Multi-channel electroencephalogram signal feature extraction and classification method and device for brain-computer intelligent earphone

Abstract

The invention relates to the technical field of brain-computer interfaces, in particular to a brain-computer intelligent earphone multi-channel electroencephalogram signal feature extraction and classification method and device. The method comprises the following steps: acquiring a channel identifier corresponding to a multi-channel signal of the intelligent earphone, carrying out component extraction and dividing an analysis domain, outputting component features to carry out spatial positioning, acquiring a functional region to construct a regional network, and distributing a weight through the regional network; outputting a regional attribute data packet according to a weight distribution result corresponding to the distribution weight, establishing a structural model, and outputting a channel association network by a root; performing hierarchical construction on the channel association network, defining feature hierarchies, performing scale fusion on the feature hierarchies to form a feature group, generating a feature element set, performing classification combination to determine a classification dimension, and obtaining corresponding classification marks to perform credibility evaluation so as to establish an evaluation system and obtain a classification result; and multi-channel electroencephalogram signal feature extraction and classification are completed. And multi-scale characteristic decomposition and analysis domain division are realized.

Description

Technical Field

[0001] This application relates to the technical field of brain-computer interfaces, and particularly to a method and device for multi-channel electroencephalogram signal feature extraction and classification of a brain-computer intelligent headset. Background Art

[0002] The development of multi-channel electroencephalogram signal acquisition technology provides a rich source of neural information for brain-computer interface applications. However, in the application scenario of a portable brain-computer intelligent headset, multi-channel electroencephalogram data faces special technical challenges. First, the electrode distribution in the periauricular area is restricted by space, with a limited number of channels and dense distribution, resulting in serious inter-channel interference and information redundancy in the acquired signals. Second, the position offset during headset wearing will cause changes in the electrode-skin contact impedance, affecting the stability of signal quality. In addition, physiological noises such as electromyogram and electrooculogram and environmental electromagnetic interference in daily activities will also significantly reduce the signal-to-noise ratio.

[0003] In terms of signal processing, there are several key problems in the existing technologies. In signal acquisition and spatial positioning, traditional methods lack targeted consideration of the special distribution of periauricular electrodes, making it difficult to accurately locate and separate the signal sources of each channel. In the feature extraction link, a single data processing strategy is often adopted, failing to fully integrate the information in the time, frequency, and spatial domains, resulting in the extracted features lacking discriminability and stability. In the process of functional area recognition and feature classification, it is difficult to balance the real-time performance and accuracy of the algorithm, and there is a lack of a systematic feature fusion and screening mechanism.

[0004] Due to the portability requirements of intelligent headsets, the system needs to implement complex signal processing and feature analysis with limited computing resources. Especially in the classification and discrimination link, the existing methods lack a complete credibility evaluation system, making it difficult to quantitatively evaluate the reliability of classification results, which is prone to misjudgment in practical applications. At the same time, individual differences and the diversity of usage environments also pose great challenges to feature extraction and classification. Currently, there is a lack of a systematic solution that can simultaneously solve signal quality optimization, feature extraction enhancement, classification performance improvement, and credibility evaluation.

[0005] Therefore, there is an urgent need for a method to solve at least one of the above problems. Summary of the Invention

[0006] The embodiments of the present application provide a method and device for multi-channel electroencephalogram signal feature extraction and classification of a brain-computer intelligent headset. The method aims to solve the problem that due to the portability requirements of intelligent headsets, the system needs to implement complex signal processing and feature analysis with limited computing resources. Especially in the classification and discrimination link, the existing methods lack a complete credibility evaluation system and it is difficult to quantitatively evaluate the reliability of classification results, which is likely to lead to misjudgments in practical applications. At the same time, individual differences and the diversity of usage environments also pose great challenges to feature extraction and classification. Currently, there is a lack of a systematic solution that can simultaneously solve signal quality optimization, feature extraction enhancement, classification performance improvement, and credibility evaluation.

[0007] In a first aspect, the embodiments of the present application provide a method for multi-channel electroencephalogram signal feature extraction and classification of a brain-computer intelligent headset, including:

[0008] Collect multi-channel signals of the intelligent headset and determine the channel distribution, and obtain the channel identifiers corresponding to the channel distribution; perform component extraction on the channel identifiers and divide the analysis domain, and output component features based on the analysis domain;

[0009] Perform spatial positioning on the component features to obtain functional regions, construct a regional network according to the functional regions, and assign weights through the regional network; output a regional attribute data packet according to the weight assignment result corresponding to the assigned weight, establish a structural model according to the regional attribute data packet, so as to output a channel association network according to the structural model;

[0010] Perform hierarchical construction on the channel association network, define feature levels, perform scale fusion on the feature levels to form a feature group, output fusion elements according to the feature group, and generate a feature element set according to the fusion elements; perform classification combination on the feature element set to determine the classification dimension, and obtain the classification label corresponding to the classification dimension;

[0011] Perform credibility evaluation on the classification label to establish an evaluation system, output a discrimination label according to the evaluation system; perform data integration on the discrimination label to obtain a classification result, and complete multi-channel electroencephalogram signal feature extraction and classification.

[0012] In a second aspect, the present application also provides a computer device, including a processor and a memory, where the memory is used to store a computer program, and when the computer program is executed by the processor, it implements the method for multi-channel electroencephalogram signal feature extraction and classification of a brain-computer intelligent headset as described in the first aspect.

[0013] This method collects multi-channel electroencephalogram (EEG) signals through intelligent earphones, determines the channel distribution, and obtains the channel identifiers. Component extraction is performed on the channel identifiers, and after dividing the analysis domain, the component features are output. Spatial localization is performed on the component features to determine the functional regions and construct the regional network. Through dynamic weight allocation, regional attribute data packets are generated, and then the structural model of the channel association network is established. Hierarchical construction is performed on the channel association network. After defining the feature levels, feature groups are formed through multi-scale fusion to generate the feature element set. Classification and combination are performed on the feature element set to determine the classification dimensions and labels, and discriminant labels are generated through credibility evaluation. Finally, the data is integrated to output the classification result. By constructing the channel distribution and association network, the spatial correlation problem of multi-channel signals is solved. Based on the dynamic optimization of the network weights in the functional regions, the efficiency of feature extraction is improved. Combining multi-scale fusion technology enhances the correlation of time-frequency-spatial domain features.

[0014] Through the spatial localization and weight allocation of multi-channel signals, noise interference is reduced, and the accuracy of feature extraction is enhanced. Dynamic weight allocation and hierarchical feature fusion reduce redundant calculations and are suitable for real-time processing on embedded devices (such as intelligent earphones). The credibility evaluation system combines multi-dimensional classification labels to improve the robustness of EEG signal classification (such as intention recognition and disease diagnosis). It is applicable to medical and consumer electronics fields such as brain-computer interfaces, neurofeedback training, and epilepsy warning.

[0015] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit this application. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic flowchart of the method for extracting and classifying multi-channel EEG signal features of the brain-computer intelligent earphone shown in the embodiments of this application;

[0017] Figure 2 It is a schematic structural diagram of the device for extracting and classifying multi-channel EEG signal features of the brain-computer intelligent earphone shown in the embodiments of this application;

[0018] Figure 3 It is a schematic structural diagram of the computer device shown in the embodiments of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] In the following description, specific details such as specific system structures and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of this application. However, those skilled in the art should clearly understand that this application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from interfering with the description of this application.

[0020] It should be understood that, as used in the specification of this application and the appended claims, the term "comprising" indicates the presence of the described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or their groups.

[0021] It should also be understood that the term "and / or" as used in the specification of this application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0022] As used in the specification of this application and the appended claims, the term "if" may be construed, depending on the context, as "when" or "once" or "in response to determining" or "in response to detecting". Similarly, the phrase "if determined" or "if [the described condition or event] is detected" may be construed, depending on the context, as meaning "once determined" or "in response to determining" or "once [the described condition or event] is detected" or "in response to detecting [the described condition or event]".

[0023] In addition, in the description of the specification of this application and the appended claims, the terms "first", "second", "third", etc. are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0024] Reference to "one embodiment" or "some embodiments" or the like described in the specification of this application means that a particular feature, structure, or characteristic described in connection with the embodiment is included in one or more embodiments of this application. Thus, statements such as "in one embodiment", "in some embodiments", "in other some embodiments", "in still other embodiments", etc. that appear in different places in this specification are not necessarily all referring to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "comprising", "including", "having" and their variants all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0025] The technical solutions of the embodiments of this application will be introduced below.

