Method, device and equipment for adaptive processing of electroencephalogram signals of attention glasses

Abstract

The present application discloses a method, device and equipment for adaptive processing of electroencephalogram (EEG) signals of attention glasses, the method comprising: obtaining original electroencephalogram (EEG) signals collected by the attention glasses to generate a standardized signal sequence for interference detection, obtaining a variation interval corresponding to signal fluctuations and constructing an interference pattern library; determining a processing interval according to the standardized signal sequence and the interference pattern library, generating a correction sequence according to the processing interval, and constructing an adaptive benchmark according to the correction sequence; reconstructing based on the adaptive benchmark to obtain an EEG feature sequence and a corresponding time-varying feature set; obtaining an attention index corresponding to the time-varying feature, performing feature extraction on the attention index, and obtaining a core feature set; constructing a mapping rule according to the core feature set, generating a state feature according to the mapping rule, and constructing a pattern for the state feature; adaptively processing a construction result corresponding to the pattern construction to generate a state tag sequence, and outputting an adaptive signal according to the state tag sequence.

Description

Technical Field

[0001] The present application relates to the field of brain-computer interface technology, and in particular to a method, device and equipment for adaptively processing electroencephalogram (EEG) signals of attention glasses. Background Art

[0002] As a new type of brain-computer interface device, attention glasses assess the user's attention level by collecting and analyzing the user's EEG signals. They have broad application prospects in the fields of improving work and study efficiency and training cognitive abilities. Existing attention glasses have many technical difficulties in processing EEG signals: First, traditional signal acquisition and processing methods use fixed parameters and processing procedures, lack the ability to evaluate the quality of the original signal and analyze the channel, and also lack personalized parameter adjustment mechanisms for different users. Secondly, attention glasses are easily affected by various interference sources during daily portable use, such as motion artifacts, electromyographic interference, environmental electromagnetic noise, etc. Existing technologies lack systematic interference identification and pattern analysis methods.

[0003] During long-term attention training, the signal quality will be unstable due to factors such as changes in electrode impedance and changes in user fatigue. The existing technology lacks a long-term monitoring and quality maintenance mechanism for electrode status. At the same time, in the signal processing process, it is difficult to balance the requirements of real-time and accuracy, and processing delays often cannot meet the needs of fast feedback. Especially in the case of rapid changes in attention status, traditional fixed processing methods can neither adjust processing parameters in time nor achieve adaptive optimization of signals, resulting in inaccurate attention evaluation results. These problems seriously restrict the performance of attention glasses in practical applications.

[0004] Therefore, a method is urgently needed to solve at least one of the above problems. Summary of the invention

[0005] The embodiments of the present application provide an electroencephalogram (EEG) signal adaptive processing method, device and apparatus for attention glasses, aiming to solve the problem that the traditional fixed processing method can neither adjust the processing parameters in time nor achieve adaptive optimization of the signal, resulting in inaccurate attention evaluation results. These problems seriously restrict the performance of attention glasses in practical applications.

[0006] In a first aspect, an embodiment of the present application provides a method for adaptively processing EEG signals of attention glasses, comprising:

[0007] Acquire the original EEG signals collected by the multi-channel electrodes of the attention glasses, and generate a standardized signal sequence after performing channel analysis on the original EEG signals;

[0008] Performing interference detection on the standardized signal sequence, obtaining a variation interval corresponding to the signal fluctuation and constructing an interference pattern library according to the variation interval;

[0009] Determine a processing interval according to the standardized signal sequence and the interference pattern library, generate a correction sequence according to the processing interval, and construct an adaptive benchmark according to the correction sequence;

[0010] Reconstruct based on the adaptive benchmark to obtain an EEG feature sequence, and obtain a time-varying feature set corresponding to the EEG feature sequence;

[0011] Obtaining an attention index corresponding to the time-varying feature, performing feature extraction on the attention index, and obtaining a core feature set;

[0012] Constructing a mapping rule according to the core feature set, generating a state feature according to the mapping rule, and constructing a pattern for the state feature;

[0013] Adaptive processing is performed on the construction result corresponding to the pattern construction to generate a state mark sequence, and an adaptive signal is output according to the state mark sequence to complete the adaptive processing of the original EEG signal.

[0014] In a second aspect, the present application also provides an electroencephalogram signal adaptive processing device, comprising:

[0015] A signal acquisition module, used to acquire the original EEG signals collected by the multi-channel electrodes of the attention glasses, and generate a standardized signal sequence after performing channel analysis on the original EEG signals;

[0016] An interference detection module, used to perform interference detection on the standardized signal sequence, obtain a variation interval corresponding to the signal fluctuation and construct an interference pattern library according to the variation interval;

[0017] A sequence generation module, configured to determine a processing interval according to the standardized signal sequence and the interference pattern library, generate a correction sequence according to the processing interval, and construct an adaptive benchmark according to the correction sequence;

[0018] A reference reconstruction module, used for reconstructing based on the adaptive reference, obtaining an EEG feature sequence, and obtaining a time-varying feature set corresponding to the EEG feature sequence;

[0019] A set acquisition module, used to acquire the attention index corresponding to the time-varying feature set, perform feature extraction on the attention index, and acquire a core feature set;

[0020] A rule construction module, used to construct a mapping rule according to the core feature set, generate a state feature according to the mapping rule, and perform pattern construction on the state feature;

[0021] The processing completion module is used to perform adaptive processing on the construction result corresponding to the pattern construction, generate a state mark sequence, output an adaptive signal according to the state mark sequence, and complete the adaptive processing of the original EEG signal.

[0022] In a third aspect, the present application also provides a computer device, including a processor and a memory, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, the method for adaptively processing EEG signals of the attention glasses as described in the first aspect is implemented.

[0023] In a fourth aspect, the present application further provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method for adaptively processing EEG signals of the attention glasses as described in the first aspect.

[0024] Compared with the prior art, this application has at least the following beneficial effects:

[0025] 1. Through multi-channel signal quality evaluation and hierarchical adaptive parameter adjustment mechanism, the method monitors and dynamically optimizes the signal quality in real time, and establishes a complete interference identification and classification system while ensuring signal reliability. For different types of interference sources such as motion artifacts, electromyographic interference, and environmental electromagnetic noise, the method adopts differentiated processing strategies, which can effectively suppress various types of interference while maintaining the effective components of the signal. Especially in complex scenarios such as user movement and noisy environment, the method significantly improves the environmental adaptability and signal stability of the attention glasses through adaptive parameter adjustment and multi-level signal correction.

[0026] 2. The method adopts a multi-scale signal decomposition and fine frequency band division scheme, combined with a dynamic construction mechanism of adaptive benchmarks, to achieve a comprehensive extraction of EEG signal features. Through time-varying feature generation and attention feature recognition, the method establishes a complete feature selection and state recognition framework. When the attention state changes rapidly, such as when the work task is switched or the concentration of attention fluctuates, the method can capture the state change characteristics in time and accurately identify different attention levels. This dynamic feature extraction and adaptive state recognition mechanism effectively overcomes the problem of slow response to state changes in traditional fixed template methods.

[0027] 3. Based on a multi-level scenario-based adaptive processing mechanism, the method achieves accurate identification and parameter optimization for different usage scenarios. Through multi-dimensional signal processing and state marking strategies, the method can accurately distinguish the attention characteristics in different scenarios such as learning, work, and research. Especially during long-term use, the method not only ensures the real-time nature of attention assessment, but also significantly improves the accuracy of the assessment results through dynamically updated marking rules and adaptive signal output mechanisms. This scenario-based adaptive processing solution enables the attention glasses to provide users in different scenarios with accurate attention state assessment and timely feedback guidance.

[0028] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A schematic diagram of a flow chart of an electroencephalogram signal adaptive processing method of attention glasses according to an embodiment of the present application;

[0030] Figure 2 This is a schematic diagram of the structure of an electroencephalogram signal adaptive processing device shown in an embodiment of the present application;

[0031] Figure 3 A schematic diagram of the structure of a computer device shown in an embodiment of the present application. DETAILED DESCRIPTION

[0032] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may 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 prevent unnecessary details from obstructing the description of the present application.

[0033] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of 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 combinations thereof.

[0034] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0035] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0036] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0037] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0038] The technical solution of the embodiment of the present application is introduced below.

