An intelligent head ring and method for fatigue monitoring based on electroencephalogram differential entropy and spectral features

Abstract

The application discloses a kind of fatigue monitoring intelligent head ring and method based on electroencephalogram differential entropy and spectral feature, fatigue monitoring intelligent head ring includes electroencephalogram signal acquisition module, pre-processing unit, feature extraction unit, fatigue judging unit and neural feedback unit.Electroencephalogram signal acquisition module is used to collect user forehead single channel electroencephalogram signal and carry out preliminary filtering amplification;Pre-processing unit carries out pre-processing to signal and obtains pure electroencephalogram signal;Feature extraction unit extracts frequency domain feature and time domain feature from the electroencephalogram signal after pre-processing, fatigue judging unit is based on individualized baseline model and dynamic threshold, calculates fatigue score and judges whether in fatigue state;Neural feedback unit triggers audiovisual alarm when judging fatigue, guide user self-regulation brain state.The application can monitor user fatigue state in real time, accurately, individualizedly, realize fatigue intervention by closed-loop neural feedback, and be applicable to driving, operation monitoring and other high safety requirement scene.

Description

Technical Field

[0001] This invention relates to the field of physiological signal monitoring and human-computer interaction technology, specifically to a smart headband and method for fatigue monitoring based on EEG differential entropy and spectral characteristics. Background Technology

[0002] Fatigue has gradually become a significant factor impacting people's health and work efficiency, especially in high-safety-requirement work scenarios such as driving, aviation, and nuclear power, where fatigue can lead to serious accidents. Traditional fatigue monitoring methods largely rely on visual analysis, such as cameras capturing eye movements and yawning, but these methods are easily affected by factors such as lighting, occlusion, and individual differences, resulting in insufficient robustness. Monitoring methods based on physiological signals, such as electrocardiograms and electromyograms, can indirectly reflect fatigue states, but their sensitivity and specificity are limited.

[0003] Electroencephalogram (EEG) signals directly reflect the neural activity of the cerebral cortex and are considered the "gold standard" for fatigue monitoring. In a state of fatigue, the power of the theta band (4-8 Hz) of the EEG signal increases, while the power of the alpha band (8-13 Hz) decreases, resulting in a significantly increased theta / alpha power ratio. Simultaneously, the signal complexity decreases, and the differential entropy value drops. These characteristics can be used to effectively identify fatigue states, especially microsleep states.

[0004] However, existing EEG fatigue monitoring systems mostly use multi-channel devices, which are bulky, costly, and inconvenient to wear, making them difficult to use in daily or work scenarios and thus limiting their practical application value. Summary of the Invention

[0005] In view of the above-mentioned deficiencies of the prior art, the technical problem to be solved by the present invention is to provide a portable, real-time, accurate fatigue monitoring smart headband with closed-loop intervention capability and a monitoring method thereof.

[0006] To achieve the above objectives, the first aspect of the present invention provides a fatigue monitoring smart headband based on EEG differential entropy and spectral characteristics, comprising an EEG signal acquisition module, a preprocessing unit, a feature extraction unit, a fatigue judgment unit, and a neural feedback unit.

[0007] The EEG signal acquisition module is used to acquire the user's frontal single-channel EEG signal and perform preliminary filtering and amplification, then transmit the acquired EEG signal to the preprocessing unit. As a preferred embodiment, the EEG signal acquisition module is an integrated smart headband, including three electrodes: the middle electrode is the frontal FP2 channel signal acquisition electrode, and the two side electrodes are reference electrodes; the headband integrates a preamplifier, bandpass filter, analog-to-digital converter, and wireless transmission module.

[0008] The preprocessing unit sequentially performs baseline drift removal, bandpass filtering, wavelet denoising, and spike detection on the received EEG signals to ensure the accuracy and reliability of the EEG signals. As a preferred embodiment, the wavelet denoising uses a db4 wavelet basis for single-level decomposition, and the signal is reconstructed after soft thresholding of the detail coefficients.

