Method, device and equipment for predicting user behavior based on EEG in smart glasses

Abstract

The present application relates to the field of brain-computer interface technology, and in particular to a method, device and apparatus for predicting user behavior based on EEG in smart glasses. The method comprises: collecting multi-channel EEG data of smart glasses to obtain original signals; segmenting data based on original signals, obtaining behavioral features for sequence analysis, obtaining behavioral samples to construct a behavior library; performing time series mapping on the behavior library, obtaining behavioral paths to extract dependencies and construct sequence rules; generating prediction tags based on sequence rules to perform environmental analysis, obtaining scene features for correlation matching to construct fusion sequences to generate prediction results, perform feature decomposition, mark prediction intervals for rule construction to determine confidence, generate judgment sequences and extract behavioral patterns; obtain prediction parameters based on behavioral patterns; construct update rules based on prediction parameters, generate prediction strategies based on update rules, and output behavior prediction results based on prediction strategies. The accurate extraction of behavioral intention information in EEG signals is achieved.

Description

Technical Field

[0001] The present application relates to the field of brain-computer interface technology, and in particular to a method, device and equipment for predicting user behavior based on EEG in smart glasses. Background Art

[0002] Smart glasses show enormous potential for understanding user behavior and enabling intelligent interaction. Traditional behavior prediction methods rely primarily on historical user behavior data and explicit operation records, failing to predict user intentions in advance. Although EEG signals contain rich information about behavioral intentions, current analysis methods focus primarily on simple intention recognition and struggle to predict complex behaviors. These challenges urgently require breakthroughs in multiple technical areas, including signal processing, feature extraction, and predictive modeling. Furthermore, existing technologies lack the real-time, accuracy, and generalizability of behavior prediction.

[0003] In real-world applications, users' behavioral intentions are often influenced by a variety of factors, including cognitive state, environmental conditions, and task requirements. Existing prediction models struggle to effectively integrate these influencing factors, resulting in unreliable predictions. Furthermore, the quality of EEG signals collected by smart glasses, limited by their portable nature, suffers from low signal-to-noise ratio and poor stability, further complicating behavior prediction. Furthermore, the individual variability and time-varying nature of user behavior patterns pose significant challenges to the design of prediction models. These technical difficulties require systematic solutions to overcome them.

[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 method, device and equipment for predicting user behavior based on EEG in smart glasses. The method is intended to solve the problem that in actual applications, the user's behavioral intention is often affected by multiple factors, including cognitive state, environmental conditions and task requirements. It is difficult for existing prediction models to effectively integrate these influencing factors, resulting in a lack of reliability in the prediction results. In addition, the quality of EEG signals collected by smart glasses is limited by the characteristics of portable devices, and there are problems such as low signal-to-noise ratio and poor stability, which further increases the difficulty of behavior prediction. In addition, the individual differences and time-varying characteristics of user behavior patterns also bring huge challenges to the design of prediction models. These technical difficulties need to be overcome through systematic solutions.

[0006] In a first aspect, an embodiment of the present application provides a method for predicting user behavior based on EEG in smart glasses, comprising:

[0007] Collecting multi-channel EEG data of the smart glasses to obtain original signals; segmenting the data based on the original signals, generating signal sequences according to segmentation results corresponding to the data segments, obtaining feature sequences corresponding to the signal sequences, and obtaining behavioral features corresponding to the feature sequences;

[0008] Perform sequence analysis based on the behavior characteristics to obtain behavior samples, and build a behavior library based on the behavior samples; perform time sequence mapping on the behavior library to obtain behavior paths; extract dependency relationships based on the behavior paths; build sequence rules based on the dependency relationships; and generate prediction tags based on the sequence rules;

[0009] Performing environmental analysis based on the prediction mark to obtain scene features; performing association matching on the scene features, constructing a fusion sequence based on the matching results of the association matching, and generating a prediction result based on the fusion sequence; performing feature decomposition on the prediction result to mark a prediction interval; constructing rules based on the prediction interval; determining a confidence level based on the rule, and generating a decision sequence based on the confidence level;

[0010] Behavior analysis is performed according to the judgment sequence to extract behavior patterns; prediction parameters are obtained based on the behavior patterns; update rules are constructed according to the prediction parameters, a prediction strategy is generated according to the update rules, and behavior prediction results are output according to the prediction strategy.

[0011] In a second aspect, the present application further provides a user behavior prediction device, comprising:

[0012] A data acquisition unit is used to collect multi-channel EEG data of the smart glasses to obtain original signals; segment the data based on the original signals, generate signal sequences according to the segmentation results corresponding to the data segments, obtain feature sequences corresponding to the signal sequences, and obtain behavioral features corresponding to the feature sequences;

[0013] a sequence analysis unit configured to perform sequence analysis based on the behavior characteristics, obtain behavior samples, and construct a behavior library based on the behavior samples; perform time sequence mapping on the behavior library to obtain behavior paths; extract dependency relationships based on the behavior paths; construct sequence rules using the dependency relationships; and generate prediction tags based on the sequence rules;

[0014] an environment analysis unit configured to perform environment analysis based on the prediction mark to obtain scene features; perform association matching on the scene features, construct a fusion sequence based on the matching results of the association matching, and generate a prediction result based on the fusion sequence; perform feature decomposition on the prediction result to mark a prediction interval; construct a rule based on the prediction interval; determine a confidence level based on the rule, and generate a decision sequence based on the confidence level;

[0015] A behavior analysis unit is used to perform behavior analysis based on the judgment sequence and extract behavior patterns; obtain prediction parameters based on the behavior patterns; construct update rules based on the prediction parameters, generate prediction strategies based on the update rules, and output behavior prediction results based on the prediction strategies.

[0016] 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 method for predicting user behavior based on EEG in smart glasses as described in the first aspect is implemented.

[0017] This method uses a multi-channel electrode array in smart glasses to collect raw EEG signals, segment them, and generate signal sequences. A high-dimensional temporal feature matrix is ​​extracted from the signal sequence, and a structured feature sequence is constructed using behavioral indicator markers (e.g., attention and cognitive load status). The feature sequence is hierarchically processed (rapid response layer, medium-term change layer, and long-term trend layer), extracting independent components (e.g., attention shift component and cognitive load component). A feature hierarchy is then constructed to generate behavioral signatures. Based on behavioral signature analysis, behavioral samples are generated, and a behavioral library containing multidimensional feature templates is constructed. Time-series mapping is performed on the behavioral library, analyzing micro-dependencies, meso-level combination patterns, and macro-level trends within behavioral paths. Temporal rules (e.g., conditional transition probabilities) are extracted to generate prediction signatures. Environmental analysis is then performed on the prediction signatures to extract scenario features (e.g., task requirements and environmental constraints). A fusion sequence is generated through association matching. The prediction results are then feature-decomposed, with prediction intervals labeled (immediate, medium-term, and long-term). Confidence rules are then constructed and a decision sequence is generated. Finally, behavioral pattern analysis (e.g., individual adaptation mechanisms) is used to extract prediction parameters, generate a dynamically updated prediction strategy, and output the behavioral prediction results.

[0018] Through multi-channel signal segmentation, hierarchical processing, and component extraction techniques (such as ICA independent component analysis), the problem of unstable signal quality in portable devices is overcome, and the temporal resolution and expressiveness of features are improved. The systematic expression of behavioral indicator markers (such as attention concentration and cognitive fatigue) and feature hierarchies enables accurate capture of complex behavioral intentions. The construction of a behavioral library based on feature templates and multidimensional temporal rules (such as conditional transition probabilities) enhances the temporal logic and adaptability of predictions. The correlation and matching of environmental features with behavioral paths (such as scenario constraint analysis) improves the reliability and real-time performance of prediction results. Feature decomposition and dynamic update rules (such as confidence determination sequences) within the prediction interval address the challenges of individual differences and time-varying characteristics of user behavior. Behavioral pattern analysis (such as pattern transition patterns) and a prediction strategy generation mechanism ensure the system's stability and generalization capabilities in complex scenarios.

[0019] 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

[0020] Figure 1 This is a flow chart of a method for predicting user behavior based on EEG in smart glasses according to an embodiment of the present application;

[0021] Figure 2 This is a schematic diagram of the structure of a user behavior prediction device shown in an embodiment of the present application;

[0022] Figure 3 This is a schematic diagram of the structure of a computer device shown in an embodiment of the present application. DETAILED DESCRIPTION

[0023] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0024] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0025] It will also be understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0026] As used in this specification and the appended claims, the term "if" can be interpreted as "when" or "upon" 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 "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0027] 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.

[0028] References to "one embodiment" or "some embodiments" in this specification mean that a particular feature, structure, or characteristic described in conjunction with that embodiment is included in one or more embodiments of the present application. Thus, phrases such as "in one embodiment," "in some embodiments," "in other embodiments," and "in other embodiments" appearing in various places in this specification do not necessarily refer to the same embodiment, but rather mean "one or more but not all embodiments," unless otherwise specifically emphasized. The terms "including," "comprising," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0029] The technical solutions of the embodiments of this application are introduced below.

