Intelligent Glass Electroencephalogram Signal Noise Reduction and Feature Extraction Method, Device and Equipment

Abstract

The present application discloses a method, device and equipment for noise reduction and feature extraction of electroencephalogram signals of smart glasses. The method includes: collecting multi-channel electroencephalogram signals of smart glasses to obtain an original data stream, and generating a standard data stream according to the original data stream; performing sub-band decomposition on the standard data stream to obtain band data, and obtaining a noise reduction signal corresponding to the band data; performing edge detection on the noise reduction signal to obtain a feature position, and obtaining a contour feature corresponding to the feature position; obtaining a feature template corresponding to the contour feature and a corresponding signal type, and constructing a signal sample according to the signal type; obtaining a frequency component corresponding to the signal sample, and performing reconstruction according to the frequency component to generate a pure waveform; obtaining a feature index corresponding to the pure waveform, generating a feature map according to the feature index, and generating a feature matrix according to the feature map; performing dimension mapping on the feature matrix to obtain a main component, and outputting a feature representation according to the main component, thereby completing the noise reduction and feature extraction of the multi-channel electroencephalogram signals.

Description

Technical Field

[0001] This application relates to the technical field of brain-computer interfaces, and particularly to a method, device, and equipment for noise reduction and feature extraction of electroencephalogram signals of smart glasses. Background Art

[0002] As a portable electroencephalogram acquisition device, the signal quality of smart glasses is vulnerable to multi-source noise interference and user behavior. The traditional signal processing methods mainly have the following limitations: First, the commonly used frequency-domain filtering and statistical noise reduction techniques are difficult to effectively process the non-stationary signals collected by smart glasses in dynamic scenarios, and it is easy to cause the loss of effective features; Second, the anomaly detection method based on a fixed threshold cannot adapt to individual differences and environmental changes, reducing the noise reduction effect; Third, the existing feature extraction methods mostly rely on pre-set feature templates and are difficult to capture the dynamic feature patterns under different cognitive tasks.

[0003] In practical applications, the electroencephalogram signals collected by smart glasses also have the following challenges: First, the signal quality differences between different channels are relatively large, and a unified quality assessment and processing standard needs to be established; Second, there is an obvious time-series dependence relationship in the signal features, and the time evolution characteristics of the features need to be considered; Third, the influence mode of environmental noise is complex and changeable, and an adaptive noise recognition and suppression method needs to be developed. In addition, the portability of smart glasses requires that the signal processing algorithm has high computational efficiency.

[0004] Therefore, there is an urgent need for a method to solve at least one of the above problems. Summary of the Invention

[0005] The embodiments of this application provide a method, device, and equipment for noise reduction and feature extraction of electroencephalogram signals of smart glasses, aiming to solve the following challenges in practical applications: First, the signal quality differences between different channels are relatively large, and a unified quality assessment and processing standard needs to be established; Second, there is an obvious time-series dependence relationship in the signal features, and the time evolution characteristics of the features need to be considered; Third, the influence mode of environmental noise is complex and changeable, and an adaptive noise recognition and suppression method needs to be developed. In addition, the portability of smart glasses requires that the signal processing algorithm has high computational efficiency.

[0006] In a first aspect, the embodiments of this application provide a method for noise reduction and feature extraction of electroencephalogram signals of smart glasses, including:

[0007] Collect multi-channel electroencephalogram signals of smart glasses to obtain an original data stream, generate an initial feature set according to the original data stream, and generate a standard data stream according to the initial feature set;

[0008] Perform sub-band decomposition on the standard data stream to obtain band data, obtain the energy characteristics corresponding to the band data, and obtain a noise reduction signal based on the energy characteristics; perform edge detection on the noise reduction signal to obtain a feature position, and obtain the contour characteristics corresponding to the feature position;

[0009] Obtain the feature template corresponding to the contour characteristics and the corresponding signal type, construct a signal sample according to the signal type; obtain the frequency components corresponding to the signal sample, and perform reconstruction according to the frequency components to generate a pure waveform;

[0010] Obtain the feature index corresponding to the pure waveform, generate a feature map according to the feature index, and generate a feature matrix according to the feature map;

[0011] Perform dimension mapping on the feature matrix to obtain the main components, output the feature representation according to the main components, and complete the noise reduction and feature extraction of the multi-channel EEG signals.

[0012] In a second aspect, the present application also provides an intelligent glasses EEG signal noise reduction and feature extraction device, including:

[0013] A signal acquisition module, configured to collect multi-channel EEG signals of the intelligent glasses, obtain an original data stream, generate an initial feature set according to the original data stream, and generate a standard data stream according to the initial feature set;

[0014] A feature acquisition module, configured to perform sub-band decomposition on the standard data stream to obtain band data, obtain the energy characteristics corresponding to the band data, and obtain a noise reduction signal based on the energy characteristics; perform edge detection on the noise reduction signal to obtain a feature position, and obtain the contour characteristics corresponding to the feature position;

[0015] A type acquisition module, configured to obtain the feature template corresponding to the contour characteristics and the corresponding signal type, construct a signal sample according to the signal type; obtain the frequency components corresponding to the signal sample, and perform reconstruction according to the frequency components to generate a pure waveform;

[0016] An index acquisition module, configured to obtain the feature index corresponding to the pure waveform, generate a feature map according to the feature index, and generate a feature matrix according to the feature map;

[0017] An extraction completion module, configured to perform dimension mapping on the feature matrix to obtain the main components, output the feature representation according to the main components, and complete the noise reduction and feature extraction of the multi-channel EEG signals.

[0018] In a third aspect, the present application further provides a computer device, including a processor and a memory. The memory is used to store a computer program, and when the computer program is executed by the processor, it implements the method for noise reduction and feature extraction of electroencephalogram signals of the smart glasses as described in the first aspect.

[0019] In a fourth aspect, the present application further provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, it implements the method for noise reduction and feature extraction of electroencephalogram signals of the smart glasses as described in the first aspect.

[0020] Compared with the prior art, the present application has at least the following beneficial effects:

[0021] 1. Through a hierarchical feature extraction and processing strategy, the method realizes the adaptive processing of the signals collected by the smart glasses. First, an initial feature set is constructed based on multi-channel signal features, including time-domain statistical features, band energy features, and spatial correlation features; then, through anomaly detection and data correction based on the probability density function, a dynamic threshold mechanism is established to achieve adaptability to individual differences; finally, by using multi-level sub-band decomposition and spectrum analysis, stable signal features are extracted, effectively solving the problem of signal instability in portable electroencephalogram acquisition.

[0022] 2. By adopting a method based on template matching and multi-dimensional feature analysis, the accurate recognition of different cognitive states is realized. The method first performs edge detection through a multi-scale derivative operator to accurately locate the key feature positions; then, by combining hierarchical clustering and spatial consistency constraints, a cross-channel feature segment alignment mechanism is established; further, through an improved wavelet packet reconstruction algorithm, the selective retention of features in different cognitive states is realized; finally, based on spectrum analysis, the feature expression is optimized, significantly improving the recognition accuracy of the user's cognitive state in a complex environment.

[0023] 3. A hierarchical feature mapping and dimensionality reduction framework is established to realize the efficient expression of features. The method constructs a multi-scale decomposition system through continuous wavelet transform technology to achieve a fine description of signal features; a low-dimensional representation of features is established based on the t-SNE algorithm and a hierarchical mapping structure; through a multi-level quality assessment mechanism, including reconstruction error analysis, temporal feature preservation, and spatial feature verification, the retention of key information in the dimensionality reduction process is ensured; finally, a feature representation method that retains both cognitive state discrimination information and has high computational efficiency is formed.

[0024] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. Description of the Drawings

[0025] Figure 1Schematic flowchart of the method for denoising and feature extraction of electroencephalogram signals of the smart glasses shown in the embodiments of the present application;

[0026] Figure 2 Schematic structural diagram of the device for denoising and feature extraction of electroencephalogram signals of the smart glasses shown in the embodiments of the present application;

[0027] Figure 3 Schematic structural diagram of the computer device shown in the embodiments of the present application. Detailed implementation manners

[0028] In the following description, specific details such as specific system architectures and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of the present application. However, those skilled in the art should clearly understand that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from interfering with the description of the present application.

[0029] It should be understood that when used in the specification of the present application and the appended claims, the term "comprising" indicates the presence of the described features, wholes, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0030] It should also be understood that the term "and / or" used in the specification of the present application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0031] As used in the specification of the present application and the appended claims, the term "if" can be interpreted as "when", "once", "in response to determining", or "in response to detecting" according to the context. Similarly, the phrase "if determined" or "if detecting [the described condition or event]" can be interpreted as meaning "once determined", "in response to determining", "once detecting [the described condition or event]", or "in response to detecting [the described condition or event]" according to the context.