[0026] The development of multi-channel electroencephalogram (EEG) signal acquisition technology has provided a rich source of neural information for brain-computer interface (BCI) applications. However, in the application scenario of portable BCI intelligent headphones, multi-channel EEG data faces special technical challenges. First, the electrode distribution in the periauricular region is restricted by space, with a limited number of channels and dense distribution, resulting in severe inter-channel interference and information redundancy in the acquired signals. Second, the position shift during headphone wearing will cause changes in the electrode-skin contact impedance, affecting the stability of signal quality. In addition, physiological noises such as electromyogram (EMG) and electrooculogram (EOG) and environmental electromagnetic interference during daily activities will also significantly reduce the signal-to-noise ratio (SNR).

[0027] In terms of signal processing, there are multiple key problems in the existing technologies. In signal acquisition and spatial localization, traditional methods lack targeted consideration of the special distribution of periauricular electrodes, making it difficult to accurately locate and separate the signal sources of each channel. In the feature extraction stage, a single data processing strategy is often adopted, failing to fully integrate the information in the time, frequency, and spatial domains, resulting in the extracted features lacking discriminability and stability. In the process of functional area recognition and feature classification, it is difficult to balance the real-time performance and accuracy of the algorithm, and there is a lack of a systematic feature fusion and screening mechanism.

[0028] Due to the portability requirements of intelligent headphones, the system needs to implement complex signal processing and feature analysis with limited computing resources. Especially in the classification and discrimination stage, the existing methods lack a complete credibility evaluation system, making it difficult to quantitatively evaluate the reliability of classification results, which is prone to misjudgment in practical applications. At the same time, individual differences and the diversity of usage environments also pose great challenges to feature extraction and classification. Currently, there is a lack of a systematic solution that can simultaneously solve signal quality optimization, feature extraction enhancement, classification performance improvement, and credibility evaluation.

[0029] Please refer to Figure 1 , Figure 1 FIG. Figure 1 is a schematic flowchart of a method for multi-channel EEG signal feature extraction and classification of a BCI intelligent headphone provided by an embodiment of the present application. The method for multi-channel EEG signal feature extraction and classification of the BCI intelligent headphone in the embodiment of the present application can be applied to a computer device, which includes but is not limited to devices such as smartphones, laptops, tablets, desktop computers, physical servers, and cloud servers. As

[0030] shown, the method for multi-channel EEG signal feature extraction and classification of this embodiment includes steps S101 to S104, which are described in detail as follows:

[0031] Specifically, during the acquisition of multi-channel signals of the intelligent earphone and the determination of channel distribution, the intelligent earphone adopts a binaural wearing mode, with 6 dry electrode sensors configured on each side. Signal amplification is achieved through an in-built high-precision operational amplifier, and the sampling rate is set within the range of 250 - 1000 Hz. The channel distribution follows the improved international 10 - 20 system, with key coverage of the temporal, parietal, and occipital regions. This layout makes full use of the limited space around the ear. The electrode array adopts a radial distribution structure, with the central electrode located above the tragus, and the remaining electrodes evenly distributed along the edge of the auricle. Each electrode is equipped with an independent elastic support structure to adapt to different ear shapes. To ensure the quality of signal acquisition, the system integrates an impedance measurement circuit to monitor the electrode-skin contact state in real time. When abnormal contact is detected, the electrode pressure is automatically adjusted through a micro-mechanical structure. The intelligent earphone is equipped with a high-precision inertial measurement unit to monitor the change of wearing position in real time. By establishing the mapping relationship between the spatial position of the electrode and the scalp anatomical landmark points, the system can compensate for the spatial error caused by position offset. This mapping uses the cubic spline interpolation method to ensure the smoothness of spatial position reconstruction. The system adopts an adaptive electrode pressure adjustment mechanism to adjust the contact pressure between the electrode and the skin in real time through a micro-pneumatic device, and automatically performs pressure compensation when abnormal contact impedance is detected. The intelligent earphone integrates an environmental noise monitoring module to evaluate the surrounding electromagnetic environment in real time and dynamically adjust the gain and filtering parameters of the signal amplifier accordingly. The system establishes a comprehensive electrode performance evaluation system, including real-time monitoring of key indicators such as contact impedance, baseline drift, and noise level. Through these innovative designs, the system generates a spatial position matrix containing accurate three-dimensional coordinate information, achieving the accurate positioning and dynamic tracking of each electrode position.

[0032] In some embodiments, obtaining the channel identifier corresponding to the channel distribution includes: separating signals according to the channel distribution and constructing a filter chain; outputting the channel identifier according to the filter chain.

[0033] Perform signal separation processing based on the obtained spatial position matrix and construct a filter chain. The signal separation algorithm introduces a spatial constraint term on the basis of traditional independent component analysis, making full use of electrode position information to improve the separation effect. The objective function of the algorithm includes an information maximization term, a spatial regularization term, and a temporal correlation constraint term. The optimization process comprehensively considers the statistical independence, spatial correlation, and temporal continuity of the signals. In actual implementation, the system adopts a segmented processing strategy, processing 2-second data segments each time, with an overlap rate of 50% for adjacent segments to ensure signal continuity. The filter chain adopts an adaptive cascade structure, including three core links: a digital notch filter, a bandpass filter bank, and an adaptive filter. The digital notch filter uses a narrowband stopband design, and its center frequency is automatically adjusted according to the local power frequency; the bandpass filter bank extracts electroencephalogram activities in four frequency bands of δ, θ, α, and β respectively, and the filter order is determined by balancing signal quality and computational load; the adaptive filter dynamically tracks and eliminates non-stationary interference, and its parameters are updated in real time through the recursive least squares algorithm. The system introduces an intelligent bypass mechanism to automatically simplify the processing flow and reduce system latency when the signal quality is good. For motion artifacts, the system adopts an adaptive filtering algorithm based on acceleration signals to effectively suppress signal interference caused by head movement. The system establishes a crosstalk compensation mechanism between channels. By estimating the coupling coefficient between channels, a compensation matrix is constructed to eliminate the crosstalk effect. Through this complete signal processing system, a multi-level adaptive filter chain structure is formed, realizing the efficient separation and purification of multi-channel electroencephalogram signals.

[0034] According to the constructed filter chain system, establish a channel identification coding system for the processed signals. The channel identification adopts a multi-level architecture, integrating spatial position, signal characteristics, and quality assessment information. The spatial position coding uses a three-level structure of "azimuth - level - serial number", directly inherited from the front-end spatial position matrix, ensuring the precise correspondence between the identification and the actual electrode position. The signal characteristic coding includes two dimensions of the dominant frequency band and the signal type. The dominant frequency band is determined based on the output of the bandpass filter bank in the filter chain, and the signal type is identified through feature analysis of the filtered waveform. The system adopts a dynamic weight mechanism to automatically adjust the importance of different feature dimensions according to the quality assessment results of each level of processing in the filter chain. The quality assessment system makes full use of the status information of each processing link in the filter chain and comprehensively evaluates from five dimensions: signal-to-noise ratio, baseline stability, myoelectric contamination rate, signal integrity, and spatial consistency, and uses fuzzy logic methods to fuse various indicators. The channel identification adopts a hierarchical storage structure, supporting fast retrieval and historical traceability. The identification data is compressed to reduce storage and transmission overhead. Based on this coding and management system, the system outputs a standardized channel identification in the format of "azimuth code - feature code - quality code", realizing the precise positioning and classification of electroencephalogram signals.

[0035] In some embodiments, outputting component features based on the analysis domain includes: mapping the analysis domain into a transformation spectrum and performing eigen - decomposition; and outputting the component features based on the eigen - decomposition.