[0030] Smart glasses show enormous potential for understanding user behavior and enabling intelligent interaction. Traditional behavior prediction methods rely primarily on historical user behavior data and explicit operation records, failing to predict user intentions in advance. Although EEG signals contain rich information about behavioral intentions, current analysis methods focus primarily on simple intention recognition and struggle to predict complex behaviors. These challenges urgently require breakthroughs in multiple technical areas, including signal processing, feature extraction, and predictive modeling. Furthermore, existing technologies lack the real-time, accuracy, and generalizability of behavior prediction.

[0031] In real-world applications, users' behavioral intentions are often influenced by a variety of factors, including cognitive state, environmental conditions, and task requirements. Existing prediction models struggle to effectively integrate these influencing factors, resulting in unreliable predictions. Furthermore, the quality of EEG signals collected by smart glasses, limited by their portable nature, suffers from low signal-to-noise ratio and poor stability, further complicating behavior prediction. Furthermore, the individual variability and time-varying nature of user behavior patterns pose significant challenges to the design of prediction models. These technical difficulties require systematic solutions to overcome them.

[0032] Please refer to Figure 1 , Figure 1 The flowchart of the method for predicting user behavior based on EEG in smart glasses provided in the embodiment of the present application is shown in FIG. The method for predicting user behavior based on EEG in smart glasses provided in the embodiment of the present application can be applied to computer devices, including but not limited to smart phones, laptops, tablet computers, desktop computers, physical servers, cloud servers and other devices. Figure 1 As shown, the EEG-based user behavior prediction method in the smart glasses of this embodiment includes steps S101 to S104, which are described in detail as follows:

[0033] Step S101, collect multi-channel EEG data of smart glasses to obtain original signals; segment the data based on the original signals, generate signal sequences according to the segmentation results corresponding to the data segments, obtain feature sequences corresponding to the signal sequences, and obtain behavioral features corresponding to the feature sequences.

[0034] Specifically, smart glasses collect multi-channel EEG data from users primarily through arrays of dry electrodes placed at strategic locations such as the temples, forehead, and behind the ears. During acquisition, a differential amplifier circuit performs preliminary signal processing, and the sampling frequency is set at 1024Hz to ensure the signal's time-frequency characteristics are not distorted. Considering the portable nature of this device, adaptive gain control technology dynamically adjusts the signal amplitude to maintain it within a ±100μV range. Furthermore, an independent reference electrode is introduced to eliminate common-mode interference, and an impedance self-detection circuit monitors the contact impedance between the electrode and the skin in real time. Data exceeding 50kΩ is automatically marked as invalid. In practical applications, the raw signal acquisition duration is typically set to 5-10 minutes to cover the user's EEG activity characteristics under different cognitive states. This acquisition scheme produces a multi-channel raw EEG data stream containing complete timestamps, electrode positions, gain coefficients, and other information. The average signal-to-noise ratio of the valid data segment exceeds 15dB, the common-mode rejection ratio exceeds 100dB, and the electrode contact impedance remains stable within the range of 20-30kΩ. These raw data are stored continuously at a frequency of 1024 Hz, and each data point contains the voltage value of 16 channels and the corresponding status mark.

[0035] Based on the acquired high-quality raw data stream, the system begins adaptive data segmentation. First, the signal is initially segmented using a sliding window method, with a window length of 2 seconds and a 50% overlap ratio to ensure signal continuity. Furthermore, a multi-scale entropy analysis method is introduced to assess changes in signal complexity. Segmentation points are marked when the entropy change rate exceeds 20%. Wavelet packet decomposition is also used to calculate the energy distribution of each frequency band, focusing on energy changes in typical EEG rhythms such as δ (1-4 Hz), θ (4-8 Hz), α (8-13 Hz), and β (13-30 Hz). The validity of the segmentation position is further confirmed when significant energy changes occur simultaneously in multiple frequency bands (exceeding two standard deviations of the mean). During the segmentation process, the system incorporates a comprehensive anomaly detection mechanism that can identify various types of interference, including myoelectric artifacts, oculoscopic artifacts, and device vibration. When an anomaly is detected, the affected data segment is automatically marked and isolated. This segmentation process divides the raw data stream into a series of data segments of uniform length and relatively stable content. Each data segment carries detailed energy distribution characteristics, entropy characteristics and anomaly marking information. The average segment length is 2 seconds, the overlapping interval of adjacent segments is 1 second, and the signal stability index within a single segment remains above 0.85.

[0036] For these normalized data segments, the system further performs signal sequence generation. First, precise time alignment techniques are used to ensure that all sequences have a consistent time base and number of sampling points. Noise reduction is then performed using a five-layer wavelet decomposition based on the sym4 wavelet basis function. The wavelet coefficients are optimized using a soft thresholding method, effectively suppressing over 90% of high-frequency interference components. Furthermore, the system introduces the Hilbert transform to calculate the instantaneous phase and frequency characteristics of the signal, obtaining a complete description of its time-varying characteristics. To account for individual variability, the signal amplitudes of all channels are Z-score-normalized to a mean of 0 and a standard deviation of 1, thus establishing a unified amplitude metric. The generated standard signal sequences, in addition to the 16-channel normalized EEG data, also include a rich set of auxiliary features, such as phase synchronization, frequency band energy ratios, and inter-channel coupling strength. The system calculates a series of quality assessment metrics for each sequence, including signal-to-noise ratio (SNR), signal stationarity index, and artifact contamination rate. The SNR threshold is set at 10dB, and the artifact contamination rate is capped at 30%. Sequences exceeding these thresholds are marked as low-quality. All sequences are numbered and stored in chronological order, forming a structured signal sequence database. After complete processing, each second of EEG data generates a high-quality signal sequence of approximately 1024 × 16 × 4 bytes, containing three levels of information: raw data, feature data, and quality indicator data. The data has a temporal resolution of milliseconds and a spatial resolution covering the entire scalp area, effectively meeting the accuracy requirements of subsequent analysis.

[0037] In some embodiments, obtaining the feature sequence corresponding to the signal sequence includes: dividing the signal sequence into windows to extract time series features; marking behavioral indicators based on the time series features; and generating the feature sequence according to the marking results corresponding to the behavioral indicator marks.

[0038] The system receives the standard signal sequence database generated in the above steps. This database contains 16 channels of normalized EEG data, auxiliary features such as phase synchronization index, frequency band energy ratio, and inter-channel coupling strength, as well as signal quality assessment metrics such as SNR, signal stationarity index, and artifact contamination rate. An adaptive multi-scale windowing strategy is used to window these signal sequences. The system first selects high-quality data segments based on SNR and artifact contamination rate. Segments with an SNR below 8dB or an artifact contamination rate exceeding 30% are marked as low-confidence. A basic window length of 500ms and a sliding step of 100ms are set, using a Hanning window function to reduce edge effects. A dynamic window adjustment mechanism is also introduced to automatically adjust the window size based on the local stability of the signal. Within each window, the system calculates time series feature parameters, including frequency-domain features such as power spectral density, sub-band energy ratio, and spectral entropy, as well as statistical features such as mean, variance, kurtosis, and skewness. The system directly integrates the phase synchronization metrics provided in the above steps to assess information flow between channels, using existing frequency band energy ratios as the basis for cognitive state assessment and analyzing coordinated patterns of brain activity based on inter-channel coupling strength. Time-frequency maps are generated using short-time Fourier transforms, and combined with wavelet packet decomposition, energy trends in the four typical frequency bands of δ, θ, α, and β are extracted. For example, when a user is focused on reading an e-book, the system can accurately capture the continuous increase in β-band energy, automatically adjusting the window length to better preserve attention-related EEG features. The processing results for each window form a feature vector, which contains both frequency-domain and statistical features. After complete window partitioning and feature extraction, the system obtains a high-dimensional time series feature matrix with a temporal resolution of 100ms.

[0039] Based on the high-dimensional time series feature matrix obtained in the previous stage, the system begins the behavioral indicator labeling process. Principal Component Analysis (PCA) is first used to reduce the dimensionality of these high-dimensional time series features, retaining the principal components that primarily explain the variance. Then, a time series pattern mining algorithm is employed to identify key behavioral indicators within the feature sequences, including typical patterns such as attentional focus, cognitive load changes, and mood swings. The system employs fuzzy inference rules to map time series features, such as energy ratios across different frequency bands and spectral entropy changes, to specific behavioral indicators. For example, a sustained increase in the β / α energy ratio and a relatively stable θ band energy are labeled as a state of focused attention; a sharp decrease in α band energy accompanied by an increase in θ band energy are labeled as a state of increased cognitive load. In practical applications, such as when users are performing complex data analysis tasks, the system can monitor the dynamic changes in these EEG features to identify when the user is entering a state of deep thought (significant suppression of α waves) or experiencing cognitive fatigue (significant enhancement of θ waves), thereby providing user status feedback to the smart glasses. Through this labeling mechanism, the system adds corresponding behavioral indicator labels to each window feature vector, including basic behavioral states (such as focus, relaxation, and fatigue) and transition states (such as attention shifting and task switching). The accuracy of labeling is directly related to the EEG variation patterns captured in the high-dimensional time series feature matrix. The richer the feature expression, the higher the accuracy of state recognition. The final output is a feature matrix containing behavioral indicator labels, in which each time window is associated with the corresponding basic behavioral state and transition state labels.

