Brain-computer interface interaction method and system based on neural oscillation mode

Abstract

The invention provides a brain-computer interface interaction method and system based on a neural oscillation mode, and relates to the technical field of brain-computer interface interaction. Clean data streams are obtained through multi-lead electroencephalogram collection and filtering artifact suppression, multi-frequency oscillation features are extracted and fused, an end-to-end time delay model is established to achieve phase compensation in combination with individualized dominant frequency and incubation period dynamic calibration, candidate actions are judged and generated according to signal quality, and the accuracy of phase compensation is improved. The external equipment is controlled in combination with state mapping and trigger execution, the problems of single feature dimension, uncompensated individual difference, uncorrected time delay error, insufficient signal quality control and the like in the existing method are solved, and the stability, accuracy and adaptability of interaction are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of brain-computer interface interaction, in particular to a brain-computer interface interaction method and system based on neural oscillation patterns. BACKGROUND

[0002] Recent studies have shown that, based on individual alpha frequency (IAF) and individual event-related potential (ERP) latency, triggering sound stimuli at specific phase points can significantly observe phase-dependent ERP amplitude changes and induced alpha rhythm modulation, thereby confirming the stable physiological link between phase and sensory processing. Such studies usually use frontal lead recording, short pulse sound stimulation and specific ISI parameters, and use improved algorithm ecHT to improve the accuracy of phase estimation. At the same time, in the prior art, adaptive two-way brain-computer interface (BCI) based on electroencephalogram (EEG) has been widely explored in the field of neural rehabilitation, and the research focus has gradually focused on real-time parameter adjustment, individualized modeling and the introduction of machine learning methods.

[0003] Current research is gradually shifting from single frequency band, single channel exploration to comprehensive utilization of multi-frequency band, multi-channel signals, and emphasizes adaptive modeling combined with individual characteristics. In addition, more and more work attempts to improve the interaction experience and rehabilitation effectiveness through multi-modal feedback, including vision, hearing and touch. At the same time, the academic and industrial communities also generally call for solutions to key landing problems such as insufficient online classification accuracy, lack of mobile devices and lack of standardized system.

[0004] However, the existing methods still mainly stay at the level of sensory processing feasibility verification or macroscopic framework review, and there are many deficiencies: first, the oscillation feature dimension is narrow, and there is a lack of in-depth modeling of multi-frequency band coupling and spatial fusion; second, a phase hit mechanism capable of online estimation and compensation of end-to-end latency and jitter has not been established; third, there is a lack of dynamic linkage between signal-to-noise ratio (SNR) gating and interaction mapping (in brain-computer interface experiments, there are usually multiple consecutive trials (trials), each trial usually corresponds to a stimulus presentation, an action intention recognition or a trigger of an external device); finally, the feedback channel is single, the performance evaluation index system is not perfect, and it is difficult to guarantee the stability and portability of brain-computer interaction in complex scenarios. SUMMARY

[0005] In order to overcome the shortcomings of the prior art, the purpose of the present application is to provide a brain-computer interface interaction method and system based on neural oscillation mode, which systematically solves the defects of the existing brain-computer interface method in single feature dimension, individual difference compensation, time delay error correction and signal quality control, etc., through the overall design of multi-lead multi-band oscillation mode construction, individual parameter dynamic calibration, end-to-end time delay compensation and quality judgment-action generation linkage, thereby significantly improving the stability, accuracy and adaptability of brain-computer interaction.

[0006] To achieve the above purpose, the present application provides the following scheme: A brain-computer interface interaction method based on neural oscillation mode, comprising: Synchronously collecting electroencephalogram data at multiple lead positions on the scalp, establishing a unified time base, and obtaining clean data streams covering theta, alpha, mu and beta bands through filtering and artifact suppression; Extracting the instantaneous phase, amplitude and phase consistency index of each frequency band from the clean data stream, combining spatial weighting and coherence measurement for fusion, and forming a neural oscillation mode vector containing multi-frequency information; Estimating individual dominant frequency and task-related latency in the baseline stage, and dynamically correcting through recursive algorithm during operation to output individual parameters changing over time; Establishing a time delay model of an end-to-end processing link, deriving a phase prediction error based on the time delay model, and calculating a target phase after compensation according to the phase prediction error; According to the three quality indicators of signal-to-noise ratio, phase consistency and phase estimation confidence, when the result of the judgment is lower than the threshold, adjust the preset oscillation frequency band or replace the feature parameter used; when the judgment result reaches the threshold, generate the corresponding candidate interaction action or parameter according to the current neural oscillation mode; According to the individual parameters, neural oscillation mode and the results obtained based on signal-to-noise ratio, phase consistency and phase estimation confidence, select the corresponding state-action pair from the preset mapping library, and determine the triggering time by combining the phase prediction error, so that the external actuator executes the action within the interval of the target phase.

[0007] Preferably, it further comprises: Record three level indicators of phase hit rate, command hit rate and task success rate respectively, and dynamically update the weight coefficients of each electroencephalogram data in spatial fusion, the priority of oscillation frequency band and state mapping parameters according to the statistical results of the level indicators, so as to realize the continuous optimization of interaction performance.

[0008] Preferably, the method comprises the following steps of: synchronously collecting electroencephalogram data at multiple lead positions on the scalp, establishing a unified time base, and obtaining clean data streams covering theta, alpha, mu and beta bands through filtering and artifact suppression. electrodes are arranged at a plurality of electrode lead positions set on the scalp, and raw electroencephalogram signals are simultaneously collected by a data acquisition module at a uniform sampling rate; The electroencephalogram signals of each channel are time-stamped during the collection process, and a uniform time base is established to achieve multi-channel synchronization; The collected raw electroencephalogram signals are sequentially subjected to band-pass filtering and notch filtering to retain theta, alpha, mu, and beta bands and suppress power frequency interference and high-frequency noise, thereby obtaining filtered signals; The filtered signals are subjected to artifact detection and rejection, including eye movement artifacts, electromyographic artifacts, and electrode falling artifacts, and artifact correction is performed using independent component analysis, thereby obtaining processed multi-channel clean data streams.

[0009] Preferably, the instantaneous phase, amplitude, and phase consistency index of each frequency band are extracted from the clean data stream, combined with spatial weighting and coherence measurement for fusion, to form a neural oscillation pattern vector containing multi-frequency information, including: The clean data stream is input into a narrowband filter group corresponding to each frequency band to obtain band-limited signals of theta, alpha, mu, and beta bands; The band-limited signals are subjected to Hilbert transform to extract instantaneous phase and instantaneous amplitude, and the phase consistency index is calculated based on the phase difference between adjacent leads; Based on the signal-to-noise ratio or coherence measurement of each lead signal, the weighting coefficient of the lead in spatial fusion is determined, and based on the weighting coefficient, the instantaneous phase of each lead is spatially integrated using a weighted average fusion method to obtain a spatially robust phase of each frequency band; The spatially robust phase, corresponding instantaneous amplitude, and phase consistency index are combined by frequency band to form the neural oscillation pattern vector containing multi-frequency information.

