A method and system for detecting eye fatigue of an electroencephalogram glasses

Abstract

The application discloses an eye fatigue detection method and system of electroencephalogram glasses, comprising the following steps: collecting biological original signals directly related to eye fatigue; based on the physiological correlation of the biological original signals, jointly denoising the biological original signals to obtain effective signals reflecting the eye fatigue state; extracting feature parameters strongly related to eye fatigue from the effective signals, learning the contribution of each feature to fatigue through a feature weight distribution mechanism, and generating a weighted feature vector focusing on eye fatigue; using a pre-trained multi-modal fusion model, combining user individual physiological baseline calibration, analyzing the weighted feature vector, and outputting an eye fatigue level adapted to individual differences.

Description

Technical Field

[0001] This invention belongs to the field of intelligent device technology and relates to a method and system for detecting eye fatigue using brainwave glasses. Background Technology

[0002] With the widespread use of electronic devices, eye strain caused by prolonged screen time has become a significant issue affecting public vision health. Eye strain not only causes discomfort such as dry and swollen eyes, but long-term accumulation can also lead to increased myopia, dry eye syndrome, and other eye diseases. Therefore, real-time and accurate monitoring of eye strain is crucial for prevention and intervention. Eyeglasses, as everyday wearable tools that come into direct contact with the eyes, are widely explored for use in daily life. However, traditional smart glasses often only have basic optical correction functions or integrate simple sensors (such as accelerometers and ambient light sensors), without integrating miniature cameras within the frame to collect dynamic eye signals in real time. Existing methods for detecting eye strain rely on subjective questionnaires or scales, which are easily affected by user subjective perception biases, resulting in insufficient accuracy. Alternatively, they may simply judge eye fatigue and the need for rest based on the duration of glasses use. These methods cannot provide a more precise detection of eye strain. Summary of the Invention

[0003] The present invention provides a method and system for detecting eye fatigue using EEG glasses. By synchronously acquiring and jointly denoising multimodal biosignals, weighted extraction of fatigue-related features, and personalized multimodal fusion model analysis, the method aims to accurately detect eye fatigue states in a way that is tailored to individual differences.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: A method for detecting eye fatigue using EEG glasses includes the following steps: S1. Collect raw biological signals directly related to eye fatigue, including electroencephalogram (EEG) signals, eye movement and pupil signals, and heart rate variability signals; S2. Based on the physiological correlation of biological raw signals, joint denoising is performed on biological raw signals to obtain effective signals that can reflect the state of eye fatigue; S3. Extract feature parameters that are strongly correlated with eye fatigue from the effective signal, learn the contribution of each feature to fatigue through a feature weight allocation mechanism, and generate a weighted feature vector of focusing eye fatigue; S4. Using a pre-trained multimodal fusion model and combined with user individual physiological baseline calibration, the weighted feature vector is analyzed to output an eye fatigue level adapted to individual differences.

[0005] Preferably, the EEG signals are acquired using flexible dry electrodes on the temples of the mirror, covering the occipital and temporal lobe regions; The eye movement and pupil signals are acquired by capturing eye images using a miniature infrared camera in the lens frame; The heart rate variability signal was acquired by collecting the pulse wave behind the ear using a PPG sensor on the temple of the glasses. The electroencephalogram (EEG) signal, the eye movement and pupillary signals, and the heart rate variability signal are sampled synchronously using the same clock source to ensure timestamp alignment.

[0006] Preferably, the step of jointly denoising the original biological signals based on their physiological correlation to obtain an effective signal reflecting eye fatigue includes the following steps: S21. Detect physiological events such as rapid blinking, fixation point drift, and head movement through eye movement and pupil signals, and simultaneously mark the corresponding time windows in EEG signals and heart rate variability signals that are affected by these events; S22. Based on the interference characteristics of each modality signal, denoising is performed using marked interference time windows: For EEG signals, independent component analysis is used to separate and remove EMG artifacts, while signal segments affected by EMG artifacts are corrected or removed; for eye movement and pupil signals, filtering is used to eliminate ambient light interference, and image registration algorithms are used to correct image shifts caused by head movements; for heart rate variability signals, bandpass filtering is used to remove motion artifacts, while smoothing out abnormal heartbeat intervals that are interfered with. S23. Compare the correlation of multimodal signals before and after denoising. If the deviation exceeds the preset threshold, repeat steps S21 to S22 until the requirements are met, and finally obtain an effective signal that can accurately reflect the state of eye fatigue.

[0007] Preferably, the step of extracting feature parameters strongly correlated with eye fatigue from the effective signal, learning the contribution of each feature to fatigue through a feature weight allocation mechanism, and generating a weighted feature vector of focusing eye fatigue includes the following steps: S31. Extract feature parameters directly related to eye fatigue from the effective signals respectively. For EEG signals, extract frequency domain features reflecting the degree of visual cortex fatigue. For eye movement and pupil signals, extract temporal features reflecting the fatigue state of eye muscles and pupil accommodation function. For heart rate variability signals, extract frequency domain features reflecting the functional disorder of the autonomic nervous system caused by fatigue. S32. Learn the contribution of each feature to eye fatigue through a feature weight allocation mechanism, wherein the weight of each feature is dynamically adjusted according to the strength of the correlation between each feature and eye fatigue. S33. Multiply each feature by its corresponding weight and then concatenate them to generate a weighted feature vector of eye fatigue caused by focusing.

[0008] Preferably, the characteristic parameters strongly correlated with eye fatigue include: The ratio of alpha wave power spectral density to beta wave power spectral density, and the ratio of theta wave power spectral density to beta wave power spectral density in EEG signals; Blink frequency, pupil diameter change rate, and fixation point drift amplitude of eye movement and pupil signals; The ratio of low-frequency to high-frequency power and the total power of the heart rate variability signal.

[0009] Preferably, the step of using a pre-trained multimodal fusion model, combined with user-specific physiological baseline calibration, to analyze the weighted feature vector and output an eye fatigue level adapted to individual differences includes the following steps: S41. Input the generated weighted feature vector of eye fatigue caused by focusing on the screen into the input layer of the multimodal fusion model as the initial data for model analysis; S42. Individual physiological baseline calibration includes initial calibration and online learning. During initial calibration, multimodal signals under normal eye use are collected when the user uses the device for the first time, features are extracted and an initial model is trained to establish the user's individual physiological baseline. During the online learning phase, multimodal signals are continuously collected and features are extracted during the user's daily use. The model parameters are dynamically updated through an incremental learning algorithm to adapt the model to changes in the user's individual eye use habits. S43. The calibrated multimodal fusion model learns the correlation between multimodal features through the interaction layer, performs in-depth analysis on the input weighted feature vector, and mines the potential correlation between features to evaluate fatigue state; S44. The multimodal fusion model classifies the analysis results through the output layer and finally outputs an eye fatigue level that is adapted to individual differences.