[0040] Based on the feature matrix containing behavioral indicator labels, the system begins constructing a structured feature sequence. It first performs temporal correlation analysis on the feature vectors, calculating the feature change rate and behavioral indicator transition probability between adjacent windows. The system specifically focuses on the basic behavioral state and transition state labels in the matrix, analyzing the temporal distribution patterns and transition characteristics of these states. A dynamic time warping (DTW) algorithm is used to align feature patterns at different time scales while preserving the labels of the basic behavioral state and transition state. Sequence encoding methods are then used to convert the features and these state labels into a unified sequence representation. During the encoding process, a hierarchical feature organization structure is introduced, where the principal component features are layered according to their information contribution. Within each layer, behavioral indicator labels are embedded, including complete information about the basic and transition states. This hierarchical structure demonstrates good adaptability in practical applications. For example, in a video conferencing scenario, the system can flexibly shift the focus of feature hierarchies: when the user is listening (basic behavioral state), the system prioritizes features related to attention maintenance; when the user needs to transition from listening to speaking (transitional state), the system shifts to features related to cognitive processing, effectively capturing the user's behavioral intentions. A sequence indexing mechanism is also established to record the temporal dependencies between features and the evolution of behavioral indicators. For example, the system can track the gradual transition (transition state) of a user from focused work (baseline state) to fatigue (baseline state), capturing characteristics such as the changing trends of alpha wave energy and beta wave energy. The resulting feature sequence is stored in a sparse matrix format. Each sequence element contains three components: a feature vector, a probability distribution of behavioral indicators (reflecting the likelihood of base and transition states), and temporal correlation information, forming a structured feature representation.

[0041] In some embodiments, obtaining the behavioral features corresponding to the feature sequence includes: performing hierarchical processing on the feature sequence to obtain basic features; performing component extraction based on the basic features; constructing a feature hierarchy using the components obtained by the component extraction; and generating the behavioral features according to the feature hierarchy.

[0042] The system receives the feature sequence generated in the above steps, which includes feature vectors, behavioral indicator probability distributions, and temporal correlation information. Based on this sequence information, the system first performs hierarchical processing to obtain basic features. Specifically, the hierarchical division is performed using the temporal variation characteristics of the feature vectors, with the behavioral indicator probability distribution as the basis for stratification. An adaptive hierarchical strategy is used to divide the feature sequence into three time-scale layers: a rapid response layer processes transient feature changes within 100ms, such as EEG oscillation pattern transitions; a medium-term change layer focuses on feature evolution within 1-5 seconds, such as changes in attention level; and a long-term trend layer tracks sustained changes over 10 seconds, such as increased fatigue. In video learning scenarios, the system uses the rapid response layer to capture transient changes in user attention on key knowledge points, the medium-term change layer to assess knowledge comprehension progress, and the long-term trend layer to monitor overall learning status. The system also establishes an inter-layer information transfer mechanism to enable the coordinated updating of features at different time scales. Based on this hierarchical processing, the system obtains a set of hierarchical basic features that incorporate temporal correlations.

[0043] Based on the three basic features described above, the system independently extracts components for each layer. For the rapid response layer, independent component analysis (ICA) is used to isolate independent components related to instantaneous behavior, such as attention shift and decision control. For the medium-term change layer, components related to sustained behavior, such as cognitive load change and task engagement, are extracted. For the long-term trend layer, components related to state evolution, such as fatigue accumulation and emotional tone, are obtained. In office scenarios, the system can extract the attention component of text scanning (rapid layer), the memory load component of text comprehension (medium-term layer), and the sustained component of cognitive engagement (long-term layer) from the user's multi-document processing. The system also calculates the correlation coefficient matrix between components at different levels to quantify the strength of associations between components, providing a data foundation for subsequent feature hierarchy construction. Through this component extraction process, the system obtains a set of independent components and their associated relationships that describe behavioral characteristics at different time scales.

[0044] Based on the extracted independent component sets and their correlation strength matrices, the system begins constructing hierarchical feature relationships. Using a hierarchical clustering algorithm, components with correlation strengths above a threshold are organized into feature clusters. Each feature cluster contains a combination of components from different timescales but related behaviors. The system calculates the hierarchical distances between components based on the correlation coefficient matrix and stores the clustering results in a tree structure. For example, the system constructs an attention feature cluster based on the high correlation (correlation coefficient > 0.8) between the attention shift component at the rapid layer and the attention persistence component at the medium layer. Similarly, a task execution feature cluster is constructed based on the correlation (correlation coefficient > 0.75) between the decision control component and the task engagement component. For each feature cluster, the system calculates its internal average correlation strength and external discrimination for subsequent feature fusion. For example, in an online meeting scenario, the system uses the correlation relationships between components to construct a complete interactive behavior feature hierarchy: the voice attention component and the expression change component (rapid layer, correlation coefficient 0.82) are grouped together with the conversation engagement component (medium layer, correlation coefficient 0.78) and the meeting engagement component (long layer, correlation coefficient 0.71) to form a feature cluster reflecting meeting engagement. The system records the hierarchical position of each feature cluster, the list of components it contains, and the correlation strength matrix, providing structured input for the construction of the feature fusion network.

[0045] Using the constructed feature clusters and their complete hierarchical organization, the system begins generating behavioral features. First, a local fusion network is established based on the internal structure of each feature cluster. The system uses the component lists recorded in the feature cluster as network nodes and sets the connection weights between nodes using the coefficients in the correlation strength matrix. For example, for the attention feature cluster, the connection weight between the attention shift component in the rapid layer and the attention persistence component in the intermediate layer is set to 0.8 (corresponding to their correlation coefficient). Then, based on the hierarchical position of the feature clusters, cross-cluster connections are constructed. When two feature clusters are close in the hierarchical structure (hierarchical distance < 2) and share common components, a connection pathway is established between them, with the connection strength determined by the correlation coefficient of the common components. The system also incorporates an adaptive weighting mechanism that dynamically adjusts the connection strength within the network based on the current task scenario. For example, in a reading comprehension task, the system increases the connection weight between the attention feature cluster and the cognitive processing feature cluster (from a baseline value of 0.6 to 0.8); in a visual search task, the system emphasizes the connection between the attention feature cluster and the visual processing feature cluster (from 0.5 to 0.7). During a two-hour programming class, the system can capture in real time the student's changes in focus while writing code (attention feature cluster), their mental activity while understanding algorithmic difficulties (cognitive processing feature cluster), and their overall learning motivation (emotional state feature cluster). When students are debugging complex programs, the system increases the weight of the cognitive processing feature cluster; when students participate in class interaction, it emphasizes the role of the emotional state feature cluster, thereby accurately assessing students' learning status and engagement. Furthermore, the system incorporates a context-aware mechanism to analyze the dynamic changes of the feature fusion network within a continuous time window, capturing the evolutionary trends of behavioral features. Through this multi-level fusion and dynamic analysis based on feature clusters, the system ultimately generates a complete set of behavioral features that describe behavioral states and predict trends.

[0046] Step S102: perform sequence analysis based on behavioral features to obtain behavioral samples, and build a behavior library based on the behavioral samples; perform time series mapping on the behavior library to obtain behavioral paths; extract dependencies based on the behavioral paths; build sequence rules using the dependencies; and generate prediction tags based on the sequence rules.

[0047] Specifically, the system receives the behavioral feature set generated in the above steps, uses the behavioral state description to determine the initial behavior type, and provides basic labels for sequence analysis; at the same time, it integrates trend prediction information to guide the selection of window length, and uses shorter windows to capture subtle changes in areas where the predicted trend changes significantly. Based on these multi-level features, the system begins to perform sequence analysis to label behavioral samples. A multi-scale sliding window is used for feature segmentation, and the window length is adaptively adjusted between 2 seconds and 10 seconds according to the behavior type. For the feature sequence within each window, the system calculates the stability index and jump characteristics of the feature cluster separately, and identifies the key transition points of the behavioral pattern. At the same time, a multi-dimensional analysis method is introduced to combine the feature changes in the three dimensions of attention, cognition, and emotion to determine the boundaries of the behavioral sample. In real-world teaching scenarios, such as when students watch online video courses, the system accurately captures typical learning behavior segments. When students encounter difficulty and rewatch a video, it can identify characteristic combinations: a high degree of focus in the attention feature cluster, an increased cognitive processing load in the cognitive processing feature cluster, and increased learning engagement in the emotional state feature cluster. When students fully grasp a key point, it can detect characteristic changes: a moderate relaxation in the attention feature cluster, a decrease in the cognitive processing load in the cognitive processing feature cluster, and a stable emotional state feature cluster. The system also incorporates a behavioral continuity verification mechanism, ensuring the temporal integrity of labeled samples by analyzing the trend of feature changes between adjacent samples. Through this multi-dimensional sequence analysis, the system generates a series of behavioral sample units containing complete feature labels.

