Personalized electroencephalogram pattern recognition method, device and equipment based on AI glasses

Abstract

The invention relates to the technical field of brain-computer interfaces, in particular to a personalized electroencephalogram mode recognition method, device and equipment based on AI glasses. The method comprises the following steps: acquiring operation data of attention glasses, and generating a monitoring table to determine an abnormal standard; generating a change track according to the abnormal standard; sorting problem sources based on the change track to form a fault spectrum; testing an operation index according to the fault spectrum, determining a health degree according to the operation index, drawing a performance diagram based on the health degree, and generating an evaluation table according to the performance diagram; counting an attenuation trend according to the evaluation table, calculating performance degradation according to the attenuation trend, predicting an influence range based on the performance degradation, and determining an optimization measure according to the influence range; determining an operation scene according to the optimization measure, determining an intervention opportunity according to the operation scene, formulating a diagnosis rule, generating a fault domain, and establishing a repair scheme; and generating a diagnosis report according to the repair scheme, and completing personalized electroencephalogram pattern recognition based on the AI glasses. And the accuracy of personalized electroencephalogram pattern recognition of the attention glasses is improved.

Description

Technical Field

[0001] The present application relates to the field of brain-computer interface technology, and in particular to a personalized EEG pattern recognition method, device and equipment based on AI glasses. Background Art

[0002] AI glasses face significant individual differences when performing EEG pattern recognition. Traditional EEG pattern recognition methods usually use universal models that cannot effectively adapt to the unique EEG characteristics and usage habits of different users. Although some studies have attempted to achieve personalization through data-driven methods, these methods often require a large amount of personal training data and a long adaptation process, which seriously affects the user experience. At the same time, existing personalization methods lack effective modeling of users' long-term EEG pattern changes, and are difficult to adapt to the gradual changes in users' cognitive states and habits.

[0003] There are several key problems in the existing technology when dealing with personalized EEG pattern recognition. First, the user's EEG pattern will be affected by many factors such as physical and mental state, environmental factors, etc., and exhibits complex dynamic characteristics. It is difficult for existing methods to achieve stable recognition in multiple scenarios. Secondly, obtaining sufficient personal training data is often time-consuming and costly, especially for new users, and it is impossible to quickly establish an effective personalized model. In addition, while maintaining the personalization of the model, the existing methods find it difficult to effectively utilize the common features in the group data, resulting in limited generalization capabilities of the model. The existing technology also lacks systematic feature management and dynamic optimization mechanisms, and is unable to automatically adjust the recognition strategy according to changes in user habits, making it difficult to maintain stable recognition performance during long-term use.

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

[0005] The embodiments of the present application provide a personalized EEG pattern recognition method, device and equipment based on AI glasses, which aims to solve several key problems in the prior art when dealing with personalized EEG pattern recognition. First, the user's EEG pattern will be affected by many aspects such as physical and mental state, environmental factors, etc., and exhibits complex dynamic characteristics. It is difficult for existing methods to achieve stable recognition in multiple scenarios. Secondly, obtaining sufficient personal training data is often time-consuming and costly, especially for new users, and it is impossible to quickly establish an effective personalized model. In addition, while maintaining the personalization of the model, the existing methods are difficult to effectively utilize the common features in the group data, resulting in limited generalization capabilities of the model. The prior art also lacks systematic feature management and dynamic optimization mechanisms, and is unable to automatically adjust the recognition strategy according to changes in user habits, and is difficult to maintain stable recognition performance during long-term use.

[0006] In a first aspect, the present application provides a personalized EEG pattern recognition method based on AI glasses, comprising:

[0007] Mark the EEG data collected by the AI ​​glasses and construct a personal benchmark; perform pattern analysis based on the personal benchmark to obtain a feature map; use the analysis result corresponding to the pattern analysis to mark the feature interval to generate a user portrait, and obtain an initial template corresponding to the user portrait;

[0008] Perform feature mapping according to the initial template to obtain common features, and mark individual features based on the common features; construct fusion rules using the individual features, generate a feature set according to the fusion rules, and obtain an identification benchmark corresponding to the feature set;

[0009] Classify the scene according to the recognition benchmark and mark the use environment; construct a state sequence based on the use environment, select a feature template according to the state sequence, generate recognition parameters according to the feature template, and generate a recognition scheme according to the recognition parameters;

[0010] Performing trend analysis on the identification scheme to obtain statistical features, marking pattern intervals based on the statistical features; constructing a feature chain according to the pattern intervals, and generating a feature combination according to the feature chain;

[0011] Performing pattern analysis on the feature combination to mark an identification sequence; constructing a mapping relationship based on the identification sequence; performing feature recombination using the mapping relationship, and generating a personalized pattern based on the feature recombination.

[0012] In a second aspect, the present application also provides a personalized EEG pattern recognition device, comprising:

[0013] A benchmark establishment unit is used to perform feature marking on the EEG data collected by the AI ​​glasses and construct a personal benchmark; perform pattern analysis based on the personal benchmark to obtain a feature map; use the analysis result corresponding to the pattern analysis to mark the feature interval to generate a user portrait, and obtain an initial template corresponding to the user portrait;

[0014] A feature mapping unit, configured to perform feature mapping according to the initial template, obtain common features, and mark individual features based on the common features; construct fusion rules using the individual features, generate a feature set according to the fusion rules, and obtain an identification benchmark corresponding to the feature set;

[0015] A scene classification unit, configured to classify scenes according to the recognition benchmark and mark the use environment; construct a state sequence based on the use environment, select a feature template according to the state sequence, generate recognition parameters according to the feature template, and generate a recognition scheme according to the recognition parameters;

[0016] A trend analysis unit, configured to perform trend analysis on the identification scheme, obtain statistical features, mark pattern intervals based on the statistical features, construct a feature chain according to the pattern intervals, and generate a feature combination according to the feature chain;

[0017] The pattern analysis unit is used to perform pattern analysis on the feature combination and mark an identification sequence; construct a mapping relationship based on the identification sequence; use the mapping relationship to perform feature recombination, and generate a personalized pattern based on the feature recombination.

[0018] In a third aspect, the present application also provides a computer device, comprising a processor and a memory, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, the personalized EEG pattern recognition method based on AI glasses as described in the first aspect is implemented.

[0019] This method realizes personalized recognition of EEG signals through dynamic feature modeling and scene adaptation technology. The built-in sensor of AI glasses is used to collect user EEG data, and the time domain, frequency domain and time-frequency domain features (such as wavelet transform energy spectrum) are extracted through preprocessing (denoising, artifact removal). Combined with the stratified sampling data of users in different cognitive states (focus, relaxation, etc.), a personal benchmark database containing feature parameters, probability distribution and time series correlation characteristics is constructed. Based on the personal benchmark, pattern analysis is performed to extract common features (such as α wave energy) and individual features (such as specific frequency band fluctuation patterns), and fusion rules (such as weighted fusion or neural network fusion) are generated. User portraits are constructed through feature interval markings (such as high-frequency band sensitivity intervals), and initial templates (such as standardized feature weight templates) are generated. The use environment (such as noisy environment, quiet environment) is classified according to the scene sensitivity level (high / medium / low), and the corresponding feature templates (such as noise suppression templates) are matched. The recognition parameters (such as thresholds, weights) are dynamically adjusted through state sequences (such as changes in EEG features within a continuous time window) to generate a scene-adapted recognition scheme. Perform trend analysis on the recognition scheme (such as stability index calculation), mark pattern intervals (such as abnormal fluctuation intervals), and build feature chains (time-series correlation feature sets). Generate the final personalized pattern through feature reorganization (such as principal component analysis or deep learning optimization) to achieve dynamic calibration and high-precision recognition.

[0020] Through personal benchmark database and dynamic feature mapping, the misjudgment problem caused by individual differences in traditional EEG recognition is solved, and the classification accuracy is improved (for example, the recognition accuracy of user focus state is improved by more than 30%). Based on scene sensitivity classification and state sequence analysis, real-time adjustment of recognition parameters is achieved (such as automatic enhancement of high-frequency suppression in noisy environments) to enhance system robustness. Combined with the hardware computing power of AI glasses (such as edge computing chips), feature extraction and pattern generation are completed locally, reducing cloud dependence and response delay (<50ms). Through user portraits and mode interval markings, personalized feedback (such as attention reminders and fatigue warnings) is supported to improve interactive friendliness.

[0021] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic diagram of a personalized EEG pattern recognition method based on AI glasses according to an embodiment of the present application;

[0023] Figure 2 This is a schematic diagram of the structure of a personalized EEG pattern recognition device shown in an embodiment of the present application;

[0024] Figure 3 A schematic diagram of the structure of a computer device shown in an embodiment of the present application. DETAILED DESCRIPTION

[0025] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0026] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0027] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0028] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0029] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0030] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0031] The technical solution of the embodiment of the present application is introduced below.

[0032] AI glasses face significant individual differences when performing EEG pattern recognition. Traditional EEG pattern recognition methods usually use universal models that cannot effectively adapt to the unique EEG characteristics and usage habits of different users. Although some studies have attempted to achieve personalization through data-driven methods, these methods often require a large amount of personal training data and a long adaptation process, which seriously affects the user experience. At the same time, existing personalization methods lack effective modeling of users' long-term EEG pattern changes, and are difficult to adapt to the gradual changes in users' cognitive states and habits.