[0010] Preferably, the following steps are also included: S5. Trigger intervention feedback based on the output fatigue level, where: Mild fatigue is indicated by vibration from the frame's vibration motor, prompting a rest. For moderate fatigue, voice prompts are played via bone conduction headphones. In cases of severe eye fatigue, an alert will be sent to the linked mobile app, and an eye fatigue trend report will be generated.

[0011] Preferably, the eye fatigue trend report includes: Fatigue time distribution: Record the specific time periods and frequency of fatigue occurrences within each day / week; Fatigue intensity analysis, statistical analysis of the duration and changing trends of mild, moderate and severe fatigue; Correlation factor analysis, combined with multimodal signal characteristics, identifies eye-use behaviors strongly correlated with fatigue; Personalized suggestions generate targeted eye care guidance based on fatigue trends and related factors.

[0012] An eye fatigue detection system for EEG glasses, used to perform the eye fatigue detection method for the EEG glasses, comprising: Biosignal acquisition module: used to acquire raw biological signals directly related to eye fatigue, including electroencephalogram (EEG) signals, eye movement and pupil signals, and heart rate variability signals; The joint denoising module is used to perform joint denoising on the original biological signals based on the physiological correlation of the original biological signals, so as to obtain an effective signal that can reflect the state of eye fatigue. The feature extraction and weighting module is used to extract feature parameters that are strongly correlated with eye fatigue from the effective signal, learn the contribution of each feature to fatigue through a feature weight allocation mechanism, and generate a weighted feature vector of focusing eye fatigue. The fatigue assessment module is used to analyze the weighted feature vector by using a pre-trained multimodal fusion model and calibrating the user's individual physiological baseline, and output an eye fatigue level that is adapted to individual differences.

[0013] The beneficial effects of this invention are as follows: By collecting EEG, eye movement, pupil, and heart rate variability signals, this invention overcomes the limitations of traditional single-modality detection, which is susceptible to individual differences or environmental interference. It can capture collaborative fatigue patterns that cannot be reflected by a single signal, significantly improving detection accuracy. Based on a joint denoising mechanism of physiological correlation, through cross-modal event labeling and submodal denoising processing, it effectively removes noise such as EMG artifacts, ambient light, and motion interference, ensuring signal quality. The feature extraction and dynamic weight allocation mechanism focuses on the core features of fatigue and dynamically adjusts the weights according to the contribution of features to fatigue, enhancing the relevance of the model input. Combined with personalized model training through initial calibration and online learning, it adapts to the physiological baselines and changes in eye habits of different users, avoiding the problem of general models not being universal. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the method steps of the present invention.

[0015] Figure 2 This is a schematic diagram of the steps of the method for obtaining an effective signal according to the present invention.

[0016] Figure 3 This is a schematic diagram of the steps involved in generating weighted feature vectors according to the present invention.

[0017] Figure 4 This is a schematic diagram of the steps in the method for outputting the eye fatigue level of the present invention. Detailed Implementation

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the specification is for the purpose of describing particular embodiments only and is not intended to limit the invention; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings are used to distinguish different objects and not to describe a particular order.

[0019] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0021] This invention provides an appendix Figures 1 to 4 In this embodiment of the invention, a method and system for detecting eye fatigue using EEG glasses achieves the goal of accurately detecting eye fatigue states in a way that is adapted to individual differences by simultaneously acquiring and jointly denoising multimodal biosignals, weighted extraction of fatigue-related features, and personalized multimodal fusion model analysis. Specific Implementation Example 1 A method for detecting eye fatigue using EEG glasses includes the following steps: S1. Collect raw biological signals directly related to eye fatigue, including electroencephalogram (EEG) signals, eye movement and pupil signals, and heart rate variability signals; Specifically, when a user wears the EEG glasses, the device synchronously collects raw biological signals directly related to eye fatigue through multimodal sensors: flexible dry electrodes built into the temples fit into areas such as the temporal and occipital lobes of the head, collecting EEG signals reflecting visual cortical activity in real time; a miniature infrared camera at the front of the frame captures images of the eyes, capturing eyelid movements (such as blinking) and pupil dynamics (such as contraction / dilation) to obtain eye movement and pupil signals; a photoplethysmography (PPG) sensor at the end of the temples collects pulse waves behind the ears, calculating the variability between adjacent heartbeats to obtain heart rate variability (HRV) signals; the three modal signals are synchronously sampled through a unified clock source to ensure timestamp alignment, providing a time-related data basis for subsequent analysis.

[0022] Furthermore, the EEG glasses also include behavioral data acquisition and physiological signal acquisition. Behavioral data acquisition includes lighting environment (ambient light sensor), sitting posture habits (six-axis gyroscope), wearing time (timer), infrared distance (infrared sensor detects screen distance), outdoor activities (GPS positioning), and environmental noise (microphone sampling rate 16kHz). Physiological signal acquisition includes EEG signals. Four flexible dry electrodes are distributed on the temples (covering the occipital and temporal lobes) to collect attention, fatigue and mood values, eye movement and pupil signals (miniature infrared camera), and heart rate variability signals (PPG sensor) at a certain sampling rate. All data is uploaded to the cloud in real time via wireless network (WiFi / 4G), with behavioral data compressed by the minute and physiological signals streamed.

[0023] S2. Based on the physiological correlation of biological raw signals, joint denoising is performed on biological raw signals to obtain effective signals that can reflect the state of eye fatigue; Specifically, based on the physiological correlation of multimodal signals, a joint denoising process is used to remove interference and retain effective information: First, physiological events such as rapid blinking and fixation point drift are detected through eye movement and pupil signals, and the time windows of EEG and HRV signals that are interfered with by these events are marked simultaneously; then, based on the interference characteristics of each modality signal (such as EEG artifacts, ambient light from eye movement, and motion interference from HRV), denoising processing is performed in conjunction with the marked time windows (such as independent component analysis of EEG, image filtering and registration of eye movement, and bandpass filtering of HRV); finally, the correlation of multimodal signals before and after denoising is compared, and if the deviation exceeds a preset threshold, the denoising parameters are iteratively optimized until an effective signal that can accurately reflect the state of eye fatigue is obtained.