[0048] In some embodiments, constructing a behavior library based on the behavior samples includes: constructing a feature template based on the behavior samples; performing data matching using the feature template; and generating the behavior library based on matching results corresponding to the data matching.

[0049] Based on these fully labeled behavioral sample units, the system begins constructing feature templates. First, a hierarchical clustering method is used to analyze the behavioral samples, clustering them into different behavioral prototypes based on the similarity of their feature combinations. Each behavioral prototype contains the typical feature sequences and patterns of change for that type of behavior across the three dimensions of attention, cognition, and emotion. The system then analyzes the temporal dependencies between the feature clusters within the behavioral prototype, extracting key feature combinations that characterize the behavior and constructing template feature vectors. In remote work scenarios, the system extracts various feature templates from users' daily work behaviors: focused work templates (characterized by sustained focus in the attention feature cluster, stable load in the cognitive processing feature cluster, and positive engagement in the emotional state feature cluster); task switching templates (characterized by brief distractions in the attention feature cluster, fluctuating load in the cognitive processing feature cluster, and mild fluctuations in the emotional state feature cluster); and rest and recovery templates (characterized by relaxed attention feature clusters, reduced load in the cognitive processing feature cluster, and a more balanced emotional state feature cluster). The system also incorporates a template optimization mechanism that continuously adjusts and refines template features by analyzing the degree of match between different behavioral samples and the same template. Through this iterative optimization, a library of feature templates accurately describes different behavioral states has been established.

[0050] Using the constructed feature template library, the system performs data matching. A multidimensional feature matching algorithm based on dynamic time warping (DTW) is designed to compare behavioral feature sequences collected in real time with feature templates in the template library. The matching process not only considers the similarity of feature values ​​but also incorporates matching of temporal variation patterns. The optimal match is determined by calculating the degree of temporal alignment between the feature sequence and the template. A fuzzy matching mechanism is also established, employing a confidence-weighted approach to handle cases where features are partially missing or uncertain. For example, in a smart classroom environment, as students perform experiments, the system can match their current multidimensional feature sequence with existing experimental templates in real time. By comparing the attention allocation feature sequence (reflecting operational focus), cognitive processing feature sequence (reflecting understanding), and emotional state feature sequence (reflecting operational confidence), the system can determine whether the student has mastered the experimental techniques correctly and whether they are encountering operational difficulties. The system also establishes a behavioral pattern warning mechanism based on temporal variations in the matching results. When a feature sequence deviates from the standard template by exceeding a threshold, it promptly flags potential problem areas. Ultimately, the system generates a set of real-time matching results with confidence assessments.

[0051] Based on these real-time matching results, the system begins generating a comprehensive behavioral database. First, matching results are graded based on their confidence level. High-confidence matches (>0.8) are directly incorporated into the core behavioral database. Medium-confidence matches (0.6-0.8) are considered candidate patterns for further verification. Low-confidence matches (<0.6) are used for reference only and not directly incorporated into the database. Statistical analysis is then performed on the matched behavioral patterns, calculating the frequency, duration, and transition probability of each pattern to establish a multidimensional index structure. The temporal correlations and causal relationships between different behavioral patterns are also analyzed, and a behavioral state transition network is constructed to describe the evolution of behavioral patterns. In university classroom settings, the system has accumulated a wealth of behavioral pattern data through long-term observation of student learning behavior. These include active engagement patterns (characterized by high concentration, active cognitive processing, and positive emotional states), deep thinking patterns (characterized by sustained and stable attention, high cognitive processing load, and focused emotional states), and distracted attention patterns (characterized by frequent attention shifts, shallow cognitive processing, and fluctuating emotional states). The system systematically organizes these behavioral patterns into a multi-layered behavioral knowledge structure: the bottom layer stores original feature templates and matching rules, the middle layer records the transformation patterns and triggering conditions, and the top layer summarizes the long-term evolution trends and influencing factors of behavioral patterns. This hierarchical behavioral library not only supports real-time behavior recognition but also provides knowledge support for long-term behavior analysis and prediction.

[0052] In some embodiments, the time series mapping of the behavior library to obtain the behavior path includes: constructing a multidimensional time series association map based on the behavior library to perform micro-dependency analysis, meso-combination pattern recognition and macro-trend tracking of the behavior path; establishing a time series causal network based on the multidimensional time series association map to calculate the conditional transition probability and time series correlation coefficient of adjacent behavior patterns; identifying typical combination patterns and variation forms in the behavior path based on the conditional transition probability and time series correlation coefficient, and generating a behavior path containing direct dependency relationships and long-term evolution laws.

[0053] After receiving the multi-level behavior library generated in the above steps, the system begins performing temporal mapping to obtain behavioral paths. First, it uses low-level feature templates and matching rules to identify the current behavior pattern and mark its position in the behavioral sequence. It then analyzes the temporal transition characteristics between patterns by using the behavioral pattern transition patterns and triggering conditions recorded at the middle level. Finally, it combines high-level long-term evolution trends and influencing factors to predict the overall development direction of the behavioral sequence. Using time series analysis, the system calculates the transition probability matrix and time interval distribution of the behavioral pattern, establishing a dynamic behavior index based on time windows. In distance learning scenarios, the system performs a detailed temporal analysis of students' learning behavior throughout the day. For example, the morning study phase typically exhibits a behavioral path of "preparation (15-20 minutes) - deep learning (45-60 minutes) - a short break (5-10 minutes) - knowledge consolidation (30 minutes)." In the afternoon, the behavioral path often follows: "shallow learning (20-30 minutes) - distraction (5-10 minutes) - forced focus (15-20 minutes)." The system also integrates the knowledge structure in the behavior library to analyze behavior distribution across different timescales: from hourly changes in learning focus, to daily fluctuations in learning efficiency, to weekly learning habit formation. By building a multidimensional temporal correlation map of behavior, the system not only captures the immediate transitions in behavioral patterns but also discovers underlying cyclical patterns, such as windows of efficient learning occurring at specific times of the day or patterns of attention fluctuations occurring on specific days of the week.

[0054] Based on the acquired behavioral paths and their multidimensional temporal correlation maps, the system begins to deeply extract dependencies. A three-tiered dependency analysis framework is constructed, targeting behavioral correlations at the micro, meso, and macro levels. At the micro level, the system applies association rule mining algorithms to analyze the direct dependencies between adjacent patterns in the behavioral paths. For example, after a focused learning mode, the system calculates the conditional probabilities of different subsequent behaviors, such as a short break, knowledge consolidation, and distraction, taking into account the influence of time intervals. At the meso level, the system employs sequential pattern mining to identify typical combination patterns within the behavioral paths. Sliding time window analysis reveals recurring patterns and variations in behavioral sequences. For example, the standard "pre-study-review" pattern can evolve into a "pre-study-interruption-forced review" pattern due to fatigue. At the macro level, the system establishes a temporal causal network to track the evolution of long-term behavioral trends. In smart office environments, the system has discovered a series of key dependency chains through long-term observation. For example, high-quality morning meetings (8:30-9:00) improve focus during the morning work (9:00-11:30), which in turn boosts task completion efficiency during the afternoon (14:00-17:00). This chain reaction has been verified on 80% of workdays. For each dependency, the system calculates the temporal correlation coefficient and conditional transition probability, establishing a quantitative dependency strength assessment system.

[0055] Using the extracted multi-level dependencies, the system begins to construct sequence rules. First, direct dependencies at the micro level are converted into a basic rule set. Typical combination patterns at the meso level are used to construct composite rules. Finally, the macro-level temporal causal network forms long-term prediction rules. The system assigns an initial weight to each rule based on the temporal correlation coefficient and conditional transition probability. Strong dependencies with a correlation coefficient above 0.7 are preferentially converted into core rules. Each rule contains three core elements: triggering conditions (current behavioral state and environmental factors), time constraints (effective action time window), and expected results (probability distribution of subsequent behaviors). The rule set is then streamlined and enhanced through a rule optimization algorithm: rules with similar triggering conditions and expected results are merged, redundant rules with confidence levels below a threshold are deleted, and necessary environmental constraints are supplemented. In classroom teaching scenarios, the system summarizes a series of high-confidence behavioral sequence rules based on a large amount of teaching observation data. For example, "After students have been focused on studying for 45 ± 5 minutes without a recent break, the probability of their attention waning will increase to 80% within the next 10 minutes. It is recommended that they immediately arrange a 5-10 minute relaxation activity." "After 20 minutes of theoretical knowledge explanation, if relevant practical exercises are immediately arranged, the probability of students' understanding depth improving by 75% will be increased, and this improvement will last for the next 2 hours." The system maintains a complete evaluation profile for each rule, recording the number of times the rule was triggered, the success rate, applicable conditions, and failure cases. Through continuous data accumulation and feedback analysis, the accuracy and practicality of the rules are continuously optimized.