[0032] In addition, in the description of the specification of the present application and the appended claims, the terms "first", "second", "third", etc. are only used for distinguishing descriptions and cannot be understood as indicating or implying relative importance.

[0033] References to "an embodiment" or "some embodiments" etc. described in the specification of this application mean that a specific feature, structure, or characteristic described in connection with that embodiment is included in one or more embodiments of this application. Thus, statements such as "in an embodiment", "in some embodiments", "in other some embodiments", "in still other embodiments", etc. that appear in different places in this specification do not necessarily all refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "comprising", "including", "having" and their variants all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0034] The technical solutions of the embodiments of this application will be introduced below.

[0035] As a portable electroencephalogram (EEG) acquisition device, the signal quality of smart glasses is vulnerable to multi-source noise interference and user behavior. The traditional signal processing methods mainly have the following limitations: First, the commonly used frequency-domain filtering and statistical noise reduction techniques are difficult to effectively process the non-stationary signals collected by smart glasses in dynamic scenarios, and it is easy to cause the loss of effective features; Second, the anomaly detection method based on a fixed threshold cannot adapt to individual differences and environmental changes, reducing the noise reduction effect; Third, the existing feature extraction methods mostly rely on preset feature templates and are difficult to capture the dynamic feature patterns under different cognitive tasks.

[0036] In practical applications, the EEG signals collected by smart glasses also have the following challenges: First, the signal quality varies greatly among different channels, and it is necessary to establish a unified quality assessment and processing standard; Second, there is an obvious temporal dependence relationship in signal features, and it is necessary to consider the time evolution characteristics of features; Third, the influence mode of environmental noise is complex and changeable, and it is necessary to develop an adaptive noise recognition and suppression method. In addition, the portability of smart glasses requires that the signal processing algorithm has high computational efficiency. Therefore, it is necessary to develop a signal processing scheme that adapts to the application characteristics of smart glasses to achieve the full-process optimization from signal acquisition, noise reduction to feature extraction.

[0037] To solve the above problems, please refer to Figure 1 , Figure 1 is a schematic flow chart of a method for noise reduction and feature extraction of EEG signals of smart glasses provided by an embodiment of this application. The method for noise reduction and feature extraction of EEG signals of smart glasses in the embodiments of this application can be applied to computer devices, and the computer devices include but are not limited to devices such as smart phones, laptop computers, tablet computers, desktop computers, physical servers, and cloud servers. As Figure 1 shown, the method for noise reduction and feature extraction of EEG signals of smart glasses in this embodiment includes steps S101 to S105, which are described in detail as follows:

[0038] Step S101: Collect multi-channel electroencephalogram (EEG) signals of the smart glasses to obtain an original data stream, generate an initial feature set based on the original data stream, and generate a standard data stream based on the initial feature set.

[0039] Specifically, perform a collection operation on the multi-channel EEG signals of the smart glasses. The smart glasses are built with an eight-channel dry electrode array. Among them, the Fp1 and Fp2 electrodes in the frontal lobe region are used to collect EEG activities related to prefrontal executive control. The T3 and T4 electrodes in the temporal lobe region collect EEG signals related to auditory cognition. The O1 and O2 electrodes in the occipital lobe region collect EEG data related to visual processing. The P3 and P4 electrodes in the parietal lobe region collect EEG activities related to spatial attention. The sampling frequency is set to 1000 HZ to capture the rapid changes of EEG signals, and the sampling accuracy is 24 bits to ensure the accurate collection of weak signals. For example, in an office scenario, when the user is concentrating on reading a document, the P3 and P4 electrodes record the state of spatial attention allocation, the O1 and O2 electrodes record the visual processing process, and the Fp1 and Fp2 electrodes record the cognitive load level. The method monitors the contact impedance value between the electrode and the skin in real time. When it is detected that the channel impedance exceeds five thousand ohms, a quality mark is added to this channel. For example, in a remote meeting scenario, the T3 and T4 electrodes mainly collect the EEG changes during the voice interaction process, the Fp1 and Fp2 electrodes record the attention regulation state, and the O1 and O2 electrodes monitor the change of visual concentration. After the EEG signals collected by each channel are synchronously aligned in real time, together with the channel quality mark information, they form an original data stream, which contains the multi-channel EEG time series features of the user in different cognitive states.

[0040] In some embodiments, the generating an initial feature set based on the original data stream includes: performing time window segmentation on the original data stream to obtain time series segments; and generating the initial feature set based on the time series segments.

[0041] Since the original data stream contains long-term signals collected continuously, appropriate segmentation processing is required to extract local cognitive state features. At the same time, there are differences in the signal amplitude ranges of different channels, and unified amplitude mapping is needed for subsequent analysis. Therefore, the method performs time window slicing and normalization processing on the original data stream. Based on the time characteristics of human cognitive processing, the window length is set to 2 seconds to include a complete cognitive activity process, and each window contains 2000 sampling points; the window sliding step is 1 second, and adjacent windows overlap by 1000 sampling points to maintain the continuity of signal analysis. For example, in a driving scenario, when the driver is looking straight ahead at the road, the O1 and O2 electrodes record continuous visual monitoring activities, the P3 and P4 electrodes collect spatial alertness states, and the Fp1 and Fp2 electrodes monitor the level of cognitive fatigue; when the driver checks the rearview mirror, the O1, O2, P3, and P4 electrodes jointly record the process of visual attention conversion, and the T3 and T4 electrodes collect spatial perception states. The method labels specific driving behavior types for each time window, including forward gaze, rearview mirror check, dashboard observation, etc. For example, in a walking scenario, window slicing can separate and record cognitive processes such as user motor imagery, spatial orientation, and environmental perception. The P3 and P4 electrodes record EEG activities related to spatial navigation, the O1 and O2 electrodes collect the process of environmental visual search, the T3 and T4 electrodes monitor the sound localization state, and the Fp1 and Fp2 electrodes record the level of executive control. Amplitude normalization processing is performed on the eight-channel signals within all sliced windows, and the signals are uniformly mapped to the interval from -1 to +1 to generate a group of standardized time series segments.

[0042] The signals in the standardized time series segment group already have a unified time scale and amplitude range, which enables the method to process the data of all channels using a consistent feature extraction method. Based on this, the method further constructs an initial feature set containing multi-dimensional features. First, basic time-domain statistical features are extracted from the signals in each window. The signal average value is calculated to represent the baseline level, the variance reflects the signal fluctuation intensity, the waveform length represents the signal complexity, and the zero-crossing rate reflects the signal oscillation frequency. For example, in the reading scenario, the energy ratio of the 4-7HZ theta band is extracted from the Fp1 and Fp2 channels to represent the working memory load, the energy ratio of the 8-13HZ alpha band represents the attention suppression level, and the energy ratio of the 14-30HZ beta band represents the cognitive activity intensity; by calculating these band energy features, the changes in the cognitive state of the user during reading can be quantified. For example, in the video viewing scenario, the alpha band energy change of the O1 and O2 channels is mainly extracted to represent the visual attention intensity, the theta / alpha energy ratio of the P3 and P4 channels is analyzed to represent the attention allocation efficiency, and the beta band energy of the T3 and T4 channels is calculated to reflect the auditory processing intensity. The cross-correlation coefficient matrix is calculated for the signals of each channel, and the phase synchronization features between the left and right brain symmetric channels, the information transfer features between the front and back brain regions, and the local connection features between adjacent electrodes are recorded. For example, in the sports scenario, the alpha band power asymmetry index of the Fp1 and Fp2 channels is extracted to represent the attention bias, the gamma band energy change of the T3 and T4 channels is analyzed to reflect the spatial perception intensity, and the beta / theta energy ratio of the P3 and P4 channels is calculated to evaluate the motor control level. The method normalizes all the extracted feature parameters and assigns unique identifiers, organizes the time-domain features, frequency-domain features, and spatial features into a structured feature array, and finally forms an initial feature set representing the multi-dimensional cognitive state of the user.

[0043] In some embodiments, generating the standard data stream according to the initial feature set includes: performing amplitude statistics on the initial feature set to obtain an amplitude distribution; obtaining the position of the outlier corresponding to the initial feature set according to the amplitude distribution; and correcting the initial feature set according to the position of the outlier to generate the standard data stream.