[0024] S3. Extract feature parameters that are strongly correlated with eye fatigue from the effective signal, learn the contribution of each feature to fatigue through a feature weight allocation mechanism, and generate a weighted feature vector of focusing eye fatigue; Specifically, features strongly correlated with eye fatigue are extracted from the denoised effective signal, and a weighted feature vector of focusing fatigue is generated through a weight allocation mechanism: frequency domain features reflecting visual cortex fatigue (such as the power ratio of alpha waves to beta waves and theta waves to beta waves) are extracted from EEG signals; temporal features reflecting eye muscle and pupillary accommodation fatigue (such as blink frequency and pupil diameter change rate) are extracted from eye movement and pupillary signals; and frequency domain features reflecting autonomic nervous system dysfunction (such as the power ratio of low frequency to high frequency) are extracted from HRV signals. The contribution of each feature to fatigue is learned through the feature weight allocation mechanism (such as the dynamic increase of the weight of certain features in the later stage of fatigue), and the weight of each feature is dynamically adjusted. Finally, each feature is multiplied by its corresponding weight and concatenated to generate a weighted feature vector of focusing eye fatigue, which serves as the input for subsequent model analysis.

[0025] S4. Using a pre-trained multimodal fusion model and combined with user individual physiological baseline calibration, the weighted feature vector is analyzed to output an eye fatigue level adapted to individual differences.

[0026] Specifically, using a pre-trained multimodal fusion model, combined with user-specific physiological baseline calibration, the weighted feature vector is analyzed and an eye fatigue level adapted to individual differences is output: the weighted feature vector is input into the input layer of the multimodal fusion model; when the user uses the model for the first time, multimodal signals under normal eye use conditions are collected to train the initial model and establish an individual physiological baseline; during daily use, the model parameters are continuously updated through online learning to adapt to changes in the user's eye use habits; the model learns the correlation between multimodal features (such as the synergy between EEG and HRV features) through the interaction layer, and outputs a fatigue level (such as mild, moderate, and severe) adapted to individual differences after classification processing.

[0027] This embodiment simultaneously acquires multimodal biosignals, covering multidimensional physiological information from the visual cortex, eye muscles, and autonomic nervous system, thus solving the problem of single-modal detection being susceptible to interference. The joint denoising process, based on the correlation of physiological events, effectively eliminates environmental and motion interference, improving signal quality. Feature extraction and weighting mechanisms focus on core fatigue features, and dynamic weight allocation highlights key indicators, enhancing the relevance of model input. Individual physiological baseline calibration, combined with initial training and online learning, adapts to the physiological differences and habit changes of different users, avoiding the defect of "general-purpose models not being universal." In summary, this embodiment achieves high-precision, personalized detection of eye fatigue, providing a reliable basis for real-time intervention. Specific Implementation Example 2 The EEG signals were acquired using flexible dry electrodes on the temples of the mirror, covering the occipital and temporal lobe regions. The eye movement and pupil signals are acquired by capturing eye images using a miniature infrared camera in the lens frame; The heart rate variability signal was acquired by collecting the pulse wave behind the ear using a PPG sensor on the temple of the glasses. The electroencephalogram (EEG) signal, the eye movement and pupillary signals, and the heart rate variability signal are sampled synchronously using the same clock source to ensure timestamp alignment.

[0029] Specifically, three flexible dry electrodes are installed on the inner side of the temples, with the electrode array distributed along the extension direction of the temples. When worn, they fit the user's bilateral temporal lobes (2cm above and in front of the ears) and occipital lobes (1cm in front of the posterior hairline), respectively. The electrodes are connected to the signal processing module via low-noise wires to acquire EEG signals (raw amplitude range 1-100μV) at a sampling rate of 250Hz. The flexible dry electrodes do not require conductive gel and can adapt to the skin contour, ensuring comfort during long-term wear.

[0030] Two miniature infrared cameras are embedded in the front of the frame (on both sides of the bridge of the nose), with the lenses aimed at the user's eyes. The cameras capture images of the eyes using the principle of near-infrared reflection, capturing eyelid movements (blinking) and pupil contraction / dilation dynamics in real time, and outputting raw signals including blink frequency, pupil diameter, and fixation point coordinates.

[0031] The inner side of the temple end of the glasses integrates a PPG sensor, which fits against the skin behind the ear (the depression between the tragus and the mastoid process) when worn. The sensor collects the pulse wave signal behind the ear at a sampling rate of 100Hz, calculates the interval between adjacent heartbeats (RR interval) using the photoplethysmography method, and generates the raw heart rate variability (HRV) signal.

[0032] The acquisition modules for EEG signals, eye movement and pupillary signals, and HRV signals are all connected to the same clock source (high-precision crystal oscillator, frequency error <0.01%). Each sensor synchronously acquires an initial timestamp upon startup, and the timestamp is calibrated every 100ms through the clock source during sampling to ensure that the time alignment error of the three-modal signals is ≤1ms. Specific Implementation Example 3 The method of jointly denoising raw biological signals based on their physiological correlation to obtain effective signals reflecting eye fatigue includes the following steps: S21. Detect physiological events such as rapid blinking, fixation point drift, and head movement through eye movement and pupil signals, and simultaneously mark the corresponding time windows in EEG signals and heart rate variability signals that are affected by these events; Specifically, physiological events such as rapid blinking, fixation drift, and head movements in users are detected through eye movement and pupil signals, and the time windows affected by these events in electroencephalogram (EEG) and heart rate variability (HRV) signals are simultaneously marked. Rapid blink event detection analyzes the blink duration of eye movement and pupil signals (calculated from eye images captured by an infrared camera to the time from eyelid closure to opening). When a blink duration of <0.2s is detected, it is marked as a "rapid blink" event, and the start timestamp (t_start) and end timestamp (t_end) of the event are recorded simultaneously. The gaze point drift event detection uses an image registration algorithm to calculate the gaze point coordinate offset (Δx, Δy) between two adjacent eye images. When Δx or Δy exceeds 5 pixels (corresponding to an actual gaze offset >2°), it is marked as a "gaze point drift" event, with a time window of [t_start-0.3s, t_end+0.3s]. Head motion event detection: Combining the accelerometer built into the temple, when the rate of change of acceleration is >0.5g, it is marked as a "head motion" event, with a time window of [t_start-0.5s, t_end+0.5s]; Synchronization marking maps the time window of the above events to the corresponding data segments of EEG signals (sampling rate 250Hz) and HRV signals (sampling rate 100Hz). For example, if the time window of the rapid blinking event is [10.2s, 10.4s], then the 2550th-2600th sampling point in the EEG signal (250Hz×10.2s=2550, 250Hz×10.4s=2600) and the 1020th-1040th sampling point in the HRV signal (100Hz×10.2s=1020, 100Hz×10.4s=1040) are marked as interference segments.