[0056] Based on the constructed sequence rule library, the system begins generating prediction tags. During the rule selection process, the system determines reliability weights based on the number of rule triggers, evaluates prediction accuracy based on historical success rates, filters relevant rules for the current scenario using applicable conditions, and analyzes failure cases to avoid potential misjudgments. The system has designed a multi-layered prediction and evaluation framework that combines rule matching with historical data analysis. At the bottom layer, the system monitors the current behavior state in real time and calculates the degree of match with each rule. At the middle layer, the system combines historical behavior data and uses a dynamic programming algorithm to search for the optimal prediction path among multiple possible behavior paths. At the top layer, the system considers environmental factors and individual differences to adaptively adjust the prediction results. In practical teaching applications, such as online programming courses, the system is capable of multi-dimensional behavior prediction. For example, if it detects that a student's attention is beginning to wander while solving a complex algorithmic problem (rule match >90%), the system predicts the likely behavior sequence over the next 30 minutes: first, a period of mental block (85% probability, lasting 10-15 minutes), followed by a period of code debugging difficulty (75% probability, lasting 15-20 minutes), and ultimately, possible learning frustration (65% probability). Each prediction tag contains four basic attributes: behavior type, probability of occurrence, expected duration, and prediction confidence. Through this rule-based prediction mechanism, the system can generate structured prediction tags for each time window, providing basic data support for subsequent environmental analysis and behavior prediction.

[0057] Step S103: Perform environmental analysis based on the prediction tags to obtain scene features; perform association matching on the scene features, construct a fusion sequence based on the matching results of the association matching, and generate a prediction result based on the fusion sequence; perform feature decomposition on the prediction result and mark the prediction interval; construct rules based on the prediction interval; determine the confidence level based on the rules, and generate a decision sequence based on the confidence level.

[0058] Specifically, the system receives the prediction tag information generated in the above steps, including the behavior type, probability of occurrence, expected duration, and prediction confidence. Based on these prediction tags, the system begins to analyze the environment to obtain scene features.

[0059] In some embodiments, the environmental analysis is performed based on the predictive markers to obtain scene features, including: constructing an environmental feature library to extract associated features between environmental conditions and task requirements; establishing a mapping relationship between environmental parameters and behavioral indicator labels to analyze the impact weights of scene elements on behavioral paths; matching real-time environmental data with historical scene patterns in a feature template library through an association matching algorithm to generate a scene feature vector containing environmental constraints and behavior triggering mechanisms.

[0060] A multidimensional environmental perception framework was established, encompassing physical environment monitoring (lighting, temperature, noise, etc.), social environment analysis (personnel distribution, interaction intensity, etc.), and task environment assessment (difficulty, urgency, importance, etc.). The system employed a stratified sampling strategy, dynamically adjusting the frequency of collecting environmental information from different dimensions based on the type and duration of the behavior in the prediction tag. An adaptive sampling algorithm was employed for physical environmental parameters, increasing the sampling frequency at critical moments (e.g., reducing the sampling interval to 100ms before the predicted behavior occurred) and appropriately reducing the frequency (increasing the sampling interval to 1s) during the stable phase. Event-triggered sampling was employed for social environmental characteristics, with data collected immediately upon detecting key events such as human interaction and location changes. Periodic sampling was employed for task environment characteristics, with task progress and difficulty assessed every five minutes. The system also incorporated a data quality control mechanism, screening reliable environmental features through temporal correlation analysis and spatial consistency checks of sensor data. The resulting constructed scene feature vector employed a multi-layered encoding structure, with the bottom layer recording raw sensor data, the middle layer storing feature statistics, and the top layer storing environmental change trends, forming a comprehensive description of the scene.

[0061] Based on the acquired scene feature vectors and predicted tag information, the system begins performing association matching. A hierarchical processing strategy is designed to address the multi-level encoding structure of scene features: noise filtering and anomaly detection algorithms are applied to the bottom-level raw sensor data to ensure data quality; normalization and dimensionality reduction are performed on the middle-level feature statistics to extract core features; and temporal pattern recognition is applied to the top-level environmental change trends to capture the dynamic characteristics of the environment. First, a scene-behavior mapping model is constructed. An improved Kalman filter algorithm is applied to the bottom-level sensor data to update the environmental state estimate in real time. A particle filter method is then used to predict the evolution of environmental parameters using the middle-level statistical features and top-level trend data. The system also designs an adaptive weighting function based on the confidence level of the predicted tag, assigning a higher weight (0.6-0.8) to predicted tags with a confidence level above 0.8 and a lower weight (0.2-0.4) to predicted tags with a confidence level below 0.5. During fuzzy reasoning, the system employs a multi-level fuzzy rule base, encompassing environmental threshold rules (such as suitable temperature ranges and noise tolerance thresholds), combined effect rules (interactions between multiple factors), and temporal evolution rules (the delayed effects of environmental changes on behavior). For each environmental factor, the system calculates its direct and indirect influence coefficients on the predicted behavior, constructing a complete influence propagation network. Scenario adaptability assessment utilizes a deep learning model based on an attention mechanism. By analyzing scenario-behavior correspondence patterns in historical data, it learns the mapping relationship between environmental features and behavior predictions, enabling rapid adaptation to new scenarios.

[0062] Leveraging the scenario-behavior mapping results from the second phase, the system begins constructing fused sequences. The system uses the environmental state estimation and evolution trend prediction results from the scenario-behavior mapping model as the foundational data for sequence construction. The system also determines the fusion weights for different features based on the output of an adaptive weighting function: features corresponding to high-confidence prediction markers (weights 0.6-0.8) receive higher fusion priority. Simultaneously, the system applies rule sets from a multi-level fuzzy rule library to the sequence construction process. It uses environmental threshold rules to screen valid features, handles feature interactions through combined effect rules, and controls the temporal evolution of the sequence based on temporal evolution rules. Furthermore, the system integrates direct and indirect influence coefficients within the influence propagation network, establishing a hierarchical fusion relationship between features, prioritizing the integration of features with high influence coefficients. The results of the scenario adaptability assessment guide the dynamic adjustment of the fusion strategy, selecting the most appropriate fusion parameters for different scenario characteristics. Based on this, the system designs a hierarchical feature fusion framework that integrates environmental features and predicted behaviors across different temporal and spatial dimensions. The analysis of the time dimension uses a multi-scale wavelet transform to decompose the environmental feature sequence into different frequency components, and extract fast-changing features (seconds), medium-term evolution features (minutes), and long-term trend features (hours). In the spatial dimension, a spatial interpolation algorithm is used to establish a continuous distribution model of environmental parameters and analyze the spatial propagation characteristics of environmental factors. During the feature fusion process, the system uses an improved LSTM network to process time series features and controls the fusion ratio of features at different time scales through a gating mechanism. At the same time, a graph neural network is introduced to process spatial correlation features and establish spatial dependencies between environmental factors. The system has also developed a feature importance evaluation module that uses the XGBoost algorithm to analyze the contribution of different features to behavior prediction and dynamically adjust the feature fusion weights.

[0063] Based on the constructed fusion sequence, the system generates the final prediction result. A multidimensional prediction evaluation matrix is ​​established, with matrix elements encompassing three dimensions: feature contribution, temporal correlation, and spatial association. A hierarchical and progressive prediction strategy is employed: short-term predictions (5-15 minutes) rely primarily on physical environment features and immediate behavioral data, using ensemble learning methods (a combination of random forests and gradient boosting trees) to generate high-precision predictions; medium-term predictions (15-60 minutes) comprehensively consider environmental trends and behavioral patterns, employing recurrent neural networks to predict the evolution of behavioral sequences; long-term predictions (1-4 hours) incorporate historical data patterns, using a Transformer model with an attention mechanism to capture long-term dependencies. Different evaluation metrics are used for different types of prediction results: a confusion matrix is ​​used to assess accuracy for categorical predictions; mean squared error and mean absolute error are used for precision for continuous predictions; and log-likelihood loss is used to assess reliability for probabilistic predictions. Through this multi-layered prediction mechanism, the system generates both accurate and practical behavioral predictions, providing strong support for decision-making in intelligent environments. In a smart classroom environment, the system can combine the physical parameter status of the classroom (temperature, humidity, light, noise), student group characteristics (location distribution, activity intensity) and learning task attributes (difficulty level, completion progress) to generate accurate predictions at multiple time scales and equip each prediction result with a reliability evaluation indicator.

[0064] The system receives the short-term, medium-term, and long-term forecast results generated in the above steps, which include behavioral prediction probabilities and environmental impact weights. Based on these hierarchical forecast results, the system begins feature decomposition to label the forecast intervals. Using a multi-scale analysis approach, the forecast results are decomposed temporally into a rapid response interval (5-15 minutes), a medium-term change interval (15-60 minutes), and a long-term trend interval (1-4 hours). For each time interval, the system extracts the main behavioral patterns and environmental impact characteristics. During the decomposition process, the system incorporates an adaptive threshold mechanism that dynamically adjusts the decomposition parameters based on the reliability of the forecast at different time scales. For example, for the rapid response interval, the system focuses on instantaneous changes in behavioral characteristics; for the medium-term change interval, it emphasizes the gradual evolution of behavioral patterns; and for the long-term trend interval, it primarily extracts stable features of behavioral evolution. Through this multi-level decomposition strategy, the system ultimately obtains a series of forecast intervals with time labels and reliability indicators.