[0039] As a new type of brain-computer interface device, attention glasses assess the user's attention level by collecting and analyzing the user's EEG signals. They have broad application prospects in the fields of improving work and study efficiency and cognitive ability training. Existing attention glasses have many technical difficulties in EEG signal processing: First, traditional signal acquisition and processing methods use fixed parameters and processing procedures, lack the ability to evaluate the quality of the original signal and analyze the channel, and also lack personalized parameter adjustment mechanisms for different users. Secondly, attention glasses are easily affected by various interference sources during daily portable use, such as motion artifacts, electromyographic interference, environmental electromagnetic noise, etc. Existing technologies lack methodical interference identification and pattern analysis methods.

[0040] During long-term attention training, the signal quality will be unstable due to factors such as changes in electrode impedance and changes in user fatigue. The existing technology lacks a long-term monitoring and quality maintenance mechanism for electrode status. At the same time, in the signal processing process, it is difficult to balance the requirements of real-time and accuracy, and processing delays often cannot meet the needs of fast feedback. Especially in the case of rapid changes in attention status, traditional fixed processing methods can neither adjust processing parameters in time nor achieve adaptive optimization of signals, resulting in inaccurate attention evaluation results. These problems seriously restrict the performance of attention glasses in practical applications.

[0041] To solve the above problems, please refer to Figure 1 , Figure 1 The flowchart of the method for adaptively processing EEG signals of the attention glasses provided in the embodiment of the present application is shown in FIG. The method for adaptively processing EEG signals of the attention glasses provided in the embodiment of the present application can be applied to computer devices, including but not limited to smart phones, laptops, tablet computers, desktop computers, physical servers, cloud servers and other devices. Figure 1 As shown, the electroencephalogram signal adaptive processing method of the attention glasses of this embodiment includes steps S101 to S107, which are described in detail as follows:

[0042] Step S101, obtaining the original EEG signal collected by the multi-channel electrodes of the attention glasses, and generating a standardized signal sequence after performing channel analysis on the original EEG signal.

[0043] Specifically, when the attention glasses collect multi-channel EEG signals, they are collected through six high-precision flexible dry electrode arrays integrated on the glasses frame. The four main collection electrodes are made of silver / silver chloride material and covered with conductive nanomaterials. They are distributed in key EEG detection positions such as the left and right foreheads and temples. The contact area between each electrode and the skin is 1 square centimeter, and the surface adopts a honeycomb microstructure design to enhance flexibility and fit. The other two reference electrodes are set at the back of the glasses for signal reference correction, and elastic brackets are used to achieve position fine-tuning. The sampling frequency is adjustable in the range of 256 to 1024 Hz, and generally works at a sampling frequency of 512 Hz. The sampling accuracy reaches 24 bits to capture weak EEG changes. Each electrode records the working parameters such as EEG signal amplitude, electrode temperature, contact pressure, etc. in real time. The temperature monitoring accuracy is 0.1 degrees Celsius, and the pressure monitoring range is 0-200 grams of force. The sampling time is accurate to milliseconds, and the timestamp is based on the reference clock provided by the onboard high-precision crystal oscillator. The method integrates a low-power Bluetooth 5.0 chip and adopts an enhanced data packet design. The original data is transmitted to the processing unit in real time. The transmission delay is controlled within 10 milliseconds. The size of a single data packet is 512 bytes, which contains signal data and status information, forming a continuous and complete original data stream.

[0044] In some embodiments, the generating of a standardized signal sequence after performing channel analysis on the raw EEG signals includes: acquiring the contact impedance between the electrode and the skin in each acquisition channel according to the raw EEG signals; judging whether the working state of each acquisition channel is normal according to the contact impedance; and performing channel analysis on the raw EEG signals corresponding to the acquisition channels with normal working states to generate the standardized signal sequence.

[0045] After the raw data stream enters the method, channel analysis is performed immediately, and the working status of each acquisition channel is monitored through the electrode parameters recorded in the data stream. The method calculates the contact impedance between the electrode and the skin in real time. The normal working impedance range is 5-50 kiloohms. When the contact impedance of an electrode abnormally increases and exceeds the threshold, the method starts the optimization program to adjust the electrode contact pressure and change the pressure distribution of the elastic bracket through the micro-actuator.

[0046] Exemplarily, performing channel analysis on the original EEG signal corresponding to the acquisition channel in normal working state to generate the standardized signal sequence includes: performing signal integrity analysis on the original EEG signal corresponding to the acquisition channel in normal working state; the signal integrity analysis includes at least signal continuity detection, amplitude stability assessment and baseline drift monitoring; and generating the standardized signal sequence according to the analysis result corresponding to the signal integrity analysis.

[0047] For channels with normal contact status, the method conducts in-depth signal integrity analysis, including signal continuity detection, amplitude stability assessment, and baseline drift monitoring. Continuity detection mainly identifies signal interruption or jump points. The method uses an adaptive threshold algorithm to judge abnormalities. Amplitude stability is evaluated by calculating the signal variance within a fixed time window, and baseline drift is quantified by the change trend of the signal mean within a long time window. At the same time, the method uses timestamp information to accurately locate the occurrence time and duration of various abnormalities, combines signal-to-noise ratio calculation and multi-scale spectrum analysis to evaluate signal purity, considers the energy distribution characteristics of typical EEG rhythm segments such as delta, theta, alpha, and beta, and finally generates a comprehensive quality score reflecting the working status of each channel.

[0048] Based on the quality scores obtained from channel analysis, the method adopts differentiated processing strategies for different signals and finally generates a standardized signal sequence. The channels with the highest scores are used as reference channels. The signals of these channels are first time-aligned to eliminate sampling delays, and then amplitude normalized to unify the signal amplitude range. The normalized signal amplitude is mapped to the range of plus or minus 100 microvolts. For channels with medium scores, the method first applies an adaptive notch filter to remove power frequency interference, and then uses the wavelet transform method to remove broadband interference such as electromyography. The processed signals are time-synchronized with the reference channel to ensure the timing consistency of multi-channel data. Channels with lower scores require more complex processing procedures. The method analyzes the signal characteristics of adjacent high-quality channels, extracts key feature parameters to establish a signal reconstruction model, and uses this model to correct and enhance the degraded signals. Special attention is paid to maintaining the physiological characteristics of the signals during the reconstruction process. After all channels are processed, the method reorganizes the signals into a signal sequence in a standard format. The sequence not only contains the processed EEG data, but also records the quality score, processing history, and feature markers of each channel. The signal sequence is stored in a standardized data structure. Each data frame contains the synchronization signal values ​​of all channels. The sampling point time interval is kept constant, and calibration information is added at the beginning of the sequence. The method uses a sliding window mechanism with a length of 2 seconds and a 50% overlap rate to continuously update the sequence content. Each processing window inherits the status information of the previous window to ensure the continuity of the signal sequence.

[0049] Step S102: perform interference detection on the standardized signal sequence, obtain the variation interval corresponding to the signal fluctuation, and construct an interference pattern library according to the variation interval.

[0050] Specifically, the method adopts a dual-window sliding analysis strategy when performing interference detection on the step signal sequence. The 50-millisecond short-term window captures fast fluctuation characteristics in real time, the 1-second long-term window establishes the signal baseline and trend characteristics, and the window sliding step is 10 milliseconds to ensure detection continuity. The time domain features calculated in the short-term window include: mean deviation (relative baseline), standard deviation change rate, kurtosis coefficient (range 2-5), skewness coefficient (range ±1), zero crossing rate, waveform factor, peak factor, time domain fluctuation, amplitude distribution entropy, electrode impedance stability, etc. The frequency domain features in the range of 0-64Hz are obtained by 8-layer wavelet decomposition, and parameters such as energy proportion of each frequency band, center of gravity frequency, number of spectral peaks, spectrum entropy, frequency band power ratio, and energy accumulation rate are calculated. A 50% overlap rate is set between windows, and the trend of characteristic parameter changes is compared and calculated in real time. The method synchronously analyzes the spatial characteristics of adjacent channels such as the mutual correlation coefficient (threshold 0.6), phase consistency (tolerance ±30 degrees), propagation delay, spatial gradient, and channel coupling. Based on these multidimensional features, the method constructs a real-time updated anomaly detection model, which dynamically adjusts the judgment threshold according to the statistical distribution of historical data. When any indicator exceeds the normal range, the method marks it as a potential interference point. Through cluster analysis of abnormal features in continuous time windows, the method finally identifies the complete signal fluctuation sequence, each of which contains attribute information such as type identification, impact range, and intensity level.