[0034] S22. Based on the interference characteristics of each modality signal, denoising is performed using marked interference time windows: For EEG signals, independent component analysis is used to separate and remove EMG artifacts, while signal segments affected by EMG artifacts are corrected or removed; for eye movement and pupil signals, filtering is used to eliminate ambient light interference, and image registration algorithms are used to correct image shifts caused by head movements; for heart rate variability signals, bandpass filtering is used to remove motion artifacts, while smoothing out abnormal heartbeat intervals that are interfered with. Specifically, denoising is performed by combining the interference characteristics of each modal signal with marked time windows: EEG signal denoising: Independent component analysis (ICA) was used to separate artifact components from the EEG signals within the marked interference time window. 1. Input the raw EEG signal into the ICA algorithm (based on FastICA) and decompose it into multiple independent components (ICs); 2. Identify ICs corresponding to electromyographic artifacts (EMG, high-frequency components) and electrooculography artifacts (EOG, strongly correlated with blink events) through manual or automatic classification (e.g., based on kurtosis threshold >3); 3. After removing the aforementioned artifact ICs, the EEG signal is reconstructed, and the signal of the unlabeled interference segment is bandpass filtered (0.5-30Hz) to remove DC drift and high-frequency noise; Eye movement and pupil signal denoising: 1. Handling of ambient light interference: Median filtering (3×3 window) is applied to the eye images captured by the infrared camera to eliminate random noise points; 2. Head motion offset correction: The current frame image is aligned with the previous frame reference image using an image registration algorithm (based on SIFT feature matching) to correct the image offset caused by head motion (maximum allowable offset of 10 pixels). 3. Remove abnormal data: If the pupil diameter change rate is >50% / s (exceeding the normal physiological range), then mark the frame data as abnormal and replace it with interpolation (using the average value of the previous and next frames); HRV signal denoising: 1. Motion artifact removal: Bandpass filtering (0.5-3Hz) is applied to the PPG signal to preserve heart rate-related frequency components; 2. Correction for abnormal heartbeat intervals: Calculate the difference (ΔRR) between adjacent RR intervals. If ΔRR > 200ms (exceeding the normal variability range), smooth the abnormal point by linear interpolation (taking the average of the two normal RR intervals before and after).

[0035] S23. Compare the correlation of multimodal signals before and after denoising. If the deviation exceeds the preset threshold, repeat steps S21 to S22 until the requirements are met, and finally obtain an effective signal that can accurately reflect the state of eye fatigue.

[0036] Specifically, the correlation of multimodal signals before and after denoising is compared to verify the denoising effect: Among them, the correlation calculation is to calculate the Pearson correlation coefficient (ρ) between the original signal before denoising and the effective signal after denoising. For EEG signals, the ρ value is taken as the alpha power spectral density (PSD), for eye movement signals, the ρ value is taken as the blink frequency, and for HRV signals, the ρ value is taken as the LF / HF ratio. The threshold judgment is based on a preset correlation threshold of ρ≥0.8 (preserving the main physiological information). If the ρ value of any modal signal is <0.8, then steps S21-S22 are re-executed (for example, adjusting the artifact recognition threshold of ICA or the number of feature matches in image registration). When the ρ value of all modal signals is ≥0.8, the iteration terminates and an effective signal that accurately reflects the state of eye fatigue is output. Specific Implementation Example 4 The step of extracting feature parameters strongly correlated with eye fatigue from the effective signal, learning the contribution of each feature to fatigue through a feature weight allocation mechanism, and generating a weighted feature vector for focusing on eye fatigue includes the following steps: S31. Extract feature parameters directly related to eye fatigue from the effective signals respectively. For EEG signals, extract frequency domain features reflecting the degree of visual cortex fatigue. For eye movement and pupil signals, extract temporal features reflecting the fatigue state of eye muscles and pupil accommodation function. For heart rate variability signals, extract frequency domain features reflecting the functional disorder of the autonomic nervous system caused by fatigue. Specifically, based on the effective signal after joint denoising, feature parameters directly related to eye fatigue are extracted from the three modal signals of electroencephalography (EEG), eye movement and pupillary light, and heart rate variability (HRV). The specific steps include: EEG signal characteristics (reflecting visual cortex fatigue): Frequency domain analysis was performed on the denoised EEG signals (sampling rate 250Hz, covering the occipital and temporal lobes): 1. Calculate the power spectral density (PSD) using the Welch algorithm (Hamming window, window length 2s, overlap 1s) and extract the PSD values ​​for alpha waves (8-13Hz), theta waves (4-7Hz), and beta waves (13-30Hz); 2. Calculate the α / β ratio (α wave PSD / β wave PSD) and the θ / β ratio (θ wave PSD / β wave PSD) as core frequency domain features reflecting the degree of visual cortex fatigue (when fatigued, θ waves increase, β waves decrease, and the ratio increases).

[0038] Eye movement and pupillary signal characteristics (reflecting eye muscle and pupillary accommodation fatigue): Temporal analysis was performed on the denoised eye-tracking image sequence (30fps): 1. Blinking frequency: Count the number of blinks in 1 minute (times / minute). The blinking frequency decreases when fatigued (normal is 15-20 times / minute, and when fatigued it is <10 times / minute). 2. Pupil diameter change rate: Calculate the change in pupil diameter over 5 consecutive frames (ΔD = D(t+1) - D(t)), and take the average of the absolute values. When fatigued, pupil accommodation ability decreases, and the change rate decreases. 3. Fixation point drift amplitude: Calculate the maximum offset (in pixels) of fixation point coordinates (x, y) within 10 seconds. When fatigued, eye muscle control weakens, and the drift amplitude increases (normal < 5 pixels, fatigued > 10 pixels).

[0039] HRV signal characteristics (reflecting autonomic nervous system dysfunction): Frequency domain analysis was performed on the denoised HRV signal (RR interval sequence): 1. Calculate the power spectrum at low frequencies (LF, 0.04-0.15 Hz) and high frequencies (HF, 0.15-0.4 Hz) using Fast Fourier Transform (FFT); 2. Calculate the LF / HF ratio (low-frequency power / high-frequency power) and total power (LF+HF). When fatigued, the sympathetic nervous system is activated, and the LF / HF ratio increases (normal 1.5-2.5, >3.0 when fatigued).

[0040] S32. Learn the contribution of each feature to eye fatigue through a feature weight allocation mechanism, wherein the weight of each feature is dynamically adjusted according to the strength of the correlation between each feature and eye fatigue. Specifically, the contribution of each feature to eye fatigue is learned through a feature weight allocation mechanism, which is implemented using a dynamic attention mechanism and includes the following steps: Among them, the initial weight setting is based on physiological knowledge (such as θ / β ratio weight 0.3, blink frequency weight 0.2, LF / HF ratio weight 0.2). The dynamic adjustment rule involves dynamically updating the weights based on the correlation between the current feature and the fatigue label using an online learning algorithm (such as stochastic gradient descent with a learning rate of 0.01). 1. Collect the Pearson correlation coefficients (ρ) between various characteristics and fatigue levels in historical data. For example, the θ / β ratio and severe fatigue have a ρ = 0.75, and blink frequency and severe fatigue have a ρ = -0.68. 2. Weight update formula: W_i(t+1) = W_i(t) + η (ρ_i - avg(ρ)), where η is the learning rate and avg(ρ) is the average correlation coefficient of all features; 3. Constrain the weight range (0≤W_i≤1) to avoid excessively high weights for a single feature (e.g., the weight of the θ / β ratio should not exceed 0.4).