[0036] When performing component extraction on the channel identifiers generated in the above steps and dividing the analysis domain, the system adopts a selective processing strategy based on channel identifiers. For high - quality channels such as those labeled "L23 - AB - Q1", the system directly uses a fine 8 - layer wavelet packet decomposition; for channels with a lower label quality level, a rough 4 - layer decomposition is used to reduce the influence of noise. Component extraction uses the wavelet packet transform method, with the db4 wavelet basis for multi - layer decomposition to form a complete time - frequency analysis tree structure. In practical applications, for example, when a user is performing concentration training, the system focuses on high - quality channels in the frontal and parietal regions. By performing a fine decomposition on the alpha band (8 - 13 Hz) of these channels, the significant suppression of alpha - wave energy when the user enters a focused state can be clearly captured. During the decomposition process, the system uses the soft - threshold method to denoise the coefficients at each scale layer, and the threshold parameter is determined by maximum likelihood estimation. The analysis domain division adopts an adaptive segmentation strategy, dividing the data into sub - intervals with similar statistical characteristics based on the non - stationary characteristics of the signal. The system calculates the local statistics of the signal, including mean, variance, skewness, and kurtosis, through a sliding time window, and marks the segmentation points when the statistical characteristics change significantly. To improve the robustness of segmentation, a joint decision mechanism of multiple statistics is introduced to ensure the reliability of the segmentation points. In an emotion recognition task, this segmentation strategy can accurately locate the time points of emotion changes because different emotional states often exhibit significantly different statistical characteristics. Through this adaptive decomposition and segmentation method, the system establishes a hierarchical signal analysis structure, and each analysis domain contains detailed time - frequency features and statistical attributes within a specific time window.

[0037] The obtained analysis domain, along with the time-frequency features and statistical attributes contained therein, is transformed into a transformation spectrum through non-linear mapping and feature decomposition is completed. The system selects an appropriate mapping strategy according to the characteristics of the analysis domain divided in the previous stage. The construction of the transformation spectrum adopts the kernel principal component analysis method, and the radial basis kernel function is used to map the original feature space to a high-dimensional feature space. During the mapping process, the kernel function parameters are optimized by the cross-validation method to balance the accuracy of feature extraction and the computational complexity. Taking the concentration assessment as an example, after the system divides 1 minute of EEG data into multiple analysis domains, the trajectory of feature migration during the change of concentration can be clearly observed through the non-linear mapping of time-frequency features and statistical attributes, which is of great significance for real-time assessment of the user's cognitive state. The system applies singular value decomposition to the transformation spectrum to extract the main feature directions and the corresponding eigenvalues. The energy threshold criterion is adopted during the feature decomposition process, and the feature components with a cumulative energy contribution rate reaching 95% are retained. In the attention persistence test, this decomposition method can effectively distinguish the EEG features in the concentrated and dispersed attention states, because the EEG signals during concentrated attention usually show stronger time-frequency structure, which is reflected in the decomposition result as a higher energy proportion of the main feature components. At the same time, the system calculates the spatial distribution pattern of the feature vectors to analyze the contribution degree of signals in different brain regions. At this stage, the transformation spectrum is decomposed into a series of orthogonal feature components, and each component carries information about a specific neural activity pattern.

[0038] Based on the orthogonal feature components obtained from the feature decomposition, the system constructs a multi-dimensional feature description system. First, time-domain features are extracted, including statistics such as average amplitude, root mean square value, kurtosis, and skewness; then frequency-domain features are calculated, including parameters such as energy distribution in each frequency band, spectral centroid, and bandwidth; finally, time-frequency joint features are analyzed, including instantaneous frequency, spectral entropy, and modulation characteristics. In the fatigue detection application, the time-domain features can reflect the overall intensity change of brain activity, the frequency-domain features can capture the typical phenomenon of enhanced theta waves and weakened alpha waves in the fatigue state, and the time-frequency joint features can describe the dynamic process of this energy transfer. When it is detected that the user is in a fatigue state, the system will observe a significant increase in the energy of delta waves (0.5 - 4 Hz), accompanied by a decrease in the energy of alpha waves and an increase in spectral entropy. The system normalizes all the extracted features to eliminate the influence of dimension and scale differences. When processing the motor imagery task, the normalized features can clearly reflect the differences in EEG features under different motor imagery modes. For example, there are significant differences in the suppression degree of the mu rhythm (8 - 13 Hz) in the corresponding motor cortex regions during the motor imagery of the left and right hands. Through this complete feature extraction process, the system finally outputs a set of standardized component features, and each feature is marked with the corresponding physical meaning and reliability index.

[0039] Step S102: Perform spatial localization on the component features to obtain functional regions, construct a regional network based on the functional regions, and assign weights through the regional network; output a regional attribute data packet according to the weight assignment result corresponding to the assigned weight, and establish a structural model based on the regional attribute data packet to output a channel association network according to the structural model.

[0040] Specifically, when performing spatial localization on the component features output in the above step, the system combines the physical meaning and reliability index of each feature and adopts an improved spatial filtering algorithm. First, based on the spatial distribution characteristics of the bilateral electrodes of the earphone, a local spatial filter is constructed, focusing on the signal source localization in the superficial brain regions of the temporal lobe, parietal lobe, and occipital lobe. During the localization process, the system uses the time-frequency features and phase information in the component features and adopts a spatial filtering method based on the Laplace operator to calculate the spatial projection of the signal source. The localization algorithm is specifically optimized for the special electrode distribution pattern of the earphone, and the localization accuracy of the signal source in the periaural region is improved by introducing a distance weighting term. In the music appreciation task, this localization method can capture the rhythm processing activities in the temporal lobe region. When the user focuses on listening to music, the system can detect the enhanced synchronization of the alpha band in the superior temporal gyrus region. The system continuously corrects the spatial filtering parameters through an iterative optimization method until the residual error is less than the preset threshold, and finally calculates the spatial distribution characteristics of each signal component in the periaural region. During the optimization process, the system dynamically adjusts the filtering parameters according to the reliability index of the features, assigns a greater processing weight to the features with high reliability, and adopts a more conservative processing strategy for the features with low reliability. At the same time, the system also fully considers the physical meaning of each component feature to ensure that the spatial localization result is consistent with the principles of neurophysiology.

[0041] Utilize the obtained spatial distribution characteristics to carry out functional area identification and construct a regional network. The functional area identification adopts a pattern matching-based method, which matches the located signal features with the standardized functional templates of periauricular brain regions. The system has pre-established a functional area template library for the acquisition characteristics of the headset, including multiple functional network patterns such as auditory processing, attention network, and emotion regulation that can be reliably detected by periauricular electrodes. During the matching process, an adaptive weight algorithm is used to compensate for individual differences to ensure the reliability of identification. The system introduces a spatial similarity calculation method, measures the matching degree between the signal features and the template through weighted cosine similarity, and sets an adaptive threshold for regional division. The construction of the regional network adopts a local connection graph method, takes the identified functional areas as nodes, constructs connection edges based on signal correlation, and forms a network structure reflecting the functional connection of the periauricular region. The connection strength is calculated by two metric indicators, mutual information and phase synchronization value, to capture the linear and non-linear associations between different regions. During concentration training, the system can accurately identify the activity states of periauricular brain regions related to attention such as the temporal lobe and posterior parietal lobe, and depict the functional connection patterns between these regions, providing neural indicators for evaluating the user's concentration level. The system has established a dynamic update mechanism, automatically adjusts the matching parameters of the functional template according to the task type, and improves the specificity of identification. Through this local network construction method, the system finally obtains a functional area network structure based on periauricular electrodes.

[0042] Perform weight assignment based on the constructed functional area network. The system uses an improved local network analysis method to calculate the importance weights of each regional node. The weight assignment considers three key indicators: local connection strength, signal stability, and functional specificity. The local connection strength reflects the signal synchronization degree of a certain brain region with other detected regions in the periauricular area, and is quantified by calculating the weighted connection degree and clustering coefficient; the signal stability characterizes the temporal consistency of the activity pattern of this brain region, and is evaluated by the stability of signal variance and statistical characteristics; the functional specificity measures the response selectivity of this brain region to specific cognitive tasks, and is determined by comparing the activation differences under different task states. The system introduces a multi-level weight fusion framework to evaluate the regional importance at different time scales, including short-term changes (second level), medium-term trends (minute level), and long-term patterns (session level). In the music appreciation task, the system identifies the temporal lobe region that plays a key role in the music processing network according to these indicators. For example, when the user appreciates classical music, the system can accurately calculate the weights of the superior temporal gyrus region in the music processing network, and these weights reflect the processing degree of the user on different components such as music rhythm and melody. The system has established a dynamic adjustment mechanism for weights, and updates the importance evaluation of each region in real time according to the task progress and user state changes. Through this weight calculation method based on the acquisition characteristics of the headset, the system finally obtains the importance weight values of each detectable functional area.