[0051] In some embodiments, obtaining the change interval corresponding to the signal fluctuation and constructing the interference pattern library based on the change interval includes: obtaining the signal fluctuation position corresponding to the interference detection; determining the change interval according to the signal fluctuation position in the standardized signal sequence; performing feature extraction on the change interval to obtain change features; and constructing the interference pattern library based on the change features.

[0052] For the identified signal fluctuation sequence, the method accurately marks the change interval. Taking each fluctuation sequence as the benchmark, the method expands the range of 2 seconds forward and backward to search for the signal stability point. The signal stability judgment adopts a multi-index joint evaluation: the time domain parameter fluctuation is required to be less than 2 times the baseline standard deviation within 100 milliseconds, the frequency domain feature and the baseline spectrum similarity is greater than 0.9, the channel correlation coefficient rises to above 0.8, and the main rhythm energy ratio recovers to the normal range. When the interval between adjacent fluctuations is less than 200 milliseconds, the method merges them into a single change interval. The merging process retains the main feature points and records the intermediate transition characteristics. For each determined change interval, the method establishes a five-layer description structure: time positioning layer (start and end time, duration, stable interval), spatial distribution layer (affected channel, propagation range, spatial gradient), intensity characterization layer (maximum deviation, mean square deviation, cumulative energy), feature evolution layer (development process, steady-state characteristics, recovery mode), and correlation analysis layer (precursor correlation, channel correlation, type correlation). Based on this hierarchical description, the method forms a change interval marking sequence containing complete information such as start and end points, range, intensity, and evolution.

[0053] Methods Multi-dimensional feature extraction is performed on the change interval sequence of the marker. First, waveform features are extracted in the time domain, including the fluctuation amplitude feature group (peak value, root mean square, standard deviation, skewness, kurtosis), the timing feature group (rise time, peak time, recovery time, duration period), and the morphological feature group (symmetry, sharpness, stability, complexity). Then, the energy distribution features are analyzed in the frequency domain, including the main frequency feature group (center frequency, main frequency bandwidth, number of spectrum peaks, spectrum peak spacing), the energy feature group (band energy ratio, energy concentration, energy mobility), and the time-frequency feature group (time-frequency joint distribution, energy flow diagram, instantaneous frequency). At the same time, spatial propagation features are extracted, including the spatial distribution feature group (spatial range, propagation speed, attenuation coefficient), the synchronization feature group (phase difference, coupling strength, synchronization metric), and the regional feature group (local features, global influence, boundary effect). Finally, probability features are extracted through statistical analysis, including the distribution feature group (probability density, cumulative distribution, moment estimation), the entropy feature group (information entropy, sample entropy, approximate entropy), and the clustering feature group (clustering coefficient, separation, purity index). All extracted features are normalized and assigned different weights according to the signal quality scores, ultimately forming a 400-dimensional feature vector set.

[0054] Based on the extracted feature vector set, the method constructs a hierarchical interference pattern library. The bottom layer contains 16 basic interference templates, such as power frequency interference, electromyographic interference, motion artifacts, poor electrode contact, electrooculographic artifacts, electrocardiographic artifacts, respiratory artifacts, and environmental electromagnetic interference. Each basic template has a unique feature distribution model, evolutionary characteristics, and spatial distribution law. The middle layer stores 64 composite interference patterns, which describe the combined characteristics when multiple interference sources act simultaneously, including the superposition effect, enhancement inhibition relationship, and propagation coupling characteristics between interferences. The top layer is expanded to 256 scene modes, reflecting the variant forms of interference under different usage environments, activity states, and individual differences. The method uses the Euclidean distance algorithm to calculate the similarity between the new feature sequence and the existing pattern. When the similarity exceeds 0.85, it is determined to be a known pattern. For unmatched new feature combinations, the method uses the K-means clustering method to classify them, and then extracts the template features through principal component analysis to form a new pattern to be added to the pattern library. The pattern library is optimized every 12 hours, merging patterns with a similarity of more than 0.95 through a hierarchical clustering method, while deleting patterns with a usage frequency of less than 1%, so that the pattern library remains streamlined and efficient. This multi-level, adaptive pattern library structure can accurately identify and classify various types of interference features, significantly improving the method's ability to adapt to complex interference environments.

[0055] Step S103, determining a processing interval according to the standardized signal sequence and the interference pattern library, generating a correction sequence according to the processing interval, and constructing an adaptive benchmark according to the correction sequence.

[0056] Specifically, the method first matches the signal sequence generated by step 1 with the interference pattern library constructed by step 1 to determine the signal interval to be processed. The matching adopts a sliding window mechanism with a length of 1 second and a step size of 100 milliseconds, and the windows overlap by 90% to ensure the continuity of the analysis. In each sliding window, the method calculates the characteristic parameters of the signal: the time domain features include parameters reflecting the waveform characteristics such as mean, variance, waveform factor, kurtosis factor, zero-crossing rate, etc.; the frequency domain features include parameters such as the energy proportion, main frequency position, and bandwidth of each frequency band (δ, θ, α, β); the channel features include spatial correlation information such as the correlation coefficient, phase difference, and propagation delay between channels. These features are compared with the templates in the interference pattern library in multiple dimensions, and the matching degree of the feature vector is calculated using cosine similarity. When the similarity exceeds the preset threshold of 0.8, the window is marked as a to-be-processed interval. For the continuously marked windows, the method analyzes their feature change trends. When the feature correlation of adjacent windows is greater than 0.7, they are merged into a complete processing interval. Each processing interval records detailed attribute information: the start and end time points are accurate to the millisecond level, the dominant interference type and its confidence, and the impact degree is divided into three levels: mild (0-0.3), moderate (0.3-0.7) and severe (0.7-1.0). The method also records the normal signal characteristics within 200 milliseconds on both sides of the processing interval, including amplitude distribution, spectral structure, channel relationship, etc., as a reference standard for signal processing. For multi-channel data, when a channel is marked as a processing interval, the method automatically analyzes the signal correlation within ±100 milliseconds of the adjacent channels, and the area with a correlation coefficient exceeding 0.6 is also included in the processing range, thus forming a complete set of processing intervals containing time and space information.

[0057] In some embodiments, generating the correction sequence according to the processing interval includes: adaptively segmenting the processing interval; determining a plurality of processing tags according to segmentation results corresponding to the adaptive segmentation; and generating the correction sequence according to the processing tags.

[0058] Adaptive segmentation is performed on the determined set of processing intervals. The method adopts a multi-scale analysis strategy and performs feature analysis simultaneously on three time scales of 100 milliseconds, 250 milliseconds and 500 milliseconds. The short time scale (100 milliseconds) mainly identifies transient change characteristics: calculates the first-order difference and second-order difference of the signal, detects amplitude mutation points (mutation amplitude exceeds 3 times the mean), extreme point distribution (adjacent extreme point interval is less than 50 milliseconds) and fast oscillation (frequency is greater than 40Hz). The medium scale (250 milliseconds) focuses on the transition characteristics of the signal: analyzes waveform morphology changes (waveform factor change rate exceeds 20%), energy distribution migration (main frequency position offset exceeds 2Hz) and baseline drift trend (drift rate exceeds 2μV / ms). The long time scale (500 milliseconds) analyzes the overall characteristics of the signal: evaluates the steady-state properties of the signal (variance stability index), spectral structure (energy distribution entropy) and channel correlation (cross-correlation function change). At each scale, when the rate of change of the characteristic parameter exceeds the set threshold (30% for short scale, 25% for medium scale, and 20% for long scale), the method marks it as a potential segmentation point. Then the segmentation points of the three scales are integrated through the density clustering algorithm (clustering radius 50 milliseconds), and the cluster center is determined as the final segmentation position. Each segment inherits the characteristic information of the original processing interval and adds the feature description of the segmentation level, including the local statistical characteristics of the signal, the frequency domain structure characteristics, and the spatial distribution characteristics.