[0041] S33. Multiply each feature by its corresponding weight and then concatenate them to generate a weighted feature vector of eye fatigue caused by focusing.

[0042] Specifically, each feature is multiplied by its corresponding dynamic weight and then concatenated to generate a weighted feature vector for focusing eye fatigue. This process includes the following steps: Feature-weight mapping: Assume the current features and their weights are: θ / β ratio (0.35), α / β ratio (0.25), blink frequency (0.20), pupil diameter change rate (0.15), LF / HF ratio (0.25); Weighted calculation: Multiply each feature value by its corresponding weight. For example, if the measured value of the θ / β ratio is 1.8, then the weighted value is 1.8 × 0.35 = 0.63. Vector concatenation: The weighted feature values ​​are concatenated in a fixed order (e.g., [θ / β weighted value, α / β weighted value, blink frequency weighted value, pupil diameter change rate weighted value, LF / HF weighted value]) to form a 5-dimensional weighted feature vector [0.63, 0.50, 1.60, 0.30, 0.75], which is used as the input to the subsequent multimodal fusion model. Specific Implementation Example 5 The characteristic parameters strongly correlated with eye fatigue include: The ratio of alpha wave power spectral density to beta wave power spectral density, and the ratio of theta wave power spectral density to beta wave power spectral density in EEG signals; Specifically, electroencephalogram (EEG) signals were acquired using flexible dry electrodes on the temples of the spectacle (covering the occipital and temporal lobes). After combined denoising, the following frequency domain features were extracted to reflect the degree of visual cortical fatigue. The calculation method for the power spectral density ratio of alpha waves to beta waves (α / β ratio) is as follows: frequency domain analysis was performed on the denoised 250Hz EEG signal, and the Welch algorithm with a Hamming window (window length 2s, overlap 1s) was used to calculate the power spectral density (PSD). Among them, alpha waves are 8-13Hz (corresponding to the rhythm of the visual cortex in a relaxed state), and beta waves are 13-30Hz (corresponding to high-frequency activities related to alertness). During normal eye use, the power of alpha waves and beta waves is balanced (α / β ratio approximately 0.8-1). 2) When the eyes are fatigued, the excitability of the visual cortex decreases, resulting in a decrease in the power of beta waves and a relative increase in the power of alpha waves, leading to an increase in the alpha / beta ratio (>1.5). The calculation method for the power spectral density ratio of theta waves to beta waves (θ / β ratio) is the same as above. The PSD value of theta waves (4-7Hz) is extracted. Theta waves are closely related to fatigue-related slow wave activity. When fatigued, the power of theta waves increases significantly (reflecting the inhibition of neural activity), while the power of beta waves continues to decrease, resulting in an increase in the θ / β ratio (normal <0.5, >0.8 when fatigued).

[0044] Blink frequency, pupil diameter change rate, and fixation point drift amplitude of eye movement and pupil signals; Specifically, eye movement and pupil signals were acquired using a miniature infrared camera in the eyeglass frame (30fps) and denoised. The following temporal features were extracted to reflect the fatigue state of the eye muscles and pupillary accommodation function: blink frequency (times / minute) was calculated by statistically analyzing the number of complete eyelid closure-opening cycles recognized by infrared images within one minute. Under normal eye use, it is 15-20 times / minute (to maintain corneal moisture). When fatigued, it decreases to <10 times / minute due to weakness of the eye muscles, which can easily lead to dry eyes and eye irritation; pupil diameter change rate (% / s) was the absolute change in pupil diameter within 5 consecutive frames (approximately 0.17s) compared to the initial diameter. The average diameter ratio is calculated as follows: Under normal eye use, the pupil automatically adjusts with light or the object being gazed upon (change rate approximately 5-10% / s). When fatigued, the pupillary sphincter and dilator muscles weaken, leading to a decrease in accommodative ability, and the change rate decreases to <3% / s. The gaze point drift amplitude (pixels) is calculated by determining the gaze point position for each frame using an image registration algorithm, and then calculating the maximum offset of the gaze point coordinates (x, y) within 10 seconds. Under normal eye use, the eye muscles stably control the gaze point (drift amplitude <5 pixels). When fatigued, due to a decrease in the eye muscle control ability, the gaze point is prone to involuntary drift (amplitude >10 pixels), manifesting as "double vision".

[0045] The ratio of low-frequency to high-frequency power and the total power of the heart rate variability signal.

[0046] Specifically, heart rate variability (HRV) signals were acquired using a PPG sensor (100Hz) on the temple of the glasses and, after denoising, the following frequency domain features were extracted to reflect autonomic nervous system dysfunction caused by fatigue: the low-frequency to high-frequency power ratio (LF / HF ratio) was calculated by performing a Fast Fourier Transform (FFT) on the RR interval sequence to determine the low-frequency (LF, 0.04-0.15Hz, reflecting sympathetic nerve activity) and high-frequency (HF) power ratios. The power spectrum (0.15-0.4Hz, reflecting parasympathetic nerve activity) is obtained by taking the ratio. Under normal eye use, the LF / HF ratio is 1.5-2.5 (sympathetic-parasympathetic balance). When fatigued, the continuous activation of the sympathetic nerves leads to an increase in LF power and a decrease in HF power, resulting in an increase in the LF / HF ratio (>3.0). The total power (LF+HF) is the sum of the LF and HF powers. Normally, it reflects the overall level of heart rate variability (500-1500ms²). When fatigued, the total power decreases (<300ms²) due to the disorder of autonomic nerve regulation, indicating the accumulation of physiological stress. Specific Implementation Example Six The process of using a pre-trained multimodal fusion model, combined with user-specific physiological baseline calibration, to analyze the weighted feature vector and output an eye fatigue level adapted to individual differences includes the following steps: S41. Input the generated weighted feature vector of eye fatigue caused by focusing on the screen into the input layer of the multimodal fusion model as the initial data for model analysis; Specifically, the weighted feature vector of focusing eye fatigue generated in Example 4 (such as [θ / β weighted value, blink frequency weighted value, LF / HF weighted value]) is input into the input layer of the multimodal fusion model. The input layer eliminates dimensional differences through normalization (mapping each feature value to the [0,1] interval). For example, the θ / β weighted value is 0.63 (original value 1.8 × weight 0.35), and after normalization, it becomes 0.63 / 1.5 (maximum theoretical value) = 0.42, ensuring the stability of the model input data.