[0010] Preferably, based on the signal-to-noise ratio or coherence measurement of each lead signal, the weighting coefficient of the lead in spatial fusion is determined, and based on the weighting coefficient, the instantaneous phase of each lead is spatially integrated using a weighted average fusion method to obtain a spatially robust phase of each frequency band, including: The instantaneous phase and instantaneous amplitude of each lead are extracted from the clean data stream in each pre-set oscillation frequency band to obtain and where is the lead number, is the frequency band, is the time sampling point; In each frequency band , the phase consistency of the lead is calculated in a sliding time window , which is used to characterize the stability of lead phase relative to population phase, and is formulated as: ; wherein, is the phase consistency of lead in frequency band ; is the number of sampling points in the time window centered at ; is the imaginary unit; is the instantaneous phase of lead in frequency band at time ; is the population average phase at the time; is the total number of leads; On each frequency band , the normalized signal-to-noise ratio mapping quantity of lead is constructed from the instantaneous amplitude in the sliding time window , which is used to measure signal reliability, and is formulated as: ; wherein, is the signal-to-noise ratio mapping quantity of lead in frequency band ; is the instantaneous amplitude; is the noise variance estimated by the median absolute deviation of the instantaneous amplitude squared in the time window ; denotes the median absolute deviation operator; On each frequency band , the unnormalized weight of lead is obtained by combining the phase consistency and the signal-to-noise ratio mapping quantity, and the spatial fusion weighting coefficient is obtained by probabilistic normalization, and is formulated as: , ; wherein, is the unnormalized weight of lead ; is the spatial fusion weighting coefficient of lead and satisfies ; The weighted circular mean is adopted to spatially integrate the instantaneous phases of each lead, and the spatial robust phase of frequency band is obtained, and is formulated as: ; wherein, is the spatial robust phase of frequency band ; is the phase value operator.

[0011] Preferably, individualized dominant frequency and task-related latency are estimated at baseline stage and dynamically corrected by recursive algorithm during running, outputting time-varying individual parameters, including: At baseline stage, spatially robust phase is calculated for each frequency band. Frequency matching optimization is performed to estimate individualized dominant angular frequency, which is converted to dominant frequency, formula is: ; wherein, is baseline dominant angular frequency estimation of frequency band ; is candidate angular frequency set of frequency band ; is baseline duration; is total number of leads; is instantaneous phase of lead in frequency band ; is imaginary unit; is dominant frequency estimation; Baseline task-related latency estimation is performed on task baseline trial to estimate individualized task-related latency in template matching manner, formula is: ; wherein, is baseline latency estimation; is latency search set; is number of trials used for baseline estimation; is time domain segment of lead in frequency band of the th trial; is template function constructed by average signal of baseline trials; is inner product; is two-norm; Enter running stage, angular frequency measurement is obtained according to discrete time derivative of spatially robust phase , and single-trial latency measurement is obtained in template matching manner, formula is: ; wherein, is angular frequency measurement; is phase unwrapping operator; is sampling interval; is single-trial latency measurement; is time domain segment of lead in frequency band of current trial; Individual parameters are expressed as , and measurement vector drives point-by-point update, update formula is: ; wherein, is individual parameter vector at time ; is the diagonal gain matrix; are the are the adaptive gains of the angular frequency and latency channels, respectively; is the dimensional running variance estimate of the state; is the dimensional measurement variance, derived from the measurement residual variance within a time window is the number of samples within the window; are the are the dimensional measurement and the previous time step state estimate, respectively; is converted to the dominant frequency together as the individualized parameter that varies over time.

[0012] Preferably, a delay model of the end-to-end processing link is established, on the basis of which a phase prediction error is derived, and a compensated target phase is calculated according to the phase prediction error, including: Each processing node of acquisition, filtering, feature extraction, communication and external execution is calibrated, time stamps are written for the input and output events of each node under a unified time base, and the reference sampling time of phase evaluation and the actual effective time of the external actuator are recorded; Taking the reference sampling time of phase evaluation as the starting point and the actual effective time of the external actuator as the ending point, the end-to-end delay value corresponding to a single interaction is calculated, and an end-to-end delay sequence is formed in time sequence; The end-to-end delay sequence is statistically processed within a sliding time window to obtain a delay model for describing the delay level and jitter range of the processing link; the delay model at least includes a delay estimation value within the current time window and a corresponding fluctuation metric; Based on the individualized dominant angular frequency, the time error is converted to a phase prediction error using the delay estimation value of the delay model, and a confidence range of the phase prediction error is given in combination with the fluctuation metric; For the interval of the target phase, the compensated target phase is calculated based on the phase prediction error, so that the actual effective time of the external actuator falls within the compensated target phase after triggering execution.

[0013] Preferably, the three quality indicators of signal-to-noise ratio, phase consistency and phase estimation confidence are determined, when the result of the determination is lower than a threshold value, the preset oscillation frequency range or the alternative feature parameter is adjusted; when the determination result reaches the threshold value, the corresponding candidate interaction action or parameter is generated according to the current neural oscillation mode, including: ​​​a) calculating the signal-to-noise ratio, phase consistency and phase estimation confidence of each brain electrical signal in each preset oscillation frequency band, respectively, and outputting the calculation results after spatial fusion as three quality indicators corresponding to the preset oscillation frequency band; b) normalizing the three quality indicators and forming a single quality judgment score according to a preset rule; c) when the single quality judgment score is lower than a preset threshold, adjusting the used oscillation frequency band according to a preset strategy, or calling an alternative feature parameter to ensure signal reliability, and re-executing steps a) and b); d) when the single quality judgment score reaches the threshold, taking the neural oscillation mode vector at the current time as input, generating a corresponding candidate interactive action or parameter from a preset mapping relationship as input for subsequent state mapping and trigger execution.

[0014] Preferably, according to the individual parameters, neural oscillation mode, and results based on signal-to-noise ratio, phase consistency and phase estimation confidence, a corresponding state-action pair is selected from a preset mapping library, and the trigger time is determined by combining the phase prediction error, so that the external actuator performs the action within the interval of the target phase, including: calling individual parameters, neural oscillation mode vector, and quality judgment results as state mapping input; finding a state entry corresponding to the state mapping input in a preset mapping library, and reading the action entry matched by the state entry to form a target state-action pair; determining the interval of the corresponding target phase according to the preset requirements of the target action, combining the individual parameters and neural oscillation mode; using the phase prediction error to convert the target phase interval into a trigger advance time to generate a trigger plan for the external actuator; sending an execution signal at the trigger advance time to make the actual effective time of the external actuator fall within the interval of the target phase, thereby completing the execution of the target action.

[0015] A brain-computer interface interactive system based on neural oscillation mode, comprising: a signal acquisition and preprocessing module for synchronously acquiring electroencephalogram data at multiple head positions, establishing a unified time base, filtering and suppressing artifacts to obtain clean data streams covering θ, α, μ and β frequency bands; an oscillation mode construction module for extracting the instantaneous phase, amplitude and phase consistency index of each frequency band from the clean data stream, combining spatial weighting and coherence measurement for fusion to form a neural oscillation mode vector containing multi-frequency information; An individual parameter calibration and tracking module is configured to estimate individual dominant frequency and task-related latency in a baseline stage and dynamically correct the individual parameters by a recursive algorithm in a running process, and output individual parameters changing over time. A delay modeling and phase compensation module is configured to establish a delay model of an end-to-end processing link, derive a phase prediction error based on the delay model, and calculate a compensated target phase according to the phase prediction error. A signal quality determination and candidate action generation module is configured to determine according to three quality indexes of signal-to-noise ratio, phase consistency and phase estimation confidence, adjust a preset oscillation frequency band or replace a feature parameter when the determination result is lower than a threshold, and generate a corresponding candidate interactive action or parameter according to a current neural oscillation mode when the determination result reaches the threshold. A state mapping and trigger execution module is configured to select a corresponding state-action pair from a preset mapping library according to the individual parameters, the neural oscillation mode and the determination result based on the signal-to-noise ratio, the phase consistency and the phase estimation confidence, and determine a trigger time according to the phase prediction error, so that an external actuator executes the action in an interval of the target phase.