[0033] There are several key problems in the existing technology when dealing with personalized EEG pattern recognition. First, the user's EEG pattern will be affected by many factors such as physical and mental state, environmental factors, etc., and exhibits complex dynamic characteristics. It is difficult for existing methods to achieve stable recognition in multiple scenarios. Secondly, obtaining sufficient personal training data is often time-consuming and costly, especially for new users, and it is impossible to quickly establish an effective personalized model. In addition, while maintaining the personalization of the model, the existing methods find it difficult to effectively utilize the common features in the group data, resulting in limited generalization capabilities of the model. The existing technology also lacks systematic feature management and dynamic optimization mechanisms, and is unable to automatically adjust the recognition strategy according to changes in user habits, making it difficult to maintain stable recognition performance during long-term use.

[0034] Please refer to Figure 1 , Figure 1 The flowchart of a personalized EEG pattern recognition method based on AI glasses provided in the embodiment of the present application is shown in FIG. The personalized EEG pattern recognition method based on AI glasses in the embodiment of the present application can be applied to computer devices, including but not limited to smart phones, laptops, tablet computers, desktop computers, physical servers, cloud servers and other devices. Figure 1 As shown, the personalized EEG pattern recognition method based on AI glasses of this embodiment includes steps S101 to S105, which are described in detail as follows:

[0035] Step S101, feature marking of EEG data collected by AI glasses and building a personal benchmark; performing pattern analysis based on the personal benchmark to obtain feature mapping; using the analysis results corresponding to the pattern analysis to mark the feature interval to generate a user portrait, and obtaining an initial template corresponding to the user portrait.

[0036] Specifically, the EEG data collected by AI glasses contains multi-channel raw potential signals, which need to be feature-labeled first and a personal benchmark constructed.

[0037] In some embodiments, the feature marking of the EEG data collected by the AI ​​glasses and the construction of a personal benchmark include: preprocessing the EEG data to remove physiological artifacts and interference noise; extracting the time domain features, frequency domain features and time-frequency domain features of the EEG data and performing dimensionality reduction processing; establishing a feature statistical parameter library of the EEG data through standardization processing; analyzing the time series correlation characteristics and fluctuation characteristics of the feature statistical parameter library in combination with stratified sampling data under different cognitive states; calculating the feature stability index of the feature statistical parameter library based on the time series correlation characteristics and fluctuation characteristics and constructing a personal benchmark database containing feature parameters, probability distribution models and time series correlation characteristics.

[0038] The preprocessing stage of the original signal includes the use of a bandpass filter (0.5-45Hz) to remove power frequency interference and baseline drift, and the independent component analysis (ICA) method to remove physiological artifacts such as electrooculography and electromyography. In the feature extraction stage, time domain features (average amplitude, root mean square value, peak-to-peak value), frequency domain features (energy of five frequency bands of δ, θ, α, β, and γ) and time-frequency domain features (wavelet transform time-varying features) are extracted respectively. The extracted features are reduced in dimension by principal component analysis, and the principal components with 95% explained variance are retained. All features are z-score standardized to establish a feature statistical parameter library. At the same time, the kernel density estimation method is used to fit the probability distribution of each feature to capture the distribution characteristics of the feature. In the data collection process, a stratified sampling strategy is adopted, and different cognitive states such as wakefulness, concentration, and relaxation are uniformly sampled, and no less than 1000 sample points are selected for each state. By analyzing the autocorrelation (lag=10) and cross-correlation (threshold 0.6) of the features, feature combinations with significant temporal patterns are identified. Combined with fixed time window analysis (window width 2s, step length 1s), the short-term fluctuation characteristics of the features were quantified. The stability index of each extracted feature was calculated, including variance stability (σ^2<0.3) and time consistency (τ>0.7). The final personal benchmark database contains 70 standardized feature parameters, probability distribution models between features, and time series correlation characteristics, which together constitute a personalized benchmark model of EEG features.

[0039] Pattern analysis was performed based on 70 standardized feature parameters, probability distribution models between features, and time series correlation characteristics in the personal benchmark database. First, the statistical characteristics of each dimension were calculated using the 70 standardized feature parameters in the benchmark database, and the feature importance was calculated in combination with the probability distribution model. Features with high probability distribution dispersion (H>0.8) and strong time series correlation (r>0.6) were given higher weights. The weighted Pearson correlation coefficient matrix was constructed using the time series correlation characteristics to identify feature combinations with absolute correlation coefficients exceeding 0.8 and stable time series patterns. Cluster analysis was performed using the improved K-means algorithm, and the cluster centers were initialized as the distribution centers of the high stability features among the 70 feature parameters in the benchmark model. The optimal number of clusters (K=5) was determined by calculating the Davies-Bouldin index and silhouette coefficient, and the number of iterations was set to 500 and the convergence threshold was 1×10^-4. The probability distribution information of the benchmark model was introduced to weight the distance calculation in the clustering process. The Bootstrap method was used to perform 100 repeated clusterings, and the stable structures with a frequency of more than 80% were screened in combination with temporal stability. The probability distribution characteristics were calculated for each cluster, including multidimensional Gaussian distribution parameters and kernel density estimation curves. Based on the distribution characteristics of 70 characteristic parameters and temporal correlation characteristics, a Mahalanobis distance metric considering dynamic changes was constructed, and an adaptive threshold (τ=0.6) was set. The analysis process made full use of all the elements in the benchmark database, and finally a set of stable characteristic patterns were obtained, including the characteristic distribution parameters of each pattern, the dynamic characteristic vectors, and the multidimensional similarity relationship between patterns.

[0040] The feature intervals and user portraits were constructed using the 70 standardized feature parameters of the benchmark model, the probability distribution model, and the clustering results of pattern analysis (including feature distribution parameters, dynamic feature vectors, and multidimensional similarity relationships between patterns). Based on the probability distribution of feature parameters in the benchmark model, the 3σ principle combined with kernel density estimation was used to determine the range of change of each feature. For feature parameters with high temporal stability, the basic levels were divided by the quantile method (Q1=25%, Q2=50%, Q3=75%). Based on the pattern clustering results, the feature distribution parameters and dynamic feature vectors of each pattern were used to construct adaptive feature intervals. Among them, the dynamic feature vector was used to capture the transient characteristics of feature changes, and the dynamic characteristics of state transition were quantified by calculating the direction change and amplitude fluctuation of the vector sequence. Combined with the multidimensional similarity relationship between patterns, the overlap threshold (θ=0.3) was set to identify independent intervals and transition intervals. The interval transition law was analyzed based on the temporal correlation characteristics, and a Markov transition model considering probability distribution was established. The calculation of transition probability made full use of the temporal stability index in the benchmark model, and gave greater weight to features with high stability. Multiple constraints are introduced in the interval division process: information gain ratio based on probability distribution (≥0.6), Gini coefficient of sample distribution (<0.3), autocorrelation coefficient of time series characteristics (r>0.4). By comprehensively analyzing the distribution characteristics of 70 feature parameters, pattern clustering results (including feature distribution parameters, dynamic feature vectors and multidimensional similarity relationships between patterns) and state transition laws, a multi-level user portrait is constructed, including: feature preference description based on probability distribution, cognitive state characteristics based on clustering, dynamic behavior patterns based on time series analysis, and the interrelationships and evolution laws between these features. This three-dimensional user portrait fully reflects the uniqueness of individuals in terms of EEG characteristics.

[0041] In some embodiments, obtaining the initial template corresponding to the user portrait includes: performing region decomposition on the user portrait and marking key features to construct a feature space; performing sample matching based on the feature space; and generating an initial template by analyzing feature combinations based on matching results of the sample matching.

[0042] The user profiles generated in the above steps are regionally decomposed. According to the feature preference description based on probability distribution, the cognitive state characteristics based on clustering, the dynamic behavior pattern based on time series analysis, and the inter-correlation relationship and evolution law between the features in the user profile, the key feature combination is extracted. In the decomposition process of the feature preference description based on probability distribution, the existing probability distribution parameters are used to retain the characteristics of stable distribution shape (kurtosis range 2.8-3.2) and low volatility (coefficient of variation <15%). In the processing of cognitive state characteristics based on clustering, the original cluster center and transition probability matrix are inherited, and the t-SNE method is used for dimensionality reduction mapping. The perplexity in the mapping process is set to 30 and the maximum number of iterations is 2000. When processing the dynamic behavior pattern based on time series analysis, the established state transition network is used as the basis to extract significant feature vectors (eigenvalue>0.1) through graph decomposition. Multiple stability constraints are introduced in the feature selection process: the prediction variance of the feature under the original probability model is required to be less than 15% of the total variance; the time consistency index of the feature is greater than 0.7; the coefficient of variation of the feature in multiple repeated measurements is less than 10%. For physiological features such as EEG rhythm, its stability at different time scales is analyzed, and the fluctuation range in the short term (within 1 hour) does not exceed ±5% of the mean, and the drift in the long term (within 24 hours) does not exceed ±10%. The final constructed feature space includes the following elements: the physical meaning of the original features (such as the value range of the cognitive load-related dimension is [0.2, 0.8], the response time threshold of the attention level-related dimension is 200ms, and the update cycle of the emotional state-related dimension is not less than 2s), probability distribution characteristics (including mean, standard deviation, covariance matrix, probability density function), feature distribution threshold (including upper and lower limits of the normal range and abnormal detection threshold) and time series dependency model.