[0048] S42. Individual physiological baseline calibration includes initial calibration and online learning. During initial calibration, multimodal signals under normal eye use are collected when the user uses the device for the first time, features are extracted and an initial model is trained to establish the user's individual physiological baseline. During the online learning phase, multimodal signals are continuously collected and features are extracted during the user's daily use. The model parameters are dynamically updated through an incremental learning algorithm to adapt the model to changes in the user's individual eye use habits. Specifically, the initial calibration phase (when the user wears the EEG glasses for the first time) requires a 25-minute calibration process: For the first 10 minutes, the user reads a paper book in natural light (avoiding blue light interference from electronic screens), and the device simultaneously collects EEG, eye movement, and HRV signals, extracting features such as the θ / β ratio (average 0.4), blink frequency (18 times / minute), and LF / HF ratio (2.0) as "fatigue-free" baseline data; For the next 15 minutes, the user continuously uses a mobile phone (screen brightness 50%) to browse short videos, and the device collects features under fatigue conditions (θ / β ratio rises to 0.9, blink frequency drops to 10 times / minute, LF / HF ratio rises to 3.2) as "fatigue" label data; Subsequently, the fatigue-free (label 0) and fatigue (labels 1-3) data are input into a multimodal fusion model with randomly initialized initial parameters, and the model weights are optimized through the cross-entropy loss function (learning rate 0.001, 50 iterations), finally establishing an individual physiological baseline model for the user.

[0049] During the online learning phase (in daily user use), the model updates its parameters hourly using an incremental learning algorithm (such as FTRL) suitable for small-batch data updates: First, it collects multimodal signals from the current 30 minutes and extracts features to label the actual fatigue state (either manually confirmed by the user via the app or automatically predicted by the model); then, it calculates the deviation between the current features and the baseline features (e.g., user A's baseline θ / β ratio is 0.4, the current measured value is 0.6, the deviation is +0.2); finally, it adjusts the weights of the attention interaction layer through gradient descent (e.g., reducing the weight of the θ / β ratio by 0.05 and increasing the weight of blinking frequency by 0.03) to adapt the model to the individual physiological fluctuations of the user (e.g., when user B's baseline LF / HF ratio is high, the model automatically reduces the contribution of this feature to fatigue assessment).

[0050] S43. The calibrated multimodal fusion model learns the correlation between multimodal features through the interaction layer, performs in-depth analysis on the input weighted feature vector, and mines the potential correlation between features to evaluate fatigue state; Specifically, the calibrated multimodal fusion model learns the correlation between multimodal features through an interaction layer based on the Transformer self-attention mechanism: the interaction layer assigns query (Q), key (K), and value (V) vectors to each feature and calculates the attention score between features (e.g., the attention score of the θ / β ratio and the LF / HF ratio is 0.7, indicating a strong correlation between the two). Then, the attention score and feature value are weighted and summed to generate a fusion feature (e.g., fusion value = 0.7 × θ / β weighted value + 0.2 × blink frequency weighted value + 0.1 × LF / HF weighted value), thereby uncovering collaborative fatigue patterns that cannot be reflected by single-modal features (e.g., when θ / β increases and LF / HF increases simultaneously, the fatigue risk increases by 2 times).

[0051] S44. The multimodal fusion model classifies the analysis results through the output layer and finally outputs an eye fatigue level that is adapted to individual differences.

[0052] Specifically, after the fused features are dimensionality-reduced by a fully connected layer (256 neurons), the softmax activation function outputs four probability distributions (Level 0: no fatigue, Level 1: mild fatigue, Level 2: moderate fatigue, Level 3: severe fatigue). The category with the highest probability is taken as the final fatigue level. For example, if a user's current fused feature value is 0.85 (threshold: Level 0 < 0.3, Level 1 0.3-0.5, Level 2 0.5-0.8, Level 3 > 0.8), the model outputs probabilities of [0.05, 0.15, 0.6, 0.2], and the final determination is Level 2 (moderate fatigue). Specific Implementation Example 7 It also includes the following steps: S5. Trigger intervention feedback based on the output fatigue level, where: Mild fatigue is indicated by vibration from the frame's vibration motor, prompting a rest. For moderate fatigue, voice prompts are played via bone conduction headphones. In cases of severe eye fatigue, an alert will be sent to the linked mobile app, and an eye fatigue trend report will be generated.

[0054] The eye fatigue trend report includes: Fatigue time distribution: Record the specific time periods and frequency of fatigue occurrences within each day / week; Fatigue intensity analysis, statistical analysis of the duration and changing trends of mild, moderate and severe fatigue; Correlation factor analysis, combined with multimodal signal characteristics, identifies eye-use behaviors strongly correlated with fatigue; Personalized suggestions generate targeted eye care guidance based on fatigue trends and related factors.

[0055] Specifically, after the multimodal fusion model outputs the eye fatigue level (Level 0: no fatigue, Level 1: mild, Level 2: moderate, Level 3: severe), the system automatically triggers the corresponding feedback based on the level threshold (intervention is initiated when Level ≥ 1).

[0056] The trigger condition for mild fatigue feedback is a fatigue level of 1 (mild), characterized by a θ / β ratio of 0.5-0.8, a blinking frequency of 10-15 times / minute, and an LF / HF ratio of 2.5-3.0 (based on the user's individual baseline calibration). The feedback method is that the built-in vibration motor in the frame (located at the connection between the temple and the frame, with an amplitude of 0.1mm) vibrates at a frequency of 50Hz for 1 second, and repeats once every 30 seconds (to avoid excessive interference). For example, if a student uses a computer continuously for 40 minutes, the model outputs a fatigue level of 1, and the temple vibrates to indicate "Current mild fatigue, it is recommended to rest after 5 minutes".

[0057] The trigger condition for moderate fatigue feedback is a fatigue level of 2 (moderate), characterized by a θ / β ratio of 0.8-1.2, a blinking frequency of 8-10 times / minute, and an LF / HF ratio of 3.0-3.5. The feedback method is to play a preset voice prompt through the bone conduction headphones integrated into the temples (frequency response range 200-8000Hz, output sound pressure level 60dB), which reads "Currently experiencing moderate fatigue, it is recommended to close your eyes and rest for 5 minutes, and massage around the eyes to relieve the symptoms." This prompt is repeated every 2 minutes (to ensure user attention). If the user continuously uses their phone for 1 hour, the model determines it to be level 2 fatigue, the bone conduction headphones play the voice prompt, and the vibration motor simultaneously provides a short vibration (20Hz, 0.5 seconds) to enhance the reminder.