[0016] According to the specific embodiments of the present application, the following technical effects are provided. Firstly, the present application extracts instantaneous phase, amplitude and phase consistency index in multi-lead electroencephalogram signals, and combines spatial weighting and coherence measurement for fusion, effectively overcoming the limitation of existing methods relying on single frequency band and single channel features, and constructing a more robust neural oscillation mode, thereby improving the robustness and reliability of the interactive process.

[0017] Secondly, the present application estimates individual dominant frequency and task-related latency in a baseline stage, and dynamically corrects by a recursive algorithm in a running process, so that the system can track individual neural state changes over time, solves the problem that the prior art cannot effectively correct individual differences and frequency drift, and improves the adaptability and generalization ability of the method.

[0018] Thirdly, the present application establishes a delay model of an end-to-end processing link, and derives a phase prediction error for target phase compensation, which can make the external actuator fall into the set target phase interval at the real effective time, avoiding the phase deviation caused by device delay and jitter in the prior art, thereby significantly improving the timeliness and accuracy of action triggering.

[0019] Fourthly, the application performs oscillation frequency band or characteristic parameter adjustment when the quality is insufficient, and generates a candidate interactive action when the quality meets the standard, and further combines state mapping and trigger execution mechanism, realizes organic unity of signal quality control and interactive action generation, and makes up for the deficiencies of the prior art in signal quality gating and interactive stability. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0021] Figure 1 The method flowchart provided for the embodiments of the present application is as follows: Figure 2 The system structure schematic diagram provided for the embodiments of the present application is as follows. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0023] The purpose of the present application is to provide a neural oscillation pattern-based brain-computer interface interaction method and system, which significantly improves the reliability and accuracy of brain-computer interface interaction.

[0024] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0025] Figure 1 The method flowchart provided for the embodiments of the present application is as follows: Figure 1 As shown in the figure, the present application provides a neural oscillation pattern-based brain-computer interface interaction method, which comprises: Step 100: synchronously collecting electroencephalogram data at multiple lead positions on the scalp, establishing a unified time base, and obtaining clean data streams covering theta, alpha, mu and beta frequency bands through filtering and artifact suppression; Step 200: extracting instantaneous phase, amplitude and phase consistency index of each frequency band from the clean data stream, combining spatial weighting and coherence measurement for fusion, and forming a neural oscillation pattern vector containing multi-frequency information; Step 300: Estimate individual dominant frequency and task-related latency at baseline stage, and dynamically correct by recursive algorithm during running process, output individual parameters changing over time; Step 400: Establish delay model of end-to-end processing link, derive phase prediction error based on delay model, and calculate target phase after compensation according to phase prediction error; Step 500: Determine according to three quality indicators of signal-to-noise ratio, phase consistency and phase estimation confidence, when the result of determination is lower than the threshold, adjust the preset oscillation frequency range or replace the characteristic parameters used; when the determination result reaches the threshold, generate the corresponding candidate interactive action or parameter according to the current neural oscillation mode; Step 600: According to individual parameters, neural oscillation mode and the results obtained by judging based on signal-to-noise ratio, phase consistency and phase estimation confidence, select the corresponding state-action pair from the preset mapping library, and determine the triggering time combined with the phase prediction error, so that the external actuator executes the action within the interval of the target phase.

[0026] Specifically, in step 100 of the embodiment, the electrodes are arranged at the preset multiple electroencephalogram lead positions of the scalp, and the data acquisition module is used to start collecting all leads at the same time at a uniform sampling rate. To ensure the time alignment between the leads, a uniform time base is established before the collection starts, a monotonically increasing timestamp is attached to each frame of data during the collection process, and the data buffer maintains the order of arrival consistent with the timestamp. When there is cache playback, packet retransmission or channel reconnection, the timestamp is used as the only basis for reordering and alignment to ensure the strict consistency of multi-lead signals on the time axis. To reduce the pickup of power frequency and environmental interference, the embodiment enters the digital processing process after preliminary suppression in the analog front end.

[0027] The embodiment sequentially performs band-pass filtering and notch processing on the collected raw electroencephalogram signals. Band-pass filtering is used to retain the energy of four types of target frequency bands θ, α, μ and β and suppress direct current drift and high frequency noise; notch processing is used to suppress the local power grid power frequency and its adjacent frequency band energy to avoid the power frequency component covering the effective electroencephalogram activity. The filtering order adopts a serial structure of band-pass first and notch second to prevent the amplitude and phase distortion generated by notch at the edge of the spectrum from expanding into the passband. After this process, the output is the filtered signal of each channel, which is used as the input for artifact processing and quality assessment.

[0028] The embodiment performs artifact detection and elimination on the filtered signal, covering eye movement and blinking artifacts, electromyographic artifacts, and electrode falling off and poor contact artifacts. Eye movement and blinking artifacts are identified by the time domain peak value and waveform duration of the frontal lead; electromyographic artifacts are identified by high frequency power ratio and waveform roughness; electrode falling off and poor contact are identified by short flat section, saturated segment and impedance monitoring results. For the identified artifact segment, the embodiment preferentially adopts the component elimination and reconstruction strategy of independent component analysis, and the residual segment that cannot be reliably reconstructed adopts paragraph masking or interpolation smoothing to avoid bringing artifacts into subsequent feature extraction. After the above processing, the output is a multi-channel clean data stream covering the target frequency bands of θ, α, μ and β, meeting the input quality requirements of subsequent oscillation mode construction and phase estimation.

[0029] Specifically, the "unified time base" refers to the time reference established by the embodiment at the start of acquisition, which is common to all channels and all processing nodes. All data frames and events are time-stamped and aligned with this time reference; "electrode lead" refers to each electrode channel arranged at the standard position of the scalp or confirmed by clinical engineering. Each lead corresponds to an independent electroencephalogram signal; "clean data stream" refers to the multi-channel electroencephalogram signal sequence obtained after bandpass and notch processing, artifact detection and correction. The signal sequence is aligned on the time axis, meets the target range in the frequency band, and meets the quality requirements of subsequent feature extraction in artifact control. The embodiment preferentially uses 16 to 32 electrode leads, which comply with the international standard scalp arrangement specification; the sampling rate is preferably not less than 500, supporting the timing accuracy of phase and frequency; the input impedance of the analog front end is preferably greater than 1 x 10 6 , the single-channel baseline noise root mean square value is preferably less than 2; the lower limit of the bandpass filter passband is preferably between 0.5 and 1, and the upper limit of the passband is preferably between 40 and 45; the center frequency of the power frequency notch is selected according to the local power grid standard, and the bandwidth is preferably 1 to 2; the time window length used for each artifact determination is preferably 1 to 2 seconds, and the continuous window overlap ratio is preferably 50%; the time domain duration threshold of the blinking artifact is preferably set in the range of 0.1 to 0.5 seconds; the high frequency power ratio threshold of the electromyographic artifact is preferably set in the frequency band higher than 30; the electrode falling off alarm trigger condition is that the single-channel flat duration exceeds 2 seconds or the impedance mutation exceeds the specified upper limit; the reference segment of independent component analysis is preferably not less than 60 seconds to ensure the stability of the decomposition; the single-channel residual power frequency component of the processed clean data stream has a relative suppression of preferably not less than 20 decibels.

[0030] Specifically, step 200 of the embodiment includes: The cleaned data stream output by step 100 is input into a narrow-band filter set in this embodiment. The filter set is provided with independent band-pass channels for four types of target frequency bands of θ, α, μ and β, and each channel is synchronously operated for each brain electrical lead. The θ frequency band is used to extract low-frequency oscillation between four and seven, the α frequency band is used to extract dominant oscillation of the occipital-parietal region between eight and thirteen, the μ frequency band is used to extract dominant oscillation of the sensory-motor region between eight and twelve and is preferentially output on the central region related lead, and the β frequency band is used to extract rhythmic activity between thirteen and thirty. Each channel adopts a linear phase design to reduce group time delay distortion, and the filter boundary adopts a slow transition band and performs end point mirror extension processing to avoid boundary ripples caused by segmentation processing. The output after filtering is a band-limited signal sequence of each lead in each target frequency band, which enters the feature extraction link.