[0059] Based on the characteristic distribution of segmented sequences, the method selects corresponding processing markers for each segment. The marker information is divided into three levels: the first level interference suppression parameters, including filter type selection (FIR / IIR), cutoff frequency setting (automatically adjusted according to the interference frequency band, 50Hz notch for power frequency interference, 30-100Hz band stop for electromyographic interference), filter order (8-32 orders adjustable), window function type (Hanning / Hamming / Blackman), filter bandwidth (affected by signal quality, range 0.5-5Hz). The second level signal enhancement parameters, according to the segment signal quality (scoring range 0-1), set the gain adjustment coefficient (0.8-1.2 times adjustable), baseline drift correction threshold (±50μV), phase compensation angle (±30 degrees range). The third level correction control parameters, control the smooth transition of processing intensity between segments, including processing weight factor (0-1 linear change), smoothing window length (20-100ms adjustable), transition zone overlap ratio (30%-50% adjustable). The method extracts the best matching processing template from the interference pattern library, and locally optimizes the template parameters according to the actual characteristics of the current segment: the parameter adjustment step size changes dynamically with the signal stability, and the adjustment range is limited to the typical characteristic range of the physiological signal.

[0060] According to the segmented processing tag sequence, the method generates the final correction sequence. For each segment, the basic processing specified by the tag is first performed: an adaptive filter is used to remove interference components, and the filter coefficient is updated every 10ms; wavelet transform is used for multi-scale denoising, and the optimal number of decomposition layers (3-5 layers) is selected; narrowband interference is processed by spectral line suppression technology. Then the processed signal segment is boundary smoothed: a 20ms transition interval is established between adjacent segments using cubic spline interpolation, and the interpolation node density is adaptively adjusted with the signal change rate (5-20 / ms). During the processing process, the method ensures that the basic characteristics of the signal remain in a reasonable range by real-time monitoring of eight physiological characteristic indicators of the signal (such as α wave proportion, β wave fluctuation, etc.). For multi-channel data, spatial correlation is maintained during correction through cross-correlation analysis between channels (correlation window 100ms) and phase synchronization evaluation (synchronization tolerance ±5ms). The final correction sequence removes interference while maintaining the temporal continuity and spatial consistency of the signal. Each data point in the correction sequence is accompanied by processing quality indicators, including quantitative parameters such as signal-to-noise ratio improvement (≥6dB), waveform distortion (≤5%), and phase offset (≤10 degrees), providing an objective evaluation standard for the processing effect.

[0061] In some embodiments, constructing the adaptive benchmark based on the correction sequence includes: performing multi-scale decomposition on the correction sequence to obtain signal components; performing frequency band division on the signal components, and obtaining multiple feature markers according to the division results corresponding to the frequency band division; and constructing the adaptive benchmark according to the feature markers.

[0062] Methods Multi-scale decomposition is performed on the correction sequence output by the step. The decomposition process first analyzes the quality indicators of the correction sequence: the signal segment with a signal-to-noise ratio higher than 6dB uses 8-layer wavelet packet decomposition to fully retain the signal details; the signal segment with a signal-to-noise ratio of 3-6dB uses 7-layer decomposition to achieve a balance between signal reconstruction and noise control; the signal segment with a signal-to-noise ratio lower than 3dB uses 6-layer decomposition to avoid the introduction of additional noise in the decomposition process. Methods The db4 wavelet basis function is used for decomposition, and the decomposition parameters are dynamically adjusted according to the signal characteristics during the processing process: when the signal amplitude change rate exceeds 30% / ms, the time window is reduced to 50ms, and the threshold coefficient is set to 0.6 to enhance the ability to capture rapid changes; when the signal is relatively stable (amplitude fluctuations are less than 10%), the time window is expanded to 200ms, and the threshold coefficient is increased to 0.8 to improve the frequency resolution accuracy. For the transition area in the correction sequence, a double-window decomposition strategy is adopted, with a main window length of 100ms, an auxiliary window of 50ms, and an overlap rate of 75%, to achieve a smooth transition of the decomposition results. In each layer of decomposition, the method combines the phase offset information of the correction sequence (from step 1) for phase calibration, and starts phase compensation when the offset is greater than 5 degrees. Finally, 256 signal components of frequency subbands are obtained, each of which contains a complete set of parameters such as center frequency, bandwidth, energy distribution, phase characteristics, and time-frequency resolution.

[0063] Based on 256 signal components, the method performs fine frequency band division. The division process first applies the step correction quality score to screen the components. Components with scores higher than 0.8 are directly involved in the division, and components with scores between 0.6 and 0.8 need to undergo additional verification. The method constructs the basic frequency band structure: the δ band (0.5-4Hz) uses a division accuracy of 0.5Hz, and each sub-band is configured with 4 signal components; the θ band (4-8Hz) uses a division accuracy of 0.8Hz, and each sub-band is configured with 6 components; the α band (8-13Hz) uses a division accuracy of 1Hz, and each sub-band is configured with 8 components; the β band (13-30Hz) uses a division accuracy of 2Hz, and each sub-band is configured with 12 components; the γ band (30-50Hz) uses a division accuracy of 5Hz, and each sub-band is configured with 16 components. For energy-concentrated areas (energy proportion exceeds 15%), the method automatically improves the division accuracy: the δ band accuracy is improved to 0.25Hz, the θ and α bands are improved to 0.5Hz, the β band is improved to 1Hz, and the γ band is improved to 2.5Hz. The method calculates the energy transfer characteristics between each frequency band in real time, and uses a short-time window of 100ms and a long-time window of 500ms to double track the energy flow. Each subdivided frequency band obtains a dynamic weight according to its signal quality, and the weight calculation considers three indicators: signal-to-noise ratio, phase stability, and energy concentration. In the frequency band transition area, the method introduces a 20% overlap interval and achieves a smooth transition through energy-weighted averaging.

[0064] Combined with the results of frequency band division, the method constructs a multi-level feature labeling system. The dynamic characteristics of each frequency band signal are analyzed at the time domain level: 20 feature points are extracted from the envelope curve, including the main peak position, peak spacing, peak ratio, and waveform factor; the morphological analysis calculates the symmetry index (range 0-1), sharpness coefficient (range 1-5), and flatness factor (range 0.5-2); the stability evaluation includes the fluctuation period (minimum resolution 20ms) and amplitude change rate (accuracy 0.1% / ms). The energy characteristics are extracted at the frequency domain level: the absolute energy value of each frequency band, the energy ratio of adjacent frequency bands (set 5 ratio thresholds), the energy concentration index (3 scoring levels), and the main frequency drift rate (accuracy 0.1Hz / s) are calculated. The time-frequency joint analysis includes: energy transfer rate (8 directional components), migration path characteristics (16 typical modes), and accumulation curve parameters (4 key inflection points). The spatial feature analysis includes: propagation velocity vector (accuracy 0.1m / s), phase difference spectrum (resolution 1 degree), and synchronization strength index (quantized into 10 levels). The method normalizes and orthogonalizes all features and constructs a 600-dimensional feature space. Each feature contains three attributes: eigenvalue, time-varying degree, and credibility. Based on the step quality score, the method establishes a three-level dynamic weighting mechanism for the features, and the weight update cycle is 50ms.

[0065] Based on the complete feature labeling, the method establishes an adaptive benchmark structure. In the 600-dimensional feature space, 32 core reference points are first determined by cluster analysis, and each reference point represents a typical signal state combination. The distribution of reference points takes into account multiple influencing factors: cognitive state (divided into 6 levels), environmental conditions (8 typical scenarios), and individual characteristics (12 combination modes). The method uses an improved radial basis function to establish a mapping network, and the mapping parameters are adjusted in real time as the features change: stable features (variance <10%) use a large radius (0.8) to obtain a smooth transition; drastically changing features (variance >30%) use a small radius (0.4) to improve the response speed; and medium-changing features use a linear interpolation radius. The method updates the feature distance matrix every 50ms, and the matrix dimension is 600×32. The adaptive weight is calculated in combination with the step quality index. Through multi-layer weighted combination, a complete benchmark system including time domain benchmark values ​​(128 sampling points), frequency domain benchmark spectra (64 frequency points), and spatial benchmark patterns (16 channel relationships) is generated. The benchmark update adopts a sliding average strategy and sets three time scales: fast response (weight 0.4, period 100ms), medium-term stability (weight 0.35, period 500ms), and long-term trend (weight 0.25, period 2s). This multi-time scale adaptive mechanism enables the method to accurately track the dynamic changes of signal characteristics while maintaining a reliable memory of stable characteristics.