[0058] The triggering condition for severe fatigue feedback is a fatigue level of 3 (severe), characterized by a theta / β ratio > 1.2, a blinking frequency < 8 times / minute, and an LF / HF ratio > 3.5, with a duration of ≥ 10 minutes (to avoid occasional interference and misjudgment). The feedback method is to push a notification to the bound mobile APP via Bluetooth (BLE 5.0 protocol). The notification bar displays a red warning icon and the text "Warning: Current severe fatigue, you need to stop using your eyes immediately!", and the APP triggers a ringtone (maximum volume, lasting 10 seconds).

[0059] The severe fatigue feedback also includes the generation of an eye fatigue trend report. The system automatically summarizes fatigue data from the past 7 days and generates a report containing the following: Fatigue time distribution: Mark the peak time of fatigue each day (e.g., 20:00-22:00 is the peak). Fatigue intensity analysis: Statistics on the duration and weekly trends of mild (30%), moderate (50%), and severe (20%) fatigue (e.g., the duration of severe fatigue this week increased by 15% compared to last week). Correlation factor analysis: Combining multimodal characteristics (such as the correlation coefficient of θ / β ratio and continuous screen viewing time of 0.82), "continuous use of electronic screens for more than 1 hour" was identified as the main cause of fatigue. Personalized suggestions: Based on the analysis results, specific guidance is generated such as "take a 10-minute break after every 40 minutes of using electronic devices" and "turn on warm light mode when studying at night".

[0060] Furthermore, the cloud-based backend receives behavioral data (including lighting environment, posture habits, wearing time, infrared distance, outdoor activity trajectory, and environmental noise) and physiological signals (EEG attention / fatigue / emotion values, eye movement and pupil signals, and heart rate variability signals) uploaded by the glasses. It then uses a timestamp alignment algorithm to correlate multi-source information (e.g., simultaneously analyzing the correlation between posture tilting periods and EEG fatigue fluctuations) and generates structured node reports (output once per minute) based on a pre-trained multimodal fusion model. The report content includes: Fatigue level classification (mild / moderate / severe); Attention score (0-100) and emotional state (positive / neutral / negative); Behavioral-physiological correlation analysis conclusions (e.g., "Screen distance of less than 30cm for more than 10 minutes leads to an increase in fatigue level to moderate"). Users can view historical fatigue trend heatmaps and attention change curves in real time through mobile APP or web backend, and receive personalized intervention suggestions (e.g., "Environmental noise > 65dB, it is recommended to wear noise-canceling headphones"). Specific Implementation Example 8 An eye fatigue detection system for EEG glasses, used to perform the eye fatigue detection method for the EEG glasses, comprising: Biosignal acquisition module: used to acquire raw biological signals directly related to eye fatigue, including electroencephalogram (EEG) signals, eye movement and pupil signals, and heart rate variability signals; The joint denoising module is used to perform joint denoising on the original biological signals based on the physiological correlation of the original biological signals, so as to obtain an effective signal that can reflect the state of eye fatigue. The feature extraction and weighting module is used to extract feature parameters that are strongly correlated with eye fatigue from the effective signal, learn the contribution of each feature to fatigue through a feature weight allocation mechanism, and generate a weighted feature vector of focusing eye fatigue. The fatigue assessment module is used to analyze the weighted feature vector by using a pre-trained multimodal fusion model and calibrating the user's individual physiological baseline, and output an eye fatigue level that is adapted to individual differences.

[0062] This invention can be used in a wide range of general-purpose or special-purpose computer system environments or configurations.

[0063] Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments that include any of the above systems or devices.

[0064] This invention can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules.

[0065] Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via communication networks.

[0066] In a distributed computing environment, program modules can reside on local and remote computer storage media, including storage devices.

[0067] Specifically, those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0068] It should be understood that although the steps in the flowcharts in the accompanying drawings are shown sequentially as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise expressly stated herein, there is no strict order in which these steps are performed, and they may be performed in other orders.

[0069] Moreover, at least some steps in the flowchart of the attached figure may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. Their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0070] Obviously, the embodiments described above are only some embodiments of the present invention, and not all embodiments. The accompanying drawings show preferred embodiments of the present invention, but do not limit the scope of the invention. The present invention can be implemented in many different forms; rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the present invention.

[0071] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this specification and drawings, whether directly or indirectly applied to other related technical fields, are similarly within the scope of protection of this patent.

Claims

1. A method for detecting eye fatigue using brainwave glasses, characterized in that, Includes the following steps: S1. Collect raw biological signals directly related to eye fatigue, including electroencephalogram (EEG) signals, eye movement and pupil signals, and heart rate variability signals; S2. Based on the physiological correlation of biological raw signals, joint denoising is performed on biological raw signals to obtain effective signals that can reflect the state of eye fatigue; S3. Extract feature parameters that are strongly correlated with eye fatigue from the effective signal, learn the contribution of each feature to fatigue through a feature weight allocation mechanism, and generate a weighted feature vector for focusing on eye fatigue. The weight of each feature is dynamically adjusted according to the strength of the correlation between each feature and eye fatigue. S4. Using a pre-trained multimodal fusion model, combined with user individual physiological baseline calibration, the weighted feature vector is analyzed to output an eye fatigue level adapted to individual differences; The method of jointly denoising raw biological signals based on their physiological correlation to obtain effective signals reflecting eye fatigue includes the following steps: S21. Detect physiological events such as rapid blinking, fixation point drift, and head movement through eye movement and pupil signals, and simultaneously mark the corresponding time windows in EEG signals and heart rate variability signals that are interfered with by any of the physiological events; S22. Based on the interference characteristics of each modality signal, denoising is performed using marked interference time windows: For EEG signals, independent component analysis is used to separate and remove EMG artifacts, while signal segments affected by EMG artifacts are corrected or removed; for eye movement and pupil signals, filtering is used to eliminate ambient light interference, and image registration algorithms are used to correct image shifts caused by head movements; for heart rate variability signals, bandpass filtering is used to remove motion artifacts, while smoothing out abnormal heartbeat intervals that are interfered with. S23. Compare the correlation of multimodal signals before and after denoising. If the deviation exceeds the preset threshold, repeat steps S21 to S22 until the requirements are met, and finally obtain an effective signal that can accurately reflect the state of eye fatigue. The step of extracting feature parameters strongly correlated with eye fatigue from the effective signal, learning the contribution of each feature to fatigue through a feature weight allocation mechanism, and generating a weighted feature vector for focusing on eye fatigue includes: S31. Extract feature parameters directly related to eye fatigue from the effective signals respectively. For EEG signals, extract frequency domain features reflecting the degree of visual cortex fatigue. For eye movement and pupil signals, extract temporal features reflecting the fatigue state of eye muscles and pupillary accommodation function. For heart rate variability signals, extract frequency domain features reflecting the functional disorder of the autonomic nervous system caused by fatigue.