[0043] Sample matching analysis is performed based on the constructed feature space. According to the probability distribution characteristics in the feature space, distance metrics are designed for different types of features: for features that obey normal distribution, Mahalanobis distance is used, using the mean and covariance matrix stored in the feature space; for features that present skewed distribution, distance based on KL divergence is used, referencing the probability density function of the feature space; for discrete distribution features, improved Hamming distance is used. The weight coefficient of the distance metric is determined by minimizing the cross-validation error, and different weight ranges are set for cognitive load-related dimensions (value range [0.2, 0.8]) and emotional state-related dimensions (update cycle not less than 2s). A multi-level similarity calculation framework is constructed. Local similarity calculation uses the feature distribution threshold defined in the feature space, and global similarity is weighted and integrated through the attention mechanism. The weight update frequency is consistent with the response time threshold (200ms) of the attention level-related dimension in the feature space. The calculation of feature similarity takes into account the influence of time scale. Short-term similarity uses exponentially weighted moving average, and long-term similarity uses multi-scale decomposition method. At the same time, an anomaly detection mechanism based on the probability distribution characteristics in the feature space is established. When the feature deviation exceeds the feature distribution threshold defined in the feature space, recalculation is triggered. The stability of the similarity calculation results is verified by Monte Carlo simulation. When 10% random noise is added, the coefficient of variation of the calculation results is kept within 0.15, which meets the stability requirements set in the feature space. The final sample matching results include similarity matrix, anomaly detection mark and time series similarity pattern.

[0044] The initial template is generated based on the similarity matrix, anomaly detection markers and temporal similarity pattern obtained by sample matching. The hierarchical clustering method is used to process the similarity matrix, and the Ward minimum variance method is used to construct the cluster tree. The physical constraints defined in the feature space are considered during clustering, such as the value range of the cognitive load-related dimensions and the minimum transition time between adjacent cognitive states. The anomaly detection markers are used to exclude unstable samples to ensure the reliability of the clustering results. The sample clusters are identified by the pruning algorithm, and the feature statistics of each cluster must meet the value range defined in the feature space. In the process of cluster formation, the dynamic programming algorithm is introduced to optimize the selection of split points, referring to the stable interval in the temporal similarity pattern. For the identified sample clusters, a multi-component Gaussian mixture model is constructed, and the initial values ​​of the model parameters are determined by the probability distribution characteristics of the feature space. The expectation maximization algorithm is used to optimize the model parameters to ensure that the model prediction results are consistent with the physical meaning of the original features in the feature space. In order to improve the stability of the model, the bagging ensemble method is used to train multiple sub-models and fuse the results through a voting mechanism. The voting threshold refers to the feature distribution threshold in the feature space. For dynamic features, a recurrent neural network model is constructed, and the network structure takes into account the temporal dependency model defined in the feature space. The network training adopts a teacher-forcing strategy and introduces an attention mechanism. The attention update rate is consistent with the response time threshold of the dimension related to the attention level in the feature space. The initial template generated in this way includes static feature templates and dynamic feature templates. Each template inherits the main properties of the feature space and also reflects the similarity pattern, abnormal feature distribution and temporal change law found in the sample matching process.

[0045] Step S102, perform feature mapping according to the initial template, obtain common features, and mark individual features based on the common features; construct fusion rules using the individual features, generate a feature set according to the fusion rules, and obtain the recognition benchmark corresponding to the feature set.

[0046] Specifically, feature mapping is performed in combination with the initial template provided in the above steps. The static feature distribution parameters (such as mean vector, covariance matrix) and the recurrent neural network model of dynamic features contained in the template are first normalized. Principal component analysis is applied to the static features, and the covariance structure in the template is used to extract the main feature directions with a cumulative contribution rate of more than 85%. During the feature extraction process, verification is performed according to the physical constraints defined by the template, such as the value range of the cognitive load indicator [0.2, 0.8] and the response time threshold of the attention level of 200ms. To ensure the stability of the mapping, a multiple verification mechanism is introduced: first, the temporal consistency of the feature is verified, requiring the coefficient of variation of the feature within the continuous time window (window length 5min) to be less than 0.15; second, the spatial consistency of the feature is verified, requiring the performance difference of the feature on different data subsets to be less than 10%; finally, the physical consistency of the feature is verified to ensure that the extracted feature direction is consistent with the physiological meaning defined in the template. In the processing of dynamic features, the recurrent neural network structure of the template is used to analyze the weight matrix and extract the main mode with a eigenvalue greater than 0.2. Special attention is paid to maintaining key timing constraints in the template, including the minimum time interval for state transitions (>500ms) and the upper limit of response delay (200ms). The Gaussian mixture model of the template is analyzed, and the distribution patterns with a frequency of more than 75% are extracted to construct a general probability model including mean, covariance and mixture weight. This mapping process evaluates its generalization performance through cross-validation (10 folds) to ensure stable performance in different scenarios. The final general features obtained include the main feature direction, the main mode, the general probability model (mean, covariance and mixture weight) and the generalization performance indicators of cross-validation.

[0047] Use the universal features obtained in the previous step to mark personalized features. Project the user data onto the main feature direction and calculate the deviation in each direction. Use the Z-score normalization method to evaluate the degree of deviation. For feature points with significant deviation (Z score>1.96), further use the mean, covariance and mixed weight of the universal probability model to calculate their probability density, and screen out points with probability density below the threshold (P<0.05). These feature points have significant personalized features under the universal probability model. In order to ensure the reliability of feature labeling, time window analysis (window length 30s, step length 5s) is introduced, and the generalization performance index of cross-validation is used to evaluate stability. Only feature points that are stable in multiple consecutive windows are retained. The DBSCAN algorithm (ε=0.15, MinPts=5) is used to cluster the selected feature points, and at the same time, it is checked whether each cluster meets the temporal constraints extracted in the first step. In the process of forming feature clusters, not only spatial distance but also temporal correlation is considered to ensure the temporal continuity of feature points within the cluster. For each feature cluster formed, its statistical characteristics (mean, variance, skewness, kurtosis) and time series characteristics (autocorrelation coefficient, cross-correlation coefficient) are calculated to establish the feature description vector of the cluster. At the same time, the detailed mapping relationship of each feature cluster in the universal feature space (projection coordinates with the main feature direction and similarity with the main mode) and reliability evaluation indicators (sample size, intra-cluster variance, inter-cluster distance) are recorded. This process ultimately obtains a set of feature clusters that meet universal constraints but have significant personalization, and each cluster has complete descriptive information.

[0048] Based on the personalized feature clusters marked in the previous step, the fusion rules are constructed. For each feature cluster, the degree of deviation from the corresponding general feature direction is calculated using its detailed mapping relationship in the general feature space, and the fusion weight is set accordingly. The weight setting is adjusted according to the reliability evaluation indicators (sample size, intra-cluster variance, and inter-cluster distance). The adaptive mechanism is adopted. The feature cluster with larger deviation and higher reliability has higher initial weight, but it must be ensured that the total deviation does not exceed the range allowed by the general model. A multi-level weight adjustment mechanism is introduced: at the feature level, the weight is adjusted according to the stability index of the cluster (variance ratio <1.5); at the time series level, the weight is adjusted according to the time continuity of the feature (autocorrelation coefficient>0.6); at the system level, the weight distribution is adjusted according to the overall consistency of the feature combination and the integrity of the mapping relationship. For static features, a probability density weighting mechanism based on the statistical characteristics of the feature cluster (mean, variance, skewness, kurtosis) is constructed. The probability density is calculated by the kernel density estimation method, and the bandwidth parameter is optimized by cross-validation. For dynamic features, the state transition rules considering the time series characteristics (autocorrelation coefficient, cross-correlation coefficient) are designed, and the transition probability matrix is ​​constructed based on the time series statistical characteristics of the feature cluster. These rules directly use the characteristics of personalized feature clusters to ensure that the significance of personalized features is maintained during the fusion process. At the same time, a rule conflict detection and resolution mechanism is established. When different rules conflict, the rule with a high reliability evaluation index is given priority. The final fusion rules include weight adjustment mechanism, state transfer rules and rule conflict detection and resolution mechanism.

[0049] The fusion rules established in the previous step are applied to integrate the common features and personalized features. Each feature cluster is weighted and combined with the corresponding common features according to the weight set by the weight adjustment mechanism. The combination process adopts a hierarchical strategy: first, basic fusion is performed at the feature level, and linear combination is performed using the set weights; then, adjustments are made at the pattern level to ensure that the fusion results meet the preset physical constraints; finally, optimization is performed at the system level to ensure the consistency of the overall features. The combination process strictly follows the constraints in the fusion rules, the feature correlation does not exceed the preset threshold (0.7), and the feature variance ratio is kept within a reasonable range (0.5-2.0). For the integration of dynamic features, the state transition rules defined in the fusion rules are used to adjust the feature weights according to the observed data at each time point (sampling rate 100Hz). When conflicts between rules are detected, the rule conflict detection and resolution mechanism is activated, and the best rule is selected according to the reliability evaluation index. In order to ensure the stability of the feature set, a sliding window mechanism (window length 5min, step length 1min) is introduced for smoothing. For detected anomalies (deviating from the mean by more than 3 standard deviations), the anomaly handling mechanism is activated to temporarily reduce the weight of the corresponding feature. The final feature set includes the weighted fusion feature distribution, dynamic update strategy and complete constraints, forming a comprehensive feature expression system that maintains the general feature framework and embodies personalized features.

[0050] In some embodiments, obtaining the recognition benchmark corresponding to the feature set includes: performing incremental analysis on the feature set to extract change features; marking dynamic points based on the change features; constructing an update sequence using the dynamic points, and generating a recognition benchmark based on the update sequence.