[0066] Step S104, reconstructing based on the adaptive benchmark to obtain an EEG feature sequence and a time-varying feature set corresponding to the EEG feature sequence.

[0067] Specifically, the method uses the adaptive benchmark established in the steps to reconstruct the signal. The reconstruction process first calls the time domain benchmark value, frequency domain benchmark spectrum and spatial benchmark mode in the benchmark. These benchmark parameters guide the overall direction of reconstruction. The method uses a 100ms main processing window and a 50ms auxiliary window to hierarchically reconstruct the signal: at the time domain level, the signal profile is established based on the 128 sampling points of the benchmark, and the continuity of the reconstructed signal is ensured by cubic spline interpolation. The density of interpolation nodes is adjusted between 5-20 / ms according to the signal change rate; at the frequency domain level, the energy distribution of 64 frequency points in the benchmark is used to reconstruct the spectral structure, focusing on maintaining the energy ratio of each frequency band of δ, θ, α, β, and γ, where the energy deviation of the α band is controlled within 5%, and the energy deviation of other bands is controlled within 10%; at the spatial level, the spatial distribution characteristics of the signal are reconstructed based on 16 channel relationship modes, and the phase difference between channels is controlled within ±10 degrees. During the reconstruction process, the method dynamically adjusts the reconstruction parameters according to the credibility of the benchmark: the parameters with credibility higher than 0.8 are directly applied, and the parameters with credibility between 0.6-0.8 need to be corrected in combination with historical data, and the correction weight is determined by the current signal quality. The method improves the reconstruction quality through iterative optimization, and sets three iteration cycles: the first focuses on the overall outline of the signal, the second optimizes the spectrum structure, and the third adjusts the spatial distribution. After each iteration, the reconstruction effect is evaluated by the matching degree with the benchmark mode, and the matching degree threshold is set to 0.85. The feature sequence generated after reconstruction not only contains the complete time-frequency-space signal information, but also records the key parameters and quality indicators of the reconstruction process.

[0068] The reconstructed feature sequence was screened. The method first analyzed the time-frequency components in the feature sequence: the time domain parameters (peak, root mean square, standard deviation, etc.) were calculated using a 50ms short-time window, the spectral structure was obtained using 256-point FFT, and the time-frequency distribution was obtained by wavelet transform. Based on the analysis results, the method performed three-level screening: first, the reconstruction quality index was used for preliminary screening, and the components with quality index lower than 0.6 were marked as pending; then the time-frequency characteristics of the components were analyzed, and the components with energy concentration higher than 0.7 and stable spectral structure were retained; finally, the correlation between components was evaluated, and the components with correlation coefficient higher than 0.85 were merged. For the five main frequency bands of δ, θ, α, β, and γ, the method set differentiated screening criteria: the δ band requires a signal-to-noise ratio higher than 15dB, the θ band focuses on spatial correlation (inter-channel correlation coefficient>0.6), the α band focuses on stability (fluctuation coefficient<0.2), the β band mainly evaluates the energy proportion (>25%), and the γ band focuses on time continuity (breakpoint rate<5%). The screening process uses a 250ms main window and a 125ms auxiliary window for sliding analysis, and the window overlap rate is kept at 50%. For the components at the window boundary, a 20ms transition interval is introduced to maintain the continuity of the screening results through weighted averaging.

[0069] Time-varying features were generated based on the component screening results. Methods A time-varying feature system was constructed for the retained high-quality components: the base layer recorded the instantaneous distribution of energy in each frequency band with a period of 50 ms, including absolute energy value (μV²), relative energy ratio (%) and energy change rate (% / ms). Ten monitoring points were set for each frequency band, and a continuous energy change curve was obtained by piecewise linear interpolation. The middle layer used a 200 ms analysis window to examine the dynamic relationship between frequency bands: parameters such as energy ratio, phase difference, and synchronization strength of each frequency band were calculated to identify the key moment of energy transfer, and the transfer threshold was set to 20% of the average energy. The top layer used a 500 ms observation window to extract macroscopic features: dominant frequency band switching mode (switching time and sequence), rhythm stability (duration and fluctuation range) and state transition characteristics (conversion speed and integrity). The features of each layer were integrated through adaptive weights, and the weight coefficients were dynamically adjusted with the signal state: the base layer weight was increased in the rapid change stage (feature change rate>20% / 50ms); the top layer weight was increased in the stable stage (feature change rate<5% / 200ms). The method calculates three key indicators for each time-varying feature: feature value (reflecting the current state), change rate (indicating the change trend) and credibility (evaluating the reliability of the feature). During the feature extraction process, an anomaly detection algorithm (based on mean ± 3 times standard deviation) is used to identify and correct outliers to ensure the continuity and reliability of the time-varying features.

[0070] Step S105, obtaining the attention index corresponding to the time-varying feature set, performing feature extraction on the attention index, and obtaining the core feature set.

[0071] Specifically, the method performs multi-level pattern analysis on the time-varying features of the step output. The analysis process first uses 200ms as the basic window to scan the time-varying features, and the window overlap rate is set to 50% to ensure feature continuity. The method focuses on the energy distribution changes of α waves (8-13Hz) and β waves (13-30Hz) that reflect cognitive activities, and uses θ waves (4-8Hz) as a reference for interference judgment. In each scanning window, the method calculates the following characteristic parameters: α / β energy ratio (the value is in the range of 0.8-1.2 when attention is focused), β wave energy change rate (the slope is greater than 0.1μV² / ms when concentration is improved), energy proportion of each frequency band (α+β wave energy proportion greater than 60% indicates active cognitive activity), and θ wave suppression rate (θ wave energy proportion less than 15% indicates that interference is effectively controlled). At the same time, a weighted analysis is performed in combination with the step feature credibility index. The feature weight with a credibility higher than 0.8 is set to 1.0, the feature weight in the range of 0.6-0.8 is linearly attenuated, and the feature weight lower than 0.6 is set to 0. Methods The feature mutation points were identified by sliding comparison, and three-level judgment criteria were set: when the feature change rate of three consecutive windows exceeded 30%, it was marked as a strong mutation point; when the change rate was between 15% and 30%, it was marked as a medium mutation point; and when the change rate was between 5% and 15%, it was marked as a weak mutation point. The identified attention features were subjected to improved K-means clustering analysis, and the number of cluster centers was set to 5, corresponding to five states: high concentration, general concentration, relaxation, mild fatigue, and severe fatigue. Each state has a specific combination of spectral features. For example, the high concentration state is characterized by the dominance of β wave energy and a stable α / β ratio, while the relaxation state is dominated by α waves. The method determines the attribution of the current state by calculating the Euclidean distance between the feature vector and the cluster center, and finally obtains a complete attention feature including state type, feature combination, significance, and stability.

[0072] In some embodiments, obtaining the attention index corresponding to the time-varying feature includes: performing pattern analysis on the time-varying feature to obtain the attention feature; constructing a state sequence based on the attention feature; partitioning and marking the state sequence; and generating the attention index based on the marking result corresponding to the partition mark.

[0073] Based on the identified attention features, the method constructs a dynamic state sequence. The construction process uses a 500ms main integration window and a 100ms auxiliary window to combine the attention features in continuous time into state units. Each state unit contains: dominant feature type (one of five basic types), energy distribution pattern of each frequency band (16-dimensional feature vector), state duration, state stability index (feature variance) and transition trend prediction. The method analyzes the transition relationship between adjacent state units and establishes a 5×5 state transition probability matrix. The matrix elements represent the transition tendency between different states, and the probability value is updated every 2 seconds. The state sequence is generated using a dual time scale: the short-term scale (1 second) focuses on analyzing feature fluctuations and sets a mean drift threshold of ±15%; the long-term scale (10 seconds) focuses on the state evolution trend and calculates the trend coefficient through linear regression. The method uses a second-order Markov chain model to describe the dynamic characteristics of state transitions, and the transition probability threshold is set to 0.3. When a sudden change in state is detected (the feature deviates from the mean by more than 2 standard deviations), the method initiates three levels of verification: the first level requires that the new state lasts for more than 1 second, the second level checks the stability of the feature (the fluctuation is less than 10%), and the third level verifies the spatial consistency (the correlation coefficient between channels is greater than 0.7). Each state in the state sequence carries a complete time series mark, including the start time, duration, previous state, and predicted trend.