[0044] Based on the initial feature set formed in step S101, the method first performs amplitude statistical analysis on the multi-channel EEG features therein. For the time-domain eigenvalue of each channel in the initial feature set, including the zero-crossing rate, waveform length, root mean square value, etc., calculate its statistical distribution parameters. For example, in the classroom concentration detection scenario, the method extracts statistical quantities such as the maximum value, minimum value, mean value, standard deviation, skewness, and kurtosis of the features of the Fp1 and Fp2 channels, where the mean value reflects the attention baseline level, the standard deviation characterizes the fluctuation range, the skewness describes the asymmetry of the attention distribution, and the kurtosis characterizes the steepness of the attention fluctuation. For the visual evoked potential features of the O1 and O2 channels, respectively, statistically analyze the amplitude distributions of key components such as the peak value of the P100 wave, the trough value of the N170 wave, and the peak value of the P300 wave. At the same time, analyze the probability distributions of the eigenvalue of early components such as N100 and P200 related to auditory cognition in the T3 and T4 channels. For the frequency-domain features, the method performs hierarchical statistics on the energy features of frequency bands such as theta, alpha, and beta, and establishes a distribution model of the frequency band energy ratio. For example, in the intelligent care scenario, the method establishes a probability density function for the attenuation feature of the alpha frequency band energy in the frontal lobe of the elderly to depict the statistical law of cognitive ability. For the spatial features, analyze the distribution characteristics of the correlation coefficient between channels and establish a probability model of the spatial connection strength. The method organizes all statistical quantities into a multi-dimensional matrix, where the row vector of the matrix represents different electrode channels, and the column vector contains various statistical parameters. At the same time, based on the time window, construct the probability density function of each channel, and finally obtain a complete amplitude distribution feature space including time-domain, frequency-domain, and spatial-domain features.

[0045] For the obtained amplitude distribution feature space, the method adopts a multi-level detection strategy to locate abnormal points. First, an adaptive detection threshold is calculated based on the probability density function of each channel. The threshold parameters are jointly determined by the interquartile range and variance to achieve sensitive detection of abnormalities at different scales. For example, in the driving fatigue monitoring scenario, the method detects the abnormal shift of the alpha-band energy distribution in the Fp1 and Fp2 channels, calculates the Mahalanobis distance between the current feature and the standard distribution, and marks the points deviating more than three standard deviations as fatigue risk points. For the visual evoked features of the O1 and O2 channels, a dynamic threshold based on the amplitude probability distribution is set to identify abnormal fluctuations caused by distracted attention. Second, the method establishes a joint probability model between channels to analyze the spatial correlation of feature distributions. For example, in the emotion recognition scenario, by calculating the correlation coefficient matrix of the symmetric channels of the left and right brains, the abnormal distribution asymmetry caused by emotional fluctuations is identified. For time series features, a sliding window is used to calculate the local probability density function, and the features significantly deviating from the historical distribution are marked as time series abnormalities. The method also constructs an anomaly detection model based on higher-order statistics to analyze higher-order moments such as skewness and kurtosis of feature distributions and capture subtle distribution abnormalities. For example, in the motor imagery scenario, by calculating the higher-order statistical features of the mu-rhythm energy distribution in the C3 and C4 channels, the abnormal points during the intention conversion process are accurately located. Finally, all detection results are integrated into an abnormal point location table, which records the timestamp, channel number, and degree of abnormality of each abnormal point.

[0046] Signal correction is performed based on the location information and degree of abnormality recorded in the abnormal point location table. The method determines the correction strategy according to the spatial distribution characteristics of the abnormal points. For the points marked as single-channel abnormalities, the average value of adjacent time windows is used for correction; for the points with multiple channels abnormal simultaneously, data reconstruction is performed based on the spatial correlation information in the abnormal point location table. For example, in the multi-person collaboration scenario, when it is detected that the Fp1 and Fp2 channels of a certain user simultaneously show the abnormalities marked in the location table, the method combines the abnormality degree scores of the two channels for joint correction. The correction amplitude is determined by the abnormality metric value recorded in the location table. The larger the abnormality metric value, the stronger the corresponding correction strength. For example, in the gait monitoring scenario, for the periodic abnormal points marked in the location table, the method analyzes their occurrence patterns and selects an appropriate correction algorithm according to the time series pattern of the abnormal features. For each corrected data point, the method retains the feature mapping relationship in the abnormal point location table to ensure the continuity and stability of the signal. During the signal correction process, the method also refers to the channel correlation information recorded in the location table and adopts a consistent correction strategy for the related channels to maintain the phase relationship between channels. After the correction process based on the abnormal point location table, a standard data stream is finally generated.

[0047] Step S102: Perform sub-band decomposition on the standard data stream to obtain band data, acquire the energy features corresponding to the band data, and obtain a noise reduction signal based on the energy features; perform edge detection on the noise reduction signal to obtain feature positions, and acquire the contour features corresponding to the feature positions.

[0048] Specifically, perform sub-band decomposition processing on the standard data stream obtained in step S101. The method uses a 6th-order Butterworth filter to construct a multi-stage filter bank, and decomposes the electroencephalogram signal into a delta band (0.5 to 4 Hz), a theta band (4 to 8 Hz), an alpha band (8 to 13 Hz), a beta band (13 to 30 Hz), and a gamma band (30 to 50 Hz). To ensure the accuracy of filtering, a transition bandwidth of 0.5 Hz is set for each band, and the stopband attenuation is set to 60 dB. For example, in the concentration detection scenario, the method performs sub-band decomposition on the signals of the Fp1 and Fp2 channels in the standard data stream, and extracts the theta band activity related to prefrontal cognitive load; at the same time, calculate the correlation coefficient before and after the signal passes through the filter to ensure signal integrity. When decomposing the signals of the O1 and O2 channels, focus on the alpha band, extract the rhythmic activity during visual processing, and verify the decomposition effect by calculating the proportion of band energy. In the decomposition of the signals of the T3 and T4 channels, focus on the gamma band to obtain the high-frequency features of auditory information processing. The sub-band decomposition uses zero-phase filtering technology, and eliminates phase distortion through forward and reverse filtering to maintain the timing characteristics of the signal. For example, in the visual evoked scenario, the method uses a sliding window decomposition for the signals of the P3 and P4 channels, the window length is set to 1 second, and the sliding step is 0.5 second, and the instantaneous features of each band are extracted respectively. In the motor imagery scenario, an adaptive filter bank is introduced for the signals of the C3 and C4 channels, and the filter parameters are dynamically adjusted according to the spectral characteristics of the signal. The method calculates the signal-to-noise ratio for each decomposed band signal, and marks the bands with a signal-to-noise ratio lower than 10 dB as objects to be processed. Finally, a multi-channel, multi-band band data matrix is obtained, and each element of the matrix contains the time-domain sequence of the corresponding channel and the corresponding band.

[0049] In some embodiments, the obtaining the noise reduction signal according to the energy features includes: performing threshold selection on the energy features according to a preset threshold to obtain the effective signal corresponding to the energy features; obtaining the reconstructed waveform corresponding to the effective signal; and obtaining the noise reduction signal according to the reconstructed waveform.

[0050] For the obtained frequency band data matrix, the method conducts multi-scale energy analysis. First, the short-time Fourier transform is used to calculate the instantaneous energy of each frequency band. The window length is set to 0.5 seconds, the sliding step is 0.25 seconds, and the Hanning window is used as the windowing function to reduce spectral leakage. For example, in the motor imagery scene, the method analyzes the energy change characteristics of the C3 and C4 channels in the 8 to 13 Hz frequency band, and evaluates the intensity of motor imagery by calculating the energy difference between the left and right motor cortices. When calculating the theta band energy of the Fz channel, the multi-resolution analysis method is used to extract the energy change pattern in the attention regulation process at different time scales. The method establishes an energy envelope curve for each frequency band, extracts the energy modulation characteristics through the Hilbert transform, and calculates the first-order derivative of the envelope to reflect the energy change rate. For example, in the working memory task, the theta band energy envelope of the Fp1 and Fp2 channels is decomposed by wavelet to extract the energy fluctuation characteristics at different scales; the phase synchronization of the alpha band energy envelope of the P3 and P4 channels is calculated to evaluate the spatiotemporal coordination of information processing. In the emotion recognition scenario, the asymmetry of the alpha band energy of the F3 and F4 channels is analyzed to construct an energy feature vector reflecting the emotional activity of the prefrontal lobe. The method also establishes an energy coupling model between frequency bands, and analyzes the energy interaction characteristics across frequency bands by calculating the cross-correlation function of the energy of different frequency bands. For example, in the attention switching task, the theta-gamma coupling intensity and the alpha-beta energy correlation are analyzed simultaneously to construct multi-band energy synergy characteristics. The method finally generates a multi-dimensional energy feature matrix containing the energy distribution, energy ratio, energy modulation characteristics, and frequency band coupling characteristics of each frequency band.