[0031] In this embodiment, Hilbert transform (HT) is performed on the band-limited signal of each target frequency band to obtain time sequences of instantaneous phase and instantaneous amplitude on the same time reference. In order to represent the phase stability in the spatial dimension, the phase difference distribution between adjacent brain electrical leads is calculated in a sliding time window, and the proportion of samples with phase difference falling within a predetermined small range is taken as the phase consistency index of the time window; when multiple pairs of adjacent leads are crossed, their weighted average is taken as the phase consistency of the frequency band in the time window. The above process is pushed along the time axis with a fixed step size, and the instantaneous phase representative value, the instantaneous amplitude representative value and the phase consistency representative value of the corresponding window are output at the end of each window.

[0032] In this embodiment, each brain electrical lead signal in each frequency band is weighted and distributed based on the signal-to-noise ratio and coherence measure. The signal-to-noise ratio is estimated by the ratio of effective power to noise power in the same time window, and the coherence measure is estimated by the consistency strength between lead pairs in the same time window; after the two quantities are unified, the fusion weight of each lead is obtained, and is normalized in the frequency band. Then, the instantaneous phase of each lead is weighted and circularly averaged to obtain the spatial robust phase of the frequency band, and the average instantaneous amplitude and the final phase consistency index of the corresponding time window of the frequency band are calculated. Finally, the spatial robust phase, the average instantaneous amplitude and the phase consistency of each frequency band are spliced in the order of the frequency band to form a neural oscillation mode vector containing multi-frequency information, which is called for subsequent individual parameter calibration and state mapping.

[0033] Specifically, the "narrow-band filter bank" refers to a parallel filter structure used to decompose the clean data stream into multiple target frequency band-limited signals under the same time reference; the "Hilbert transform (HT)" refers to a transform process that maps a real-valued band-limited signal into an analytic signal so as to directly read the instantaneous phase and the instantaneous amplitude; the "phase consistency index" refers to a non-dimensional quantity between zero and one calculated according to the concentration degree of the phase difference between adjacent leads within a sliding time window, and used to reflect the stability of the spatial phase; the "magnitude-squared coherence (MSC)" refers to a non-dimensional quantity used to measure the correlation strength between lead pairs within a sliding time window; the "spatial robust phase" refers to a representative phase obtained by weighted circular mean fusion of the multi-lead instantaneous phases within the same frequency band; and the "neural oscillation pattern vector" refers to a multi-dimensional feature vector composed of the spatial robust phases, the average instantaneous amplitudes and the phase consistencies of each frequency band in a predetermined order. In the embodiment, θ is preferably four to seven, α is preferably eight to thirteen, μ is preferably eight to twelve, and β is preferably thirteen to thirty; the sampling rate is preferably five hundred to one thousand; the transition bandwidth of each channel of the narrow-band filter bank is preferably zero point five to one, the stopband attenuation is preferably twenty to forty decibels, and the group delay deviation is preferably not higher than one sampling period; the sliding time window length is preferably one to two seconds, and the time window overlap ratio is preferably fifty percent to seventy percent; the small-range threshold of the phase consistency is preferably thirty-one to sixty-one percent of a circle; the coherence determination threshold is preferably zero point four to zero point six; the working dynamic range of the signal-to-noise ratio is preferably eight to twenty-six decibels; the weight normalization error during spatial fusion is preferably not higher than one thousandth; and the update step of the pattern vector is preferably less than half of the time window length, so as to ensure the time sequence continuity of adjacent vectors.

[0034] Specifically, after completing the feature extraction of step 200, the embodiment enters the data acquisition of the baseline stage. The subject maintains a specified resting or light task state, continuously acquires clean data streams covering the target frequency band, and writes the data in time sequence into a unified time base timestamp and quality label. To ensure the stability of subsequent estimation, the embodiment fixes the lead set, the filter parameters and the sliding time window setting in the baseline stage, and masks or replaces abnormal segments, forming a baseline data set that can be used for dominant frequency estimation and latency estimation.

[0035] The embodiment adopts a frequency matching optimization method to obtain individual dominant frequency. Specifically, each candidate frequency is matched with spatial robust phase in the same time window to score, and the score is based on the joint measurement of the concentration and consistency strength of the phase after alignment in the time window; the maximum score in all candidate frequencies is selected as the initial value of the dominant frequency of the frequency band. To improve noise resistance, the score curve is smoothed, low-quality time windows are removed, and consistency test is performed on the initial value in adjacent time windows, and finally the individual dominant frequency initial value of each frequency band is output.

[0036] The embodiment collects several event-locked data under baseline conditions to construct a template signal for matching with the same time reference. For each single trial data, the template is shifted point by point within the defined latency search range, the normalized correlation score of the template and the single trial is calculated, and the time delay corresponding to the correlation peak is taken as the latency of the trial. To reduce the influence of incidental disturbances, the latency results of multiple trials are robustly statistically processed, and the initial value of the individual task-related latency is obtained after removing outliers, and its stability is verified in subsequent short-time retests.

[0037] After entering the running phase, the embodiment generates two types of observation quantities with a fixed step size: one is obtained by the time difference of spatial robust phase, which is used to represent the instantaneous angular frequency; the other is obtained by the normalized correlation of each event-locked data and the template, which is used to represent the current latency. For the above two types of observation quantities, the embodiment adopts a recursive strategy with adaptive variance to update them jointly: when the fluctuation of the observation is small and close to the last estimate, the weight of the observation is increased; when the fluctuation of the observation is increased or deviates from the last estimate, the weight of the observation is reduced or the update is temporarily suspended. The variance of the observation noise is obtained by residual statistics in a sliding time window and is updated rolling in time. To ensure stability, the embodiment sets an upper limit of the change rate and a confidence interval constraint, freezes or reduces the weight of the update in the time window with poor quality label, and avoids the transmission of short-term dramatic fluctuations to the final estimate.

[0038] The embodiment converts the angular frequency estimate obtained in the running phase into the dominant frequency, which is output together with the recursively obtained latency as individual parameters that change over time, for subsequent state mapping and trigger execution call. At the same time, the results of each update are written into the cache and log together with the time stamp, quality label and observation residual, which facilitates tracing and offline review; when the cumulative samples reach a set threshold, the effective range of the template and candidate frequency set is automatically refreshed, so that the individual parameters remain consistent with the state of the subject in the long-term running.

[0039] Specifically, the "frequency matching optimization" of the embodiment refers to point-by-point retrieval within a pre-set candidate frequency set, scoring based on the concentration degree and consistency strength after phase alignment, and taking the candidate frequency with the highest score as the dominant frequency of the frequency band; the "template normalization correlation" refers to point-by-point alignment and similarity calculation of the template signal obtained in the baseline stage and the single trial data within the latency search range, so as to obtain the time delay position corresponding to the latency; the "task-related latency" refers to the time difference from the presentation of the stimulus to the peak or typical component of the evoked response in the event-locked paradigm, which is used as a physiological reference for subsequent trigger timing and phase compensation. The duration of the baseline stage is preferably 60 to 120 seconds; the candidate frequency step is preferably 0.1 or 0.2, and the search range is set within the standard range of each frequency band, for example, eight to thirteen for the dominant frequency of the occipital region, eight to twelve for the dominant frequency of the sensory motor region, and thirteen to thirty for the general rhythm; the event-locked window for template construction is preferably 100 milliseconds before the stimulus to 400 milliseconds after the stimulus, the latency search range is preferably 50 to 250 milliseconds, and the effective threshold of the correlation score is preferably 0.3 to 0.6; the sliding time window length observed in the running stage is preferably 1 to 2 seconds, and the time window overlap ratio is preferably five to seven; the output refresh rate of the recursive update is preferably ten to fifty times per second, the upper limit of the change rate of the dominant frequency is preferably not more than one per second, and the upper limit of the change rate of the latency is preferably not more than twenty per second; the outlier rejection criterion can use a robust threshold based on time window statistics, and the threshold multiple is preferably three; when the quality marker indicates that the signal-to-noise ratio is lower than eight decibels or the phase consistency is lower than zero point six, the parameter update is suspended or the weight processing is reduced.