[0074] Combined with the state sequence, the method performs fine partition labeling. The labeling process uses a 1-second main processing window and a 200-ms transition window with a window overlap rate of 75%. The method first identifies the characteristic intervals in the state sequence: stable intervals (state retention time exceeds 2 seconds and characteristic fluctuations are less than 15%), transition intervals (state transition process, characteristic continuous change) and fluctuation intervals (characteristic fluctuations are drastic but do not meet the state transition criteria). For stable intervals, the method performs five-level attention labeling: high concentration (α / β ratio 1.0±0.1, β wave energy share >40%), high concentration (ratio 0.9-1.1, β wave share 30-40%), general concentration (ratio 0.7-1.3, β wave share 20-30%), distraction (ratio unstable or β wave share <20%), and inattention (theta wave share significantly increased >25%). The transition interval focuses on marking the state change characteristics: change rate (fast: <500ms, medium: 0.5-1s, slow: >1s), change direction (increase, decrease, fluctuation), transition mode (jump, gradual change, oscillation). The method also records the key events that affect the state change: external interference (based on the abnormality of theta wave), fatigue accumulation (based on the long-term trend of the α / β ratio), and state recovery (the process of feature regression baseline). Each partition marker contains information such as timestamp, state attributes, change characteristics, and event association.

[0075] According to the results of partition labeling, the method generates a comprehensive attention index system. The index design adopts a three-layer architecture: the instantaneous index sampling period is 100ms, and the weight coefficient is determined by the feature credibility through weighted average calculation of α / β energy ratio; the cumulative index update period is 1 second, and the state duration (weight 0.4), conversion frequency (weight 0.3) and fluctuation amplitude (weight 0.3) are comprehensively considered; the trend index statistical period is 10 seconds, combining the conversion characteristics of the state sequence and the linear prediction results. The method adopts a standardized scale of 0-100 to convert the indicators of each level into quantitative values: 90-100 indicates extremely high attention, 80-90 indicates high attention, 70-80 indicates good level, 60-70 indicates general level, and below 60 indicates insufficient attention. The dynamic weight mechanism is introduced in the index calculation: when the feature credibility is greater than 0.9, the weight is 1.0, the weight in the range of 0.7-0.9 is linearly decayed, and the features below 0.7 are not included in the calculation. Methods: We established a correlation analysis between attention level and external conditions, recording influencing factors such as environmental factors (based on theta wave abnormality), physical state (based on the trend of alpha wave changes), cognitive load (assessed by beta wave energy), etc. Through this multi-dimensional indicator system, we generated a complete attention index including numerical value, credibility and change rate.

[0076] The method extracts multi-dimensional features from the attention indicators generated by the steps. The extraction process uses a 500ms main analysis window and a 100ms auxiliary window, and the window overlap rate is set to 75%. At the instantaneous level, the method extracts the statistical characteristics of the indicators: mean level (based on 0-100 standardized values), maximum fluctuation amplitude (difference between adjacent sampling points), average change rate (change per unit time), fluctuation period (peak interval), zero crossing rate (number of zero crossing points per unit time), waveform factor (ratio of effective value to average value), peak factor (ratio of peak value to effective value), skewness coefficient (distribution deviation), kurtosis coefficient (distribution steepness), cumulative growth rate (integral value per unit time). At the state level, the stability parameters are extracted: duration (indicator stability time), recovery time (return to benchmark time), fluctuation frequency (number of times crossing the threshold), energy concentration (distribution density), rise time (time to reach steady state), fall time (time to decay to threshold), steady-state error (steady-state deviation value), and adjustment time (time to adapt to the new state). At the same time, the timestamp, credibility, and original attention indicator value of each feature are recorded for subsequent selection.

[0077] The results of feature extraction were selected methodologically. The selection process first set the screening criteria according to the credibility of the extracted features: features with a credibility higher than 0.8 were directly included in the feature set, which mainly included stability indicators such as mean level, fluctuation period, duration, and fluctuation frequency; features with a credibility between 0.6 and 0.8 were included in the correlation analysis, including transition features such as rise time, adjustment time, and energy concentration; features with a credibility lower than 0.6 were directly eliminated, such as some instantaneous fluctuation parameters. Methods The correlation coefficient matrix between the retained features was calculated. When the correlation coefficient between any two features exceeded 0.85, the features with higher credibility were retained. At the same time, the feature distribution matrix was constructed, and the discrimination ability was evaluated by variance analysis. Features with a variance contribution rate of more than 0.6 were retained first. Methods A 1-second sliding window was used to analyze the temporal stability of the features, and the coefficient of variation within 10 consecutive windows was calculated. Features with a coefficient of variation less than 0.3 were retained. The selection process adopted a 100-point scoring system: credibility score (40 points), discrimination score (30 points), and stability score (30 points). The specific score and selection basis of each retained feature were recorded.

[0078] The feature set is constructed based on the selected features. The feature set adopts a three-layer architecture, and all features come from the selection results. The first layer of time series features contains 10 selected core statistical parameters: mean level, fluctuation amplitude, change rate, fluctuation period, zero crossing rate, waveform factor, peak factor, skewness coefficient, kurtosis coefficient, and energy concentration. The credibility of these features is higher than 0.8, which is used to reflect the immediate state change. The second layer of statistical features contains 8 selected state parameters: duration, recovery duration, fluctuation frequency, energy concentration, rise time, fall time, steady-state error, and adjustment time. These features have passed correlation analysis and stability evaluation and are used to describe the medium-term performance. The third layer of trend features is formed by combining the selected features: the fatigue accumulation feature is obtained by combining the duration and steady-state error, the recovery time and rise time are combined to obtain the recovery ability feature, and the fluctuation period and adjustment time are combined to obtain the adaptability feature. The method achieves deep fusion between features: the stability of attention is evaluated by the product and ratio combination of time series features, the medium-term performance is reflected by the weighted combination of statistical features, and the change trend is predicted based on the linear combination of trend features. Each feature combination inherits the credibility attribute of the original feature, and the credibility of the combined feature is the weighted average of the credibility of the original feature. The method performs normalization on all features and feature combinations to establish a unified numerical range. At the same time, the method records the complete evolution process of each feature, including the change trajectory of the feature value over time, the distribution of mutation points, and periodic patterns. For high-order feature combinations, the method establishes a feature map to describe the correlation structure and influence level between features, and tracks the dynamic changes of these correlations through a sliding time window. This multi-level and multi-dimensional feature fusion structure enables the method to comprehensively and accurately characterize the attention change characteristics in various scenarios.

[0079] Step S106, constructing a mapping rule according to the core feature set, generating a state feature according to the mapping rule, and constructing a pattern for the state feature.

[0080] Specifically, the method performs multi-level transformation processing on the step feature set. The transformation uses a 300ms main processing window and a 100ms auxiliary window with a window overlap rate of 75%. The time series feature transformation uses four basic forms for the 10 core statistical parameters selected in the step: amplitude normalization maps the feature value to the [-1,1] interval, and the normalization threshold of each feature is determined according to the historical distribution; phase adjustment is based on FFT phase correction to deal with the delay effect between features; frequency mapping converts the feature period to a standard frequency for easy comparison across features; and energy conversion reconstructs the energy distribution of the feature. Structural transformation is performed on 8 statistical features: linear transformation adjusts the feature distribution characteristics; nonlinear transformation enhances feature discrimination; and orthogonal transformation eliminates feature redundancy. Combination transformation is performed on trend features formed by feature combinations: cumulative transformation reflects long-term change trends, differential transformation captures mutation characteristics, and integral transformation smoothes fluctuations. The method generates transformation tags for each transformation process, recording information such as transformation type, parameter settings, and effect evaluation. The transformation tag adopts a three-layer structure: the basic tag records the transformation parameters of a single feature, the associated tag describes the transformation relationship between features, and the combined tag characterizes the overall transformation characteristics of the feature group.

[0081] In some embodiments, constructing a mapping rule according to the core feature set includes: performing transformation processing on the core feature set to obtain a transformation mark; and constructing a mapping rule according to the transformation mark.