[0051] Adaptive threshold selection based on energy feature matrix. The improved OTSU algorithm is used to calculate the energy threshold of each frequency band, and the signal segments with energy significantly higher than the background are marked as active. The improvements include introducing inter-channel correlation weights and time continuity constraints to improve the stability of threshold selection. For example, in the visual attention task, the method sets a dynamic threshold for the alpha band energy of the O1 and O2 channels. When the energy exceeds the threshold, the corresponding time window is marked as the active period of visual processing; at the same time, the energy distribution of the P3 and P4 channels is considered, and a threshold adjustment mechanism based on spatial correlation is established. For the case of simultaneous activity of multiple channels, the method establishes a joint threshold criterion based on inter-channel energy correlation, and optimizes the threshold parameters by calculating the energy covariance matrix between channels. For example, in the motion control scenario, the beta band energy of the C3, C4 and FCz channels is monitored simultaneously, and the energy principal components of the three channels are calculated, and the joint threshold is set according to the principal component score. For each signal segment that passes the threshold test, its start and end time, channel, band type, energy intensity and spatial distribution characteristics are recorded to form a valid signal table. At the same time, the corresponding original band information in the band data matrix is retained for subsequent signal reconstruction.

[0052] Signal reconstruction is performed using the spatio-temporal marking information of the valid signal table and the reserved band data. First, the method performs amplitude normalization on the band data marked in the valid signal table, and uses a piecewise linear mapping method to eliminate the energy differences between different bands. The normalization parameters are optimized according to the physiological significance of each band. For example, the weight of the cognitive-related band is enhanced, and the contribution of the electromyogram interference band is suppressed. Then, the signals of each band are reconstructed into a complete waveform by weighted superposition, and the weight coefficients are determined by the energy contribution degrees of each band. For example, in the cognitive load assessment scenario, when the method reconstructs the signal of the Fz channel, higher weight coefficients are assigned to the theta band and the beta band to prominently reflect the electroencephalogram characteristics related to working memory. In spatial attention analysis, the spatial characteristics of the alpha band of the P3 and P4 channels are mainly retained, and the phase correction is used to ensure the spatial consistency of the reconstructed signal. The method uses an improved wavelet packet reconstruction algorithm to selectively reconstruct each band according to the guidance of the valid signal table. For example, in visual evoked potential analysis, when reconstructing the signals of the O1 and O2 channels, the time-domain characteristics of the alpha band and the gamma band are mainly retained, and the reconstruction quality is evaluated by calculating the cross-correlation function before and after reconstruction, and finally a noise-reduced signal is generated.

[0053] In some embodiments, obtaining the contour feature corresponding to the feature position includes: obtaining the signal boundary corresponding to the noise-reduced signal according to the feature position; dividing the noise-reduced signal according to the signal boundary to obtain a feature segment; and performing feature extraction on the feature segment to obtain the contour feature.

[0054] Edge detection processing is performed on the noise-reduced signal. The method uses a multi-scale derivative operator to calculate the first-order derivative and the second-order derivative of the signal, and establishes an edge detection factor library. When calculating the derivative, Gaussian smoothing kernels of different scales are used, and the scale parameters are set to 2 milliseconds, 5 milliseconds, and 10 milliseconds respectively to capture the signal change characteristics of different time spans. For example, in visual evoked potential analysis, the local maximum gradient positions of the noise-reduced signals of the O1 and O2 channels are calculated to identify the fast-changing edges of components such as P100 and N170. At the same time, an adaptive threshold mechanism is introduced, and the threshold parameter is determined by the local standard deviation and mean of the signal. For example, in the motor imagery task, the method extracts the position information of the peaks and valleys of the noise-reduced signals of the C3 and C4 channels and marks the edge points at the start and end of the motor imagery. The method calculates a significance score for each detected edge point, and the score is jointly determined by the gradient amplitude and the local contrast. In the attention switching task, the edge position of the energy change in the alpha band is mainly detected, and a feature position sequence including the edge position, the edge intensity, and the edge type is formed through weighted fusion of the multi-scale edge detection results.

[0055] Mark the signal boundaries based on the obtained sequence of feature positions. The method performs clustering analysis on the edge points in the sequence of feature positions, using a hierarchical clustering method based on time distance to group edge points with close time into the same boundary region. The clustering threshold is dynamically adjusted according to the sampling rate of the signal. For a signal with a sampling rate of 1000 Hz, edge points with a time interval less than 20 milliseconds are grouped into the same class. For example, in an attention-switching task, the method analyzes the boundary features related to attention changes in the Fp1 and Fp2 channels, and determines the start and end positions of the boundary by calculating the time distribution density of the edge points. For cross-channel boundary marking, a spatial consistency constraint is introduced, requiring that the deviation of the boundary positions between adjacent channels does not exceed a set threshold. For example, in an emotion recognition scenario, the method performs boundary marking on the emotion-related waveforms in the F3 and F4 channels, and establishes the corresponding relationship between the left and right brain region boundaries. The method also calculates the stability index of each boundary, which is determined by the density and intensity distribution of the edge points within the boundary region. Finally, a signal boundary table is generated, recording the time position, duration, channel belonging, and boundary intensity information of each boundary.

[0056] Divide the feature segments according to the signal boundary table and the noise-reduced signal. The method first validates the effectiveness of the signal boundaries of each channel and eliminates the false boundaries caused by instantaneous noise. For example, in a working memory task, the method divides the feature segments of the encoding, maintenance, and retrieval stages according to the boundary information of the theta rhythm in the Fz channel. For cross-channel collaborative activities, a segment alignment mechanism based on boundary synchronization is established. For example, in a visual search task, the method synchronously divides the signal segments of the P3, P4, O1, and O2 channels to ensure the temporal correspondence of spatial attention and visual processing features. In an emotion induction scenario, the method grades the segments according to the boundary intensity, and divides the key analysis segments at boundaries with an intensity above 75%. The method establishes an index for the divided feature segments to form a feature segment set containing the start and end times of the segments, segment types, and channel distributions.

[0057] Feature extraction is performed based on the time attributes and channel distribution information recorded in the feature segment set. The method adopts a differentiated extraction strategy for different types of feature segments: for the segments marked as the encoding stage, the energy features in the theta frequency band are mainly extracted; for the segments marked as the retention stage, the suppression features in the alpha frequency band are mainly analyzed; for the segments marked as the extraction stage, the activity features in the beta frequency band are concerned. In the time-domain feature layer, statistical quantities such as kurtosis, skewness, and zero-crossing rate of the signal are calculated according to the duration information of the segment; in the frequency-domain feature layer, spectral features such as power spectral density and frequency band energy ratio are extracted based on the channel attributes of the segment; in the morphological feature layer, morphological parameters such as waveform symmetry, smoothness, and complexity are analyzed based on the boundary features of the segment. For example, in the remote meeting scenario, the method analyzes the theta / beta energy ratio of the Fp1 and Fp2 channel segments marked as the focused state in the feature segment set to extract the feature index reflecting cognitive engagement. In the document reading task, for the visual processing segments of the O1 and O2 channels in the feature segment set, the P100-N170-P300 waveform sequence features are extracted based on their time span. The method also establishes a combined extraction mechanism for multi-channel features according to the spatial distribution information recorded in the segment set. For example, in the attention switching task, the alpha band energy changes of the frontal and parietal lobe segments are synchronously analyzed to extract the collaborative features of the spatial attention network. Finally, a contour feature set including time-domain, frequency-domain, and morphological features is generated.

[0058] Step S103: Obtain the feature template corresponding to the contour feature and the corresponding signal type, and construct a signal sample according to the signal type; obtain the frequency components corresponding to the signal sample, and perform reconstruction according to the frequency components to generate a pure waveform.