[0043] Output the regional attributes according to the assigned weight values. The system designs a set of attribute grading mechanisms based on weights. Regions with high weights obtain more detailed attribute descriptions, including multiple dimensions such as function types, activation modes, and connection characteristics; regions with medium weights record basic functional attributes and main connection relationships; regions with low weights only retain core function labels. In the attention training application, this grading mechanism can highlight the key brain region characteristics related to the task. For example, when the user performs a sustained attention task, the system will focus on describing the activation characteristics of attention-related brain regions such as the temporal lobe and the posterior parietal lobe, including phenomena such as enhanced theta wave activity in these regions and enhanced functional coupling with other nodes in the attention network. At the same time, the system will also record the changes in the degree of alpha wave suppression in these regions, and these characteristics can reflect the user's attention investment level. For brain regions with weak task relevance, such as the peripheral regions covered by the edge electrodes, only their basic activity states are recorded. During music appreciation, the system can accurately characterize the characteristics of each node in the auditory processing network according to the weights. The high-weight temporal lobe region will record detailed rhythm processing patterns, such as the time-frequency coupling characteristics of delta waves and gamma waves, and these characteristics can reflect the degree of processing of music rhythm by the user; while the low-weight peripheral regions only need to record the basic activation states. Through this attribute assignment scheme linked to weights, the system finally outputs a structured regional attribute data packet.

[0044] In some embodiments, establishing a structural model according to the regional attribute data packet includes: performing a spatial mapping on the regional attribute data packet to construct a topological distribution map; detecting connection nodes based on the topological distribution map; and establishing the structural model according to the connection nodes.

[0045] When implementing spatial mapping on the regional attribute data packets output from the above steps, the system adopts an improved topological mapping algorithm. In view of the characteristics of headphone collection, a special local spatial mapping strategy is designed to project the functional regions in the regional attribute data packets onto a two-dimensional plane, forming a topological structure diagram of the functional regions. The multidimensional scaling analysis method is used in the mapping process to maintain the relative distance relationship between the functional regions in the original space. In the concentration training scenario, this mapping method can clearly present the distribution patterns of attention-related regions such as the temporal lobe and parietal lobe on the two-dimensional plane, as well as their spatial relationships with the surrounding functional regions. The system adopts an iterative optimization strategy to dynamically adjust the mapping parameters to ensure the minimization of information loss during the projection process. To improve the mapping accuracy, a local preservation constraint is introduced to ensure that adjacent functional regions still maintain similar spatial relationships after projection. At the same time, the system also establishes a mapping quality evaluation mechanism to evaluate the mapping effect by calculating the topological preservation degree before and after projection. In the music appreciation task, this mapping method can accurately reflect the spatial distribution characteristics of each functional region in the auditory processing network, providing a reliable spatial basis for subsequent node detection. Through this precise spatial mapping process, the system finally obtains a complete topological structure diagram of the functional regions.

[0046] Based on the generated topological structure diagram, connection node detection is carried out. The system designs a multi-level node recognition strategy to gradually find potential connection nodes from local to global. At the local level, by calculating the connection strength and information flow characteristics between functional regions, regions with significant connection patterns are identified; at the global level, the importance of nodes is evaluated based on the network topological characteristics. The node detection process focuses on three key indicators: signal transmission efficiency, connection stability, and functional selectivity. In the concentration training application, the system can accurately identify the key nodes responsible for attention regulation in the temporal lobe region, and these nodes often show strong signal synchronization and stable connection patterns. The system also establishes an adaptive threshold mechanism to dynamically adjust the node screening criteria according to the task characteristics. To improve the detection reliability, a cross-validation mechanism is introduced to ensure the stability of node recognition through multiple repeated detections. In the music processing task, the system can effectively identify the key functional nodes involved in rhythm processing, and these nodes usually show significant phase synchronization characteristics. Through this systematic node detection method, a group of connection nodes with a core role in the functional network are finally determined.

[0047] Construct a structural model around the detected connection nodes. The model construction adopts a hierarchical and progressive strategy. First, a skeletal network is constructed based on the core connection nodes, and then secondary connections and functional pathways are gradually added to finally form a complete network structure. The setting of model parameters fully considers the characteristics of headphone acquisition, and quantitatively describes the connection strength and directionality between nodes. The system designs a parameter optimization mechanism to adjust the model parameters by minimizing the prediction error, and at the same time introduces a dynamic update strategy to enable the model to adjust its internal structure according to real-time data. In the concentration training, the structural model can clearly display the change process of the user's attention level. For example, when the user maintains a highly focused state during listening to a class, the model can capture the significant suppression of alpha wave activity in the temporal lobe region and the strong connection pattern between other attention-related nodes. Through continuous verification and optimization, the system finally establishes a functional network structure model that can accurately reflect the cognitive state.

[0048] Conduct channel correlation analysis based on the established structural model. The system designs an accurate channel mapping mechanism to establish a corresponding relationship between the functional nodes in the model and the actual acquisition channels. The mapping process considers the spatial distribution, signal characteristics, and functional selectivity of the channels to ensure the accurate correspondence between the functional network and the actual signal acquisition. The system adopts a multiple verification strategy, selects the optimal mapping relationship by comparing the effects of different mapping schemes, and establishes a weight assignment mechanism based on the importance of channels. In the concentration training scenario, the system focuses on analyzing the correlation patterns between the T3 and T4 channels in the temporal lobe region and the P7 and P8 channels in the parietal lobe region. When the user maintains a focused state, these channels exhibit specific collaborative characteristics: the T3 and T4 channels in the temporal lobe region show significant suppression of alpha waves (8 - 13 Hz) and enhancement of beta waves (13 - 30 Hz), and at the same time form stable functional connections with the P7 and P8 channels in the parietal lobe region; when the attention level drops, there will be a synchronous increase in the energy of theta waves (4 - 7 Hz) in these four channels, and the phase synchrony between the channels is significantly reduced. The system quantitatively calculates the correlation strength of each channel pair, and integrates features such as time series correlation, frequency band energy ratio, and phase synchronization index into a standardized correlation index. These correlation indexes are weighted and combined to form a multi-dimensional channel correlation matrix, where the matrix elements represent the degree of functional coupling between channels. Finally, the system organizes these quantified correlation features into a hierarchical network structure and outputs a complete channel functional correlation network.

[0049] Step S103, perform hierarchical construction on the channel correlation network, define feature levels, perform scale fusion on the feature levels to form feature groups, output fusion elements according to the feature groups, and generate a set of feature elements based on the fusion elements; perform classification combination on the set of feature elements to determine the classification dimension and obtain the classification label corresponding to the classification dimension.

[0050] When performing hierarchical construction on the channel function association network output by the above steps, the system first analyzes the functional connection strength and direction information of each channel node in the network. Based on these association features, a hierarchical progressive processing strategy is adopted for multi-scale decomposition, and hierarchical structures are established in three dimensions: time, frequency, and space. In the time dimension, a multi-scale sliding window is used to divide the signal into processing units of different time lengths; in the frequency dimension, wavelet transform is used to construct an 8-layer decomposition tree from details to approximation; in the space dimension, based on the functional connection strength between channels, a hierarchical network structure from local to global is formed. In the application of concentration training, this hierarchical method can capture both fast-changing transient attention features and slow-changing persistent attention features. For example, when a user is listening to a lecture in class, the system can simultaneously detect short-term attention shifts to sudden sounds (reflected in millisecond-level beta wave changes) and continuous changes in overall concentration (reflected in minute-level alpha wave modulation). The system determines the time-frequency-space parameters of each layer through iterative optimization, and finally constructs a feature hierarchy system including three levels: micro (millisecond level), meso (second level), and macro (minute level).

[0051] Using the constructed feature hierarchy system, the system performs scale fusion on the features at each level. For micro-level features, it mainly fuses millisecond-level neuron firing patterns and local field potential changes; for meso-level features, it fuses second-level EEG rhythm changes and functional connection strength; for macro-level features, it integrates minute-level cognitive state transitions and network topology evolution. The fusion process adopts a multi-level cascade structure, starting from the finest scale and integrating feature information layer by layer. At each fusion node, the system combines features of different scales in a weighted summation manner, and the weight coefficients are optimized through the maximum mutual information criterion. The system also introduces a time consistency constraint to ensure the continuity of temporal information during the feature fusion process. An adaptive interpolation algorithm is designed for time alignment in view of the time resolution differences of features at different scales. When fusing transient features at the micro level and trend features at the macro level, the system uses wavelet reconstruction method to maintain the integrity of the signal. Through this adaptive fusion mechanism, the system obtains a set of multi-scale feature representations covering fine-grained neural activities to coarse-grained cognitive states.