[0051] Incremental analysis is performed on the feature set generated in the above steps. The benchmark is set according to the feature distribution after weighted fusion in the feature set. For static features, the probability density function is used as the judgment standard, and for dynamic features, the state transition probability is used as the benchmark. According to the dynamic update strategy defined in the feature set, a sliding window (window length 10min, step length 1min) is set for segmentation, and the weighted contribution of each feature is calculated using the weighting scheme in the feature set. In the feature extraction process, the complete constraints in the feature set are strictly followed, including the feature correlation threshold (0.7) and the feature variance ratio range (0.5-2.0) to ensure that the analysis process does not violate the original physical constraints. Special attention is paid to the key state intervals defined in the feature set. Changes in these intervals often indicate important state transitions. Multidimensional statistical indicators are calculated for each window: mean, standard deviation, skewness and kurtosis, and compared with the standard distribution in the feature set. When the statistical indicator of a feature deviates from its standard distribution by more than 2 standard deviations, and this deviation is consistent in multiple consecutive windows, it is marked as a potential change interval. An adaptive change point detection algorithm is applied to these intervals using the update rule of the feature set. The detection parameters are dynamically adjusted according to the historical stability of the feature: a higher detection sensitivity (α=0.01) is used for features with high stability (coefficient of variation <0.1), and the sensitivity is reduced (α=0.05) for features with large fluctuations. At the same time, multi-scale analysis is introduced to verify the consistency of changes at different time scales (1min, 5min, 10min). Finally, a set of change features is obtained, each of which contains its timestamp, change amplitude, duration and reliability score.

[0052] A dynamic point marking system is constructed based on the detected change features (including timestamp, change amplitude, duration and reliability score). First, an initial time series framework is established based on the timestamp of the change feature. For each potential change interval detected by the change feature, the feature change trend within the window before and after (±5min) is calculated. Special attention is paid to the intervals with a long duration (>30s) and significant change amplitude (>2σ) in the change feature, which are likely to correspond to important state transition points. In the analysis process, the reliability score of the change feature is combined for weighting, and the change feature with high reliability has a greater influence in the dynamic point determination. When the comprehensive change index of a certain time point exceeds the threshold (0.3) and its influence range is consistent with the range predicted by the change feature, it is marked as a dynamic point. A feature description vector is constructed for each dynamic point, which contains all the statistical properties of the original change feature and supplements the results of the time series correlation analysis. At the same time, a hierarchical evaluation system for dynamic points is established: first, the change significance of a single feature dimension is evaluated, then the synergistic change pattern of the feature combination is analyzed, and finally the time series consistency of the change is verified. This process ultimately generates a set of dynamic point sequences, each of which contains a complete feature description and reliability evaluation.

[0053] The marked dynamic point sequence (including complete feature description and reliability evaluation) is used for in-depth analysis. Based on the feature description vector of the dynamic point, the similarity matrix between the points is calculated. The similarity calculation not only considers the Euclidean distance of the feature value, but also the temporal properties and reliability evaluation of the dynamic point. Based on this similarity matrix, a temporal correlation graph of the dynamic points is constructed, and the edge weights in the graph reflect both feature similarity and temporal correlation. For dynamic points with high reliability scores (score>0.8), their connections in the graph have higher weights. Graph analysis algorithms, such as PageRank variants, are applied to identify key dynamic point links, requiring that the similarity of adjacent nodes on the link is higher than 0.7 and the time interval meets the preset constraints. The extracted links are subjected to pattern analysis, and the sequence mining algorithm is used to discover typical state transition patterns. On this basis, a state transition probability matrix is ​​constructed, and the matrix elements contain not only the transition probability, but also the typical time scale and reliability evaluation of the transition. This analysis process ultimately generates a structured update sequence, in which each node in the sequence retains the complete feature information and transition relationship from the dynamic point.

[0054] The recognition benchmark is generated based on the constructed update sequence (containing complete feature information and conversion relations). The conversion patterns in the update sequence are classified, and high-frequency and stable conversion patterns (occurrence frequency>0.1, reliability>0.8) are used as basic recognition units. Each recognition unit inherits the feature threshold, timing constraints and reliability indicators in the update sequence. In terms of static features, an adaptive threshold update mechanism is established based on the complete feature information retained by the nodes in the update sequence, including statistical distribution parameters and timing change characteristics. The adjustment step of the threshold is positively correlated with the stability of the data observed in the sequence: when the sequence shows that the feature is relatively stable (coefficient of variation<0.15), a smaller adjustment step (0.05σ) is used; when the feature fluctuates greatly, the adjustment step (0.1σ) is increased. For dynamic features, the conversion relations in the update sequence are directly used to build a prediction model, and the model parameters are determined by maximum likelihood estimation. At the same time, a multi-level anomaly detection mechanism is established: the abnormal fluctuation of a single feature is detected at the feature level, the abnormal change of the feature combination is detected at the pattern level, and the abnormal pattern of the conversion sequence is detected at the system level. When an anomaly is detected, the reliability score in the update sequence is used to decide whether to trigger a model update. The generation process of the identification benchmark also includes a self-verification mechanism, which verifies the validity of the benchmark by backtesting the historical data in the update sequence. The identification benchmark finally generated is an adaptive dynamic system that can adjust the identification strategy in time according to the characteristic change trend reflected in the update sequence.

[0055] Step S103, classify the scenes according to the recognition benchmark and mark the use environment; construct a state sequence based on the use environment, select a feature template according to the state sequence, generate recognition parameters according to the feature template, and generate a recognition scheme according to the recognition parameters.

[0056] Specifically, the recognition benchmark generated in the above steps is analyzed for scene adaptability. The system classifies the input data stream in multiple dimensions based on the feature threshold system and state transition rules defined in the benchmark.

[0057] In some embodiments, the scene classification according to the recognition benchmark includes: dividing the scene into high-sensitivity scene, medium-sensitivity scene and low-sensitivity scene types based on the scene sensitivity level corresponding to the recognition benchmark; constructing a mapping relationship table between the scene and feature configuration corresponding to the recognition benchmark; matching the corresponding scene classification label according to the physical environment parameters and cognitive state parameters of the scene; establishing scene classification and feature template loading rules to complete the scene classification.

[0058] In terms of static features, the adaptive threshold mechanism in the benchmark (adjustment step size 0.05σ-0.1σ) is used to calculate the degree of match between the feature value and the benchmark interval. When the statistical distribution of the feature is highly consistent with the benchmark model (>0.85), the scene features of the time period are extracted. In terms of dynamic features, the state prediction model in the benchmark is used to analyze the feature sequence and identify the data segments that meet the expected transition pattern. Special attention is paid to the anomaly detection results of the benchmark, and the detected anomalies (deviation >3σ) and anomaly sequences are used as key markers for scene transitions. In the multidimensional feature space, the feature combinations are weighted according to the reliability scoring system defined in the benchmark. The weight distribution follows the multi-layer evaluation mechanism established in the benchmark: the feature layer score accounts for 40%, the pattern layer accounts for 35%, and the system layer accounts for 25%. Through multi-dimensional correlation analysis, time periods with similar feature patterns and transition characteristics are clustered to form preliminary scene categories. In the clustering process, the DBSCAN algorithm (ε=0.15, MinPts=5) is used to extract stable categories, and the category validity is verified by the silhouette coefficient (>0.6). For each scene category, a complete feature description vector is calculated, including static distribution characteristics (mean, variance, skewness, kurtosis) and dynamic characteristics (autocorrelation coefficient, periodicity intensity, switching frequency). These scene categories form a structured scene classification library.

[0059] The usage environment modeling was performed for the scene classification library. The feature description vectors (including static distribution characteristics and dynamic characteristics) in each scene category were deeply analyzed, focusing on the feature combinations that showed consistency. For static features, the discriminative index (Fisher discriminant ratio>1.5) and information gain (>0.4) of the feature value distribution between different scenes were calculated. For dynamic features, the consistency of feature sequences within the scene (autocorrelation coefficient>0.6) and the difference between scenes (intercorrelation coefficient<0.4) were analyzed. During the feature extraction process, the multi-layer evaluation weights (40% for feature layer, 35% for pattern layer, and 25% for system layer) were maintained, and the sliding window method (window length 5min, step length 30s) was used to capture the time-varying characteristics, and the wavelet transform was applied to extract the energy distribution of different frequency bands (δ, θ, α, β, γ). Features with strong discriminativeness and meeting the multi-layer evaluation criteria were selected to construct the environment description vector, and kernel density estimation was used for probability distribution modeling. The kernel function bandwidth was determined by maximum likelihood cross-validation (range 0.05-0.2). The weighted Mahalanobis distance is used to calculate the environmental similarity, and the weights are directly inherited from the feature discriminability index in scene classification. For scenes with similarity exceeding the threshold (0.75) and consistent feature patterns, hierarchical clustering is performed, and the clustering tree is constructed using the Ward minimum variance method. The optimal number of environmental categories is determined by the dynamic tree cutting algorithm. Finally, an environmental type library is formed, in which each environmental type contains four key elements: a standard feature distribution model, a typical conversion rule set, a stability evaluation index, and an environmental similarity matrix.

[0060] The state sequence is constructed according to the environment type library. The state identification node is extracted from the environment type, and the environment feature distribution model is used as the state judgment benchmark. The typical transition rule set is used to determine the state duration constraint. The stability evaluation index in the environment type library is directly used for state reliability assessment. The environment features with high stability (score>0.8) are given a higher state judgment weight. The state recognition adopts a multi-feature fusion method, combining time domain features (average amplitude, volatility) and frequency domain features (band energy ratio) for comprehensive judgment. The state duration strictly follows the environment definition. The stable environment (coefficient of variation <0.15) adopts a longer duration (>30s), and the changing environment adopts a short-term constraint (>10s). The Markov transition chain is constructed using the transition rule set of the environment type. The transition probability matrix is ​​obtained by maximum likelihood estimation, and its stability is evaluated by Monte Carlo simulation (1000 iterations). The state transition must meet the constraints defined by the environment similarity matrix to ensure the rationality of the transition. The system dynamically adjusts the state judgment threshold according to the stability evaluation index of the environment type library. For environments with high stability indicators, a strict threshold (±1σ) is adopted, and for volatile environments, the threshold standard is relaxed (±1.5σ). An additional verification mechanism is introduced for low-probability transitions (<0.1), requiring that multiple consecutive time windows (≥3) meet the transition conditions. In the process of generating state sequences, a three-layer state evaluation system is established at the same time: the micro-layer evaluates feature consistency (weight 0.3), the meso-layer evaluates the rationality of state transitions (weight 0.4), and the macro-layer evaluates the overall stability of the sequence (weight 0.3). Through this process, a complete state template set containing state definitions, transition rules, and evaluation criteria is formed.