[0059] Specifically, pattern matching processing is performed on the contour features obtained in step S102. The method first constructs a feature template library, which contains standard contour features of typical cognitive activities such as visual reading, conference conversation, and daily office work. The visual evoked potential template records the standard latency and amplitude thresholds of components such as P100, N170, and P300. The latency range of P100 is set to 80 to 120 milliseconds, and the amplitude threshold is 3 to 8 microvolts. The latency range of N170 is 150 to 200 milliseconds, and the amplitude threshold is 5 to 10 microvolts. Different matching strategies are set for different cognitive tasks. For example, video viewing uses a matching criterion based on latency and amplitude, and the matching score is determined by the weighted sum of time deviation and amplitude deviation. The attention task uses a matching method based on band energy to calculate the similarity of the energy contour in the 8 to 13 Hz frequency band. For example, in a remote conference, the method extracts the P300 component in the contour features of the O1 and O2 channels, calculates the latency deviation score, amplitude deviation score, and waveform width deviation score respectively, and uses a weight distribution of 0.4, 0.4, and 0.2 to calculate the comprehensive matching score. The matching threshold is dynamically adjusted according to the task difficulty. The simple browsing task requires a matching degree of more than 85%, while the complex document reading task is reduced to 70%. In the daily office scenario, the attention mode features are calculated for the contour features of the Fp1 and Fp2 channels, including timing parameters such as the concentrated attention time, the degree of distracted attention, and the task switching frequency. The method uses the dynamic time warping algorithm to align the features to be matched and the template features, and the maximum allowable time deformation range is 20% of the original length. Finally, a feature template set containing the optimal matching template identifier, matching metric score, time alignment parameter, and deformation coefficient is generated for each contour feature.

[0060] Signal type marking is performed using the matching information in the feature template set. The method analyzes the matching metric scores of each contour feature with its optimal matching template, and when the score exceeds a preset threshold, the cognitive type label corresponding to the template is assigned to the feature. To improve the reliability of the marking, the method adopts a three-layer evaluation mechanism for the matching metric scores: the first layer evaluates the time-domain matching degree, including waveform correlation coefficient and root mean square error; the second layer evaluates the frequency-domain matching degree, including power spectrum similarity and phase consistency; the third layer evaluates the morphological matching degree, including peak-valley distribution and waveform complexity. For example, in the evaluation of meeting concentration, the method determines that features with high matching degrees (score > 0.8) are highly concentrated, features with medium matching degrees (0.6 - 0.8) are generally concerned, and features with low matching degrees (< 0.6) are distracted based on the matching scores of the features in channels Fp1 and Fp2 with the attention template. At the same time, considering the combined scores of the three matching dimensions, a high-reliability marking is confirmed only when the time-domain matching degree > 0.75, the frequency-domain matching degree > 0.7, and the morphological matching degree > 0.8. Based on the time alignment parameters in the feature template set, the method establishes cross-channel synchronization constraints. For example, in a multimedia interaction scenario, the matching time difference between the features in the visual channel and the auditory channel and their respective templates is analyzed, and it is required that the alignment time difference between the associated channels does not exceed 50 milliseconds. The method also evaluates the reliability of the type marking according to the deformation coefficient. For every 0.1 increase in the deformation coefficient, the reliability of the marking decreases by 10%. In continuous office tasks, the method establishes a transition constraint for the type markings in consecutive time windows to limit the mutation probability of the marking types. When the types in adjacent windows change, sufficient matching evidence is required to support it. Finally, a signal type table containing feature types, three-layer matching reliabilities, and temporal relationships is generated.

[0061] Construct signal samples based on the matching features of the feature template set and the marking information of the signal type table. The method first screens high-quality samples according to the matching metric scores, and selects the features with the top 30% comprehensive matching scores as the reference samples. For each reference sample, a weighted feature expression is constructed by combining its type marking in the signal type table and three-layer reliability information. The sample weight is jointly determined by the matching metric score and the reliability, and the calculation formula is: weight = matching score × (0.4 × time-domain reliability + 0.3 × frequency-domain reliability + 0.3 × morphological reliability). For example, in the document reading task, the method selects the O1 and O2 channel features with a high matching degree (>0.85) with the visual attention template, and organizes them into a continuous attention sample sequence according to the timing information in the signal type table. In the remote meeting scenario, by analyzing the matching stability and type conversion rules of the Fp1 and Fp2 channel features with the concentration template, a complete attention sample set is established. The sample set contains feature samples of different states such as concentration, distraction, and conversation, and each sample inherits the template matching information, type marking attribute, and reliability index of its corresponding feature. For the cognitive indicators that need to be accurately quantified, the method establishes a sample scoring mechanism by combining the deformation parameters of the feature template set and the timing relationship of the type markings. The scoring formula is: sample score = basic score × (1 - deformation coefficient) × timing consistency, where the basic score is determined by the matching metric, and the timing consistency reflects the logical rationality of the sample sequence. The method implements quality control on the sample set, regularly calculates the representativeness and diversity indicators of the samples, and dynamically updates the sample library according to these indicators. Finally, a standardized signal sample set is formed.

[0062] In some embodiments, the reconstructing according to the frequency components to generate a pure waveform includes: performing band-pass screening on the frequency components; obtaining the frequency components corresponding to the frequency components according to the screening result corresponding to the band-pass screening; and reconstructing a signal according to the frequency components to generate the pure waveform.

[0063] Perform spectral analysis on the acquired signal samples. The method uses multi-scale derivative operators to calculate the first and second derivatives of the signal, and establishes an edge detection factor library. When calculating the derivatives, Gaussian smoothing kernels of different scales are used, and the scale parameters are set to 2 milliseconds, 5 milliseconds, and 10 milliseconds respectively to capture the signal change characteristics of different time spans. The first derivative is used to detect the fast-changing regions of the signal, and the second derivative is used to locate the positions of the peaks and valleys. The combination of the two can effectively identify the key change points of the signal. For example, in the analysis of visual evoked potentials, calculate the local maximum gradient positions of the signals in the O1 and O2 channels to identify the fast-changing edges of components such as P100 and N170. The method sets an adaptive gradient threshold, and the threshold size is jointly determined by the mean and variance within a local 50-millisecond window of the signal. At the same time, a morphological filter is introduced to optimize the edge detection results and eliminate the false edges caused by small fluctuations. In the analysis process, the short-time Fourier transform is used to calculate the spectral characteristics of the signal. The window length is 0.5 seconds, the sliding step is 0.25 seconds, and the Hann window is selected as the window function to reduce spectral leakage. The method calculates the power spectral density estimate for each analysis window and extracts the energy distribution characteristics of different frequency bands. Finally, a frequency component matrix is generated, and each element in the matrix contains information such as frequency value, energy value, phase value, and spectral peak position.

[0064] Design a band-pass screening strategy based on the energy distribution and spectral peak information in the frequency component matrix. The method determines the key frequency band range according to the peak distribution of the power spectral density, and sets a band-pass filter in the frequency band where the energy is concentrated. For each detected main spectral peak, design a band-pass filter with the center frequency corresponding to the spectral peak position, and the filter bandwidth is determined according to the energy distribution range of the spectral peak. For example, in the working memory task, when the frequency component matrix shows a significant energy peak in the theta frequency band within the range of 4 - 8 Hz, the method adjusts the band-pass range accordingly to cover the energy concentration area. The method analyzes the energy transition characteristics of adjacent frequency bands in the frequency component matrix, and designs the transition band parameters based on this. For the frequency band junctions with steep energy changes, a narrower transition band is used. For example, in the concentration training scenario, based on the energy distribution characteristics of the alpha frequency band, dynamically adjust the passband ripple and stopband attenuation parameters of the filter. In the sleep quality monitoring scenario, analyze the phase characteristics of each frequency band in the frequency component matrix, and use zero-phase filtering design for the frequency bands sensitive to phase. The method evaluates the filter performance by calculating indicators such as signal-to-noise ratio, frequency band energy ratio, and out-of-band suppression ratio, and generates a screening result table containing filter parameters and performance indicators.

[0065] According to the filter parameters in the screening result table, selective recombination of the spectral components in the frequency component matrix is performed. The method first sets the screening thresholds for each frequency band based on the filter performance indicators, with a focus on retaining the frequency components that exhibit high signal-to-noise ratio and energy stability in the corresponding frequency bands. For example, in the classroom learning scenario, for the theta frequency band of the Fp1 and Fp2 channels, the method extracts the frequency components within the range of 4 - 8 Hz according to the frequency band parameters in the screening result table and assigns recombination weights according to the performance indicators in the table. In the visual search task, the method screens the spectral components of 8 - 13 Hz according to the filter characteristics of the alpha frequency band of the O1 and O2 channels, and ensures that the selected frequency components have sufficient signal-to-noise ratio through performance indicator evaluation. For the cross-frequency band collaborative features, the method designs a frequency band combination strategy with reference to the band-pass parameters in the screening result table. For example, in the attention switching task, the frequency components of the theta and gamma frequency bands are simultaneously selected based on the pass-band characteristics of the filter, and their phase relationship is maintained. The method generates a detailed frequency component selection table based on the screening and selection results, recording the parameter information of each retained frequency component.