[0009] The feature extraction unit extracts frequency domain and time domain features from the preprocessed EEG signal. The frequency domain features include the power spectral density in the θ band (4-8Hz), the power spectral density in the α band (8-13Hz), and the power ratio θ / α. The time domain feature is the differential entropy DE. As a preferred embodiment, the feature extraction unit analyzes each 3-second segment of the preprocessed signal, calculates the power spectral density using a fast Fourier transform, and assumes the signal follows a Gaussian distribution, using the formula... Estimate the differential entropy.

[0010] The fatigue assessment unit calculates a fatigue evaluation score based on the features extracted by the feature extraction unit, combined with an individualized baseline model and a dynamic threshold, and determines whether the user is in a state of fatigue. This fatigue assessment unit includes a baseline modeling module, a dynamic threshold setting module, and a fatigue scoring module. Its operation is divided into two stages: individualized baseline modeling and real-time fatigue assessment. Each module performs its corresponding workflow:

[0011] The baseline modeling module performs the individualized baseline modeling phase. Upon first system use, it collects EEG signals from the user while they are awake, establishes an individualized baseline model, and calculates the mean θ / α power ratio. Standard deviation and the mean of differential entropy Standard deviation .

[0012] The dynamic threshold setting module and the fatigue scoring module work together in the real-time fatigue assessment phase. First, the dynamic threshold setting module sets the dynamic threshold based on the baseline model.

[0013]

[0014] Where k and m are adjustable sensitivity coefficients, with the preferred values ​​being k=1.0 and m=0.1.

[0015] The fatigue scoring module then processes the feature values ​​extracted in real time. and To perform the calculation, first calculate its deviation from the threshold:

[0016]

[0017] Then calculate the Euclidean distance between the current state and the fatigue and health states:

[0018]

[0019] Finally, the fatigue assessment score is calculated using the formula:

[0020]

[0021] The fatigue scoring module limits the score to the range of [0, 100], and a score below 60 is considered a state of fatigue.

[0022] When the fatigue judgment unit determines that the user is in a fatigued state, the neurofeedback unit triggers an audiovisual alarm to provide neurofeedback intervention to the user. As a preferred embodiment, the neurofeedback unit includes small speakers integrated on both sides of the headband and a visual alarm interface linked to the host computer software. When fatigue is detected, the interface background flashes red and white, and simultaneously, the headband speakers play an alarm sound, forming dual audiovisual feedback.

[0023] A second aspect of this invention provides a fatigue monitoring method based on EEG differential entropy and spectral characteristics, applied to the aforementioned smart headband, comprising the following steps:

[0024] Step S1: Acquire the user's prefrontal cortex EEG signals using the EEG signal acquisition module;

[0025] Step S2: The preprocessing unit performs baseline drift removal, bandpass filtering, wavelet denoising, and spike detection on the acquired EEG signals to obtain clean EEG signals;

[0026] Step S3: Extract frequency domain features and time domain features from the preprocessed EEG signal through the feature extraction unit. The frequency domain features include the power spectral density of the θ band, the power spectral density of the α band, and the power ratio θ / α between the two. The time domain features are differential entropy DE.

[0027] Step S4: Establish an individualized baseline model through the fatigue judgment unit, set a dynamic threshold, calculate the fatigue assessment score based on the extracted features, and determine whether the user is in a state of fatigue.

[0028] Step S5: When fatigue is detected, audiovisual feedback signals are triggered through the neurofeedback unit to guide the user to adjust their brain state.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] 1. The device adopts a single-channel frontal EEG acquisition solution, which is lightweight, comfortable to wear, and low in cost, making it suitable for long-term use in daily and work scenarios.

[0031] 2. A fatigue judgment algorithm based on individualized baseline modeling and dynamic threshold is proposed, which effectively overcomes individual differences and improves monitoring accuracy and robustness.

[0032] 3. By introducing differential entropy as a time-domain feature and combining it with the frequency-domain feature θ / α power ratio, fatigue state is comprehensively evaluated from two dimensions: signal complexity and spectral distribution, resulting in a more comprehensive identification.