[0052] Based on the fused multi-scale feature representation, the system establishes a feature organization framework. For micro-features at the millisecond level, the system extracts the synchrony and causal relationships of neuronal population activities; for meso-features at the second level, it analyzes the energy distribution and phase coupling of electroencephalogram rhythms; for macro-features at the minute level, it focuses on the dynamic reorganization patterns of functional networks. The feature organization adopts a tree-like structure, where the top-level nodes represent the global cognitive state, the middle-level nodes correspond to different cognitive sub-functions, and the bottom-level nodes retain the detailed time-frequency features. In the scenario of concentration monitoring, this organizational structure can accurately depict the changes in the user's cognitive state. For example, in a 45-minute online course, the system can organize and record the instantaneous attention fluctuations of students (reflected in the rapid changes in β-wave energy), the sustained concentration level (reflected in the degree of α-wave suppression), and the overall learning engagement (manifested as the co-variation pattern of energies in different frequency bands). Through the systematic organization of multi-scale features, a multi-level structured feature set is finally formed.

[0053] Fusion element generation is performed on the formed structured feature set. The system first evaluates the discriminative ability of each feature in the feature set for different cognitive tasks, including the neuronal firing patterns at the micro level, the rhythm features at the meso level, and the network topology features at the macro level. Then, the most representative feature combinations are screened through the information gain criterion while maintaining the balance of features at different time scales. The system designs a hierarchical feature evaluation mechanism to screen key feature indicators at each time scale. At the micro level, it mainly focuses on instantaneous spectral features and phase synchrony; at the meso level, it extracts the energy change trend and the modulation relationship between frequency bands; at the macro level, it emphasizes the stability features of functional connection patterns. According to the characteristics of different cognitive tasks, the system dynamically adjusts the combination method of features. For example, in the task of sustained attention, the system focuses on the dynamic balance between α-wave suppression and θ-wave enhancement; in the working memory task, it pays more attention to the correspondence between γ-wave activity and memory load. These screened and combined features are standardized and finally a set of fusion elements containing multi-scale cognitive features is output.

[0054] In some embodiments, generating a feature element set according to the fusion elements includes: performing feature screening according to the fusion elements to extract key elements; integrating the key elements into a feature system for classifying features according to the feature system; and outputting the feature element set based on the feature classification results corresponding to the classified features.

[0055] When conducting feature screening on the fused elements (including multi-dimensional feature information and feature data at different time scales) output from the above steps, the system established a multi-level evaluation index system. The index system includes four dimensions: the information amount, stability, noise resistance, and discrimination of features. Multiple quantitative evaluation parameters are set under each dimension. The system uses an entropy-based feature evaluation method to calculate the information gain ratio of each fused element and screen out feature subsets with low information redundancy. In the attention assessment task, the feature screening process focuses on significant features such as the suppression of alpha waves and the enhancement of beta waves in the temporal lobe region. For example, when the user maintains a focused state during an important meeting, the system can accurately screen out key indicators reflecting the attention level from complex time-frequency features, such as the energy ratio of prefrontal theta waves to temporal alpha waves, and the phase synchronization intensity between different brain regions. At the same time, the system can also identify the changes in the user's cognitive engagement when listening to key information, which is reflected in the dynamic modulation of beta wave energy and the transient enhancement of gamma wave activity. Through repeated verification and optimization, the system finally determined a set of highly discriminative key elements including time-frequency features, spatial features, and cognitive features.

[0056] Based on the key elements obtained through screening, the system constructs a feature integration framework. In the integration process, according to the nature of each type of key element, a hierarchical organizational structure is adopted, and time-frequency features, spatial features, and cognitive features are grouped according to functional attributes and time scales. For time-frequency features, the system established a multi-scale representation from instantaneous states to long-term trends; for spatial features, a multi-level description from local channels to global networks was constructed; for cognitive features, a functional hierarchy from basic perception to advanced cognition was formed. In the feature integration process, the system introduced an adaptive weight mechanism to dynamically adjust the contribution of features to the overall system according to their importance and reliability. For example, in the assessment of attention persistence, the system will focus on the stability of alpha wave suppression and the periodic changes in theta wave activity, which can often more accurately reflect the user's cognitive engagement state. According to the characteristics of different cognitive tasks, the system can flexibly adjust the feature combination method. During the concentration training process, if the system detects fluctuations in the user's attention, it will automatically adjust the feature weights and increase the monitoring ratio of short-term attention indicators. Through this dynamic integration strategy, the system finally formed an adaptive feature system including multi-scale representations of time-frequency features, multi-level descriptions of spatial features, and functional hierarchies of cognitive features.

[0057] Feature classification is performed using the constructed adaptive feature system. Based on the established time scale, spatial hierarchy, and cognitive function framework in the adaptive feature system, a multi-level classification strategy is adopted in the classification process. First, preliminary classification is carried out based on the physical attributes of the features. For example, the time-domain features, frequency-domain features, and time-frequency joint features in the multi-scale representation of time-frequency features are classified into different categories respectively. Then, further subdivision is performed according to the functional significance of the features. For example, the attention-related features in the multi-level description of spatial features, the cognitive load features in the cognitive function hierarchy of cognitive features, etc. are classified into the corresponding functional categories. The system comprehensively considers the correlation and complementarity between features during the classification process and optimizes the category division through the clustering analysis method. In the scenario of professional reading comprehension, this classification system can accurately capture the cognitive processing features of users. For example, when a user reads professional literature, the system can simultaneously monitor the sustained attention level (through the alpha wave suppression feature in the multi-scale representation of time-frequency features), the information processing load (through the theta / beta ratio in the multi-scale representation of time-frequency features), and the depth of understanding (through the gamma wave activity pattern in the cognitive function hierarchy of cognitive features), and automatically classify these features into the corresponding cognitive function groups. Through this systematic classification process, a hierarchical feature classification system including physical attribute classification and functional attribute classification is finally formed.

[0058] Based on the formed feature classification system, the system outputs feature elements in a standardized manner. For each classification level in the hierarchical feature classification system, the system designs a standardized feature output format, including information such as the basic attributes, numerical range, timestamp, and reliability index of the features. For each type of feature in the hierarchical feature classification system, the system generates a detailed description document, including the physiological significance, calculation method, and typical application scenarios of the features. In the evaluation of professional meeting concentration, the system can output a complete attention feature data packet. When the user participates in an important meeting that lasts for 2 hours, the system can track and record the complete process of cognitive state changes: during 9:00 - 10:00 in the morning, the user maintains a high level of concentration, which is reflected in the physical attribute classification in the hierarchical feature classification system showing continuous alpha wave suppression (10 - 12 Hz) and stable beta wave activity (15 - 18 Hz); during 10:00 - 10:30, the attention level fluctuates, and the functional attribute classification in the hierarchical feature classification system shows a brief increase in theta wave activity (4 - 7 Hz) and an intermittent decrease in alpha wave suppression; during 10:30 - 11:00, after a short break, the attention level gradually recovers, which is reflected in the gradual optimization of the ratio of prefrontal theta waves to temporal alpha waves in the physical attribute classification and functional attribute classification in the hierarchical feature classification system. The system organizes these time-series feature data in a standard format to form a feature element data set containing the complete cognitive process, where each feature in the hierarchical feature classification system is marked with a clear functional category and reliability score. Through this refined feature extraction and organization scheme, a multi-dimensional feature element data set containing physical features and cognitive features is finally output.

[0059] In some embodiments, obtaining the classification label corresponding to the classification dimension includes: converting the classification dimension into a classification framework; summarizing the results according to the classification framework; and outputting the classification label according to the result summary content corresponding to the summarization result.

[0060] When performing classification and combination on the multi-dimensional feature element dataset output from the above steps, the system first establishes a feature combination strategy. The combination process is based on the physical and cognitive feature attributes in the multi-dimensional feature element dataset and adopts a hierarchical combination scheme. In the bottom-level combination, the system combines features with the same time scale and similar spatial distribution in the multi-dimensional feature element dataset. For example, it combines the alpha wave energy, phase, and time-varying characteristics in the same channel into an alpha activity feature, and combines the energy ratio, phase coupling, and modulation characteristics of theta waves and beta waves into a rhythm interaction feature. In the middle-level combination, it integrates features with different time scales but related functions in the multi-dimensional feature element dataset. For example, it combines the millisecond-level gamma wave bursts, second-level alpha wave suppression, and minute-level theta wave change trends into an attention feature. In the top-level combination, it fuses the feature combinations across functional regions in the multi-dimensional feature element dataset. For example, it integrates the activity characteristics of multiple brain regions such as the temporal lobe and parietal lobe to form a cognitive network feature. The combination process sets a feature correlation threshold and an information redundancy constraint to ensure that the combined features maintain information integrity while avoiding redundancy. In the attention assessment scenario, the system combines features such as alpha wave suppression feature, theta / beta energy ratio, and prefrontal-temporal functional connection strength in the multi-dimensional feature element dataset to form a feature combination reflecting the attention level. To adapt to the dynamic changes of different cognitive states, the system also establishes an adaptive adjustment mechanism for feature combination, which can dynamically adjust the combination strategy according to signal quality and task requirements. Through multi-level feature combination optimization, the system finally determines a set of classification dimensions including basic feature combinations, functional feature combinations, and network feature combinations.