[0065] Based on the decomposed prediction interval sequence and its variation patterns, the system begins constructing a rule system. The system uses time stamps within the prediction interval sequence to organize rules in a temporal order, ensuring accurate application of the rules to the corresponding time windows. Furthermore, high-quality features are selected based on reliability metrics. Highly reliable prediction intervals provide stronger supporting evidence for rule construction. First, interval mapping rules are established, mapping key characteristics of the prediction interval (such as duration, fluctuation range, and variation trend) to behavioral patterns. In rapid-response intervals, rules primarily describe the immediate characteristics of behavioral changes; in medium-term variation intervals, rules focus on the evolution of behavioral patterns; and in long-term trend intervals, rules emphasize analyzing the cumulative effects of behavior. The system employs a fuzzy inference mechanism to handle rule decisions, effectively addressing uncertainty in the forecasting process. A rule conflict resolution mechanism is also established. When conflicting rules arise between different timescales, the system dynamically adjusts rule priorities by analyzing the applicable conditions and confidence levels of the rules. Each rule in the rule base is equipped with a complete set of conditional attributes and a rule strength metric, forming an adaptive behavior decision rule system.

[0066] Using the established multi-level rule system, the system begins to determine the confidence level of its predictions. A multi-dimensional confidence assessment framework was designed, building an evaluation system based on three dimensions: temporal correlation, spatial consistency, and rule matching. For temporal correlation, prediction reliability is assessed by analyzing feature stability within the prediction interval; for spatial consistency, the distribution of environmental features is examined; and for rule matching, the confidence level is calculated based on the triggering strength of the rules. The system employs a comprehensive evaluation strategy, weighting and integrating the evaluation results from multiple dimensions to generate a final confidence metric. In a smart classroom scenario, when the system predicts that a student may be distracted, it comprehensively evaluates multiple factors: examining the stability of the prediction within the recent time window (temporal correlation), analyzing whether the distribution of classroom environmental features supports the prediction (spatial consistency), and the degree of match between this behavior pattern and the historical rule base (rule matching). For example, if a student is detected to be continuously looking down and the ambient noise increases, and historical rules show that this combination is highly correlated with distraction, the system will assign a higher confidence level. Furthermore, a dynamic learning mechanism is introduced to adjust the weighting coefficients of each dimension based on the accuracy of historical predictions. The timeliness factor is also taken into account in the confidence calculation process, and the weight of earlier judgment results is appropriately reduced.

[0067] Based on the determined multi-dimensional confidence indicators, the system begins generating a decision sequence. A time-series decision framework is constructed, organizing the behavioral states and confidence indicators within the prediction interval into a continuous decision sequence. This decision sequence is generated using a multi-scale sliding window strategy, with the window size adaptively adjusted based on the time span of the prediction interval. The system also incorporates a sequence smoothing algorithm to effectively eliminate sudden changes and noise in the decision results. For each time point, the decision not only provides a prediction of the primary behavior type but also includes the corresponding confidence interval. In a university programming lab course, the system can generate a complete decision sequence for learning behavior. When identifying students debugging complex programs, the decision sequence reflects the changes throughout the problem-solving process. For example, the initial state might display a "deep thinking state (confidence interval 0.85-0.95)," followed by a "problem-solving state (confidence interval 0.75-0.85)" in the middle. If the problem persists, it may transition to a "attention fatigue state (confidence interval 0.80-0.90)" in the later stages. By analyzing the consistency of decisions across adjacent time windows, the system establishes a time-series correction mechanism for decision results, enabling timely identification and correction of potential decision deviations. For example, when a sudden change in the judgment result is detected, the system will comprehensively analyze the characteristics of the preceding and following time windows to determine whether the change is reasonable. This structured judgment sequence maintains temporal continuity and demonstrates the reliability of the prediction.

[0068] Step S104: perform behavior analysis based on the decision sequence to extract behavior patterns; obtain prediction parameters based on the behavior patterns; construct update rules based on the prediction parameters, generate prediction strategies based on the update rules, and output behavior prediction results based on the prediction strategies.

[0069] Specifically, the system receives the decision sequence generated in the above steps, which includes the type of behavior prediction, time window, and confidence interval. Based on this decision information, the system begins to perform behavior analysis to extract behavior patterns.

[0070] In some embodiments, the behavioral analysis is performed based on the judgment sequence to extract the behavioral pattern, including: hierarchical decomposition of the judgment sequence to identify basic behavioral units and their combination rules; obtaining individual behavioral characteristics and time-varying pattern features through pattern matching methods; constructing a style rule library based on the individual behavioral characteristics and time-varying pattern features, extracting the behavioral pattern conversion rules and stability indicators corresponding to the prediction parameters, and generating the behavioral pattern that includes the individual difference adaptation mechanism.

[0071] First, the time granularity of behavioral analysis is determined by using time windows within the decision sequence. Then, based on the distribution of prediction types and confidence intervals, high-confidence behavioral features are identified. A multi-level analysis framework is employed to extract behavioral features at the micro, meso, and macro levels. At the micro level, the system focuses on behavioral characteristics within a single time window, such as immediate states like attention level and cognitive load. At the meso level, it analyzes behavioral transition patterns between adjacent time windows to identify typical behavioral evolution paths. At the macro level, it extracts behavioral periodicity and trend characteristics from long-term series. In online learning scenarios, the system can extract complete behavioral patterns from a student's learning process throughout a single lesson, such as identifying a cyclical pattern of "focus-distraction-adjustment" or a progressive pattern of "understanding-practice-improvement." Through this multi-level behavioral analysis, the system obtains a set of structured behavioral pattern descriptions.

[0072] Based on the extracted behavioral patterns, the system begins labeling predictive parameters. An adaptive parameter labeling framework has been designed to assign corresponding sets of predictive parameters to different types of behavioral patterns. This labeling process considers three key dimensions: time-based parameters (such as behavior duration, transition timing, and cycle length); state-based parameters (such as behavior intensity, stability, and rate of change); and environmental-based parameters (such as the influence weight of environmental factors and trigger thresholds). In remote programming teaching scenarios, the system labels students' programming behaviors with parameters: the time dimension records continuous coding time, debugging frequency, and the interval between code submissions; the state dimension labels code modification rate, compilation error rate, and problem-solving efficiency; and the environmental dimension labels development tool switching frequency and reference material search frequency. When a student's programming efficiency is observed to improve significantly during a specific period, the system adjusts the weights of these parameters accordingly, strengthening its ability to identify effective behavioral patterns. For each dimension, the system has established a dynamic update mechanism that adjusts parameter values ​​based on actual observations. Furthermore, the system configures a valid range and update step size for each parameter to ensure the rationality of parameter adjustments. Through this systematic parameter labeling, a complete predictive parameter system is established for each behavioral pattern.

[0073] Using the labeled prediction parameter system, the system begins constructing update rules. First, a mapping relationship is established between parameters and behavior prediction results, analyzing the impact of different parameter combinations on prediction accuracy. Based on the analysis results, a set of parameter update rules is designed, including trigger conditions (when to initiate updates), update directions (how to adjust parameters), and update amplitudes (the size of the adjustments). The system employs a hierarchical update strategy, giving more frequent updates to core parameters that significantly impact prediction results, while updating less critical parameters less frequently. For example, in an online learning scenario, if a student's attention prediction parameters perform poorly in the afternoon, the system will first analyze the deviation pattern between actual attention states and predicted results. When the deviation primarily occurs in attention duration, the system prioritizes updating parameters related to the time dimension; when the deviation manifests itself in attention transition frequency, it prioritizes adjusting state transition parameters. For environmental influence parameters, the system periodically updates them after accumulating sufficient observation samples. The system also establishes a parameter linkage mechanism. When a core parameter undergoes significant adjustment, related subordinate parameters are also updated synchronously according to pre-set rules, ensuring the overall consistency of the parameter system. This multi-layered update rule system enables the system to continuously optimize the accuracy of prediction parameters.