[0033] 4. Design a closed-loop neurofeedback intervention mechanism. Through the speakers and audio-visual alarms integrated into the headband and the software interface, the system guides the user to adjust their brain state in real time, realizing a complete closed loop from monitoring to intervention, which effectively improves the system's practicality and safety. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the system structure of a specific embodiment of the present invention; Figure 2 This is a schematic diagram of the hardware integrated headband of the present invention, wherein 1 is the FP2 channel electrode; 2 is the right reference electrode; 3 is the left reference electrode; 4 is the right integrated speaker; 5 is the left integrated speaker; and 6 is a switch. Detailed Implementation

[0035] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0036] Example 1

[0037] like Figure 1 , Figure 2 As shown, this embodiment provides a fatigue monitoring smart headband based on EEG differential entropy and spectral characteristics, including an EEG signal acquisition module 1, a preprocessing unit 2, a feature extraction unit 3, a fatigue judgment unit 4, and a neural feedback unit 5.

[0038] I. Hardware Configuration and Signal Acquisition

[0039] like Figure 2 As shown, the smart headband in this embodiment adopts an ergonomic three-section adjustable headband design to ensure wearing comfort and signal acquisition stability. The core sensing components of the headband are three dry electrodes. The electrode located in the center of the headband serves as the signal acquisition electrode, precisely corresponding to the FP2 channel position in the international 10-20 system standard lead, and is used to acquire EEG signals from the prefrontal cortex. The two electrodes located on either side of the acquisition electrode are reference electrodes, used to acquire common-mode interference signals (such as environmental noise and EMG artifacts).

[0040] The headband integrates signal conditioning and transmission circuitry, specifically including:

[0041] (1) Preamplifier: The gain is set to 2000 times to amplify weak EEG signals (usually in the microvolt range) to a processable range.

[0042] (2) Hardware bandpass filter: The passband frequency range is 0.2Hz to 70Hz, which is used to initially filter out DC bias and some high-frequency noise.

[0043] (3) Analog-to-digital converter: A 24-bit high-precision ADC is used, and the sampling rate can be configured to 250Hz or 500Hz to ensure signal quantization accuracy.

[0044] (4) Wireless transmission module: The low-power Bluetooth 5.0 technology is used to transmit the digitized EEG signals to the host computer (such as a dedicated data processing terminal, personal computer or smartphone) in real time for further processing.

[0045] In addition, a small speaker is integrated on each side of the headband, serving as the auditory output component of the neurofeedback unit 5. The headband has a built-in rechargeable lithium battery with a battery life of no less than 8 hours.

[0046] II. Signal Preprocessing

[0047] After receiving the raw EEG signal, the host computer performs the following digital signal processing steps in sequence to obtain a clean EEG signal:

[0048] (1) Baseline drift removal: Use a detrending algorithm, such as the detrend function in MATLAB, to remove the extremely low frequency (<0.2Hz) baseline drift components in the signal caused by breathing, sweating or electrode polarization.

[0049] (2) Bandpass filtering: Design a digital bandpass filter with a passband frequency of 0.5Hz to 30Hz. This step aims to preserve the rhythmic information related to fatigue, such as δ (0.5-4Hz), θ (4-8Hz), α (8-13Hz), and β (13-30Hz), while effectively suppressing power frequency interference (50 / 60Hz) and high-frequency electromyographic noise.

[0050] (3) Wavelet denoising: The signal is decomposed into approximation coefficients and detail coefficients using the Daubechies4 (db4) wavelet basis. The detail coefficients containing the main noise are processed by a soft thresholding function to suppress random noise and transient artifacts. Finally, the denoised signal is obtained by reconstruction.

[0051] (4) Peak detection and smoothing: Identify transient peaks in the signal whose amplitude exceeds a preset threshold (e.g., ±100μV) and whose duration is short, and identify them as motion artifacts. Smooth the data segment by using cubic spline interpolation or the mean of adjacent data to prevent it from interfering with subsequent feature extraction.

[0052] After the above steps, a preprocessed pure EEG signal fragment with a length of 3 seconds is obtained, which is used as the input of feature extraction unit 3.