[0061] Based on the determined classification dimensions, the system constructs a classification framework. The classification framework adopts a multi-level structure for different types of feature combinations, including the time dimension corresponding to the basic feature combination, the space dimension corresponding to the network feature combination, and the function dimension corresponding to the functional feature combination. The time dimension reflects the dynamic change characteristics of the basic feature combination, which is divided into three time scales: instantaneous response (1 - 100 ms), short-term change (0.1 - 10 s), and long-term trend (10 - 600 s). For each time scale, corresponding feature extraction windows and update periods are set. The space dimension describes the spatial distribution characteristics of the network feature combination, including local activity (single-channel features), regional linkage (adjacent channel feature combinations), and global coordination (multi-channel network features), and detailed feature extraction ranges and spatial filtering parameters are defined for each spatial scale. The function dimension characterizes the cognitive processing attributes of the functional feature combination, covering levels such as basic perception (signal strength, stability), attention regulation (degree of alpha wave suppression, amplitude of beta wave enhancement), and advanced cognition (frequency of gamma wave bursts, depth of theta wave modulation). For each functional level, corresponding feature thresholds and state discrimination criteria are configured. In the analysis of meeting participation, the system can comprehensively characterize the changes in the user's cognitive state from the basic feature combination, functional feature combination, and network feature combination. For example, the basic feature combination reflects the fluctuation law of concentration, specifically manifested as the duration and recovery rate of alpha wave suppression; the network feature combination reveals the activation pattern of the attention network, reflected in the dynamic changes in the functional connection strength between the prefrontal lobe and the temporal lobe; the functional feature combination reflects the depth of cognitive processing, characterized by the change trend of the theta / beta ratio and the gamma wave burst pattern. By constructing such a refined classification framework, the system generates a set of standardized classification frameworks.

[0062] Relying on the established classification framework, the system conducts result summarization. The summarization process adopts a hierarchical progressive approach. First, within the time dimension of the classification framework, features are aggregated in time series, including calculating statistical measures such as the time average, variance, and skewness of features, and extracting the periodic patterns and mutation features of features. Then, within the spatial dimension of the classification framework, feature aggregation is carried out, including calculating spatial features such as the correlation coefficient between channels, the phase synchronization index, and the functional connection strength. Finally, within the functional dimension of the classification framework, feature aggregation is carried out, including calculating functional indicators such as the combined features of energies in different frequency bands and the cross-frequency band coupling features. The summarization process adopts an adaptive weight strategy, dynamically adjusting the weights of features in each dimension according to the signal-to-noise ratio, stability, and discriminability of features in the classification framework. The system also establishes the mapping relationship between dimensions in the classification framework, calculates statistical measures such as the mutual information and conditional entropy between dimensions, and analyzes the interaction of dimensional features. In the scenario of analyzing the focus of professional meetings, the system tracks the changes in the cognitive state of participants in real time. When users participate in important meeting discussions, the system simultaneously monitors the instantaneous cognitive load, the level of sustained attention, and the degree of information integration, and summarizes and integrates these features in the three dimensions of the classification framework to form a multi-dimensional result summary.

[0063] Based on the formed multi-dimensional result summary, the system outputs classification labels. The labeling system adopts a multi-level structure, including the basic state label, abnormal pattern label, and comprehensive evaluation label in the multi-dimensional result summary. The basic state label contains specific quantitative indicators in the multi-dimensional result summary, such as the alpha wave suppression ratio (0 - 100%), the theta / beta ratio (0.5 - 2.0), the prefrontal gamma wave burst frequency (0 - 10 Hz), etc.; the abnormal pattern label contains state identifications such as the attention fluctuation warning and cognitive fatigue warning in the multi-dimensional result summary, and corresponding triggering conditions and credibility evaluation rules are set for each warning; the comprehensive evaluation label integrates the state information of multiple dimensions in the multi-dimensional result summary to form an overall description of the cognitive state. In the evaluation of the participation degree in continuous meetings, the system tracks and records the changes in the cognitive state: at the beginning stage of the meeting, the alpha wave suppression ratio in the multi-dimensional result summary remains above 80%, the theta / beta ratio is stable at about 0.8, and the prefrontal gamma wave burst frequency is 8 Hz, which is labeled as the "highly focused state"; in the middle stage, the alpha wave suppression ratio in the multi-dimensional result summary drops to 60%, the theta / beta ratio rises to 1.2, and the gamma wave burst frequency drops to 5 Hz, which is labeled as the "attention fluctuation state"; in the later stage, the alpha wave suppression ratio in the multi-dimensional result summary is lower than 40%, the theta / beta ratio exceeds 1.5, and the gamma wave burst frequency is lower than 3 Hz, which is labeled as the "cognitive fatigue state". The system generates such detailed classification labels for the multi-dimensional result summary data of each time window to form a complete cognitive state evaluation sequence.

[0064] Step S104: Evaluate the credibility of the classification markers to establish an evaluation system, and output discriminant labels according to the evaluation system; integrate the discriminant labels to obtain the classification results, and complete the feature extraction and classification of multi-channel EEG signals.

[0065] Specifically, when evaluating the credibility of the multi-level classification markers (including the basic state markers, abnormal pattern markers, and comprehensive evaluation markers) output in the above steps, the system constructs a hierarchical evaluation system. The evaluation system quantifies the credibility from three dimensions of signal quality, feature stability, and state consistency in the multi-level classification markers. The signal quality evaluation includes calculating the signal-to-noise ratio, artifact detection, and baseline drift analysis of the basic state markers in the multi-level classification markers, and assigns a score to the data quality of each time window; the feature stability evaluation includes time variance analysis, trend test, and mutation detection of the abnormal pattern markers in the multi-level classification markers to evaluate the time continuity of the features; the state consistency evaluation evaluates the internal consistency of the state judgment by calculating the mutual information and conditional entropy between different dimension features of the comprehensive evaluation markers in the multi-level classification markers. Specific evaluation indicators include: signal-to-noise ratio, artifact ratio, and baseline stability in the signal quality dimension; coefficient of variation, trend coefficient, and mutation index in the feature stability dimension; mutual information amount, conditional entropy value, and correlation coefficient in the state consistency dimension, etc. In the focus monitoring scenario, the system quantitatively evaluates the credibility of each marker in the multi-level classification markers. For example, when detecting the "highly focused" state in the multi-level classification markers, the system simultaneously calculates the signal quality score, energy stability index of the alpha wave suppression feature, and the consistency coefficient with other attention-related features, and comprehensively forms the credibility score of this state marker. Through this multi-dimensional credibility evaluation, the system establishes a complete credibility evaluation system.

[0066] Based on the established evaluation system, the system conducts threshold interval division. Different threshold levels are set for the evaluation indicators in different dimensions of the credibility evaluation system. The signal quality dimension is divided into three intervals: high quality (signal-to-noise ratio > 20 dB, artifact ratio < 5%, baseline drift < 1 μV / min), medium quality (10 - 20 dB, 5 - 15%, 1 - 3 μV / min), and low quality (< 10 dB, > 15%, > 3 μV / min); the feature stability dimension is set with three levels: stable (coefficient of variation < 0.1, trend coefficient < 0.05), fluctuating (0.1 - 0.3, 0.05 - 0.15), and unstable (> 0.3, > 0.15); the state consistency dimension is divided into three levels: highly consistent (mutual information > 0.8, conditional entropy < 0.2), partially consistent (0.5 - 0.8, 0.2 - 0.5), and inconsistent (< 0.5, > 0.5). The system also establishes a threshold adaptive adjustment mechanism to dynamically optimize the threshold parameters according to the task type and environmental conditions. In the evaluation of meeting concentration, the system dynamically adjusts the thresholds in the credibility evaluation system to adapt to the characteristics of different meeting stages. For example, in the important report session, the requirement threshold for state consistency in the credibility evaluation system is increased to ensure the reliability of state judgment; in the break discussion session, the threshold requirement for signal quality in the credibility evaluation system is appropriately relaxed to adapt to more body movement interference. Through this flexible threshold division strategy, the system forms a complete multi-level threshold interval system.