[0066] Based on the parameter information in the frequency component selection table, signal reconstruction is performed on the selected frequency components. The method uses an improved inverse Fourier transform algorithm to recombine the frequency components in the selection table according to their recorded amplitude and phase information. The reconstruction process adopts a segmented processing strategy, with each segment having a length of 2 seconds and adjacent segments overlapping by 1 second, and smooth transition is achieved through weighted averaging. For example, in the speech concentration analysis, the method reconstructs the signals of the P3 and P4 channels according to the frequency component parameters of the alpha frequency band in the selection table, and uses a cosine smoothing window with a length of 500 milliseconds at the transition between the speaking and listening states. The method dynamically adjusts the reconstruction weights according to the frequency component stability indicators recorded in the selection table, and assigns a greater reconstruction weight to the frequency components with high stability. In the multi-channel EEG feedback task, the method synchronously reconstructs the signals of the Fp1, Fp2 and O1, O2 channels based on the parameter information in the selection table, and ensures the spatial characteristics by maintaining the phase relationship recorded in the table. For example, in the motor imagery analysis, the mu rhythm signal is reconstructed according to the frequency component parameters of the C3 and C4 channels in the selection table, and the phase difference is maintained within 20 degrees. The method also establishes a multi-level quality assessment system. The first level calculates the correlation coefficient between the reconstructed signal and the original sample based on the parameters in the selection table, the second level analyzes the spectral purity of the reconstructed signal, and the third level evaluates the physiological rationality of the reconstructed signal. Finally, a pure waveform retaining the key cognitive components is generated.

[0067] Step S104, obtain the characteristic indicators corresponding to the pure waveform, generate a characteristic map according to the characteristic indicators, and generate a characteristic matrix according to the characteristic map.

[0068] Specifically, sub-band division processing is performed on the pure waveform generated in step S103. The method uses continuous wavelet transform technology to construct a multi-scale wavelet decomposition system, and divides the signal into characteristic frequency bands such as delta (0.5 - 4 Hz), theta (4 - 8 Hz), alpha (8 - 13 Hz), beta (13 - 30 Hz), and gamma (30 - 50 Hz). In the division process, complex wavelet basis functions are used, and the wavelet scales are set to progress in multiples of 3, with a total of 12 scales, to achieve fine characterization of the frequency bands. For example, in the driving fatigue monitoring scenario, the method performs sub-band division on the pure waveforms of the Fp1 and Fp2 channels, focuses on the subdivision characteristics of the alpha frequency band, divides the 8 - 13 Hz alpha frequency band into two sub-bands: low alpha (8 - 10 Hz) and high alpha (10 - 13 Hz), and extracts the energy envelope and phase sequence of each sub-band. For the vision-related activities of the O1 and O2 channels, the method divides the beta frequency band into three sub-bands: low beta (13 - 20 Hz), medium beta (20 - 25 Hz), and high beta (25 - 30 Hz), and simultaneously calculates the instantaneous amplitude and phase changes of each sub-band. For the motor imagery task, the method adopts an adaptive bandwidth division strategy for the signals of the C3 and C4 channels, dynamically adjusts the sub-band boundaries according to the peak distribution of the power spectrum, and obtains the time-frequency energy distribution of each sub-band. The method calculates the power correlation coefficient between sub-bands to construct a frequency band coupling matrix, forming a set of characteristic frequency bands that includes the energy distribution, phase sequence, and coupling relationship of each sub-band.

[0069] Feature indicators are extracted based on the energy distribution and phase sequence information in the set of characteristic frequency bands. In the time domain, statistical moments are calculated for the energy envelope of each sub-band, including mean, standard deviation, skewness, and kurtosis; envelope smoothness and phase stability indicators are calculated for the phase sequence. For example, in the emotion recognition task, the method uses the energy distribution characteristics of the alpha sub-band of the F3 and F4 channels to calculate the asymmetry index, which is obtained through the normalized difference of the energy envelopes of the left and right hemispheres. In the frequency domain feature layer, the power spectral density is estimated based on the time-frequency energy distribution of the sub-band, and spectral parameters such as center frequency, frequency standard deviation, and main frequency ratio are extracted. For example, in the working memory task, the relative power and absolute power are calculated for the energy distribution of the theta sub-band of the Fz channel, and a cognitive load index is constructed using the energy ratio of the theta and alpha sub-bands. The method also extracts cross-frequency band features based on the frequency band coupling matrix, such as calculating the theta-gamma coupling strength, alpha-beta phase synchronization degree, etc. In the attention detection scenario, the instantaneous phase synchronization degree is calculated using the phase sequence of the beta sub-band of the P3 and P4 channels to evaluate the spatio-temporal coordination of the attention network. By methodically extracting various types of information in the set of characteristic frequency bands, a multi-dimensional feature index set including time domain indicators, frequency domain indicators, and coupling indicators is finally obtained.

[0070] Construct a feature mapping relationship based on the parameters of the multi-dimensional feature index set. The method first performs a correlation analysis on the feature index set, calculates the Pearson correlation coefficient matrix between the indexes, and selects the index subset with a correlation lower than 0.7 for mapping. The t-SNE algorithm is used to project the filtered high-dimensional feature indexes into a 3D feature space. The perplexity parameter in the mapping process is set to 30, the learning rate is set to 200, and the number of iterations is 1000 times. In a multi-modal learning scenario, for example, the method selects the alpha rhythm indexes of the visual channels (O1, O2), the gamma rhythm indexes of the auditory channels (T3, T4), and the theta rhythm indexes of the executive control channels (Fp1, Fp2), and constructs a unified mapping space based on their time-domain statistics and frequency-domain features. For a dynamic task scenario, the method uses the sliding window method with a window length of 2 seconds and a step size of 0.5 seconds, and organizes the corresponding feature indexes in each window for mapping to form a feature trajectory. In sleep stage analysis, a hierarchical mapping structure is established using the energy indexes and phase indexes of multiple channels, and the feature distributions of different sleep stages are reflected through the combination of feature indexes. The method evaluates the topological preservation of the feature mapping by calculating the local adjacency relationship in the mapping space. When the stress value exceeds 0.2, the mapping parameters are adjusted to reconstruct the feature space. The finally generated feature mapping not only retains the discriminant information of the original feature indexes but also realizes the low-dimensional representation of the data.

[0071] In some embodiments, generating a feature matrix according to the feature mapping includes: obtaining feature categories through grouping and clustering according to the feature mapping; constructing a feature set according to the feature categories; and generating the feature matrix according to the feature set.

[0072] Perform grouping and clustering analysis on the obtained feature mapping. The method uses the hierarchical clustering algorithm. First, calculate the Euclidean distance matrix of the sample points in the feature space, and then gradually merge the clusters based on the Ward minimum variance criterion. The number of clustering levels is dynamically set according to the scale of the data set. For a data set containing 1000 samples, 4 to 6 levels are set for progressive clustering. In a multi-task learning scenario, for example, the method clusters the cognitive load feature mapping of the Fp1 and Fp2 channels, and divides the working memory load into 3 categories: low load (concentrating attention), medium load (multitasking), and high load (cognitive overload). The center and boundary of each category are determined by the distribution characteristics of its feature mapping. In spatial attention analysis, the density clustering method is used for the feature mapping of the P3 and P4 channels. By setting the neighborhood radius to 0.5 and the minimum number of samples to 15, the main patterns of attention allocation are identified. The method calculates the within-class density and between-class distance based on the clustering results to form a feature category table containing category labels, category centers, between-class distance matrices, and within-class density distributions.

[0073] Construct a feature set using the class centers and density distribution information in the feature category table. For each category, the method extracts sample points whose distances from the class center are less than a threshold, and determines the sample weights according to the within-class density distribution. For example, in the emotion recognition task, for the three emotion categories of positive, negative, and neutral obtained by clustering the F3 and F4 channels, the 30% sample points with the highest within-class density are respectively selected as the core samples. For cross-channel feature categories, the method establishes spatial associations based on the inter-class distance matrix, and combines categories with close distances to form a joint feature representation. In the cognitive state assessment, the three working memory load features of Fp1 and Fp2 are combined with the attention category features of O1 and O2, and a multi-dimensional feature set is constructed based on within-class density weighting. Finally, a multi-modal feature set including core samples, weight coefficients, and spatial associations of each category is formed.