[0053] III. Feature Extraction

[0054] Feature extraction unit 3 analyzes each 3-second preprocessed signal segment, simultaneously extracting frequency and time domain features:

[0055] (1) Frequency domain feature extraction:

[0056] Apply the Hamming window function to the signal segment to reduce spectral leakage.

[0057] Perform a Fast Fourier Transform to convert the signal from the time domain to the frequency domain.

[0058] Calculate the power spectral density. Extract the power spectral density values ​​in the θ band (4-8Hz) and α band (8-13Hz) respectively, and denot them as... and .

[0059] Calculate the power ratio of the two as the core frequency domain feature: .

[0060] (2) Temporal feature extraction:

[0061] Calculate the differential entropy of a signal segment. Assuming the signal follows a Gaussian distribution, the differential entropy DE can be estimated from the signal variance, using the following formula:

[0062] )

[0063] in, This is the variance estimate for this signal segment. It is a natural constant. Differential entropy reflects the complexity of a signal; under fatigue, the complexity of EEG signals decreases, and the DE value decreases.

[0064] IV. Fatigue Assessment

[0065] The workflow of fatigue assessment unit 4 is divided into two stages: "individualized baseline modeling" and "real-time fatigue assessment." This fatigue assessment unit 4 includes a baseline modeling module, a dynamic threshold setting module, and a fatigue scoring module. Each module performs all operations corresponding to its respective stage.

[0066] (1) Individualized baseline modeling:

[0067] The baseline modeling module guides the user to sit quietly with eyes closed for 10 minutes in a conscious and relaxed state when using the system for the first time or when resetting the model. During this time, the system collects and processes the electroencephalogram (EEG) signals. Based on the preprocessed and feature-extracted data, the mean θ / α power ratio is calculated for the user in a conscious state. and standard deviation and the mean of the differential entropy and standard deviation These statistical parameters constitute the user's individualized baseline model.

[0068] (2) Real-time fatigue assessment:

[0069] Dynamic threshold setting: The dynamic threshold setting module sets dynamic thresholds for frequency domain and time domain features based on the individual baseline model. The threshold calculation formula is as follows: ; .

[0070] Where k and m are adjustable sensitivity coefficients used to balance the sensitivity and specificity of the detection. A total of 1788 EEG signals were collected in the experiment, with a cycle of 3 seconds and 750 sampling points, under both awake and fatigued states. 728 EEG data points were collected under awake conditions, and 1060 EEG data points were collected under fatigued conditions. The classification results of this fatigue monitoring algorithm are shown in the table below:

[0071] Serial Number k value m value accuracy Accuracy Recall rate Specificity 1 1 0.1 68.74% 72.63% 75.85% 58.38% 2 1 0.3 68.85% 72.84% 75.66% 58.93% 3 1 0.5 68.29% 72.72% 74.43% 59.34% 4 2 0.1 69.18% 76.87% 68.68% 69.92% 5 2 0.3 69.30% 77.15% 68.49% 70.47% 6 2 0.5 68.74% 77.08% 67.26% 70.88% 7 3 0.1 69.91% 83.04% 61.89% 81.59% 8 3 0.3 70.02% 83.42% 61.70% 82.14% 9 3 0.5 69.46% 83.46% 60.47% 82.55%

[0072] When k=1.0 and m=0.1, the system's recall rate for fatigue states reaches 75.85%, improving its ability to detect fatigue states.

[0073] Feature deviation and fatigue score calculation: The feature value Ratio extracted in real time by the fatigue scoring module. val and DE val Calculate its relative deviation from the corresponding threshold:

[0074]

[0075] deviation and The sign and magnitude of the value indicate the degree to which the current state deviates from a healthy state and approaches a state of fatigue, respectively.

[0076] The fatigue scoring module then calculates the Euclidean distance between the current feature vector and the "fatigue state" (all deviations are positive) and "healthy state" (all deviations are negative) in the feature space:

[0077]

[0078] Based on the above distance, the fatigue scoring module constructs an intuitive fatigue assessment score (0-100 points), with higher scores indicating greater alertness:

[0079] This formula ensures that when the characteristics perfectly match the fatigue state... large and A smaller score approaches 0; conversely, a larger score approaches 100. The initial baseline level of alertness corresponds to a score of 80.