[0067] In some embodiments, the outputting the discrimination label according to the evaluation system includes: dividing a threshold interval by the evaluation system to output a calibration result through the threshold interval; and outputting the discrimination label according to the calibration result.

[0068] According to the divided multi-level threshold interval system, the classification marking results are calibrated. The calibration process adopts a multiple verification strategy. First, calibration is carried out separately within the three evaluation dimensions of signal quality, feature stability, and state consistency in the multi-level threshold interval system to determine the credibility level of each dimension. Then, cross-verification between the dimensions of the multi-level threshold interval system is carried out to analyze the consistency of the evaluation results of different dimensions. Finally, a comprehensive calibration result is obtained through weighted fusion. The system designs a calibration compensation mechanism. When the evaluation index of a certain dimension in the multi-level threshold interval system is close to the threshold boundary (such as the signal-to-noise ratio is between 19 - 21 dB), it is corrected by referring to the evaluation results of other dimensions in the multi-level threshold interval system. In the concentration evaluation scenario, the calibration process can identify possible misjudgment situations. For example, when a sudden significant decrease in the attention level is detected, the system comprehensively analyzes the calibration results of the three dimensions of signal quality, feature stability, and state consistency in the multi-level threshold interval system to confirm whether this change is a real attention fluctuation or a misjudgment caused by external interference. Through this rigorous calibration process, the system finally obtains a set of verified signal quality dimension scores, feature stability dimension scores, state consistency dimension scores, and a comprehensive rating that combines these three dimensions.

[0069] Based on the obtained signal quality dimension scores, feature stability dimension scores, state consistency dimension scores, and comprehensive rating, the system generates and outputs discriminant labels. The label generation adopts a hierarchical structure, including basic state labels based on signal quality dimension scores, credibility level labels based on feature stability dimension scores, and comprehensive evaluation labels based on state consistency dimension scores and comprehensive ratings. The basic state labels describe specific cognitive state categories; the credibility level labels mark the reliability level of state judgment; the comprehensive evaluation labels combine state information and credibility information to provide a complete state description. In the meeting concentration monitoring, the discriminant labels output by the system contain rich state information. For example, when the "high concentration" state is recognized, the labels generated by the system include: "State category: high concentration (α-wave suppression > 80%, θ / β < 0.8); Credibility level: A level (Signal quality dimension score: 9.2 / 10, Feature stability dimension score: 0.95, State consistency dimension score: 0.92); Timestamp: 10:30:45; Duration: 15 minutes; Associated features: Enhanced prefrontal β-wave (+60%), Suppressed temporal lobe α-wave (-75%)". The system generates such detailed discriminant labels for the data of each time window and maintains the consistency of the label format, realizing the organic combination of accurate discrimination of cognitive states and credibility evaluation. These basic state labels, credibility level labels, and comprehensive evaluation labels not only contain rich state information but also come with complete credibility evaluation results, forming a complete set of discriminant labels.

[0070] In some embodiments, the data integration of the discrimination label to obtain a classification result includes: integrating the data of the discrimination label to construct an output structure; and obtaining the classification result according to the output structure.

[0071] When integrating the discrimination labels output from the above steps, the system establishes a multi-level integration strategy. In the integration process, the discrimination labels are first grouped according to the time series, with a basic time window set to 1 minute (to analyze instantaneous state changes), a medium time window set to 5 minutes (to analyze state stability), and a long-term time window set to 15 minutes (to analyze state trends). For the basic time window, the system calculates the statistical distribution and transition probability of the basic state labels, such as the occurrence frequency and duration of the "highly focused" state; for the medium time window, it analyzes the change trend and stability characteristics of the basic state label sequence, including the state retention rate and transition frequency; for the long-term time window, it combines the comprehensive evaluation labels to extract the periodicity and long-range correlation of state evolution, such as the regularity of attention fluctuations. At the same time, the credibility level labels of each time window are integrated to establish a credibility-weighted state statistics. In the concentration assessment, the system performs multi-scale integration on the continuous discrimination labels, such as calculating the weighted duration ratio of the "highly focused" state within 15 minutes (the weight comes from the credibility level label), the number of effective state transitions (eliminating fluctuations with low credibility), and the average credible level of state discrimination obtained based on the comprehensive evaluation label. Through this multi-scale data integration, the system constructs a hierarchical output structure that includes time series features (state duration, transition moment), state features (distribution features and transition features of each state), and credibility features (credibility distribution of each time scale).

[0072] Based on the temporal features, state features, and credibility features in the constructed hierarchical output structure, the system design conversion strategy converts the integrated data into a standardized classification sequence. The conversion process designs a three-level sequence structure: The first-level sequence describes the temporal evolution of the state based on temporal features, including state category, duration, transition moment, and instantaneous credibility, specifically recording the state labels and their credibility scores per minute; the second-level sequence characterizes the stability and reliability of the state based on state features, integrating indicators such as the state distribution, average credibility, and state transition frequency within 5 minutes; the third-level sequence depicts the long-term change features of the state based on credibility features, including trend indicators, periodic parameters, and cumulative credibility on a 15-minute scale. The system defines standard data formats and update rules based on credibility features for each level of the sequence. In the meeting concentration analysis, the first-level sequence records the attention state and its credibility per minute using temporal features; the second-level sequence describes the attention stability and credibility changes within 5 minutes using state features; the third-level sequence reflects the attention change trend and overall reliability on a 15-minute scale using credibility features. Through this hierarchical conversion strategy, the system generates a set of classification sequences containing short-term, medium-term, and long-term state evolution features.

[0073] Using the short-term, medium-term, and long-term state evolution features contained in the generated classification sequence, the system conducts feature integration. The integration process adopts a credibility-weighted adaptive mechanism, dynamically adjusting their weights in feature integration according to the reliability and relevance of different levels in the classification sequence. For the first-level sequence in the classification sequence, the system extracts timeliness indicators (such as state switching delay) and accuracy indicators (such as misjudgment rate) of state transitions, and the weights are determined by the instantaneous credibility in the classification sequence; for the second-level sequence in the classification sequence, it analyzes stability indicators (such as fluctuation frequency) and predictability indicators (such as trend consistency) of the state, and the weights are determined by the average credibility in the classification sequence; for the third-level sequence in the classification sequence, it extracts long-term change trends (such as fatigue accumulation rate) and regular features (such as attention period), and the weights are determined by the cumulative credibility in the classification sequence. The system designs a feature conflict resolution mechanism. When features at different levels in the classification sequence are inconsistent, arbitration is carried out by evaluating the credibility and context information of each level of features in the classification sequence. In the attention monitoring scenario, the system can effectively integrate attention features at different time scales in the classification sequence. For example, when the first-level sequence in the classification sequence shows a sudden drop in attention, the system will combine the stability features of the second-level sequence and the trend features of the third-level sequence in the classification sequence to comprehensively judge the nature and possible duration of this fluctuation. Through this multi-level feature integration, the system forms a feature integration including the dimensions of time, state, and credibility.

[0074] Exemplarily, obtaining the classification result according to the output structure includes: converting the output structure into a classification sequence, obtaining integrated features according to the classification sequence; and outputting the classification result according to the integrated features.

[0075] According to the formed feature integration, the system outputs a classification result. The generation of the classification result makes full use of the information in each dimension of the feature integration, including the state classification in the feature integration {current state category, state duration, state stability score, key physiological indicators}, the trend analysis in the feature integration {short-term trend prediction, medium-term change direction, long-term evolution pattern}, and the credibility assessment in the feature integration {state credibility, trend credibility, overall assessment credibility}. In the application of concentration assessment, the classification result output by the system in real time specifically includes: the state classification based on feature integration {attention level: 85%, duration: 10 minutes, stability: 0.92, alpha wave suppression: -75%, beta wave enhancement: +45%}; the trend analysis based on feature integration {short-term: stable, medium-term: slightly decreasing (-5% / hour), long-term: periodic fluctuation (T = 90 minutes)}; the credibility index based on feature integration {state: 0.95, trend: 0.88, overall: 0.91}. The system generates a normalized classification result for each time window and dynamically updates it based on the credibility score in the feature integration. When a significant change in the state in the feature integration is detected (such as the change in the alpha wave suppression degree exceeds 20% and the credibility in the feature integration is greater than 0.9), the classification result is immediately updated. Through this dynamic classification mechanism based on feature integration, the system finally outputs a classification result with high time resolution and high credibility.