[0074] Based on the constructed update rules, the system begins generating prediction strategies. A hierarchical strategy generation framework integrates prediction requirements across different timescales and behavior types into a unified strategy system. During strategy generation, the system comprehensively considers multiple factors, including the historical effects of parameter updates, the characteristics of current behavior patterns, and changes in environmental conditions. A strategy evaluation mechanism is also established to dynamically adjust strategy priorities and execution methods by tracking policy execution performance. In a smart classroom environment, when the system is applied to a 180-minute programming course, it can generate refined prediction strategies for different stages. During the knowledge explanation phase (0-45 minutes), the strategy focuses on predicting comprehension level by analyzing students' code following speed and their behavior of marking key knowledge points. During the example analysis phase (45-90 minutes), the strategy shifts to problem analysis skills by monitoring students' code debugging process and problem-solving paths. During the practical programming phase (90-180 minutes), the strategy focuses on predicting programming proficiency improvement by analyzing metrics such as code quality, completion speed, and error handling to predict learning outcomes. The system also dynamically adjusts the parameter thresholds and update frequency of the prediction strategy based on individual students' programming proficiency and learning characteristics. For example, for students with weaker foundations, the system reduces the weight of programming speed and increases the weight of code correctness. For advanced students, the system places greater emphasis on predicting code optimization and problem-solving efficiency. This adaptive prediction strategy system accurately captures students' behavioral characteristics at different learning stages, providing timely and effective decision support for teaching interventions.

[0075] In some embodiments, outputting the behavior prediction result according to the prediction strategy includes: performing multi-dimensional verification on the prediction strategy to obtain verification features; marking behavior attributes based on the verification features, and constructing a prediction sequence using the behavior attributes; and outputting the behavior prediction result according to the prediction sequence.

[0076] The system receives the forecast strategy and its hierarchical strategy system generated in the above steps. Based on this strategy information, the system begins multi-dimensional verification to obtain verification features. A verification feature framework is constructed, encompassing three dimensions: temporal consistency, spatial correlation, and logical integrity. In temporal verification, the system constructs a time series similarity matrix to compare historical data with forecast data, calculates short-term volatility indices and long-term trend consistency, and uses time series decomposition techniques to decompose the forecast sequence into cyclical, trend, and random components to assess reliability. Multi-window analysis techniques are introduced during the verification process to assess forecast consistency at different time scales, and fluctuation threshold detection is used to identify anomalous forecast points. In spatial verification, an environmental feature map is created, and forecast results are projected into a multidimensional environmental space to calculate vector similarity. A spatial correlation heat map is constructed to identify significantly correlated areas and outliers. The system implements a feature space partitioning algorithm to divide the environmental feature space into multiple functional regions, calculates the distribution probability of forecast results in each region, and generates a spatial distribution feature vector. In logical verification, a behavioral rule inference engine is implemented to compare predefined rules with the current forecast, construct a state transition diagram to verify the rationality of the behavioral path, and calculate the consistency index of the transition probability matrix. At the same time, a rule conflict detection module was developed to identify and quantify the degree of logical violations in the prediction results and to construct a violation index scoring system. In distance education scenarios, the system verifies the duration distribution of attention indicators, the support of feature combinations in the learning environment, the completeness of the preceding behavior sequence, and the distribution pattern of attention resources, thereby obtaining multi-dimensional verification features.

[0077] Based on the acquired verification features, the system begins labeling behavioral attributes. An adaptive attribute labeling framework is designed to extract three key attributes: basic attributes, associated attributes, and evolving attributes. For basic attribute labeling, a feature-to-behavior mapping map is constructed. A duration prediction module is developed to analyze historical distribution data, and a multi-factor probability calculator is implemented to integrate environmental conditions and individual differences. The system introduces a differentiated attribute processing mechanism, assigning differentiated attribute weights to different types of behavior and automatically adjusting the accuracy and granularity of attribute extraction based on the quality of the verification features. For associated attribute labeling, a behavioral pre-sequence map is constructed to record trigger chains and intensities. A trigger condition extractor is developed to analyze environmental variable thresholds and generate an influencing factor weight matrix to quantify the influence of each factor. A factor linkage effect analysis module is implemented to assess the combined impact of multiple factors on behavior and identify key factor combination patterns and their triggering thresholds. For evolving attribute labeling, a trend fitting tool is developed to perform polynomial fitting on behavioral indicators. A state transition network is constructed to calculate transition probabilities and times. A stability assessment module is implemented to quantify the range and frequency of fluctuations. A property evolution trajectory tracking system is designed to record the temporal changes in attribute values, identify characteristic patterns during stable, transitional, and sudden phases, and construct a comprehensive profile of property evolution. The system defines priorities and relationships through attribute dependency graphs and calculates historical data support to generate precise credibility scores. In programming learning scenarios, the system labels the precise numerical range and fluctuation curve of focus, records the operation sequence and error handling strategies during problem solving, and analyzes the rate of skill improvement and the evolution of programming style.

[0078] Using labeled behavioral attributes, the system constructs a prediction chain. It analyzes temporal correlations and conditional dependencies between attributes, constructs directed influence diagrams to represent causal relationships, quantifies the impact strength between different attributes, and implements attribute conflict detection to resolve conflicting dependencies. The system also develops an attribute propagation model to simulate how attribute value changes propagate and evolve within the prediction network, assess the sensitivity of individual attribute changes to the overall prediction results, and identify key decision nodes and stability bottlenecks. The prediction chain employs a three-tier architecture: the bottom tier integrates behavioral attributes to generate instantaneous feature vectors, constructs feature mapping relationships to predict short-term behavioral probability distributions; the middle tier develops a state sequence generator to predict state sequences based on current states and historical patterns, constructs a trigger condition monitor to assess satisfaction in real time, and calculates the precise timing of behavioral state transitions; the top tier designs a trend simulation engine to separate short-term fluctuations from long-term trends, identify cyclical patterns in behavioral changes, and assess trend stability and potential change points. The system implements a multi-path prediction generation mechanism, simultaneously maintaining multiple possible prediction paths and adjusting path probability weights based on real-time feedback, ensuring the prediction system is sufficiently adaptable and robust to environmental changes. The system configures a dynamic weighting mechanism for each node, adjusting the weight of the prediction path based on historical accuracy, environmental compatibility, and time span. A temporal correlation mechanism is established to construct a time-dependency graph, implementing a time decay function to quantify changes in prediction accuracy over time. In a classroom setting, the prediction chain can predict recent changes and peak concentration based on current attention states, calculate possible turning points and triggering conditions for cognitive fatigue, and analyze learning improvement trends and comprehension barriers throughout the course.

[0079] Based on the constructed prediction chain, the system outputs behavioral predictions. Cross-validation and consistency comparisons are implemented across multiple layers of prediction results, automatically adjusting any logical inconsistencies to ensure coherence. The system also features a prediction result fusion engine, employing a hierarchical integration strategy to integrate prediction outputs from different layers. This engine generates a consistent and robust set of predictions through techniques such as weighted voting, Bayesian averaging, and sequence alignment. The bottom layer outputs immediate behavior predictions, including behavior type, probability of occurrence, and duration. The middle layer outputs state transition predictions, describing the evolution path, triggering conditions, and key time points. The top layer outputs trend predictions, indicating development direction, rate of change, and stability. The bottom layer's behavior classification engine maps complex patterns to standard types, generates confidence intervals for probabilities, and predicts behavior duration. The system also develops multi-granularity time window analysis tools to evaluate behavioral characteristics at different time scales, capturing the full spectrum of behaviors, from micro-changes at the second level to macro-trends at the hour level. The middle layer generates a time-dependent sequence of behavioral state evolution, defines the triggering thresholds for each state transition, and precisely locates the occurrence of key transitions. A state transition warning system is implemented, issuing warning signals before impending state transitions and providing quantitative assessments of transition probabilities and expected impacts. Top-level predictions build trend direction indicators to predict long-term development directions and inflection points, quantify the speed and acceleration of behavioral changes, and assess the stability and fluctuation range of trends. In intelligent teaching applications, the system outputs: "Current attention focus > 0.8, with a probability of 85% ± 5% of duration, expected to last 15 minutes, and low frequency of gaze switching," "Expected state transition triggered by accumulated cognitive load in 10 minutes, with a probability of 75% ± 8% and a completion time of 5 minutes, manifested by changes in eye movement patterns," and "Learning efficiency improvement rate of 0.05 / hour, with a fluctuation of ±0.02, an 80% probability of trend retention, and expected improvement in programming speed." The system integrates the results of each layer through a multi-level prediction fusion matrix, dynamically adjusts the weight coefficients, and converts the prediction results into a standard data structure output.

[0080] The provided method has the following beneficial effects:

[0081] This invention achieves accurate extraction of behavioral intention information from EEG signals through multi-scale feature analysis and adaptive hierarchical processing. Specifically, it employs multi-channel signal segmentation and temporal feature extraction to improve the temporal resolution of feature sequences; utilizes hierarchical processing and component extraction techniques to enhance the expressive power of features; and, through the construction of a feature hierarchy, achieves a systematic expression of behavioral features, effectively overcoming the problem of unstable signal quality collected by smart glasses.

[0082] This paper proposes a method for constructing a behavior library based on feature templates, combining multi-dimensional predictive tagging with environmental feature analysis to establish a complete behavior prediction framework. By tagging behavioral samples and constructing feature templates, a rich behavioral knowledge base is formed. The temporal mapping of behavioral paths and dependency analysis enhance the temporal nature of predictions. Combined with the correlation matching of environmental features, the reliability and real-time performance of predictions are improved.