[0040] Specifically, the step 400 of the embodiment includes: The embodiment calibrates the processing nodes of collection, filtering, feature extraction, communication and external execution one by one. During calibration, time stamps are written for the input events and output events of each node under a unified time base, and the reference sampling time corresponding to the phase evaluation and the actual effective time corresponding to the effective output of the external actuator are recorded. To eliminate the drift of the internal clock of different devices, the embodiment performs time synchronization before starting the flow, and re-measures the synchronization deviation at a fixed period during running; when it is found that the deviation exceeds the pre-set upper limit, immediate re-synchronization is performed and the correction amount is recorded in the log. All time stamps are written in a monotonically increasing manner, and any late data is reordered into the queue according to the time stamp, ensuring the consistency of the time series of each node.

[0041] The embodiment takes the reference sampling time as the starting point and the actual effective time of the external actuator as the ending point, and calculates the end-to-end time delay value of each interaction after the interaction is completed. The time delay value, the corresponding task number, the trigger source, and the actuator mode are written into the ring buffer and the persistent log, and form an end-to-end time delay sequence in chronological order. For interactions with packet loss, retransmission, or execution failure, the embodiment does not include them in the statistics after being marked as invalid, and compensates with adjacent valid samples in the subsequent window to prevent abnormal samples from causing deviation in the statistical results. To facilitate tracing, the embodiment also saves the quality label and phase evaluation results related to the interaction for offline verification.

[0042] The embodiment uses a sliding time window to statistically analyze the end-to-end time delay sequence, and gives two indicators in each time window: one is the time delay estimation value, which reflects the typical time delay level in the time window; the other is the fluctuation measure, which reflects the dispersion degree of the time delay in the same time window. To improve robustness, the embodiment uses a robust statistical method to suppress the influence of extreme values, and smooths the results of multiple consecutive time windows to obtain a continuously updated time delay model. The model is automatically refreshed at each window sliding, and its output provides the time delay estimation value and the corresponding fluctuation measure of the current period, which are used for subsequent phase conversion.

[0043] The embodiment calls the individualized dominant angular frequency output in step three hundred, converts the time delay estimation value in the time delay model into a phase prediction error, and simultaneously expands the phase prediction error into a confidence range with upper and lower bounds according to the fluctuation measure. The conversion considers the periodicity of the phase, so that the phase prediction error always falls within the preset main value range. When the fluctuation measure significantly increases in a short time, the embodiment triggers a protection strategy to increase the confidence range and reduce the trigger sensitivity to avoid overly aggressive early triggering when the jitter is large. The phase prediction error and its confidence range are updated synchronously after each window refresh, serving as input for subsequent compensation calculation.

[0044] The embodiment calculates the target phase after compensation based on the phase prediction error according to the interval setting of the target phase (determined by the center phase and the tolerance), and generates a trigger plan for the external actuator based on the target phase. The trigger plan calculates the specific time of the early trigger based on the reference sampling time, so that the external actuator falls within the target phase interval at the actual effective time. After the action is completed, the embodiment reads the reference sampling time and the actual effective time of this time, checks whether it hits the target phase interval, and writes the hit or miss back to the log for self-checking. If there are consecutive deviations, the embodiment automatically tightens the target phase tolerance or temporarily increases the quality threshold, and sets the time delay sample in this segment to low credibility to prevent pollution of subsequent statistics.

[0045] Specifically, the "end-to-end processing link" of the embodiment refers to a complete path from the reference sampling time corresponding to the phase evaluation, sequentially passing through the nodes of collection, filtering, feature extraction, communication, external execution, and so on, until the output of the external executor reaches a predetermined threshold; the "delay model" refers to a periodical description obtained by robustly statistically processing the end-to-end delay sequence within a sliding time window, at least including a delay estimation value and a corresponding fluctuation metric; the "reference sampling time" refers to the data time point at which the phase evaluation result is generated; the "actual effective time" refers to the time point at which the output of the external executor reaches the effective threshold; the "phase prediction error" refers to the phase deviation amount converted from the delay estimation value; and the "compensated target phase" refers to the target phase used to trigger scheduling obtained by deducting the phase prediction error from the original target phase and combining the confidence range. The time synchronization retest period is preferably ten to thirty seconds; the end-to-end delay window length is preferably five to twenty seconds, and the window sliding step is preferably zero point five to two seconds; the threshold for removing abnormal delay values is preferably three to four times the relative median deviation; the protection strategy is triggered when the delay fluctuation metric exceeds ten milliseconds, and the target phase tolerance is temporarily widened to twice the original setting; the advance resolution of the reference sampling time to the trigger signal is preferably better than one millisecond; the actual effective determination threshold is preferably set to five to ten percent of the rated output of the external executor; in the hit test, the target phase tolerance is preferably set to thirty to sixty one-hundredth of the circumference, and the tolerance is reset and the quality threshold is raised when there is no hit for three times in a row; the log retention time is preferably not less than seven days, facilitating offline review and comparison.

[0046] Specifically, the step 500 of the embodiment includes: The embodiment calculates the signal-to-noise ratio value, the phase consistency value, and the confidence value of the phase estimation for each electroencephalogram lead signal in each predetermined oscillation frequency band. After the calculation is completed, the three indicators are weighted and integrated in the lead dimension according to the spatial fusion rule obtained in step two hundred, and the output is the current time window representative value of the three quality indicators corresponding to the oscillation frequency band. To ensure time sequence continuity, the embodiment uses a fixed length sliding time window and a fixed step to advance on the time axis, and sets a quality mark and a time stamp for the three quality indicators obtained by each advancement for subsequent normalization and comprehensive judgment.

[0047] The embodiment maps the three quality indicators to the standard range of zero to one, and forms a single quality determination score according to the preset monotonic aggregation rule. The score is compared with the preset threshold to obtain the determination result of the current time window. When the score is lower than the threshold, the embodiment performs at least one of band adjustment and alternative feature parameter calling according to the established rollback strategy, including increasing the priority of the oscillation band with higher stability, enabling the feature combination dominated by phase consistency, enabling more stringent spatial fusion weight constraints, and shortening or lengthening the sliding time window to improve robustness. After the rollback action is completed, the quality indicator calculation and comprehensive determination process is returned immediately, and the adjustment reason and adjustment amplitude are recorded in the determination log for subsequent parameter adaptation.

[0048] When the single quality determination score reaches the threshold, the embodiment takes the neural oscillation mode vector at the current time as input, retrieves the corresponding state entry from the preset mapping relationship, and then reads the action entry and parameter entry associated with the state entry to obtain the candidate interaction action and candidate interaction parameter. To facilitate subsequent triggering and execution, the embodiment sorts the candidate results according to the quality-related confidence, and outputs the sorting result, the corresponding state identifier, the timestamp and the quality score to the state mapping and triggering execution process, and writes them into the running log for offline review and online adaptive update.