[0061] Generate identification parameters based on the state template set. Each state definition in the state template set is matched with the feature pattern in the constructed state sequence. The matching process adopts a multi-objective optimization method and considers three key indicators: distribution matching (required to be >0.8), conversion consistency (required to be >0.75) and timing characteristic matching (required to be >0.7). The conversion rules in the state template set are directly used to construct the state conversion verification mechanism of the identification parameters, including conversion time constraints, characteristic conditions required for conversion and stability verification methods after conversion. For composite states, a hierarchical matching strategy is introduced to first determine the dominant features and then match the secondary features. Feature selection is based on the scoring results of the three-layer state evaluation system (micro, meso and macro). The parameter configuration process fully refers to the evaluation criteria of the state template: for high-scoring states (>0.85), select high-precision feature standards (tolerance <5%) and adopt strict parameter constraints; for medium-scoring states (0.6-0.85), select adaptable standards (tolerance 5%-15%) and adopt relatively loose parameter ranges; for low-scoring states (<0.6), select robust standards (tolerance >15%) and introduce necessary fault tolerance mechanisms. Feature conflict handling adopts a priority strategy. When different features indicate contradictory states, the final judgment is determined based on the feature reliability defined in the state sequence (defined as "stability × discrimination"). The state tracking algorithm of the recognition system is set using the conversion rules of the state template set, including state continuity verification, abnormal state filtering, and state sequence smoothing. Based on the selected standards and state template characteristics, specific recognition parameter configurations are generated. Parameter settings include feature threshold ranges, weighting coefficients, update conditions, and constraint rules provided in the state sequence, which are directly inherited from the feature definitions of the state template. All parameter configurations are verified and optimized based on actual observation data in the state sequence to ensure the practicality and effectiveness of parameter settings.

[0062] In some embodiments, generating an identification scheme based on the identification parameters includes: performing personalized marking based on the identification parameters to determine the parameter range; constructing adjustment rules based on the parameter range; performing feature configuration using the adjustment rules, and generating an identification scheme based on the feature configuration.

[0063] The identification parameters generated in the above steps are analyzed by personalized labeling. First, the characteristic threshold range, weighting coefficient, update condition and constraint rules in the identification parameter set are extracted to establish a parameter characteristic evaluation system. The static parameters are subjected to sensitivity analysis, and the influence of the parameters on the recognition results is evaluated through perturbation test (±10%). The adaptability of the identification parameters to different user data is analyzed in detail, and the individual difference coefficient (CV) of each parameter is calculated using historical data. Cluster analysis is performed on parameters with high CV values ​​(>0.2), and the K-means method (K=3-5) is used to extract typical parameter distribution patterns. For dynamic parameters, their update trigger conditions and step size settings are analyzed to evaluate their stability in different usage scenarios. In the parameter analysis process, a multi-scale evaluation method is introduced to calculate the parameter stability index in the short-term (5 minutes), medium-term (1 hour) and long-term (1 day) time windows. For parameters with large volatility, exponential smoothing (α=0.3) is used to reduce the impact of short-term fluctuations. Through these analyses, each parameter is marked with a personalized degree score (0-1). Parameters with high scores (>0.7) require a larger adjustment space, while parameters with low scores (<0.3) can adopt a narrower range of variation. Finally, the personalized adjustment range of each parameter is determined, including the minimum value, maximum value, step size and priority, forming a parameter range definition library.

[0064] The adjustment rule system is constructed based on the parameter range definition library. First, for parameters with high personalized scores (>0.7), an adaptive adjustment strategy is designed. The rules include trigger conditions, adjustment directions, and step control. The trigger conditions are based on performance evaluation indicators, such as a drop in recognition accuracy of more than 5% or more than 3 consecutive misjudgments. The adjustment direction is determined by the correlation between the current parameter value and the target performance, and the optimal parameter change path is explored through the Bayesian optimization method. For parameters with medium personalized scores (0.3-0.7), conditional adjustment rules based on usage scenarios are established, and different parameter configurations are adopted in different environments. For parameters with low personalized scores (<0.3), the minimum intervention principle is formulated, and fine-tuning is only performed when the system performance is significantly reduced. The rule construction process determines the adjustment boundary based on the minimum and maximum values ​​in the parameter range definition library, the step size determines the accuracy of each adjustment, and the priority determines the execution order of the rules. A conflict detection mechanism is introduced. When multiple rules are triggered at the same time, the final executed rule is determined based on the rule priority (1-10) and the current system status. In order to improve the robustness of the rules, the fuzzy logic method is used to define the rule triggering conditions, and the membership function (trapezoidal or Gaussian) is used to replace the hard threshold to reduce the instability of the boundary conditions. At the same time, the association adjustment rules between parameters are established. When a parameter changes, the related parameters are adjusted synchronously according to the preset ratio to ensure the overall consistency of the parameter system. Finally, a multi-level adjustment rule library is formed, including adaptive rule sets, scenario rule sets and association rule sets.

[0065] The constructed adjustment rule base is used for feature configuration. For the adaptive rule set in the adjustment rule base, the priority and execution order are set, and the high priority rules (related to key features) are executed first. The scenario rule set in the adjustment rule base is used for parameter switching in different usage environments, such as using different sensitivity settings in quiet environments and noisy environments to ensure the adaptability of feature configuration to environmental changes. The association rule set ensures that the interrelated parameters maintain a reasonable proportional relationship to prevent the system from being unstable due to the adjustment of a single parameter. For each feature parameter, an appropriate adjustment rule is selected based on its personalized score and current performance index. The static feature configuration process adopts a grid search strategy to traverse the parameter range with a preset step size and select the parameter combination with the best performance. The dynamic feature configuration adopts an online learning method to dynamically update the parameter value according to the adjustment rule. The association configuration between features follows the association rule set in the rule base to ensure the overall coordination of the feature system. To improve the configuration efficiency, a hierarchical configuration strategy is adopted: first configure the high-sensitivity parameters (influence coefficient>0.8), fix these parameters, then configure the medium-sensitivity parameters, and finally the low-sensitivity parameters. For configuration items with complex calculations, approximate calculation methods are used, such as using quadratic response surfaces instead of complete performance evaluation, to reduce the calculation complexity from O(n^3) to O(n^2). A verification mechanism is introduced during the configuration process, and the robustness of the configuration is evaluated through cross-validation (5-fold). A complete feature description is generated for each configuration, including parameter values, adjustment history, and performance evaluation results, to form a feature configuration set.

[0066] Generate personalized recognition solutions based on feature configuration sets. Integrate the static and dynamic configurations in the feature configuration set, and combine the detailed feature descriptions of each configuration item (including parameter values, adjustment history, and performance evaluation results) to form a complete recognition parameter system. The performance evaluation results in the feature configuration set are directly used for the quality evaluation of the solution, including indicators such as accuracy, recall, and F1 score, to ensure that the final solution selects the configuration combination with the best performance. For the configuration effects corresponding to different scene feature combinations, a complete scene-configuration mapping table is established to ensure that the system can quickly load the optimal parameter settings in high-sensitivity scenarios (such as fatigue driving), medium-sensitivity scenarios (such as learning status), and low-sensitivity scenarios (such as daily activities) defined by the feature configuration. Build a configuration validity verification mechanism, including parameter consistency check, compatibility check, and performance effect check. In the verification process, the performance evaluation results recorded in the feature configuration set are directly used as a benchmark. The solution generation adopts a modular design, and each functional module sets a clear parameter range based on the performance indicators of the feature description. The core recognition algorithm adopts an ensemble learning method to integrate the results of multiple basic classifiers to reduce the limitations of a single model. To improve the robustness of the system, an exception handling mechanism that has been verified by feature configuration is introduced. When an abnormal input data or abnormal processing process is detected, graceful degradation can be performed to maintain the basic functions of the system. A complete recognition scheme is generated for the configuration that has passed the verification, including a feature extraction module (using optimized feature parameters), a pattern recognition module (based on configured classification rules), and a feedback adjustment module (to achieve dynamic update of parameters). In practical applications, this personalized recognition scheme enables AI glasses to accurately identify changes in the user's attention state. For example, in driving scenarios, the system can automatically adjust the warning sensitivity according to the user's fatigue level; in learning scenarios, it can distinguish between concentration, understanding, and distraction states, and provide corresponding learning suggestions.

[0067] Step S104, performing trend analysis on the identification scheme, obtaining statistical features, marking pattern intervals based on the statistical features; constructing feature chains according to the pattern intervals, and generating feature combinations according to the feature chains.

[0068] Specifically, a trend analysis is performed on the recognition scheme of the above steps to obtain statistical features; a pattern interval is marked based on the statistical features; a feature chain is constructed using the pattern interval; and a stability representation is generated based on the feature chain.

[0069] In some embodiments, the trend analysis of the identification scheme is performed to obtain statistical features, and the pattern interval is marked based on the statistical features, including: obtaining the stability index and performance consistency index of the feature combination in the identification scheme as the statistical features to complete the trend analysis; extracting the dynamic association pattern and weight distribution characteristics corresponding to the feature combination; constructing a feature chain structure with time-series correlation; identifying unstable intervals in the feature chain through an anomaly detection mechanism for marking the pattern interval.