[0074] Perform feature combination based on the core samples and weight information in the multi-modal feature set. The method analyzes the feature correlations between core samples, and pairs and combines features with a correlation coefficient lower than 0.3 to ensure feature complementarity. For example, in the concentration assessment scenario, the core samples of the theta / beta ratio of the Fp1 and Fp2 channels in the feature set are weighted and combined with the core samples of the alpha energy of the P3 and P4 channels, and the weight coefficients are determined by the density distribution of the samples in the original category. For dynamic task scenarios, the method constructs spatio-temporal constraints for feature combination according to the spatial association information in the feature set. In driving fatigue monitoring, the core feature samples of the prefrontal lobe, occipital lobe, and temporal lobe are fused according to their weight coefficients to generate a multi-dimensional fatigue index. The method calculates the stability assessment of the combined features to ensure that the within-class aggregation degree of the combined features is not lower than that of the original feature set. Finally, a set of feature combination schemes including combination methods, weight coefficients, and stability indicators is obtained.

[0075] Generate a feature matrix according to the combination rules and weight parameters in the feature combination scheme. The method first performs weighted fusion on each group of combined features, and the weight values directly adopt the normalized results of the stability indicators in the scheme. Then, a matrix structure is constructed, where the rows of the matrix represent different feature combinations, and the columns represent the dimensions of the combined features. For example, in the multi-modal learning task, the stable combinations of visual attention features, auditory processing features, and motor control features are mapped to a 100-dimensional feature space based on the combination scheme. In the attention persistence test, the method constructs a feature sequence according to the spatio-temporal constraints of the combination scheme to maintain the temporal continuity of the combined features. Perform condition number analysis on the generated feature matrix. When the condition number exceeds 100, optimize the matrix structure by adjusting the weight parameters in the combination scheme. The finally generated feature matrix not only retains the discriminative information of the original features but also reflects the combination relationship between the features.

[0076] Step S105: Perform dimensionality mapping on the feature matrix to obtain the main components, and output a feature representation based on the main components, thereby completing the noise reduction and feature extraction of the multi-channel EEG signals.

[0077] Specifically, perform dimensionality mapping on the feature matrix. The method constructs a feature projection framework using a non-linear dimensionality reduction algorithm. First, calculate the distance matrix of sample points in the high-dimensional feature space, and use the Euclidean distance to measure the similarity between features. Then, project the high-dimensional features into a 3D representation space based on the t-SNE algorithm, where the perplexity parameter is set to 30, the learning rate is set to 200, and the maximum number of iterations is 1000. For example, in a remote meeting scenario, the method reduces the dimensionality of the concentration index in the feature matrix, maps the high-dimensional representation containing features such as the theta / beta ratio and alpha band energy into a trajectory in 3D space, and realizes real-time tracking of the attention state of the participants. For a long meeting process, the method calculates the feature mapping every 30 seconds to construct an attention change curve. In a document reading task, the method combines and maps the visual processing features of the O1 and O2 channels and the cognitive load features of the Fp1 and Fp2 channels to generate a state space reflecting the reading engagement. For the emotion-related features of the F3 and F4 channels, calculate the feature aggregation degree according to the spatial distribution after mapping. The higher the aggregation degree, the more stable the emotion state. For example, in a multi-person collaboration scenario, the method synchronously maps the attention features of all participants and evaluates the team collaboration state by calculating the spatial distribution features of the mapped points. In a daily office environment, the method integrates and maps the visual attention features, cognitive load features, and emotion state features of the user to construct a multi-dimensional work state index. The method evaluates the mapping quality by calculating the local structure preservation rate before and after projection. When the preservation rate is lower than 85%, a parameter optimization mechanism is triggered. At the same time, a stress value evaluation is introduced to ensure that the topological relationship of the high-dimensional features is maintained during the dimensionality reduction process. Finally, a main component matrix containing mapping coordinates, quality indicators, and topological features is obtained.

[0078] In some embodiments, the outputting a feature representation based on the main components includes: performing a spatial transformation on the main components according to the principal component analysis method to obtain the spatial transformation structure corresponding to the main components; and performing dimensionality reduction processing on the spatial transformation structure to obtain the feature representation.

[0079] Perform a spatial transformation based on the obtained principal component matrix. The method uses the principal component analysis method to perform an orthogonal transformation on the mapping result, and selects the eigen-directions with a cumulative contribution rate reaching 95% to construct a new feature space. During the transformation process, calculate the variance contribution rate and eigenvalue of each principal component, and sort them according to the eigenvalue size. For example, in the document reading task, the method performs a spatial rotation on the principal components of the visual attention features of channels O1 and O2, aligning the maximum variance direction with the attention intensity axis to achieve a quantitative evaluation of reading concentration. In the video viewing scenario, the method performs a joint transformation on the attention components of the visual channel and the processing components of the auditory channel to establish an evaluation model for multi-modal attention. For the online learning task, the method aligns the cognitive load component and the visual processing component in the feature space, and judges the learning efficiency through the projection direction of the principal component. In the concentration evaluation scenario, perform a scaling transformation on the cognitive load components of channels Fp1 and Fp2, adjust the spatial scale according to the eigenvalue size, and establish a standardized attention scoring system. For the conference interaction scenario, the method performs a unified spatial transformation on the attention features of multiple participants, enabling the attention states of different individuals to be compared in the same metric space. The method also establishes a discriminant analysis model based on the Fisher criterion to optimize the between-class separation and within-class aggregation, improving the accuracy of state recognition. In the browsing scenario, construct a state index reflecting information acquisition efficiency through the spatial transformation of visual scanning features. Finally, obtain a spatial transformation structure containing the transformation matrix, eigenvalues, discriminant coefficients, and standardized parameters.

[0080] Apply a spatial transformation structure to perform dimensionality reduction on the main components. The method first analyzes the condition number of the transformation matrix to ensure numerical stability. Then, it determines the truncation threshold based on the eigenvalue magnitude and retains the dimensions with an explained variance contribution rate exceeding 2%. For example, in the attention persistence test, compress the fatigue features of the frontal lobe leads, determine the optimal compression ratio by calculating the information entropy, and achieve long-term monitoring of the attention level. In the rapid reading task, the method establishes a feature compression model to compress the high-dimensional features of visual processing and cognitive load into a low-dimensional representation convenient for real-time calculation. For the multi-channel collaborative features in the meeting scenario, establish a spatio-temporal dimensionality reduction model to reduce feature redundancy while maintaining temporal correlation and achieve real-time assessment of the group interaction state. In the outdoor activity scenario, the method extracts stable attention features through dimensionality reduction processing to reduce the influence of environmental noise. For the multi-task work scenario, perform adaptive dimensionality reduction on the attention features during task switching to capture the key features of task conversion. The method conducts multi-level evaluation of the reconstruction error of the dimensionality reduction result: at the first level, calculate the signal reconstruction error to ensure that key information is not lost during dimensionality reduction; at the second level, evaluate the preservation degree of temporal features to verify the continuity of state transitions; at the third level, analyze the fidelity of spatial features to ensure the consistency of inter-channel relationships. When the evaluation index of any level exceeds the threshold, the method automatically adjusts the dimensionality reduction parameters until the quality requirements are met. Through the multi-level optimization strategy, finally generate a feature representation that not only retains the key information of the cognitive state but also has high computational efficiency.

[0081] The method provided by this application has at least the following beneficial effects:

[0082] 1. Through a hierarchical feature extraction and processing strategy, the method realizes the adaptive processing of the signals collected by the smart glasses. First, construct an initial feature set based on multi-channel signal features, including time-domain statistical features, band energy features, and spatial correlation features; then, establish a dynamic threshold mechanism through anomaly detection and data correction based on the probability density function to achieve adaptability to individual differences; finally, use multi-level sub-band decomposition and spectral analysis to extract stable signal features, effectively solving the problem of signal instability in portable EEG acquisition.

[0083] 2. Adopt a method based on template matching and multi-dimensional feature analysis to achieve accurate recognition of different cognitive states. The method first performs edge detection through a multi-scale derivative operator to accurately locate the key feature positions; then, combines hierarchical clustering and spatial consistency constraints to establish a cross-channel feature segment alignment mechanism; further, realizes the selective retention of features in different cognitive states through an improved wavelet packet reconstruction algorithm; finally, optimizes the feature expression based on spectral analysis, significantly improving the recognition accuracy of the user's cognitive state in complex environments.

[0084] 3 A hierarchical feature mapping and dimensionality reduction framework is established to achieve efficient feature expression. The method constructs a multi-scale decomposition system through continuous wavelet transform technology to achieve fine characterization of signal features; a low-dimensional representation of features is established based on the t-SNE algorithm and hierarchical mapping structure; through a multi-level quality assessment mechanism, including reconstruction error analysis, temporal feature preservation, and spatial feature verification, the retention of key information during the dimensionality reduction process is ensured; finally, a feature representation method that retains both discriminant information of the cognitive state and has high computational efficiency is formed.