[0080] Fatigue state determination: The fatigue scoring module ultimately limits the calculated score to the range of [0, 100]. When the real-time evaluation score is lower than the preset threshold (e.g., 60 points), the fatigue judgment unit 4 determines that the user has entered a fatigue state and immediately sends a trigger command to the neural feedback unit 5.

[0081] V. Closed-loop neurofeedback intervention

[0082] Upon receiving the fatigue assessment signal, neurofeedback unit 5 immediately activates the closed-loop intervention mechanism:

[0083] (1) Visual alarm: The background color of the host computer software interface is switched to a mode of alternating flashing red and white to attract the user's attention with strong visual stimulation.

[0084] (2) Auditory alarm: Send control commands to the smart headband via Bluetooth to trigger the integrated speakers on both sides of the headband to play a periodic "beep-beep-beep" alarm sound.

[0085] (3) Feedback-regulation closed loop: The combined visual and auditory alarms act on the user to activate their ascending reticular activating system, prompting them to autonomously adjust their attention, take deep breaths, or change their current behavior, thereby inhibiting theta wave activity in the EEG and enhancing alpha or beta wave activity, helping the user recover from fatigue. When the system detects that the fatigue score has risen above the threshold, the neural feedback automatically stops, completing a complete "monitoring-feedback-regulation" closed loop.

[0086] Through the close integration of the aforementioned hardware and algorithms, the smart headband provided in this embodiment can monitor the user's fatigue state in real time, accurately, and individually, and provide effective closed-loop intervention, significantly improving its application value and reliability in scenarios with high safety requirements such as driving and continuous operation.

[0087] Example 2

[0088] This embodiment optimizes the parameters of feature extraction unit 3 and fatigue judgment unit 4 based on embodiment 1. Feature extraction unit 3 analyzes the preprocessed signal of 2-second segments to shorten the response time. The dynamic threshold coefficients of fatigue judgment unit 4 are adjusted to k=1.2 and m=0.15 to improve the specificity of detection, making it suitable for applications with high requirements for false alarms. The composition and working principle of the remaining modules are the same as in embodiment 1. To improve specificity, k and m can be adjusted to 3 and 0.5, but the recall rate will be very low.

[0089] Example 3

[0090] This embodiment expands upon Embodiment 1 by adding a neurofeedback unit 5. In addition to the audiovisual alarm, a tactile feedback module is added, specifically a miniature vibration motor integrated inside the headband. When fatigue is detected, the tactile feedback module simultaneously activates vibration, creating triple feedback (audiovisual and tactile) to enhance the wake-up effect. The composition and working principle of the remaining modules are the same as in Embodiment 1.

[0091] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A smart headband for fatigue monitoring based on EEG differential entropy and spectral characteristics, characterized in that, include: The EEG signal acquisition module (1) is used to acquire the user's prefrontal cortex EEG signals; The preprocessing unit (2) is connected to the EEG signal acquisition module (1) and is used to preprocess the acquired EEG signals to remove noise and artifacts. The feature extraction unit (3) is connected to the preprocessing unit (2) and is used to extract frequency domain features and time domain features from the preprocessed EEG signal; The fatigue judgment unit (4) is connected to the feature extraction unit (3) and is used to calculate the fatigue assessment score and determine whether the user is in a fatigued state based on the extracted features, combined with the individualized baseline model and dynamic threshold; the neural feedback unit (5) is connected to the fatigue judgment unit (4) and is used to trigger a feedback signal when the user is determined to be in a fatigued state, so as to guide the user to adjust the brain state.

2. The fatigue monitoring smart headband based on EEG differential entropy and spectral characteristics according to claim 1, characterized in that, The EEG signal acquisition module (1) is an integrated headband, which includes three electrodes, of which the middle electrode is the signal acquisition electrode, corresponding to the FP2 channel position of the international 10-20 system, and the two side electrodes are reference electrodes.