[0083] This invention designs a systematic prediction verification and strategy generation mechanism. By building multidimensional verification features and prediction chains, it achieves accurate prediction of user behavior. It employs feature decomposition and rule construction within the prediction interval to enhance prediction adaptability. It also utilizes behavioral pattern analysis and prediction parameter tagging to improve prediction accuracy. By extracting verification features and building prediction chains, it addresses the prediction challenges presented by individual differences and time-varying characteristics.

[0084] In order to implement the user behavior prediction method based on EEG in smart glasses corresponding to the above method embodiment, to achieve the corresponding functions and technical effects. Figure 2 , Figure 2 The following is a block diagram of a user behavior prediction device 200 provided in an embodiment of the present application. For ease of explanation, only the parts related to this embodiment are shown. The user behavior prediction device 200 provided in an embodiment of the present application includes:

[0085] The data acquisition unit 201 is used to collect multi-channel EEG data of the smart glasses and obtain original signals; segment the data based on the original signals, generate signal sequences according to the segmentation results corresponding to the data segments, obtain feature sequences corresponding to the signal sequences, and obtain behavioral features corresponding to the feature sequences;

[0086] The sequence analysis unit 202 is configured to perform sequence analysis based on the behavior characteristics, obtain behavior samples, and construct a behavior library based on the behavior samples; perform time sequence mapping on the behavior library to obtain behavior paths; extract dependency relationships based on the behavior paths; construct sequence rules based on the dependency relationships; and generate prediction tags based on the sequence rules.

[0087] The environment analysis unit 203 is configured to perform environment analysis based on the prediction tags to obtain scene features; perform association matching on the scene features, construct a fusion sequence based on the matching results of the association matching, and generate a prediction result based on the fusion sequence; perform feature decomposition on the prediction result to mark a prediction interval; construct rules based on the prediction interval; determine a confidence level based on the rule, and generate a decision sequence based on the confidence level;

[0088] The behavior analysis unit 204 is used to perform behavior analysis according to the judgment sequence and extract behavior patterns; obtain prediction parameters based on the behavior patterns; construct update rules according to the prediction parameters, generate prediction strategies according to the update rules, and output behavior prediction results according to the prediction strategies.

[0089] The user behavior prediction device 200 can implement the EEG-based user behavior prediction method in smart glasses of the above-mentioned method embodiment. The optional options in the above-mentioned method embodiment also apply to this embodiment and will not be described in detail here. The remaining contents of the embodiment of this application can refer to the contents of the above-mentioned method embodiment and will not be repeated in this embodiment.

[0090] 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.

[0091] The computer device 3 may be a computing device such as a smart phone, a tablet computer, a desktop computer, or a cloud server. The computer device may include but is not limited to a processor 30 and a memory 31. It will be understood by those skilled in the art that Figure 3 This is merely an example of the computer device 3 and does not constitute a limitation on the computer device 3 . The computer device 3 may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the computer device 3 may also include input and output devices, network access devices, etc.

[0092] 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, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor.

[0093] In some embodiments, the memory 31 may be an internal storage unit of the computer device 3, such as a hard drive 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 drive, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. equipped on the computer device 3. Furthermore, the memory 31 may include both an internal storage unit of the computer device 3 and an external storage device. The memory 31 is used to store an operating system, application programs, a boot loader, data, and other programs, such as the program code of the computer program. The memory 31 may also be used to temporarily store data that has been output or is about to be output.

[0094] 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.

[0095] 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 various method embodiments when executing the computer program product.

[0096] In several embodiments provided in the present application, it is understood that each box in the flow chart or block diagram can represent a part of a module, program segment or code, and the part of the module, program segment or 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 an order different 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, which depends on the functions involved.

[0097] 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, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: 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.

[0098] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit the scope of protection of this application. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of this application by those skilled in the art should be included within the scope of protection of this application.

Claims

1. A method for predicting user behavior based on EEG in smart glasses, characterized in that: include: Collect multi-channel EEG data from smart glasses and obtain original signals; Performing data segmentation based on the original signal, generating a signal sequence according to segmentation results corresponding to the data segmentation, obtaining a feature sequence corresponding to the signal sequence, and obtaining a behavioral feature corresponding to the feature sequence; Perform sequence analysis based on the behavior characteristics to obtain behavior samples, and build a behavior library based on the behavior samples; perform time sequence mapping on the behavior library to obtain behavior paths; and extract dependency relationships based on the behavior paths; constructing a sequence rule using the dependency relationship; generating a prediction tag according to the sequence rule; Performing environmental analysis based on the prediction markers to obtain scene features; Performing correlation matching on the scene features, constructing a fusion sequence based on the matching results of the correlation matching, and generating a prediction result according to the fusion sequence; performing feature decomposition on the prediction result, and marking a prediction interval; constructing rules based on the prediction interval; determining a confidence level according to the rule, and generating a decision sequence according to the confidence level; Behavior analysis is performed according to the judgment sequence to extract behavior patterns; prediction parameters are obtained based on the behavior patterns; update rules are constructed according to the prediction parameters, a prediction strategy is generated according to the update rules, and behavior prediction results are output according to the prediction strategy.

2. The method according to claim 1, characterized in that The acquiring of the characteristic sequence corresponding to the signal sequence includes: Performing window division on the signal sequence to extract time series features; Marking behavioral indicators based on the temporal features; The feature sequence is generated according to the marking result corresponding to the behavior indication mark.

3. The method according to claim 1, characterized in that The obtaining of the behavioral features corresponding to the feature sequence includes: Performing layered processing on the feature sequence to obtain basic features; performing component extraction based on the basic features; constructing a feature hierarchy using components obtained by the component extraction; The behavioral features are generated according to the feature hierarchy.

4. The method according to claim 1, wherein The step of constructing a behavior library according to the behavior samples includes: Building a feature template based on the behavior sample; Performing data matching using the feature template; The behavior library is generated according to the matching results corresponding to the data matching.

5. The method according to claim 1, characterized in that The performing time sequence mapping on the behavior library to obtain the behavior path includes: Constructing a multi-dimensional temporal correlation map based on the behavior library to perform micro-dependency analysis of behavior paths, meso-combination pattern recognition, and macro-trend tracking; Establishing a temporal causal network based on the multi-dimensional temporal association graph to calculate the conditional transition probability and temporal correlation coefficient of adjacent behavior patterns; Typical combination patterns and variation forms in the behavior path are identified according to the conditional transition probability and the time series correlation coefficient, and a behavior path containing direct dependency and long-term evolution rules is generated.

6. The method according to claim 1, characterized in that The performing environmental analysis based on the prediction mark to obtain scene features includes: Build an environmental feature library to extract the correlation features between environmental conditions and task requirements; Establish a mapping relationship between environmental parameters and behavioral indicator labels, and analyze the impact weight of scene elements on behavioral paths; The real-time environmental data is matched with the historical scene patterns in the feature template library through the association matching algorithm to generate a scene feature vector containing environmental constraints and behavior triggering mechanisms.

7. The method according to claim 1, characterized in that The performing behavior analysis based on the determination sequence and extracting the behavior pattern includes: The time granularity of behavioral analysis is determined by dividing the time windows in the decision sequence; high-confidence behavioral features are identified based on the distribution of prediction types and confidence intervals; a multi-level analysis framework is used to extract behavioral features from the micro, meso, and macro levels; in online learning scenarios, complete behavioral patterns are extracted from the learning process; and structured behavioral patterns are obtained through multi-level behavioral analysis.

8. The method according to claim 1, characterized in that Outputting the behavior prediction result according to the prediction strategy includes: Performing multi-dimensional verification on the prediction strategy to obtain verification features; Marking behavior attributes based on the verification features, and constructing a prediction sequence using the behavior attributes; The behavior prediction result is output according to the prediction sequence.

9. A user behavior prediction device, characterized in that: include: A data acquisition unit is used to collect multi-channel EEG data from smart glasses and obtain original signals; Performing data segmentation based on the original signal, generating a signal sequence according to segmentation results corresponding to the data segmentation, obtaining a feature sequence corresponding to the signal sequence, and obtaining a behavioral feature corresponding to the feature sequence; A sequence analysis unit is configured to perform sequence analysis based on the behavior characteristics, obtain behavior samples, and construct a behavior library based on the behavior samples; perform time sequence mapping on the behavior library to obtain behavior paths; and extract dependency relationships based on the behavior paths; constructing a sequence rule using the dependency relationship; generating a prediction tag according to the sequence rule; An environment analysis unit, configured to perform environment analysis based on the prediction markers to obtain scene features; Performing correlation matching on the scene features, constructing a fusion sequence based on the matching results of the correlation matching, and generating a prediction result based on the fusion sequence; performing feature decomposition on the prediction result, marking a prediction interval; and constructing rules based on the prediction interval; determining a confidence level according to the rule, and generating a decision sequence according to the confidence level; A behavior analysis unit is used to perform behavior analysis based on the judgment sequence and extract behavior patterns; obtain prediction parameters based on the behavior patterns; construct update rules based on the prediction parameters, generate prediction strategies based on the update rules, and output behavior prediction results based on the prediction strategies.

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