[0082] Mapping rules were constructed based on the transformation marks obtained by transformation processing. Methods The transformation effects recorded in the transformation marks were analyzed, and the distribution characteristics after feature transformation were mapped to the corresponding state meanings. State mapping rules were constructed based on basic marks: the time series features after amplitude normalization were mapped to high concentration in the interval [0.8,1], general concentration in [0.6,0.8), critical state in [0.4,0.6), mild fatigue in [0.2,0.4), and severe fatigue in [0,0.2]. Trend mapping rules were established using associated marks: feature change rate greater than 0.1 / s was mapped to rapid improvement of attention, change rate in the interval [-0.02,0.02] was mapped to state maintenance, and change rate in the interval [-0.05,-0.02] was mapped to fatigue accumulation. Mode mapping rules were constructed based on combination marks: the focus learning mode requires the stability of the time series features to be greater than 0.8 and the fluctuation of the statistical features to be less than 0.2; the corresponding feature combination of the thinking and research mode presents obvious periodic changes; the rest and relaxation mode is characterized by the regression of the feature value to the baseline level. The method uses a 500ms sliding window to update the mapping parameters in real time to ensure that the mapping rules can adapt to different scenarios. The update of mapping parameters is based on the quantitative evaluation of the previous transformation effect, and the parameter adjustment is triggered when the mapping accuracy is lower than 85%.

[0083] The state features are generated using the established mapping rules. First, short-term state features are generated based on the state mapping rules, with a sampling period of 100ms: the transformed time series features are mapped to the instantaneous attention level (0-100 standard points), and the feature stability greater than 0.8 is mapped to high concentration; the mid-term state features are obtained through the trend mapping rules, with an update period of 1s: the duration feature is mapped to the work continuity ability, the recovery duration feature is mapped to the adjustment ability, and the fluctuation feature is mapped to the anti-interference ability; the long-term state features are generated using the pattern mapping rules, with a statistical period of 10s: the continuous work ability is predicted based on the fatigue accumulation feature, the rest effect is evaluated based on the recovery ability feature, and the state development trend is predicted based on the adaptability feature. All state features inherit the credibility attributes of the original features and obtain an enhanced confidence level through feature combination. The state feature generation process sets an abnormality detection mechanism. When the feature value deviates from the historical range by more than 2 standard deviations, a 500ms review window is started for confirmation.

[0084] The generated state features are pattern constructed. Methods First, the immediate mode is identified based on the short-term state features: when the attention level is greater than 85 points and the stability is higher than 0.9, it is marked as a deep focus mode; when the attention level fluctuates between 70 and 85 points, it corresponds to a general work mode; when the attention level shows obvious periodic changes, it is identified as a thinking and research mode. Then, the continuous mode is constructed in combination with the mid-term state features: when both the work continuity and anti-interference capabilities are high, it is established as an efficient work mode; when the adjustment ability is dominant, it is determined to be an adaptive mode. Finally, a prediction model is established through long-term state features: based on the fatigue accumulation trend, the continuous work limit is predicted, the best rest time is evaluated based on the recovery ability, and the work rhythm is optimized using adaptive features. Methods Each mode is evaluated in real time: a 1s main window and a 200ms auxiliary window are used to continuously calculate the matching degree between the current state and each mode. The matching degree is based on the weighted Euclidean distance of the state features, and the weight is dynamically adjusted according to the predictive ability of the features. Pattern construction supports adaptive updates: when a new feature combination appears more than 3 times per hour and lasts for more than 10 minutes, the method summarizes it into a new sub-pattern; when the matching rate of a pattern is less than 20% for 5 consecutive days, the pattern enters the state to be optimized. Through this state-feature-based pattern construction mechanism, the method achieves accurate characterization and dynamic tracking of the user's attention state.

[0085] Step S107, performing adaptive processing according to the construction result corresponding to the pattern construction, generating a state mark sequence, outputting an adaptive signal according to the state mark sequence, and completing the adaptive processing of the original EEG signal.

[0086] Specifically, the method first adaptively processes the four types of patterns and their feature combinations output by the step. The processing uses a 200ms main window and a 50ms auxiliary window, with a window overlap rate of 80%. According to the characteristic distribution of different modes of the step, the method sets differentiated processing parameters: for the deep focus mode, when the attention level is greater than 85 points and the stability is higher than 0.9, the α / β band energy ratio threshold is set to 1.2, and the attention level fluctuation tolerance is ±8 points; for the general working mode (attention level 70-85 points), the energy ratio threshold is adjusted to 1.3, and the fluctuation tolerance is relaxed to ±10 points; in the thinking and research mode (attention level changes periodically), the minimum energy proportion of the β band is set to 25%. The method also sets an adaptive range for the processing parameters: according to the historical distribution of pattern characteristics, the attention level baseline value is adjustable in the range of 60-90 points, and the energy ratio threshold fluctuates in the range of 0.8-2.0. For continuous working conditions, the method combines the step fatigue prediction mode to automatically adjust the processing sensitivity: after 1 hour of continuous working, the fluctuation tolerance is tightened by 10%; after 2 hours, it is tightened by 20%; after 3 hours, it is tightened by 30%. The processing results include: the optimized parameter set of each mode, the adjustment threshold of the state characteristics, and the boundary conditions of the adaptive range.

[0087] Marking adjustment is performed based on the processing results. Methods The state is reclassified through the optimized parameters and thresholds, and three-layer marking rules are set. The first layer of instant state marking: when the attention level is in the high concentration interval defined by the optimized parameters and the duration exceeds 25 cycles (5 seconds) of the processing window, it is marked as "deep concentration"; when the feature combination meets the optimization conditions of the general working mode and lasts for 15 cycles, it is marked as "normal working"; when there is a periodic change that conforms to the thinking research mode (the cycle length is within the optimization range), it is marked as "thinking state". The second layer of trend marking: based on the set adaptive range, the feature change rate is analyzed. When the change rate exceeds the optimized upper threshold, it is marked as "rapid improvement", and when it is lower than the lower threshold, it is marked as "attention attenuation", and when it is in the threshold interval, it is marked as "stable state". The third layer of periodic marking: according to the optimized fatigue prediction parameters, the state characteristics of different time periods are periodically evaluated to generate "physiological cycle" markings. During the marking process, the applicable effects of the optimization parameters are recorded synchronously. When the accuracy of a certain type of marking (verified by feature stability) exceeds 90%, the marking is used to update the optimization parameters; when the accuracy is between 70% and 90%, parameter fine-tuning is triggered; when it is lower than 70%, parameter optimization is performed again. Finally, a complete state marking sequence containing timestamps, state attributes, optimization parameters, and accuracy is generated.

[0088] The method outputs adaptive signals according to the generated state marker sequence. In the temporal dimension, multi-level output is achieved based on the hierarchical structure of the marker sequence: an immediate signal is output every 100ms, which includes the attention level defined by the immediate state marker, the optimized dominant EEG features and the state judgment results; a trend signal is output every 1 second, and the attention change trend and short-term prediction results are calculated based on the trend marker sequence; a comprehensive signal is output every 10 seconds, and the stage performance is evaluated in combination with the periodic marker. In the spatial dimension, the method integrates multi-channel information according to the optimization parameters corresponding to different state markers: the prefrontal electrodes (Fp1, Fp2) provide the main attention features, and the weight ratio of the α and β band data when marked as deep concentration state is set according to the optimization results; the temporal electrodes (T3, T4) are used to monitor external interference, and the analysis weight of the θ band is increased when the marker sequence indicates state fluctuation; the signals of the occipital electrodes (O1, O2) are used for baseline correction of vision-related tasks. The signal processing strategy is dynamically adjusted according to the state indicated by the tag sequence: when in the "deep focus" tag, the method enhances the beta band weight and increases the sampling rate to 1024Hz; when marked as "thinking state", the balanced weight distribution of the alpha and beta bands is maintained; when the "attention decay" tag appears, the method quickly adjusts the processing parameters, improves the sampling accuracy, and shortens the state judgment window to 100ms. When a state mutation in the tag sequence is detected, the method automatically optimizes the processing strategy based on the tag attributes before and after the mutation. The final output adaptive signal fully records the attention change characteristics based on the state tag sequence, realizing the accurate evaluation and dynamic tracking of the user's cognitive state.