[0049] Specifically, the "single quality determination score" refers to a unique score obtained by normalizing the signal-to-noise ratio value, the phase consistency value, and the phase estimation confidence value within the range of zero to one according to a predetermined monotonic aggregation rule, which is used to give a pass or insufficient conclusion after comparison with a threshold value; the "alternative feature parameter" refers to a backup feature configuration used to improve robustness when the quality is insufficient, including a feature combination centered on phase consistency, tightening the upper and lower limits of the spatial fusion weight, adjusting the time window length and step size, etc.; the "preset mapping relationship" refers to a lookup structure from the neural oscillation mode vector to the interactive action and the interactive parameter established in the training stage or the pre-configuration stage, which can be an enumerated table structure or an index structure centered on similarity retrieval; the "candidate interactive action and candidate interactive parameter" refer to the action set and parameter set returned by the preset mapping relationship after the quality meets the standard at the current time, and have credibility ranking information, which are used for subsequent trigger execution and phase compensation scheduling. The sliding time window length is preferably one to two, the step size is preferably zero point five to one, and the time window overlap ratio is preferably five to seven; the pass lower limit of the signal-to-noise ratio is preferably set between eight and twenty, the pass lower limit of the phase consistency is preferably set between zero point six and zero point eight, and the pass lower limit of the phase estimation confidence is preferably set between zero point seven and zero point nine; the threshold value of the single quality determination score is preferably set between zero point six and zero point eight; when the score of three to five consecutive time windows is lower than the threshold value, the fallback strategy is triggered, the priority of the stable frequency band is preferentially increased, the fluctuation range of the spatial fusion weight is tightened, and the time window length is increased by two to five tenths; the size of the candidate interactive action set is preferably two to five, and the size of the candidate interactive parameter set is preferably two to three groups; the candidate result output frequency when the quality meets the standard is preferably ten to fifty times per second; the online fine-tuning amplitude of all threshold values and parameters is preferably controlled between ten to twenty percent of the original setting value to avoid excessive oscillation.

[0050] Specifically, the step 600 of the embodiment includes: The embodiment first simultaneously calls three types of inputs: one is the current dominant frequency and latency information obtained from the aforementioned individual parameters, two is the current multi-frequency state description obtained from the aforementioned neural oscillation mode vector, and three is the pass or insufficient conclusion and corresponding quality score obtained from the aforementioned quality determination. Only when the quality meets the standard, the state mapping process is entered; if the quality is insufficient, the fallback strategy is executed until the quality meets the standard. After the condition is met, the embodiment organizes the above three types of inputs into a structured state mapping input with a timestamp as the key value, and performs retrieval in the preset mapping library. The mapping library takes state entries as the basic unit, and each state entry contains the applicable frequency band range, phase interval, amplitude strength, phase consistency level, and corresponding action entries and parameter entries. The embodiment retrieves in order from high to low according to the matching degree, and when there are ties, the action entry with higher recent success rate and lower execution cost is preferentially selected, forming a target state-action pair.

[0051] After determining the target state-action pair, the embodiment generates the target phase interval according to the preset requirements of the action, combined with the individual parameters and the current neural oscillation mode. The interval is jointly defined by the center and the tolerance, the center is provided by the mapping library entry and can be fine-tuned according to the individual parameters, and the tolerance is adaptively converged or relaxed according to the quality score. Subsequently, the embodiment calls the aforementioned phase prediction error, converts the target phase interval into the trigger advance time, and generates the trigger plan of the external executor. The trigger plan includes: trigger advance, preheating time (if applicable), execution duration, retry number and retry interval, and protection strategy when the fluctuation is intensified in a short time. In order to ensure that the actual effective time falls within the target phase interval, the embodiment considers the response delay and jitter characteristics of the external executor in the plan, and reserves a safety buffer for abnormal situations.

[0052] After reaching the trigger advance time, the embodiment sends an execution signal to the external executor, and records the planned trigger time and the device feedback signal. After the execution is completed, the embodiment reads the actual effective time of the external executor, and performs hit checking with the target phase interval: if it falls within the interval, it is determined to be a hit, and the state, quality score, hit result and timing information of the action are written into the running log and the statistical field of the mapping library; if it is not hit, one or more retries are immediately initiated according to the retry strategy in the trigger plan; if the retry still does not hit, the priority of the state entry is temporarily reduced, and the target phase tolerance is tightened or the quality threshold is raised, for stable operation in the subsequent period. After continuous multiple hits, the tolerance is moderately relaxed to improve the throughput, and the cumulative results are periodically used for online updating of the action selection weight of the mapping library.

[0053] Specifically, the "preset mapping library" refers to a lookup structure established in the training or pre-configuration stage for mapping the current neural oscillation mode and individual parameters to external actions and their parameters, including state entries, action entries and parameter entries; the "state-action pair" refers to a set of one-to-one correspondence obtained by searching the mapping library, used to indicate the action and parameters to be executed in a specific state; the "interval of the target phase" refers to the phase hit range set for a certain action, jointly defined by the center and the tolerance; the "trigger advance time" refers to the trigger time issued before the reference time point to offset the device processing and response delay; the "actual effective time" refers to the time point when the output of the external executor reaches the effective threshold and can have an effect on the interaction.

[0054] Exemplarily, the refresh frequency of the mapping library retrieval and action selection is preferably ten to fifty times per second; the tolerance of the target phase interval is preferably set to one-thirtieth to one-sixtieth of a full cycle; when the quality score is near the lower limit of the threshold, the tolerance can be temporarily relaxed to one-twentieth of a full cycle; the response delay of the external executor is preferably recorded with a resolution better than one millisecond, and when the jitter exceeds ten milliseconds, a protection strategy is triggered and the quality threshold is increased by one level; the number of retries in the trigger plan is preferably one to two times, and the retry interval is preferably five to twenty milliseconds; the single execution duration is preferably fifty to two hundred milliseconds; after three consecutive hits, the tolerance can be relaxed by twenty percent to improve the throughput, and after two consecutive misses, the tolerance is tightened by thirty percent and the quality threshold is increased by one level; the log retention time is preferably no less than seven days, and the online update period of the mapping library is preferably ten to thirty seconds.

[0055] Preferably, it further comprises: Step 700: Record the phase hit rate, command hit rate and task success rate three levels of indicators respectively, and dynamically update the weight coefficient of each said electroencephalogram data in spatial fusion, the priority of the oscillation frequency band and the state mapping parameter according to the statistical results of said levels of indicators, so as to realize the continuous optimization of the interaction performance.

[0056] Specifically, the embodiment automatically records three types of indicators, i.e. phase hit rate, command hit rate and task success rate, after each external executor completes an action. The phase hit rate is used to measure whether the actual effective time of the external executor falls within the target phase interval; the command hit rate is used to measure whether the triggered candidate action is consistent with the expected state-action pair; and the task success rate is used to measure whether the established interaction task goal is achieved in multiple consecutive actions. The embodiment establishes independent counters and time stamp indexes for the three types of indicators respectively, and performs statistical update within a fixed time window, so as to grasp the interaction performance at different levels in real time.

[0057] After the statistical results are generated, the embodiment compares the three types of indicators with the preset threshold or historical average value respectively. When the phase hit rate decreases, the weight coefficient of the electroencephalogram data in spatial fusion is adjusted preferentially, the weight of the stable lead is reduced and the weight of the unstable lead is increased; when the command hit rate decreases, the priority of the oscillation frequency band is adjusted preferentially, so that the frequency band with higher stability obtains greater weight in the mapping; and when the task success rate decreases, the state mapping parameter is adjusted preferentially, including the trigger condition of the state entry, the phase tolerance and the action selection priority. The embodiment ensures that the update action is clear in pertinence through layered feedback, and the update strategies triggered by different indicators have different focuses.

