improvement in intervention effects; moderate sensitization (30 ≤ Isens < 60) corresponded to a "mild sensitization" state, indicating improvement but not significant; weak sensitization (Isens < 30) corresponded to a "negative sensitization" state, indicating poor intervention effects or adaptive behavior. Individual sensitization patterns were identified through combinations of four indicators: rapid response type (high Vsens + moderate Asens), cumulative enhancement type (moderate Vsens + high Asens), and stable maintenance type (moderate Vsens + moderate Asens + low CVsens). The structured data output forms a complete report on the enhancement effect, which includes the original values, standardized values, level determination, comprehensive status assessment and individual pattern recognition results for four dimensions.

[0065] Generate a hierarchical warning instruction based on the current fatigue level, sensitization effect data, and intervention effect score. The generation of the warning instruction comprehensively uses three key data sources: the current fatigue level reflected by the real-time BEI value, the multi-dimensional sensitization data output from the aforementioned sensitization effect analysis, and the intervention effect score S_effect of S160. The current fatigue risk assessment is based on the latest BEI value and is divided into three levels: low risk (BEI < 50), medium risk (50 ≤ BEI < 70), and high risk (BEI ≥ 70). The determination of intervention effectiveness is comprehensively determined according to the S_effect score and the sensitization effect data: highly effective (S_effect > 80 and positive sensitization), moderately effective (60 < S_effect ≤ 80 or mild sensitization), and lowly effective (S_effect ≤ 60 or negative sensitization). The warning matrix determines the warning level through the combination of the current fatigue risk and intervention effectiveness: green warning (low risk + highly effective), yellow warning (medium risk + moderately effective), orange warning (high risk + moderately effective or medium risk + lowly effective), and red warning (high risk + lowly effective). The instruction coding converts the warning level into specific control parameters and outputs a structured instruction containing the warning level, vibration mode, and intensity parameters. Control the safety helmet vibration device through the hierarchical warning instruction to output differentiated warning signals. The green warning uses a single short vibration to indicate a good state, the yellow warning uses a double vibration mode to remind of fatigue, the orange warning uses a triple vibration to warn of increased fatigue, and the red warning uses continuous vibration to require immediate rest, finally completing the detection and warning of the fatigue state.

[0066] In order to execute the method for detecting and warning the fatigue state of operating personnel corresponding to the above method embodiments to achieve the corresponding functions and technical effects. Refer to Figure 2 , Figure 2 The block diagram of a device 200 for detecting and warning the fatigue state of operating personnel provided by an embodiment of the present application is shown. For the sake of convenience of description, only the parts related to this embodiment are shown. The device 200 for detecting and warning the fatigue state of operating personnel provided by an embodiment of the present application includes: An electroencephalogram acquisition module 201, configured to acquire the electroencephalogram signal of an operating personnel, perform mirror symmetry analysis on the electroencephalogram signal to determine the activity difference between the left and right cerebral hemispheres, and generate an electroencephalogram imbalance index based on the activity difference; A spectrum analysis module 202, configured to perform spectrum analysis on the electroencephalogram signal to extract the dominant frequency distribution, collect the noise signal in the operating environment and identify its spectrum gap to obtain the noise silent frequency band, and map the dominant frequency distribution to the noise silent frequency band to generate an intervention window parameter; A signal generation module 203, configured to select an optimal carrier frequency point based on the intervention window parameter, generate a pure carrier signal based on the optimal carrier frequency point, and modulate and fuse the pure carrier signal through the electroencephalogram imbalance index to generate a composite intervention signal; The fatigue recognition module 204 is used to monitor the coupling relationship between the operator's head micro-movements and the electroencephalogram (EEG) signals to identify fatigue precursor features and determine the release timing of the composite intervention signal based on the fatigue precursor features. The intervention playback module 205 is used to play the composite intervention signal in the noise-silent frequency band according to the release timing, and monitor the change of the EEG imbalance index during the playback process to generate a fatigue relief curve. The effect evaluation module 206 is used to perform slope analysis on the fatigue relief curve to generate an intervention effect score, monitor the real-time changes of the EEG imbalance index after the composite intervention signal stops playing to establish a recovery sensitivity curve, and identify the moment when the brain's demand for intervention is strongest in the recovery sensitivity curve to determine the optimal interval duration. The early warning control module 207 is used to control the release rhythm of the composite intervention signal to form an interval playback sequence using the optimal interval duration, monitor the changes in EEG response intensity under the interval playback sequence to generate sensitization effect data, and generate graded early warning instructions based on the current fatigue level, the sensitization effect data and the intervention effect score.