3. The fatigue monitoring smart headband based on EEG differential entropy and spectral characteristics according to claim 1, characterized in that, The preprocessing unit (2) includes: The baseline drift removal module is used to remove DC components and low-frequency drift from the signal; The bandpass filter module is used to retain signals in the 0.5-30Hz frequency band and suppress high-frequency noise; The wavelet denoising module is used to remove high-frequency noise and transient artifacts; The spike detection module is used to identify and smoothly replace transient spikes with abnormal amplitudes.

4. The fatigue monitoring smart headband based on EEG differential entropy and spectral characteristics according to claim 1, characterized in that, The frequency domain features extracted by the feature extraction unit (3) include the power spectral density of the θ band (4-8Hz), the power spectral density of the α band (8-13Hz), and the power ratio θ / α between the two; the time domain feature is the differential entropy DE.

5. The fatigue monitoring smart headband based on EEG differential entropy and spectral characteristics according to claim 1, characterized in that, The fatigue judgment unit (4) includes: The baseline modeling module is used to collect EEG signals from the user while they are awake when the system is first used, to establish an individualized baseline model, and to obtain the mean θ / α power ratio. Standard deviation and the mean of differential entropy Standard deviation ; The dynamic threshold setting module is used to set dynamic thresholds based on the baseline model. Where k and m are adjustable sensitivity coefficients; The fatigue scoring module is used to calculate the deviation between the current signal characteristic value and the threshold in real-time monitoring, and then calculate the fatigue assessment score.

6. The fatigue monitoring smart headband based on EEG differential entropy and spectral characteristics according to claim 5, characterized in that, The fatigue scoring module calculates the fatigue assessment score in the following manner: Calculate the current signal eigenvalues and Deviation from the threshold: Calculate the Euclidean distance between the fatigue state and the health state: The formula for calculating fatigue assessment scores is: The score is limited to the range [0, 100], and a score below 60 is considered a state of fatigue.

7. The fatigue monitoring smart headband based on EEG differential entropy and spectral characteristics according to claim 1, characterized in that, The neurofeedback unit (5) includes an auditory feedback module and a visual feedback module; the auditory feedback module includes small speakers integrated on both sides of the headband; the visual feedback module includes an interface linked with the host computer software, which is used to trigger the red and white background of the interface to flash when the fatigue state is determined, and at the same time the speaker plays an alarm sound, forming a dual audio-visual feedback.

8. A fatigue monitoring method based on EEG differential entropy and spectral characteristics, applied to the smart headband described in any one of claims 1-7, characterized in that, Includes the following steps: S1: Collect the user's frontal brain electrical signals through the EEG signal acquisition module (1); S2: The collected EEG signals are processed by the preprocessing unit (2) to perform baseline drift removal, bandpass filtering, wavelet denoising and peak detection to obtain pure EEG signals; S3: Extract frequency domain features and time domain features from the preprocessed EEG signal through the feature extraction unit (3); S4: Establish an individualized baseline model through the fatigue judgment unit (4), set a dynamic threshold, calculate the fatigue assessment score based on the extracted features, and determine whether the user is in a state of fatigue. S5: When the user is identified as being in a state of fatigue, the audiovisual feedback signal is triggered through the neurofeedback unit (5) to guide the user to adjust their brain state.

9. The fatigue monitoring method according to claim 8, characterized in that, The frequency domain features extracted in step S3 include the power spectral density of the θ band (4-8Hz), the power spectral density of the α band (8-13Hz), and the power ratio θ / α between the two; the time domain feature is the differential entropy DE.

10. The fatigue monitoring method according to claim 8, characterized in that, The specific method for calculating the fatigue assessment score in step S4 includes: Establish an individualized baseline model to obtain the mean θ / α power ratio. Standard deviation and the mean of differential entropy Standard deviation ; Set dynamic threshold: Where k and m are adjustable sensitivity coefficients; Calculate the current signal eigenvalues and Deviation from the threshold: Calculate the Euclidean distance between the fatigue state and the health state: The formula for calculating fatigue assessment scores is: The score is limited to the range [0, 100], and a score below 60 is considered a state of fatigue.

Images