[0058] The embodiment writes the new weight coefficient, frequency band priority and mapping parameter back to the running memory after each round of statistical update, and takes effect in the next round of state mapping and trigger execution. At the same time, all update records are written into the log for offline analysis and subsequent parameter reset reference. When the three types of indicators of the continuous multiple time windows are maintained at a level higher than the preset threshold, the embodiment keeps the current parameters unchanged; when any of the indicators is continuously lower than the threshold, the embodiment increases the update amplitude and triggers the protection strategy, such as increasing the quality threshold or reducing the phase tolerance. Through dynamic updating and adaptive iteration, the embodiment can realize continuous optimization of interactive performance and ensure long-term stable operation of the system.

[0059] Corresponding to the above method, as shown in Figure 2 The embodiment also provides a brain-computer interface interaction system based on neural oscillation mode, which comprises: A signal acquisition and preprocessing module is configured to synchronously acquire electroencephalogram data at multiple positions on the scalp, establish a unified time base, and obtain clean data streams covering theta, alpha, mu and beta bands through filtering and artifact suppression; An oscillation mode construction module is configured to extract instantaneous phase, amplitude and phase consistency index of each frequency band from the clean data streams, combine spatial weighting and coherence measurement for fusion, and form a neural oscillation mode vector containing multiple frequency information; An individual parameter calibration and tracking module is configured to estimate individualized dominant frequency and task-related latency in the baseline stage, and dynamically correct through recursive algorithm in the running process, and output individual parameters changing with time; A time delay modeling and phase compensation module is configured to establish a time delay model of an end-to-end processing link, derive a phase prediction error on the basis of the time delay model, and calculate a target phase after compensation according to the phase prediction error; A signal quality determination and candidate action generation module is configured to determine according to three quality indicators of signal-to-noise ratio, phase consistency and phase estimation confidence, adjust the preset oscillation frequency band or replace the feature parameter when the determination result is lower than the threshold, and generate a corresponding candidate interactive action or parameter according to the current neural oscillation mode when the determination result reaches the threshold; A state mapping and trigger execution module is configured to select a corresponding state-action pair from a preset mapping library according to the individual parameters, neural oscillation mode and determination result based on signal-to-noise ratio, phase consistency and phase estimation confidence, and determine a trigger time in combination with the phase prediction error, so that an external actuator executes the action in the interval of the target phase.

[0060] The various embodiments described in this specification are presented for the purpose of illustrating the principles of the application and its best mode of operation. Each of the embodiments described in this specification has different advantages and disadvantages, and the embodiments disclosed in this specification are not intended to limit the scope of the application to the embodiments disclosed. Embodiments of the system disclosed in this specification are intended to be limited only by the appended claims.

[0061] The principles and implementations of the present application are described in this specification with specific examples. The above description of the embodiments is only to help understand the method of the present application and its core idea. For those skilled in the art, the specific implementation and application scope of the present application can be changed according to the idea of the present application. In summary, the content of this specification should not be understood as a limitation of the present application.

Claims

1. A brain-computer interface interaction method based on neural oscillation patterns, characterized in that, include: EEG data were simultaneously acquired at multiple leads on the scalp, a unified time base was established, and after filtering and artifact suppression, a clean data stream covering the θ, α, μ, and β frequency bands was obtained. The instantaneous phase, amplitude, and phase consistency index of each frequency band are extracted from the clean data stream and fused with spatial weighting and coherence measurement to form a neural oscillation pattern vector containing multi-frequency information. At the baseline stage, the individualized dominance frequency and task-related latency are estimated and dynamically corrected during operation using a recursive algorithm, outputting individual parameters that change over time. Establish a time delay model for the end-to-end processing link, derive the phase prediction error based on the time delay model, and calculate the compensated target phase based on the phase prediction error. The judgment is made based on three quality indicators: signal-to-noise ratio, phase consistency, and phase estimation confidence. When the judgment result is lower than the threshold, the preset oscillation frequency band or alternative characteristic parameters are adjusted. When the judgment result reaches the threshold, the corresponding candidate interaction action or parameter is generated based on the current neural oscillation pattern. Based on the individual parameters, neural oscillation patterns, and the results obtained from the signal-to-noise ratio, phase consistency, and phase estimation confidence, a corresponding state-action pair is selected from a preset mapping library, and the triggering time is determined in conjunction with the phase prediction error, so that the external actuator performs the action within the range of the target phase.

2. The brain-computer interface interaction method based on neural oscillation patterns according to claim 1, characterized in that, Also includes: The system records three levels of indicators: phase hit rate, command hit rate, and task success rate. Based on the statistical results of these indicators, it dynamically updates the weight coefficients of each EEG data in spatial fusion, the priority of the oscillation frequency band, and the state mapping parameters, thereby achieving continuous optimization of interaction performance.

3. The brain-computer interface interaction method based on neural oscillation patterns according to claim 1, characterized in that, EEG data were simultaneously acquired at multiple leads on the scalp, a unified time base was established, and after filtering and artifact suppression, a clean data stream covering the theta, alpha, μ, and beta frequency bands was obtained, including: Electrodes were placed at multiple electrode lead locations on the scalp, and raw EEG signals were simultaneously acquired at a uniform sampling rate using a data acquisition module. During the acquisition process, the EEG signals of each channel are timestamped, and a unified time base is established to achieve multi-channel synchronization; The raw EEG signals were sequentially subjected to bandpass filtering and notch filtering to preserve the θ, α, μ, and β frequency bands and suppress power line interference and high-frequency noise, resulting in filtered signals. The filtered signal is subjected to artifact detection and removal, including eye movement artifacts, electromyography artifacts and electrode detachment artifacts, and artifact correction is performed using independent component analysis to obtain the processed multi-channel clean data stream.

4. The brain-computer interface interaction method based on neural oscillation patterns according to claim 1, characterized in that, Instantaneous phase, amplitude, and phase consistency indices for each frequency band are extracted from the clean data stream and fused using spatial weighting and coherence metrics to form a neural oscillation pattern vector containing multi-frequency information, including: The clean data stream is input into the narrowband filter group corresponding to each frequency band to obtain the band-limited signals of the θ, α, μ, and β frequency bands; The instantaneous phase and instantaneous amplitude of the band-limited signal are extracted by Hilbert transform, and the phase consistency index is calculated based on the phase difference between adjacent leads; Based on the signal-to-noise ratio or coherence measure of each lead signal, the weighting coefficients of the leads in spatial fusion are determined, and based on the weighting coefficients, the instantaneous phases of each lead are spatially integrated using a weighted average fusion method to obtain the spatially robust phases of each frequency band. The spatially robust phase, the corresponding instantaneous amplitude, and the phase consistency index are combined by frequency band to form the neural oscillation mode vector containing multi-frequency information.

5. The brain-computer interface interaction method based on neural oscillation patterns according to claim 4, characterized in that, Based on the signal-to-noise ratio or coherence metric of each lead signal, the weighting coefficients for each lead in spatial fusion are determined. Then, based on these weighting coefficients, a weighted average fusion method is used to spatially integrate the instantaneous phases of each lead, resulting in spatially robust phases for each frequency band, including: The instantaneous phase and instantaneous amplitude of each lead are extracted from the clean data stream within each preset oscillation frequency band to obtain... and ,in Number the leads For frequency band, For time sampling points; In each frequency band Up, click the sliding time window Calculate leads Phase consistency The phase consistency is used to characterize the stability of the lead phase relative to the group phase, and the formula is: ;in, For leads In frequency band Phase consistency; For The number of sampling points within the central time window; It is represented by a complex exponent; The imaginary unit; For leads In frequency band time The instantaneous phase; This represents the average phase of the group at that moment; Total number of leads; In each frequency band Up, click the sliding time window Constructing leads from instantaneous amplitude Normalized signal-to-noise ratio mapping quantity It is used to measure signal reliability, and the formula is: ;in, For leads In frequency band The signal-to-noise ratio mapping quantity; Instantaneous amplitude; In the time window The noise variance is estimated based on the median absolute deviation calculated from the square of the instantaneous amplitude. The operator representing the absolute deviation of the median; In each frequency band Above, the unnormalized weights of the leads are obtained by combining the phase consistency and signal-to-noise ratio mapping. The spatial fusion weighting coefficients are obtained by probabilistic normalization. The formula is: , ;in, For leads Unnormalized weights; For leads Spatial fusion weighting coefficient and satisfying ; The instantaneous phase of each lead is spatially integrated using a weighted circular mean to obtain the frequency band. Spatial robust phase The formula is: ;in, frequency band Spatial robust phase; This is the phase value operator.