[0070] Perform trend analysis on the identification scheme generated by the above steps. Extract the complete parameter system of the three core components (feature extraction module, pattern recognition module and feedback adjustment module) in the scheme. Establish a parameter time series database, inherit the sensitivity classification setting time window defined in the identification scheme: high sensitivity parameters use a 10-day window, medium sensitivity parameters use a 20-day window, and low sensitivity parameters use a 30-day window. Perform time series analysis on static parameters and calculate key statistical indicators: mean, standard deviation, coefficient of variation and its rate of change. Analyze the adjustment frequency and amplitude of dynamic parameters, and build a parameter change model according to the scenario adaptation rules in the scheme. Use the ARIMA model to fit the parameter change trend and extract the model parameters as trend features. At the same time, apply the wavelet decomposition method to decompose the parameter changes into long-term trends and short-term fluctuation characteristics. For the feedback adjustment module, analyze its trigger frequency and adjustment effect to evaluate the system's adaptive ability. Introduce seasonal analysis to detect whether there is a periodic pattern in parameter changes. For significant cycles (p<0.01), perform spectrum extraction and calculate the main frequency and harmonic ratio. Through these analyses, a complete set of statistical characteristics is obtained, including stability indicators, periodicity indicators and trend indicators, which constitute the mathematical description of parameter evolution.

[0071] The pattern intervals are marked based on the obtained statistical features. The stability classification is based on the coefficient of variation in the stability index and the prediction error of the ARIMA model: high stability (coefficient of variation <0.1 and prediction error <5%), moderate stability (coefficient of variation 0.1-0.3 or prediction error 5%-15%) and low stability (coefficient of variation >0.3 or prediction error >15%). The periodicity classification is based on the power spectrum characteristics and seasonality intensity in the periodicity index: strong periodicity (significant seasonality and power spectrum peak / average power >5), weak periodicity (seasonality exists and power spectrum peak / average power is between 1-5) and non-periodicity (no significant seasonality and power spectrum peak / average power <1). The trend classification integrates the ARIMA trend term in the trend index and the Mann-Kendall test results, and is divided into three categories: significant increase, significant decrease and no significant trend. Based on these three dimensions, the classification space of the parameter pattern is constructed, and the interval division is determined by the cluster center of the statistical characteristics. The parameter distribution characteristics and transition probabilities are calculated for each interval to establish a complete interval feature description. The Bootstrap method is introduced to verify the reliability of the interval boundary and calculate the confidence interval (95%). For areas with fuzzy boundaries, a transition zone is defined, and the fuzzy set theory is used to describe the multi-interval membership of the parameters. The migration law of parameters between different intervals is extracted, and a conditional probability matrix considering the time series characteristics is constructed. Finally, a structured pattern interval library is formed, and each interval contains a clear boundary definition and feature description.

[0072] The feature chain is constructed using the pattern interval library. The transition sequence of parameters between different intervals is analyzed. The boundary definition in the pattern interval library is used to accurately identify the interval transition points. The main transition paths are identified based on the conditional probability matrix: critical transition paths (probability exceeds two standard deviations of the conditional probability mean), general transition paths (probability is within the range of mean ± two standard deviations), and rare transition paths (probability is lower than two standard deviations of the mean). The residence time threshold for identifying stable states is determined according to the boundary definition and feature description in the pattern interval library. The threshold for high stability interval is 7 days, the threshold for medium stability interval is 5 days, and the threshold for low stability interval is 3 days. A Markov chain model of parameter evolution is constructed. The state space is based on the pattern interval definition, and the transfer matrix inherits the transition probability between intervals. The co-evolution relationship between parameters is analyzed, and the correlation grouping is based on the cross-correlation analysis of time series features: strong correlation group (cross-correlation coefficient>0.7 and continuously stable) and weak correlation group (cross-correlation coefficient fluctuates or is between 0.3-0.7). The strong correlation group constructs a joint Markov model, and the weak correlation group maintains an independent model and sets conditional constraints based on fuzzy membership. The topological structure of the feature chain is analyzed by graph theory algorithms to extract strongly connected components, key paths and loop structures. Finally, a complete feature chain network is generated to describe the parameter evolution process.

[0073] In some embodiments, generating a feature combination according to the feature chain includes: generating a stability representation according to the feature chain; performing feature management according to the stability representation and establishing a feature library; performing pattern marking based on the feature library; constructing a recognition sequence using the pattern marking; generating a feature combination according to the recognition sequence;

[0074] Generate a stability representation based on the feature chain network. Perform a hierarchical stability analysis on the network structure, and calculate the state residence probability, state return probability, and network entropy based on the micro, meso, and macro hierarchical structures defined by the feature chain. For the strongly connected components identified in the feature chain, calculate their stability thresholds: high-frequency transition paths correspond to high stability thresholds (>0.8), general paths correspond to medium stability thresholds (>0.6), and low-frequency paths set low stability thresholds (>0.4). The key paths extracted from the feature chain are directly used to construct the backbone sequence of state transitions, determine the core stable state and key turning points of the system, and nodes on the key paths are given higher stability weights. The loop structure is used to identify the steady-state oscillation mode in the system, analyze its periodic characteristics and triggering conditions, and set the corresponding stability evaluation criteria. For the high stability area, analyze its attraction domain range and convergence speed, and establish a stable state description model corresponding to the hierarchical structure of the feature chain. In the transition area, the Markov model of the feature chain is continued to analyze its dynamic characteristics, including transition directionality and transition rate. A stability prediction model based on hierarchical feature chains is established, and the future state of the parameters and their confidence intervals are predicted using deep learning methods. At the same time, an anomaly detection algorithm is developed based on the abnormal transfer mode of the feature chain to identify abnormal states that deviate from the expected evolution path. A complete stability portrait is established for each stable area, including area range, convergence characteristics, anti-interference ability and state prediction. In the daily use scenarios of AI glasses, stability representation helps the system understand the user's usage habits, such as identifying the concentration ability curve and providing intelligent cognitive support.

[0075] Combined with the stability representation of the above steps, feature management is performed. First, three layers of stable features are extracted from the stability representation: single parameter stability characteristics at the micro level, parameter group collaborative stability at the meso level, and system stable state at the macro level. Features in the high stability region (first-order stability>0.7) are prioritized and used as the core components of the feature library. At the same time, dynamic evolution rules are extracted from the stability representation, including state transition paths, convergence characteristics, and anomaly detection patterns, to build a dynamic update mechanism for features. A hierarchical storage structure is established: high-frequency and high-stability features are placed in the fast storage area, and features that are determined to be stable by anomaly detection are given high storage priority; medium-stability features are placed in the main storage area; low-stability features and fluctuation features marked by anomaly detection are placed in the backup storage area. According to the feature access mode, an adaptive index structure is designed to optimize common query modes (such as similarity query and range query). Feature lifecycle management is introduced, and the validity period and update strategy are set for the features based on the temporal information and anomaly pattern prediction in the stability representation. The association between features is represented by a graph structure, and the edge weight is determined by the correlation coefficient and anomaly co-occurrence probability in the stability representation. Through these operations, a structured feature library is established, which includes static feature sets, dynamic evolution rules and association relationship networks. In the daily use of AI glasses, this feature management can dynamically adjust resource allocation according to the stability of user habits, such as allocating more computing resources to the user's stable attention indicator features to improve recognition accuracy and response speed.

[0076] Pattern labeling is performed based on the established feature library. Features are extracted hierarchically from the static feature set in the feature library: high stability features in the fast storage area are used first, medium stability features in the main storage area, and low stability features in the backup storage area are used last. A hierarchical clustering method (Ward minimum variance method) is applied to the selected static feature set to construct a feature tree. The clustering threshold corresponds to the storage hierarchy: a strict threshold (0.8) is used for the fast storage area features, a medium threshold (0.6) is used for the main storage area, and a loose threshold (0.4) is used for the backup storage area. Each cluster represents a potential pattern category, and a pattern descriptor is constructed by calculating the statistical distribution (mean, variance, skewness, kurtosis) and dynamic characteristics (autocorrelation coefficient, conditional entropy) of the intra-class features. Special attention is paid to the association relationship network in the feature library. The strongly associated feature groups are identified based on the graph community detection algorithm, and the detection parameters are adaptively adjusted according to the storage hierarchy. The dynamic evolution rules of the feature library are used to analyze the temporal evolution characteristics of the pattern, including pattern duration, transition frequency, and periodicity. Assign a unique identifier to each identified pattern category, and build a multi-level tagging system corresponding to the storage hierarchy: core tags for fast storage areas, extended tags for primary storage areas, and supplementary tags for backup storage areas. Verify the tagging results, and evaluate the consistency and reliability of the tags through cross-validation (10 folds). Finally, a complete set of pattern tags is formed, and each pattern contains a detailed description of feature distribution and dynamic evolution characteristics. In the learning assistance scenario of AI glasses, this pattern tagging can accurately distinguish different cognitive states, such as subdividing the "focus" state into sub-modes such as "deep thinking", "information acquisition" and "creative thinking", providing more accurate learning status feedback.