[0085] To implement the EEG signal noise reduction and feature extraction method corresponding to the above method embodiment to achieve the corresponding functions and technical effects. Refer to Figure 2 , Figure 2 FIG. shows a structural block diagram of an EEG signal noise reduction and feature extraction device 200 provided by an embodiment of the present application. For ease of explanation, only the parts related to this embodiment are shown. The EEG signal noise reduction and feature extraction device 200 provided by the embodiment of the present application includes:

[0086] A signal acquisition module 201, configured to collect multi-channel EEG signals of the smart glasses, obtain an original data stream, generate an initial feature set according to the original data stream, and generate a standard data stream according to the initial feature set;

[0087] A feature acquisition module 202, configured to perform sub-band decomposition on the standard data stream to obtain band data, obtain energy features corresponding to the band data, and obtain a noise reduction signal according to the energy features; perform edge detection on the noise reduction signal to obtain feature positions, and obtain contour features corresponding to the feature positions;

[0088] A type acquisition module 203, configured to obtain a feature template corresponding to the contour feature and a corresponding signal type, construct a signal sample according to the signal type; obtain frequency components corresponding to the signal sample, and perform reconstruction according to the frequency components to generate a pure waveform;

[0089] An index acquisition module 204, configured to obtain feature indexes corresponding to the pure waveform, generate a feature map according to the feature indexes, and generate a feature matrix according to the feature map;

[0090] An extraction completion module 205, configured to perform dimensionality mapping on the feature matrix to obtain main components, output a feature representation according to the main components, and complete the noise reduction and feature extraction of the multi-channel EEG signals.

[0091] The above-mentioned intelligent glasses EEG signal noise reduction and feature extraction device 200 can implement the intelligent glasses EEG signal noise reduction and feature extraction method of the above method embodiment. The optional items in the above method embodiment are also applicable to this embodiment, which will not be elaborated here. The remaining content of the embodiment of the present application can refer to the content of the above method embodiment, and will not be repeated in this embodiment.

[0092] Figure 3 It is a schematic structural diagram of a computer device provided by an embodiment of the present application. As Figure 3 shown, the computer device 3 of this embodiment includes: at least one processor 30 ( Figure 3 only one is shown here), a memory 31, and a computer program 32 stored in the memory 31 and executable on the at least one processor 30. When the processor 30 executes the computer program 32, it implements the steps in any of the above method embodiments.

[0093] The computer device 3 may be a computing device such as a smart phone, a tablet computer, a desktop computer, and a cloud server. The computer device may include but is not limited to the processor 30 and the memory 31. Those skilled in the art can understand that Figure 3 this is only an example of the computer device 3, and does not constitute a limitation on the computer device 3. It may include more or fewer components than shown in the figure, or combine some components, or different components. For example, it may also include input / output devices, network access devices, etc.

[0094] The so-called processor 30 may be a central processing unit (CPU). The processor 30 may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0095] In some embodiments, the memory 31 may be an internal storage unit of the computer device 3, such as the hard disk or memory of the computer device 3. In some other embodiments, the memory 31 may also be an external storage device of the computer device 3, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc., equipped on the computer device 3. Further, the memory 31 may also include both the internal storage unit and the external storage device of the computer device 3. The memory 31 is used to store an operating system, application programs, a BootLoader, data, and other programs, such as the program code of the computer program. The memory 31 may also be used to temporarily store data that has been output or is to be output.

[0096] In addition, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0097] An embodiment of the present application provides a computer program product, and when the computer program product runs on a computer device, the computer device is caused to implement the steps in each of the above method embodiments when executed.

[0098] In several embodiments provided by the present application, it can be understood that each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, the program segment, or the part of code includes one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in an order different from that marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved.

[0099] When the above-mentioned functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art or a part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0100] The specific embodiments described above further elaborate on the purpose, technical solution, and beneficial effects of this application. It should be understood that the above are only specific embodiments of this application and are not used to limit the protection scope of this application. It is particularly pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of this application should be included in the protection scope of this application.

Claims

1. An intelligent glasses electroencephalogram signal noise reduction and feature extraction method, characterized in that Including: Collecting multi-channel electroencephalogram (EEG) signals of smart glasses to obtain an original data stream, generating an initial feature set according to the original data stream, and generating a standard data stream according to the initial feature set; Performing sub-band decomposition on the standard data stream to obtain band data, obtaining energy features corresponding to the band data, and obtaining a noise reduction signal according to the energy features; Performing edge detection on the noise reduction signal to obtain a feature position, including: detecting the edge position of the corresponding alpha band energy change, and forming a feature position sequence including the edge position, edge intensity, and edge type through weighted fusion of multi-scale edge detection results; obtaining the contour feature corresponding to the feature position; Obtaining a feature template corresponding to the contour feature and the corresponding signal type, constructing a signal sample according to the signal type; obtaining the frequency components corresponding to the signal sample, and performing reconstruction according to the frequency components to generate a pure waveform; Obtaining a feature index corresponding to the pure waveform, generating a feature map according to the feature index, and generating a feature matrix according to the feature map; Performing dimension mapping on the feature matrix to obtain main components, outputting a feature representation according to the main components, and completing noise reduction and feature extraction of the multi-channel EEG signal.

2. The method according to claim 1, wherein The generating the standard data stream according to the initial feature set includes: Performing amplitude statistics on the initial feature set to obtain an amplitude distribution; Obtaining the abnormal point position corresponding to the initial feature set according to the amplitude distribution; Correcting the initial feature set according to the abnormal point position to generate the standard data stream.

3. The method according to claim 1, characterized in that The obtaining the noise reduction signal according to the energy features includes: Performing threshold selection on the energy features according to a preset threshold to obtain an effective signal corresponding to the energy features; Obtaining a reconstructed waveform corresponding to the effective signal; Obtaining the noise reduction signal according to the reconstructed waveform.

4. The method according to claim 1, wherein The obtaining the contour feature corresponding to the feature position includes: Obtaining a signal boundary corresponding to the noise reduction signal according to the feature position; Dividing the noise reduction signal according to the signal boundary to obtain feature segments; Performing feature extraction on the feature segments to obtain the contour feature.

5. The method according to claim 1, characterized in that, The performing reconstruction according to the frequency components to generate a pure waveform includes: Performing band-pass screening on the frequency components; Obtaining the frequency components corresponding to the frequency components according to the screening result corresponding to the band-pass screening; Reconstructing a signal according to the frequency components to generate the pure waveform.

6. The method according to claim 1, characterized in that, The generating the feature matrix according to the feature map includes: Performing grouped clustering according to the feature map to obtain feature categories; Constructing a feature set according to the feature categories; Generating the feature matrix according to the feature set.

7. The method according to claim 1, wherein The outputting the feature representation according to the main components includes: Performing a spatial transformation on the main components according to the principal component analysis method to obtain a spatial transformation structure corresponding to the main components; Performing dimensionality reduction processing on the spatial transformation structure to obtain the feature representation.

8. The method according to claim 1, wherein The generating the initial feature set according to the original data stream includes: Performing time window segmentation on the original data stream to obtain time series segments; Generating the initial feature set according to the time series segments.

9. An electroencephalogram signal noise reduction and feature extraction device for smart glasses, characterized in that, Including: A signal acquisition module, configured to collect multi-channel electroencephalogram signals of the smart glasses, obtain an original data stream, generate an initial feature set according to the original data stream, and generate a standard data stream according to the initial feature set; A feature acquisition module, configured to perform sub-band decomposition on the standard data stream to obtain band data, obtain energy features corresponding to the band data, and obtain a noise reduction signal according to the energy features; Performing edge detection on the noise reduction signal to obtain a feature position, including: detecting an edge position of a corresponding alpha band energy change, and forming a feature position sequence including an edge position, an edge intensity, and an edge type through weighted fusion of multi-scale edge detection results; obtaining a contour feature corresponding to the feature position; A type acquisition module, configured to obtain a feature template corresponding to the contour feature and a corresponding signal type, construct a signal sample according to the signal type; obtain a frequency component corresponding to the signal sample, and perform reconstruction according to the frequency component to generate a pure waveform; An index acquisition module, configured to obtain a feature index corresponding to the pure waveform, generate a feature map according to the feature index, and generate a feature matrix according to the feature map; An extraction completion module, configured to perform dimensionality mapping on the feature matrix to obtain main components, output a feature representation according to the main components, and complete noise reduction and feature extraction of the multi-channel electroencephalogram signals.

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

Images