6. The brain-computer interface interaction method based on neural oscillation patterns according to claim 1, characterized in that, During the baseline phase, individualized dominance frequencies and task-related latency are estimated and dynamically corrected using a recursive algorithm during operation, outputting time-varying individual parameters, including: During the baseline phase, based on the spatially robust phase of each frequency band Frequency matching optimization is performed to estimate individualized dominant angular frequencies and convert them into dominant frequencies, using the following formula: ;in, frequency band Baseline dominant angular frequency estimation; frequency band The set of candidate angular frequencies; Baseline duration; Total number of leads; For leads In frequency band The instantaneous phase; The imaginary unit; For dominant frequency estimation; Baseline task-related latency is estimated using template matching on the baseline trial of the task. The formula is as follows: ;in, Baseline latency estimation; For the incubation period search set; The number of trials used for baseline estimation; For the first The second test was in the frequency band. A band-limited time-domain segment; It is a template function constructed from the baseline trial average signal; It is the inner product; It is a norm 2; Entering the operational phase, based on the spatially robust phase The discrete-time derivative is used to obtain the angular frequency measurement, and template matching is used to obtain the single-trial latency measurement. The formula is as follows: ;in, It is a measurement of angular frequency; For phase untangling operators; The sampling interval; This is a measurement of the latency period for a single test. For the current test in the frequency band Band-limited segments; Individual parameters are expressed as to measure vector The driver updates point by point, and the update formula is: , , ;in, For a moment The individual parameter vector; It is a diagonal gain matrix; and These are the adaptive gains for the angular frequency and latency channels, respectively. For the first Estimation of the operational variance of a dimensional state; For the first Dimensional measurement variance, by time window The variance of the measurement residuals within the range is obtained; This represents the number of samples within the window. and The first Dimensional measurement and state estimation from the previous time step; Will Converted to dominant frequency ,and Together they serve as the individualized parameters that change over time.

7. The brain-computer interface interaction method based on neural oscillation patterns according to claim 6, characterized in that, Establish a time delay model for the end-to-end processing link, derive the phase prediction error based on the time delay model, and calculate the compensated target phase based on the phase prediction error, including: The processing nodes for acquisition, filtering, feature extraction, communication and external execution are calibrated. Under a unified time base, timestamps are written for the input and output events of each node, and the reference sampling time for phase evaluation and the actual effective time of the external actuator are recorded. Starting from the reference sampling time of the phase evaluation and ending at the actual effective time of the external actuator, the end-to-end delay value corresponding to a single interaction is calculated, and an end-to-end delay sequence is formed in chronological order. The end-to-end delay sequence is statistically analyzed within a sliding time window to obtain a delay model that describes the delay level and jitter range of the processing link; the delay model includes at least the delay estimate and its corresponding fluctuation metric within the current time window. Based on the individualized dominant angular frequency, the time delay estimate of the time delay model is used to convert the time error into the phase prediction error, and the confidence range of the phase prediction error is given in combination with the fluctuation metric. For the range of the target phase, the compensated target phase is calculated based on the phase prediction error, so that the actual effective time of the external actuator after triggering execution falls into the compensated target phase.

8. The brain-computer interface interaction method based on neural oscillation patterns according to claim 1, characterized in that, The judgment is made based on three quality indicators: signal-to-noise ratio, phase consistency, and phase estimation confidence. When the judgment result is lower than the threshold, the preset oscillation frequency band or alternative characteristic parameters are adjusted. When the judgment result reaches the threshold, corresponding candidate interactive actions or parameters are generated based on the current neural oscillation pattern, including: a) Within each preset oscillation frequency band, calculate the signal-to-noise ratio, phase consistency, and confidence level of each EEG signal, and output the results after spatial fusion as the three quality indicators corresponding to the preset oscillation frequency band. b) Normalize the three quality indicators and form a single quality judgment score according to preset rules; c) When the single quality judgment score is lower than the preset threshold, adjust the oscillation frequency band used according to the preset strategy, or call alternative feature parameters to ensure signal reliability, and re-execute steps a) and b). d) When the single quality judgment score reaches the threshold, the corresponding candidate interactive action or parameter is generated from the preset mapping relationship using the neural oscillation mode vector at the current moment as input, and used as the input for subsequent state mapping and trigger execution.

9. The brain-computer interface interaction method based on neural oscillation patterns according to claim 1, characterized in that, Based on the individual parameters, neural oscillation patterns, and results determined by signal-to-noise ratio, phase consistency, and phase estimation confidence, a corresponding state-action pair is selected from a preset mapping library. The triggering time is determined by combining the phase prediction error, causing the external actuator to perform the action within the target phase interval, including: Individual parameters, neural oscillation pattern vectors, and quality determination results are used as inputs for state mapping; The state entry corresponding to the state mapping input is searched in the preset mapping library, and the action entry matching the state entry is read to form a target state-action pair; Based on the preset requirements of the target action, and in combination with the individual parameters and neural oscillation patterns, the corresponding target phase interval is determined; The target phase interval is converted into a trigger advance time using the phase prediction error, and a triggering plan for the external actuator is generated. An execution signal is issued at the trigger advance time, so that the actual effective time of the external actuator falls within the range of the target phase, thereby completing the execution of the target action.

10. A brain-computer interface interaction system based on neural oscillation patterns, characterized in that, include: The signal acquisition and preprocessing module is used to simultaneously acquire EEG data at multi-lead locations on the scalp, establish a unified time base, and obtain a clean data stream covering the θ, α, μ, and β frequency bands after filtering and artifact suppression. The oscillation mode construction module is used to extract the instantaneous phase, amplitude and phase consistency index of each frequency band from the clean data stream, and fuse them with spatial weighting and coherence measurement to form a neural oscillation mode vector containing multi-frequency information. The individual parameter calibration and tracking module is used to estimate the individualized dominant frequency and task-related latency during the baseline phase, and dynamically corrects them through a recursive algorithm during operation, outputting individual parameters that change over time. The delay modeling and phase compensation module is used to establish a delay model for the end-to-end processing link, derive the phase prediction error based on the delay model, and calculate the compensated target phase based on the phase prediction error. The signal quality judgment and candidate action generation module is used to judge based on three quality indicators: signal-to-noise ratio, phase consistency, and phase estimation confidence. When the judgment result is lower than the threshold, the preset oscillation frequency band or alternative feature parameters are adjusted. When the judgment result reaches the threshold, the corresponding candidate interaction action or parameter is generated based on the current neural oscillation pattern. The state mapping and triggering execution module is used to select the corresponding state-action pair from the preset mapping library based on the individual parameters, neural oscillation mode, and the results obtained based on signal-to-noise ratio, phase consistency, and phase estimation confidence, and to determine the triggering time in combination with the phase prediction error, so that the external actuator performs the action within the range of the target phase.

Images