[0077] The recognition sequence is constructed using the pattern tag set. First, based on the feature distribution description and dynamic evolution characteristics in the pattern tag set, a hierarchical recognition framework is constructed in combination with storage hierarchy: the fast storage area features construct the core recognition chain, the main storage area features construct the auxiliary recognition chain, and the spare storage area features construct the supplementary recognition chain. The temporal correlation between the patterns at each level is analyzed, and the pattern conversion probability matrix defined in the tag set is calculated. The matrix structure corresponds to the storage hierarchy. A hierarchical Markov model is constructed based on this matrix, and each layer captures the conversion rules of different granularities. In order to improve the model's expressiveness, the feature distribution description in the pattern tag set is introduced as context information to construct a conditional random field model. The state space is organized according to the storage hierarchy, and the feature function is designed based on the feature distribution description and dynamic evolution characteristics of the pattern. The model training adopts maximum likelihood estimation and uses the L-BFGS optimization algorithm. For patterns with strong temporal dependence, the long short-term memory network (LSTM) is introduced, and the network structure corresponds to the storage hierarchy: the fast storage area features use a deep network, the main storage area uses a medium-sized network, and the spare storage area uses a shallow network. A hierarchical confidence evaluation mechanism is introduced in the recognition process. Each level calculates the confidence independently and obtains the final result through weighted combination. For low-confidence recognition results, the verification process is started layer by layer according to the storage level. Finally, a hierarchical recognition sequence is obtained, which contains multi-level pattern recognition rules, sequence constraints and prediction mechanisms. In the conference scene application of AI glasses, the recognition sequence can track the changes in the user's participation status in real time, identify the state transition sequence from "passive listening" to "active thinking" and then to "preparing to speak", and provide a summary of the meeting content or speech suggestions at the appropriate time.

[0078] Generate feature combinations based on the recognition sequence. For each key node in the recognition sequence, apply pattern recognition rules to extract its corresponding feature requirements, including required feature sets and optional feature sets. Required features are directly selected from the core features of the pattern tag, while optional features are selected based on context matching. The feature selection process adopts the forward-backward search strategy defined in the sequence constraints. First, the core features are added, then auxiliary features are added through incremental evaluation, and finally unnecessary features are removed through redundancy analysis. For dynamic scenarios, a feature adaptation mechanism is constructed to prepare the feature sets that may be needed in advance according to the prediction mechanism of the recognition sequence. A feature fusion layer is introduced to integrate the selected features. The fusion method is based on the rule selection in the sequence constraints, including feature connection, weighted average and nonlinear transformation. The fusion parameters are optimized by grid search (parameter space 10×10×10). In order to improve the robustness of the combination, multi-version feature combinations are constructed to form an integrated recognition framework. Different versions of the combination are optimized for different noise environments and user states. The generated feature combinations are comprehensively evaluated, with indicators including recognition accuracy, computational complexity, memory usage and response time. Finally, a feature combination library is formed, each of which contains a complete list of features, combination methods and applicable conditions. In the sports scenes of AI glasses, the feature combination can adapt to the changes in EEG features under different exercise intensities, such as from rest to light activity to high-intensity exercise, dynamically adjust the feature weights, maintain stable cognitive state monitoring, and help users maintain the best training state and concentration.

[0079] Step S105, performing pattern analysis on the feature combination, marking the recognition sequence; constructing a mapping relationship based on the recognition sequence; utilizing the mapping relationship to perform feature recombination, and generating a personalized pattern based on the feature recombination.

[0080] Specifically, the feature combination library generated in the above steps is subjected to pattern analysis. The constituent elements, combination methods and applicable conditions of each feature combination are extracted from the combination library, and the integration results of multiple versions of feature combinations are integrated. The performance indicators of the combination are evaluated in multiple dimensions, including recognition accuracy, computational complexity, response time, and consistency indicators between integrated versions. The adaptability of different combinations and their integrated versions in various scenarios is analyzed using historical usage data, and a scenario-combination mapping matrix is ​​established. The similarity between combinations is measured, the Jaccard coefficient is used to calculate the overlap of feature sets, and the EMD distance is used to evaluate the difference in feature weight distribution. Spectral clustering is performed based on the similarity results and the grouping results of the integrated recognition framework. The number of clusters is consistent with the number of versions of the integrated framework to identify combination families with common characteristics. For each combination family, its common core features and differential features are extracted, the noise adaptation characteristics in the multi-version feature combination are integrated, and the template structure of the feature combination is constructed. The identified patterns are arranged in time sequence, the conversion rules between patterns are analyzed, and typical recognition sequence templates are extracted. These sequence templates are marked to establish a sequence description library containing sequence structure, conversion conditions and performance characteristics. In daily interaction scenarios of AI glasses, this recognition sequence marker can capture the process of users' attention shifting from distraction to gradual focus, helping the system identify the best time for interaction and avoiding sending non-urgent notifications when users need to concentrate.

[0081] The mapping relationship is constructed based on the labeled recognition sequence description library. According to the sequence structure in the sequence description library, the feature requirements of its composition patterns are analyzed, and the sequence-feature dependency graph is established. The dependency graph adopts a directed weighted graph structure, where nodes represent features or patterns, edges represent dependencies, and weights reflect the strength of dependencies. The conversion conditions in the sequence description library are used for path analysis to extract the conversion path from basic features to advanced patterns. Based on the performance characteristics of the sequence description library, a forward mapping function from features to patterns is established, and the network structure corresponds to the classification of the combination family: each combination family corresponds to a subnetwork, and the subnetwork dimension is adaptively set according to the complexity of the features within the family, forming a hierarchical mapping structure as a whole. At the same time, the reverse mapping from pattern to feature is constructed by combining the sequence structure and the conversion conditions, maintaining the symmetric structure with the forward mapping. The design of the mapping function takes into account the nonlinear feature interaction, and introduces the attention mechanism to highlight the impact of the key features identified in the combination family. The robustness of the mapping relationship is evaluated by injecting noise that matches the distribution of the combination family features and testing based on the performance characteristics in the sequence description library. Finally, a bidirectional mapping relationship library is formed, including forward recognition mapping and reverse inference mapping, and each mapping is equipped with accuracy and reliability evaluation. In the meeting scenario of AI glasses, this mapping relationship can associate the user's attention change pattern with the user's speaking intention. When the EEG feature sequence of the user preparing to speak is detected, the speech enhancement and recording functions are automatically activated.

[0082] The established bidirectional mapping relationship library is used for feature recombination. First, the user historical data is analyzed through forward recognition mapping to extract personalized feature fingerprints, and reverse derivation mapping is applied to verify the consistency of the extraction results. Feature fingerprints contain the typical numerical range, fluctuation characteristics and combination patterns of user features. These individual features are compared with the standard patterns in the mapping relationship library, and the matching degree and deviation degree are calculated using accuracy and reliability evaluation. For features with high matching degree, the standard configuration is retained; for features with medium matching degree, parameters are fine-tuned through forward recognition mapping; for features with low matching degree, the feature extraction and combination scheme is reconstructed using reverse derivation mapping. The feature recombination process adopts evolutionary algorithm, and the parameter setting is dynamically adjusted according to the mapping complexity: the population size is proportional to the number of mapping layers, the evolutionary generation is proportional to the number of mapping nodes, and the selection pressure is proportional to the mapping accuracy requirement. The fitness function integrates the accuracy and reliability evaluation indicators in the mapping relationship library, and the weight distribution is adaptively adjusted according to the importance of the mapping. The knowledge guidance mechanism is introduced in the evolutionary process, and the search space is constrained by the rules in the mapping relationship library. The evolutionary results are analyzed for diversity, and multiple optimal solutions with their own advantages are retained as alternatives. Finally, a set of feature reorganization schemes optimized for users are obtained, each of which includes a list of features, combination methods, and expected performance. In the reading scenario of AI glasses, feature reorganization can adapt to the reading EEG patterns of different users. Some users have enhanced alpha waves when concentrating on reading, while others have dominant theta waves. The system adapts to individual differences by reorganizing features and provides personalized reading assistance.

[0083] Generate personalized patterns according to the feature recombination scheme. Apply the feature list in the recombination scheme to the real-time data processing process, use the combination method specified by the scheme to build a personalized feature extraction-combination-recognition link, and set the initial parameters according to the expected performance. Perform targeted optimization on each link: adjust the filter parameters and window length in the feature extraction stage to match the user's signal characteristics; optimize the weight configuration and fusion method in the feature combination stage to highlight the user's significant features; adjust the decision threshold and discrimination boundary in the recognition stage to adapt to the user's category distribution. The choice of feature list directly affects the initial configuration of the processing pipeline, the combination method determines the specific algorithm of feature fusion, and the expected performance is used to set the optimization goal and evaluation benchmark of the system. At the same time, establish a pattern adaptability evaluation mechanism to continuously optimize the pattern parameters through online learning methods. The learning strategy adopts incremental update to ensure that the pattern can be smoothly adjusted as the user's habits change. In order to improve the generalization ability of the pattern, domain adaptation technology is introduced to enable the pattern to migrate between different usage environments. Establish a complete description document for the generated personalized pattern, including parameter configuration, performance characteristics and applicable conditions. A / B testing is used to verify the performance improvement of personalized patterns compared with general patterns. The evaluation indicators include improved accuracy, shortened response time and user satisfaction. Finally, a complete set of personalized mode libraries is formed, which can automatically select the most suitable recognition mode according to user characteristics and usage scenarios. In the working scenario of AI glasses, the personalized mode can identify each user's unique fatigue precursor EEG characteristics. Some users show weakened beta waves, while others show enhanced delta waves. Based on this, the system provides personalized rest reminders to help users maintain the best working state and efficiency.

[0084] The provided method has at least the following beneficial effects:

[0085] 1. By constructing personal benchmarks and feature mapping, rapid personalization is achieved under low data volume conditions, reducing the user's initial adaptation cost while retaining the generalization ability of the general model, enabling the system to fully utilize the common characteristics in group data while maintaining personalized recognition accuracy.

[0086] 2. The introduction of scene classification and environmental adaptation mechanism enables the recognition system to perceive the user's environment and dynamically adjust the recognition parameters. Through the establishment of recognition benchmarks, adaptive adjustment of parameter ranges, and dynamic configuration of recognition schemes, the system can maintain stable recognition performance in complex scenarios such as user activity state transitions (such as from sitting to walking), environmental conditions changes (such as switching between indoor and outdoor), and task type changes (such as from reading to talking). This multi-level environmental adaptation mechanism solves the problem of drastic fluctuations in recognition performance of traditional methods when the environment changes.