[0076] The provided method has the following beneficial effects:

[0077] 1. An adaptive signal processing system is established. Through a multi-level acquisition mechanism and processing strategy, the inter-channel interference, electromyogram interference, and environmental noise in the periauricular area are suppressed. On this basis, multi-scale feature decomposition and analysis domain division are realized, thereby enhancing the ability to extract the time-frequency features of electroencephalogram signals while improving the signal quality and stability.

[0078] 2. A complete functional recognition framework is constructed. Using spatial mapping technology and regional network structure, the precise characterization of the functional connection of the periauricular brain region is realized, and the feature organizational structure is optimized through a multi-level feature fusion mechanism and an adaptive screening strategy, significantly improving the accuracy of feature expression and the stability of the system.

[0079] 3. A systematic classification and evaluation mechanism is formed. The classification combination optimization and multi-dimensional credibility assessment are organically combined. Through a dynamic result integration strategy and a real-time update mechanism, the accuracy and timeliness of classification markers are improved, and the reliability and adaptability of the system in the daily application environment are enhanced.

[0080] To implement the multi-channel electroencephalogram (EEG) signal feature extraction and classification method for the brain-computer intelligent headset corresponding to the above method embodiments, so as to achieve the corresponding functions and technical effects. Refer to Figure 2 , Figure 2 FIG. shows a structural block diagram of a multi-channel EEG signal feature extraction and classification device 200 for a brain-computer intelligent headset provided by an embodiment of the present application. For ease of description, only the parts related to this embodiment are shown. The multi-channel EEG signal feature extraction and classification device 200 for a brain-computer intelligent headset provided by an embodiment of the present application includes:

[0081] A distribution determination unit 201, configured to collect multi-channel signals of the intelligent headset and determine the channel distribution, obtain the channel identifiers corresponding to the channel distribution; perform component extraction on the channel identifiers and divide the analysis domain, and output component features based on the analysis domain;

[0082] A network output unit 202, configured to perform spatial positioning on the component features, obtain the functional areas, construct a regional network according to the functional areas, and allocate weights through the regional network; output a regional attribute data packet according to the weight allocation result corresponding to the allocated weights, establish a structural model according to the regional attribute data packet, so as to output a channel association network according to the structural model;

[0083] A marker acquisition unit 203, configured to perform hierarchical construction on the channel association network, define the feature hierarchy, perform scale fusion on the feature hierarchy to form a feature group, output fusion elements according to the feature group, and generate a feature element set according to the fusion elements; perform classification combination on the feature element set to determine the classification dimension, and obtain the classification marker corresponding to the classification dimension;

[0084] A classification completion unit 204, configured to perform credibility evaluation on the classification marker to establish an evaluation system, and output a discrimination label according to the evaluation system; perform data integration on the discrimination label to obtain a classification result, and complete the multi-channel EEG signal feature extraction and classification.

[0085] The above multi-channel EEG signal feature extraction and classification device 200 for a brain-computer intelligent headset can implement the multi-channel EEG signal feature extraction and classification method of the above method embodiments. The optional items in the above method embodiments are also applicable to this embodiment, which will not be elaborated here. The remaining content of the embodiments of the present application can refer to the content of the above method embodiments, and will not be repeated in this embodiment.

[0086] Figure 3 FIG. is a schematic structural diagram of a computer device provided by an embodiment of the present application. As Figure 3 shown, the computer device 3 of this embodiment includes: at least one processor 30 ( Figure 3(only one is shown), a memory 31, and a computer program 32 stored in the memory 31 and executable on the at least one processor 30. When the processor 30 executes the computer program 32, the steps in any of the above method embodiments are implemented.

[0087] The computer device 3 may be a computing device such as a smart phone, a tablet computer, a desktop computer, and a cloud server. The computer device may include, but is not limited to, a processor 30 and a memory 31. Those skilled in the art can understand that Figure 3 merely examples of the computer device 3, which do not constitute a limitation on the computer device 3. It may include more or fewer components than shown in the figure, or combine certain components, or different components. For example, it may also include input / output devices, network access devices, etc.

[0088] The so-called processor 30 may be a central processing unit (CPU), and the processor 30 may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0089] The memory 31 may be an internal storage unit of the computer device 3 in some embodiments, such as the hard disk or memory of the computer device 3. The memory 31 may also be an external storage device of the computer device 3 in other embodiments, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device 3. Further, the memory 31 may also include both the internal storage unit and the external storage device of the computer device 3. The memory 31 is used to store an operating system, application programs, a boot loader, data, and other programs, such as the program code of the computer program, etc. The memory 31 may also be used to temporarily store data that has been output or will be output.

[0090] In addition, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0091] An embodiment of the present application provides a computer program product, and when the computer program product runs on a computer device, the computer device is caused to implement the steps in each of the above method embodiments when executed.

[0092] In several embodiments provided in the present application, it can be understood that each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, the program segment, or the part of code includes one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the block may occur in an order different from that marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved.

[0093] 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 such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0094] The above specific embodiments have further elaborated on the purpose, technical solution, and beneficial effects of the present application. It should be understood that the above are only specific embodiments of the present application and are not used to limit the protection scope of the present application. It is particularly pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for extracting and classifying multi-channel EEG signals of a brain-computer intelligent headset, characterized in that: include: Collect multi-channel signals of the smart headset and determine channel distribution, and obtain channel identifiers corresponding to the channel distribution; Extracting components from the channel identifier and dividing the analysis domain, and outputting component features based on the analysis domain; Spatially locating the component features, obtaining functional regions, constructing a regional network according to the functional regions, and assigning weights through the regional network; outputting a regional attribute data packet according to a weight assignment result corresponding to the assigned weights, establishing a structural model according to the regional attribute data packet, and outputting a channel association network according to the structural model; Performing hierarchical construction on the channel association network to define a feature level, performing scale fusion on the feature level to form a feature group, outputting fusion elements according to the feature group, and generating a feature element set according to the fusion elements; performing classification combination on the feature element set to determine a classification dimension, and obtaining a classification label corresponding to the classification dimension; Performing credibility evaluation on the classification mark to establish an evaluation system, and outputting a discriminant label according to the evaluation system; The discriminant labels are integrated to obtain classification results, and multi-channel EEG signal feature extraction and classification are completed.

2. The method according to claim 1, characterized in that The obtaining the channel identifier corresponding to the channel distribution includes: Perform signal separation and construct a filter chain according to the channel distribution; The channel identification is output according to the filter chain.

3. The method according to claim 1, characterized in that The outputting component features based on the analysis domain comprises: Mapping the analysis domain into a transformation spectrum and performing eigendecomposition; The component features are output based on the eigendecomposition.

4. The method according to claim 1, characterized in that: The step of establishing a structural model according to the regional attribute data packet comprises: Implementing spatial mapping on the regional attribute data packet to construct a topological distribution map; Detecting connection nodes based on the topological distribution map; The structural model is established according to the connection nodes.

5. The method according to claim 1, characterized in that The step of generating a feature element set according to the fused elements comprises: Perform feature screening according to the fusion elements to extract key elements; integrating the key elements into a feature system for classifying features according to the feature system; The feature element set is output based on the feature classification result corresponding to the classification feature.

6. The method according to claim 1, characterized in that The obtaining of the classification mark corresponding to the classification dimension includes: converting the classification dimensions into a classification framework; Aggregate the results according to the described classification framework; Output a classification mark according to the result summary content corresponding to the summary result.

7. The method according to claim 1, characterized in that Outputting a discriminant label according to the evaluation system includes: Dividing the evaluation system into threshold intervals to output calibration results through the threshold intervals; The discrimination label is output according to the calibration result.

8. The method according to claim 1, characterized in that The step of integrating the discriminant labels to obtain classification results includes: Performing data integration on the discriminant labels to construct an output structure; The classification result is obtained according to the output structure.

9. The method according to claim 8, characterized in that The obtaining the classification result according to the output structure includes: Converting the output structure into a classification sequence, and obtaining integrated features according to the classification sequence; The classification result is output according to the integrated feature.

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