[0089] In order to implement the electroencephalogram signal adaptive processing method of the attention glasses corresponding to the above method embodiment, so as to achieve the corresponding functions and technical effects. Figure 2 , Figure 2 The structure block diagram of an electroencephalogram signal adaptive processing device 200 provided in an embodiment of the present application is shown. For the convenience of description, only the parts related to the present embodiment are shown. The electroencephalogram signal adaptive processing device 200 provided in an embodiment of the present application includes:

[0090] The signal acquisition module 201 is used to acquire the original EEG signals collected by the multi-channel electrodes of the attention glasses, and generate a standardized signal sequence after performing channel analysis on the original EEG signals;

[0091] An interference detection module 202 is used to perform interference detection on the standardized signal sequence, obtain a variation interval corresponding to the signal fluctuation, and construct an interference pattern library according to the variation interval;

[0092] A sequence generation module 203, configured to determine a processing interval according to the standardized signal sequence and the interference pattern library, generate a correction sequence according to the processing interval, and construct an adaptive benchmark according to the correction sequence;

[0093] A reference reconstruction module 204 is used to reconstruct based on the adaptive reference, obtain an EEG feature sequence, and obtain a time-varying feature set corresponding to the EEG feature sequence;

[0094] A set acquisition module 205 is used to acquire the attention index corresponding to the time-varying feature set, perform feature extraction on the attention index, and acquire a core feature set;

[0095] A rule construction module 206, configured to construct a mapping rule according to the core feature set, generate a state feature according to the mapping rule, and perform pattern construction on the state feature;

[0096] The processing completion module 207 is used to perform adaptive processing on the construction result corresponding to the pattern construction, generate a state mark sequence, output an adaptive signal according to the state mark sequence, and complete the adaptive processing of the original EEG signal.

[0097] The above-mentioned electroencephalogram signal adaptive processing device 200 can implement the electroencephalogram signal adaptive processing method of the attention glasses of the above-mentioned method embodiment. The optional items in the above-mentioned method embodiment are also applicable to this embodiment and will not be described in detail here. The rest of the contents of the embodiment of this application can refer to the contents of the above-mentioned method embodiment, and will not be repeated in this embodiment.

[0098] Figure 3 This is a schematic diagram of the structure of a computer device provided in one embodiment of the present application. Figure 3 As shown, the computer device 3 of this embodiment includes: at least one processor 30 ( Figure 3 Only one is shown in the figure), a memory 31 and a computer program 32 stored in the memory 31 and executable on the at least one processor 30, wherein the processor 30 implements the steps of any of the above method embodiments when executing the computer program 32.

[0099] The computer device 3 may be a computing device such as a smart phone, a tablet computer, a desktop computer, a cloud server, etc. The computer device may include but is not limited to a processor 30 and a memory 31. Those skilled in the art will understand that Figure 3 It is only an example of computer device 3 and does not constitute a limitation on computer device 3. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components, for example, it may also include input and output devices, network access devices, etc.

[0100] The processor 30 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0101] In some embodiments, the memory 31 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In 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 an internal storage unit and an external storage device of the computer device 3. The memory 31 is used to store an operating system, an application program, a boot loader (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.

[0102] In addition, an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0103] An embodiment of the present application provides a computer program product. When the computer program product is run on a computer device, the computer device implements the steps in the above-mentioned method embodiments when executing the computer device.

[0104] In several embodiments provided in the present application, it is understood that each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved.

[0105] If the 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 the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0106] The specific embodiments described above further describe the purpose, technical solutions and beneficial effects of the present application in detail. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the scope of protection of the present application. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of protection of the present application.

Claims

1. A method for adaptively processing EEG signals using attention glasses, characterized in that: include: Acquire the original EEG signals collected by the multi-channel electrodes of the attention glasses, and generate a standardized signal sequence after performing channel analysis on the original EEG signals; Performing interference detection on the standardized signal sequence, obtaining a variation interval corresponding to the signal fluctuation and constructing an interference pattern library according to the variation interval; Determine a processing interval according to the standardized signal sequence and the interference pattern library, generate a correction sequence according to the processing interval, and construct an adaptive benchmark according to the correction sequence; Reconstruct based on the adaptive benchmark to obtain an EEG feature sequence, and obtain a time-varying feature set corresponding to the EEG feature sequence; Obtaining an attention index corresponding to the time-varying feature set, performing feature extraction on the attention index, and obtaining a core feature set; Constructing a mapping rule according to the core feature set, generating a state feature according to the mapping rule, and constructing a pattern for the state feature; Adaptive processing is performed on the construction result corresponding to the pattern construction to generate a state mark sequence, and an adaptive signal is output according to the state mark sequence to complete the adaptive processing of the original EEG signal.

2. The method according to claim 1, characterized in that The generating a standardized signal sequence after performing channel analysis on the original EEG signal comprises: Acquire the contact impedance between the electrode and the skin in each acquisition channel according to the original EEG signal; Judging whether the working state of each of the acquisition channels is normal according to the contact impedance; Channel analysis is performed on the original EEG signal corresponding to the acquisition channel in normal working state to generate the standardized signal sequence.

3. The method according to claim 2, characterized in that Performing channel analysis on the original EEG signal corresponding to the acquisition channel in normal working state to generate the standardized signal sequence includes: Performing signal integrity analysis on the original EEG signal corresponding to the acquisition channel in normal working state; the signal integrity analysis at least includes signal continuity detection, amplitude stability assessment and baseline drift monitoring; The standardized signal sequence is generated according to the analysis result corresponding to the signal integrity analysis.

4. The method according to claim 1, characterized in that: The obtaining of a variation interval corresponding to the signal fluctuation and constructing an interference pattern library according to the variation interval includes: Obtaining a signal fluctuation position corresponding to the interference detection; Determining the variation interval according to the signal fluctuation position in the standardized signal sequence; Extracting features from the change interval to obtain change features; The interference pattern library is constructed according to the change characteristics.

5. The method according to claim 1, characterized in that: Generating a correction sequence according to the processing interval includes: Adaptively segmenting the processing interval; Determining a plurality of processing marks according to the segmentation results corresponding to the adaptive segmentation; The correction sequence is generated based on the processing mark.

6. The method according to claim 1, characterized in that The step of constructing an adaptive benchmark according to the correction sequence comprises: Performing multi-scale decomposition on the correction sequence to obtain signal components; Performing frequency band division on the signal component, and obtaining a plurality of characteristic markers according to a division result corresponding to the frequency band division; The adaptive benchmark is constructed based on the signature.

7. The method according to claim 1, characterized in that The obtaining of the attention index corresponding to the time-varying feature includes: Performing pattern analysis on the time-varying features to obtain attention features; constructing a state sequence according to the attention feature; Partition marking is performed on the state sequence; The attention index is generated according to the marking result corresponding to the partition mark.

8. The method according to claim 1, characterized in that: The constructing of mapping rules according to the core feature set includes: Performing transformation processing on the core feature set to obtain a transformation mark; A mapping rule is constructed according to the transformation mark.

9. An electroencephalogram signal adaptive processing device, characterized in that: include: A signal acquisition module, used to acquire the original EEG signals collected by the multi-channel electrodes of the attention glasses, and generate a standardized signal sequence after performing channel analysis on the original EEG signals; An interference detection module, used to perform interference detection on the standardized signal sequence, obtain a variation interval corresponding to the signal fluctuation and construct an interference pattern library according to the variation interval; A sequence generation module, configured to determine a processing interval according to the standardized signal sequence and the interference pattern library, generate a correction sequence according to the processing interval, and construct an adaptive benchmark according to the correction sequence; A reference reconstruction module, used for reconstructing based on the adaptive reference, obtaining an EEG feature sequence, and obtaining a time-varying feature set corresponding to the EEG feature sequence; A set acquisition module, used to acquire the attention index corresponding to the time-varying feature set, perform feature extraction on the attention index, and acquire a core feature set; A rule construction module, used to construct a mapping rule according to the core feature set, generate a state feature according to the mapping rule, and perform pattern construction on the state feature; The processing completion module is used to perform adaptive processing on the construction result corresponding to the pattern construction, generate a state mark sequence, output an adaptive signal according to the state mark sequence, and complete the adaptive processing of the original EEG signal.

10. A computer device, characterized in that: The method comprises 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