[0087] 3. A complete pattern evolution and feature reorganization framework has been established. The system can capture the long-term changes in the user's EEG pattern and actively adjust the recognition strategy. Through the establishment of stable representation, dynamic reconstruction of feature combinations, and continuous optimization of personalized patterns, the system has achieved adaptive tracking of long-term changes in user cognitive habits, attention patterns, fatigue characteristics, etc. This continuous evolution mechanism ensures that the system can continuously adjust as the user's usage habits change, overcomes the limitation of traditional methods that cannot adapt to the long-term changes in users' habits, and maintains a stable recognition effect.

[0088] In order to implement the personalized EEG pattern recognition method based on AI glasses corresponding to the above method embodiment, so as to achieve the corresponding functions and technical effects. Figure 2 , Figure 2 The structure block diagram of a personalized EEG pattern recognition device 200 provided in an embodiment of the present application is shown. For the convenience of description, only the parts related to the present embodiment are shown. The personalized EEG pattern recognition device 200 provided in an embodiment of the present application includes:

[0089] The benchmark establishment unit 201 is used to perform feature marking on the EEG data collected by the AI ​​glasses and construct a personal benchmark; perform pattern analysis based on the personal benchmark to obtain a feature map; use the analysis result corresponding to the pattern analysis to mark the feature interval to generate a user portrait, and obtain an initial template corresponding to the user portrait;

[0090] A feature mapping unit 202 is used to perform feature mapping according to the initial template, obtain common features, and mark individual features based on the common features; construct fusion rules using the individual features, generate a feature set according to the fusion rules, and obtain an identification benchmark corresponding to the feature set;

[0091] A scene classification unit 203 is used to classify scenes according to the recognition reference and mark the use environment; construct a state sequence based on the use environment, select a feature template according to the state sequence, generate recognition parameters according to the feature template, and generate a recognition scheme according to the recognition parameters;

[0092] A trend analysis unit 204 is used to perform trend analysis on the identification scheme, obtain statistical features, mark pattern intervals based on the statistical features, construct a feature chain according to the pattern intervals, and generate a feature combination according to the feature chain;

[0093] The pattern analysis unit 205 is used to perform pattern analysis on the feature combination and mark the recognition sequence; construct a mapping relationship based on the recognition sequence; use the mapping relationship to perform feature recombination, and generate a personalized pattern based on the feature recombination. The above-mentioned personalized EEG pattern recognition device 200 can implement the personalized EEG pattern recognition method based on AI glasses in the above-mentioned method embodiment. The optional items in the above-mentioned method embodiment are also applicable to this embodiment and will not be described in detail here. The rest of the contents of the embodiments of the present application can refer to the contents of the above-mentioned method embodiments, and will not be repeated in this embodiment.

[0094] Figure 3 This is a schematic diagram of the structure of a computer device provided in one embodiment of the present application. Figure 3 As shown, the computer device 3 of this embodiment includes: at least one processor 30 ( Figure 3 Only one is shown in the figure), a memory 31 and a computer program 32 stored in the memory 31 and executable on the at least one processor 30, wherein the processor 30 implements the steps of any of the above method embodiments when executing the computer program 32.

[0095] The computer device 3 may be a computing device such as a smart phone, a tablet computer, a desktop computer, a cloud server, etc. The computer device may include but is not limited to a processor 30 and a memory 31. Those skilled in the art will understand that Figure 3 It is only an example of computer device 3 and does not constitute a limitation on computer device 3. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components, for example, it may also include input and output devices, network access devices, etc.

[0096] The processor 30 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0097] In some embodiments, the memory 31 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 31 may also be an external storage device of the computer device 3, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device 3. Further, the memory 31 may also include both an internal storage unit and an external storage device of the computer device 3. The memory 31 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program. The memory 31 may also be used to temporarily store data that has been output or is to be output.

[0098] In addition, an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0099] An embodiment of the present application provides a computer program product. When the computer program product is run on a computer device, the computer device implements the steps in the above-mentioned method embodiments when executing the computer device.

[0100] In several embodiments provided in the present application, it is understood that each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved.

[0101] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0102] The specific embodiments described above further describe the purpose, technical solutions and beneficial effects of the present application in detail. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the scope of protection of the present application. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of protection of the present application.

Claims

1. A personalized EEG pattern recognition method based on AI glasses, characterized in that: include: Feature labeling of EEG data collected by AI glasses and building a personal benchmark; Performing pattern analysis based on the personal benchmark to obtain a feature map; Using the analysis result corresponding to the pattern analysis to mark the feature interval to generate a user portrait, and obtaining an initial template corresponding to the user portrait; Perform feature mapping according to the initial template to obtain common features, and mark individual features based on the common features; Constructing a fusion rule using the individual characteristics, generating a feature set according to the fusion rule, and obtaining an identification benchmark corresponding to the feature set; Classify the scene according to the recognition benchmark and mark the use environment; construct a state sequence based on the use environment, select a feature template according to the state sequence, generate recognition parameters according to the feature template, and generate a recognition scheme according to the recognition parameters; Performing trend analysis on the identification scheme to obtain statistical features, and marking pattern intervals based on the statistical features; constructing a feature chain according to the pattern interval, and generating a feature combination according to the feature chain; Performing pattern analysis on the feature combination to mark an identification sequence; and constructing a mapping relationship based on the identification sequence; The mapping relationship is used to perform feature reorganization, and a personalized mode is generated according to the feature reorganization.

2. The method according to claim 1, characterized in that The obtaining of the initial template corresponding to the user portrait includes: Performing regional decomposition on the user portrait and marking key features to construct a feature space; Performing sample matching based on the feature space; An initial template is generated based on the feature combination of the matching result of the sample matching.

3. The method according to claim 1, characterized in that The obtaining of the recognition benchmark corresponding to the feature set includes: Performing incremental analysis on the feature set to extract change features; Marking dynamic points based on the change characteristics; An update sequence is constructed using the dynamic points, and a recognition benchmark is generated according to the update sequence.

4. The method according to claim 1, characterized in that: The generating of the identification scheme according to the identification parameters comprises: Perform personalized marking according to the identification parameters and determine the parameter range; constructing adjustment rules based on the parameter range; The adjustment rules are used to perform feature configuration, and an identification scheme is generated according to the feature configuration.

5. The method according to claim 1, characterized in that Generating a feature combination according to the feature chain includes: generating a stability representation based on the feature chain; Perform feature management according to the stability representation and establish a feature library; Performing pattern marking based on the feature library; constructing a recognition sequence using the pattern marking; A feature combination is generated according to the recognition sequence.

6. The method according to claim 1, characterized in that The feature marking of the EEG data collected by the AI ​​glasses and the establishment of a personal benchmark include: Preprocessing the EEG data to remove physiological artifacts and interfering noise; Extracting the time domain features, frequency domain features and time-frequency domain features of the EEG data and performing dimensionality reduction processing; Establishing a characteristic statistical parameter library of the EEG data through standardization processing; analyzing the time series correlation characteristics and fluctuation characteristics of the characteristic statistical parameter library in combination with stratified sampling data under different cognitive states; The characteristic stability index of the characteristic statistical parameter library is calculated according to the time series correlation characteristics and fluctuation characteristics, and a personal benchmark database including characteristic parameters, probability distribution models and time series correlation characteristics is constructed.

7. The method according to claim 1, characterized in that The classifying the scene according to the recognition reference includes: Dividing the scene types into high-sensitivity scene, medium-sensitivity scene and low-sensitivity scene based on the scene sensitivity level corresponding to the recognition reference; Constructing a mapping relationship table between scenes and feature configurations corresponding to the recognition benchmark; The corresponding scene classification labels are matched according to the physical environment parameters and cognitive state parameters of the scene; and the loading rules of the scene classification and feature templates are established to complete the scene classification.

8. The method according to claim 1, characterized in that The performing trend analysis on the identification scheme to obtain statistical features, and marking the mode interval based on the statistical features, includes: Obtaining the stability index and the performance consistency index of the feature combination in the identification scheme as the statistical feature to complete the trend analysis; Extract dynamic association patterns and weight distribution characteristics corresponding to feature combinations; A feature chain structure with time series correlation is constructed; an unstable interval in the feature chain is identified through an anomaly detection mechanism to mark the pattern interval.

9. A personalized EEG pattern recognition device, characterized in that: include: A benchmark establishment unit, which is used to feature the EEG data collected by the AI ​​glasses and build a personal benchmark; Performing pattern analysis based on the personal benchmark to obtain a feature map; using the analysis result corresponding to the pattern analysis to mark the feature interval to generate a user portrait, and obtaining an initial template corresponding to the user portrait; A feature mapping unit, configured to perform feature mapping according to the initial template, obtain common features, and mark individual features based on the common features; Constructing a fusion rule using the individual characteristics, generating a feature set according to the fusion rule, and obtaining an identification benchmark corresponding to the feature set; A scene classification unit, configured to classify scenes according to the recognition benchmark and mark the use environment; construct a state sequence based on the use environment, select a feature template according to the state sequence, generate recognition parameters according to the feature template, and generate a recognition scheme according to the recognition parameters; A trend analysis unit, configured to perform trend analysis on the identification scheme, obtain statistical features, and mark pattern intervals based on the statistical features; constructing a feature chain according to the pattern interval, and generating a feature combination according to the feature chain; A pattern analysis unit, used to perform pattern analysis on the feature combination and mark an identification sequence; Building a mapping relationship based on the recognition sequence; The mapping relationship is used to perform feature reorganization, and a personalized mode is generated according to the feature reorganization.

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