2. The method for detecting eye fatigue using EEG glasses according to claim 1, characterized in that, The EEG signals were acquired using flexible dry electrodes on the temples of the mirror, covering the occipital and temporal lobe regions. The eye movement and pupil signals are acquired by capturing eye images using a miniature infrared camera in the lens frame; The heart rate variability signal was acquired by collecting the pulse wave behind the ear using a PPG sensor on the temple of the glasses. The electroencephalogram (EEG) signal, the eye movement and pupillary signals, and the heart rate variability signal are sampled synchronously using the same clock source to ensure timestamp alignment.

3. The method for detecting eye fatigue using EEG glasses according to claim 1, characterized in that, The step of extracting feature parameters strongly correlated with eye fatigue from the effective signal, learning the contribution of each feature to fatigue through a feature weight allocation mechanism, and generating a weighted feature vector for focusing on eye fatigue includes the following steps: S32. Learn the contribution of each feature to eye fatigue through a feature weight allocation mechanism; S33. Multiply each feature by its corresponding weight and then concatenate them to generate a weighted feature vector of eye fatigue caused by focusing.

4. The method for detecting eye fatigue using EEG glasses according to claim 1, characterized in that, The characteristic parameters strongly correlated with eye fatigue include: The ratio of alpha wave power spectral density to beta wave power spectral density, and the ratio of theta wave power spectral density to beta wave power spectral density in EEG signals; Blink frequency, pupil diameter change rate, and fixation point drift amplitude of eye movement and pupil signals; The ratio of low-frequency to high-frequency power and the total power of the heart rate variability signal.

5. The method for detecting eye fatigue using EEG glasses according to claim 4, characterized in that, The process of using a pre-trained multimodal fusion model, combined with user-specific physiological baseline calibration, to analyze the weighted feature vector and output an eye fatigue level adapted to individual differences includes the following steps: S41. Input the generated weighted feature vector of eye fatigue caused by focusing on the screen into the input layer of the multimodal fusion model as the initial data for model analysis; S42. Individual physiological baseline calibration includes initial calibration and online learning. During initial calibration, multimodal signals under normal eye use are collected when the user uses the device for the first time, features are extracted and an initial model is trained to establish the user's individual physiological baseline. During the online learning phase, multimodal signals are continuously collected and features are extracted during the user's daily use. The model parameters are dynamically updated through an incremental learning algorithm to adapt the model to changes in the user's individual eye use habits. S43. The calibrated multimodal fusion model learns the correlation between multimodal features through the interaction layer, performs in-depth analysis on the input weighted feature vector, and mines the potential correlation between features to evaluate fatigue state; S44. The multimodal fusion model classifies the analysis results through the output layer and finally outputs an eye fatigue level that is adapted to individual differences.

6. The method for detecting eye fatigue using EEG glasses according to claim 5, characterized in that, It also includes the following steps: S5. Trigger intervention feedback based on the output fatigue level, where: Mild fatigue is indicated by vibration from the frame's vibration motor, prompting a rest. For moderate fatigue, voice prompts are played via bone conduction headphones. In cases of severe eye fatigue, an alert will be sent to the linked mobile app, and an eye fatigue trend report will be generated.

7. The method for detecting eye fatigue using EEG glasses according to claim 6, characterized in that, The eye fatigue trend report includes: Fatigue time distribution: Record the specific time periods and frequency of fatigue occurrences within each day / week; Fatigue intensity analysis, statistical analysis of the duration and changing trends of mild, moderate and severe fatigue; Correlation factor analysis, combined with multimodal signal characteristics, identifies eye-use behaviors strongly correlated with fatigue; Personalized suggestions generate targeted eye care guidance based on fatigue trends and related factors.

8. An eye fatigue detection system for EEG glasses, used to perform the eye fatigue detection method for EEG glasses according to any one of claims 1 to 7, characterized in that, include: Biosignal acquisition module: used to acquire raw biological signals directly related to eye fatigue, including electroencephalogram (EEG) signals, eye movement and pupil signals, and heart rate variability signals; The joint denoising module is used to perform joint denoising on the original biological signals based on the physiological correlation of the original biological signals, so as to obtain an effective signal that can reflect the state of eye fatigue. The feature extraction and weighting module is used to extract feature parameters that are strongly correlated with eye fatigue from the effective signal, learn the contribution of each feature to fatigue through a feature weight allocation mechanism, and generate a weighted feature vector for focusing on eye fatigue. The weight of each feature is dynamically adjusted according to the strength of the correlation between each feature and eye fatigue. The fatigue assessment module is used to analyze the weighted feature vector by using a pre-trained multimodal fusion model and calibrating the user's individual physiological baseline, and output an eye fatigue level that is adapted to individual differences. The method of jointly denoising raw biological signals based on their physiological correlation to obtain effective signals reflecting eye fatigue includes the following steps: S21. Detect physiological events such as rapid blinking, fixation point drift, and head movement through eye movement and pupil signals, and simultaneously mark the corresponding time windows in EEG signals and heart rate variability signals that are interfered with by any of the physiological events; S22. Based on the interference characteristics of each modality signal, denoising is performed using marked interference time windows: For EEG signals, independent component analysis is used to separate and remove EMG artifacts, while signal segments affected by EMG artifacts are corrected or removed; for eye movement and pupil signals, filtering is used to eliminate ambient light interference, and image registration algorithms are used to correct image shifts caused by head movements; for heart rate variability signals, bandpass filtering is used to remove motion artifacts, while smoothing out abnormal heartbeat intervals that are interfered with. S23. Compare the correlation of multimodal signals before and after denoising. If the deviation exceeds the preset threshold, repeat steps S21 to S22 until the requirements are met, and finally obtain an effective signal that can accurately reflect the state of eye fatigue. The step of extracting feature parameters strongly correlated with eye fatigue from the effective signal, learning the contribution of each feature to fatigue through a feature weight allocation mechanism, and generating a weighted feature vector for focusing on eye fatigue includes: S31. Extract feature parameters directly related to eye fatigue from the effective signals respectively. For EEG signals, extract frequency domain features reflecting the degree of visual cortex fatigue. For eye movement and pupil signals, extract temporal features reflecting the fatigue state of eye muscles and pupillary accommodation function. For heart rate variability signals, extract frequency domain features reflecting the functional disorder of the autonomic nervous system caused by fatigue.

Images