[0067] The aforementioned worker fatigue detection and early warning device 200 can implement the worker fatigue detection and early warning method described in the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0068] like Figure 3 As shown, the third embodiment of the present invention also provides a computer device, including a memory 301, a processor 302, and a computer program stored in the memory 301 and executable on the processor 302, characterized in that the processor 302 executes the program to implement the steps of the method for detecting and warning of worker fatigue described in the first embodiment of the present invention.

[0069] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

[0070] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. A method for detecting and warning of worker fatigue, characterized in that, include: The brainwave signals of the workers are acquired, and the mirror symmetry analysis of the brainwave signals is performed to determine the activity difference between the left and right hemispheres. Based on the activity difference, a brainwave imbalance index is generated. The dominant frequency distribution is extracted by performing spectral analysis on the EEG signal, noise signals in the work environment are collected and their spectral gaps are identified to obtain noise silence frequency bands, and the dominant frequency distribution is mapped to the noise silence frequency bands to generate intervention window parameters. The optimal carrier frequency is selected based on the intervention window parameters, a pure carrier signal is generated based on the optimal carrier frequency, and the pure carrier signal is modulated and fused by the EEG imbalance index to generate a composite intervention signal. The coupling relationship between the operator's head micro-movements and the electroencephalogram (EEG) signals is monitored to identify fatigue precursor features, and the timing of releasing the composite intervention signal is determined based on the fatigue precursor features. According to the release timing, the composite intervention signal is played in the noise-silent frequency band, and the change of the brain electrical imbalance index during the playback process is monitored to generate a fatigue relief curve. The slope of the fatigue relief curve is analyzed to generate an intervention effect score. The real-time change of the EEG imbalance index is monitored after the composite intervention signal stops playing to establish a recovery sensitivity curve. The moment when the brain's demand for intervention is strongest in the recovery sensitivity curve is identified as the optimal interval. The optimal interval duration is used to control the release rhythm of the composite intervention signal to form an interval playback sequence. The changes in the intensity of the EEG response under the interval playback sequence are monitored to generate sensitization effect data. Based on the current fatigue level, the sensitization effect data and the intervention effect score, a graded early warning instruction is generated.

2. The method according to claim 1, characterized in that, The step of mapping the dominant frequency distribution to the noise-silent frequency band to generate intervention window parameters includes: Extract peak frequency components from the dominant frequency distribution; Find an embedding location within the noise-quiet frequency band that matches the peak frequency component; The frequency offset is determined based on the embedding location and the dominant frequency distribution; Intervention window parameters are generated based on the frequency offset.

3. The method according to claim 1, characterized in that, The process of modulating and fusing the pure carrier signal using the electroencephalogram imbalance index to generate a composite intervention signal includes: The carrier signal is amplitude modulated to generate an anti-fatigue audio sequence; The fatigue-sensitive characteristic signals are identified based on the aforementioned brainwave imbalance index; The fatigue-sensitive feature signal is subjected to time-reversal processing to form a fatigue-resistant waveform; The fatigue-resistant waveform is embedded into the fatigue-resistant audio sequence to complete the generation of the composite intervention signal.

4. The method according to claim 3, characterized in that, The process of performing time-reversal processing on the fatigue-sensitive feature signal to form a fatigue-resistant waveform includes: Extract the time-domain envelope of the fatigue-sensitive feature signal; The time-domain envelope is flipped along the time axis to determine the reverse timing sequence; The phase of the reverse timing is adjusted according to the fatigue-sensitive characteristic signal to generate a fatigue-resistant waveform.

5. The method according to claim 1, characterized in that, The optimal interval duration is determined by identifying the moment in the recovery sensitivity curve when the brain's need for intervention is strongest, including: Slope analysis is performed on the recovery sensitivity curve to generate slope change data; Identify the starting point of accelerated ascent from the slope change data; Sensitivity threshold analysis is performed on the starting point to obtain critical sensitivity characteristics; Based on the critical sensitivity feature, the time value corresponding to the starting point is extracted and determined as the optimal interval duration.

6. The method according to claim 1, characterized in that, The coupling relationship between the monitoring worker's head micro-movements and the electroencephalogram (EEG) signals is used to identify early signs of fatigue, including: The head micro-movements are collected to form a micro-movement time series; Based on the micro-movement time series, waveform segments corresponding to the time moments are extracted from the EEG signals; The coupling strength is determined by analyzing the correlation between the micro-motion time series and the waveform segment; Fatigue precursor features are identified based on the coupling strength.

7. The method according to claim 6, characterized in that, The identification of fatigue precursor features based on the coupling strength includes: Periodic analysis of the coupling strength generates a coupling rhythm; Identify abnormal breakpoints in the coupled rhythm; Feature extraction is performed on the abnormal fracture points to generate fatigue early warning signals; The fatigue warning signals are used to determine the characteristics of early signs of fatigue.

8. The method according to claim 1, characterized in that, The monitoring of changes in EEG response intensity under the interval playback sequence to generate sensitization effect data includes: The response intensity sequence is obtained by monitoring the execution process of the interval playback sequence; An incremental analysis was performed on the response intensity sequence to identify the sensitization inflection point; Based on the aforementioned sensitivity enhancement inflection point, a sensitivity enhancement quantitative index is constructed, which includes sensitivity enhancement speed, sensitivity enhancement magnitude, and sensitivity enhancement stability. The sensitization effect data is generated using the aforementioned sensitization quantification indicators.

9. A device for detecting and warning of worker fatigue, characterized in that, include: The EEG acquisition module is used to acquire the EEG signals of the operator, perform mirror symmetry analysis on the EEG signals to determine the activity difference between the left and right hemispheres, and generate an EEG imbalance index based on the activity difference. The spectrum analysis module is used to perform spectrum analysis on the EEG signal to extract the dominant frequency distribution, collect noise signals in the work environment and identify their spectral gaps to obtain noise silence frequency bands, and map the dominant frequency distribution to the noise silence frequency bands to generate intervention window parameters. The signal generation module is used to select the optimal carrier frequency point based on the intervention window parameters, generate a pure carrier signal based on the optimal carrier frequency point, and modulate and fuse the pure carrier signal through the EEG imbalance index to generate a composite intervention signal. The fatigue recognition module is used to monitor the coupling relationship between the operator's head micro-movements and the electroencephalogram (EEG) signals to identify fatigue precursor features and determine the timing of releasing the composite intervention signal based on the fatigue precursor features. An intervention playback module is used to play the composite intervention signal in the noise-silent frequency band according to the release timing, and monitor the changes in the EEG imbalance index during the playback process to generate a fatigue relief curve. The effect evaluation module is used to perform slope analysis on the fatigue relief curve to generate an intervention effect score, monitor the real-time changes of the EEG imbalance index after the composite intervention signal stops playing to establish a recovery sensitivity curve, and identify the moment in the recovery sensitivity curve when the brain's demand for intervention is strongest to determine the optimal interval duration. The early warning control module is used to control the release rhythm of the composite intervention signal to form an interval playback sequence using the optimal interval duration, monitor the changes in EEG response intensity under the interval playback sequence to generate sensitization effect data, and generate graded early warning instructions based on the current fatigue level, the sensitization effect data and the intervention effect score.

10. A computer device, characterized in that, It includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and, when executing the computer program, implement the method as described in any one of claims 1 to 8.

Images