A method and system for regulating upper limb movement based on brain-computer interface adaptive music

Abstract

This invention provides an adaptive music-controlled upper limb movement method and system based on a brain-computer interface, relating to the field of computer vision technology. The method includes: acquiring acoustic envelope sequences of multi-channel physiological electrical signals and digital audio streams; sequentially calculating the temporal synchronization features of the physiological electrical signals and acoustic envelope sequences, the spatial asymmetric features constrained by the temporal synchronization feature frequency band, and the event response delay features constrained by the channel weights of the spatial asymmetric features; calculating the consistency deviation index, iteratively correcting the features that failed verification using a dynamic confidence smoothing factor, and outputting the corrected target feature set; mapping and calculating spatial azimuth parameters, reverberation wet / dry ratio, and low-frequency oscillation modulation parameters, generating and outputting a dynamic dual-channel output data sequence to perform acoustic reconstruction; collecting the posterior target feature set after acoustic reconstruction to calculate the temporal physiological state gradient, and updating the feature extraction parameters and consistency decision threshold based on the temporal physiological state gradient.

Description

Technical Field

[0001] This application relates to the fields of computer digital signal processing and human-computer interaction technology, and more specifically, to an adaptive music-controlled upper limb movement method and system based on a brain-computer interface. Background Technology

[0002] Brain-computer interface (BCI) technology constructs a communication pathway for direct interaction between the physical world and underlying physiological information by collecting and analyzing physiological electrical signals from the central nervous system. In multimodal human-computer interaction and motion intention recognition scenarios, introducing digital audio streams as feedback stimulus signals can effectively induce specific frequency band EEG neural oscillation responses in the target object (such as event-related desynchronization). By processing the data characteristics of physiological electrical signals and mapping them to control commands for digital audio, a closed-loop information link between external audio sequences and internal physiological neural activities can be constructed, thus playing an important role in data interaction scenarios such as motion rhythm guidance and physiological state modulation.

[0003] In existing audio interaction systems based on digital signal processing, parallel feature extraction from multi-channel physiological electrical signals is typically relied upon. The general data processing flow involves independently calculating the time-domain response delay, frequency-domain energy spectral density, and timing-locked loop characteristics of the acquired signals. These independently extracted multidimensional features are then input in parallel into a preset static decision model. When a single feature or a linear combination of multiple features exceeds a preset static threshold, the system triggers a specific audio control command. These commands are usually limited to one-dimensional or basic audio parameter adjustments, such as linear scaling of the overall output gain or direct switching of digital audio files.

[0004] However, the aforementioned conventional data processing architecture has significant technical limitations when facing complex physiological interaction scenarios. Multi-channel physiological electrical signals exhibit high physical coupling in the time, frequency, and spatial domains, and the use of independent parallel feature extraction logic ignores the underlying constraints between different feature dimensions. In actual acquisition environments, input signals are highly susceptible to transient noise, leading to logical contradictions in the data manifold between isolated extracted spatial energy features and temporal phase-locked features. Existing systems generally lack consistency verification mechanisms and cascaded guidance rules among multi-source features, and cannot dynamically shrink the computational search space of subsequent features using features with high preceding confidence, resulting in weak anti-interference capabilities of the overall signal analysis model and difficulty in guaranteeing the accuracy of feature vector extraction.

[0005] Furthermore, existing systems are relatively rigid in the execution phase of mapping digital features to audio output, lacking high-dimensional dynamic modulation capabilities of the sound field space. Faced with physiological input sequences exhibiting significant left-right asymmetry, current algorithms struggle to achieve refined calculations of stereo spatial phase shift and depth physical parameters based on head-related transformations during the digital audio stream reconstruction phase. Simultaneously, existing systems generally lack update mechanisms based on long-term temporal state feedback. When an external digital audio stream is output and interacts with the target object, the system fails to verify the gradients of newly induced physiological electrical signals, nor can it adaptively update the initial feature extraction guidance rules and consistency judgment logic based on these temporal gradient verification results. This results in the algorithm model being unable to achieve iterative self-learning when handling continuous data interaction and motion guidance tasks, making it difficult to maintain highly robust tracking and accurate audio mapping in dynamic environments. Summary of the Invention

[0006] This invention provides a method for adaptive music-controlled upper limb movement based on a brain-computer interface, the method comprising: Acquire the acoustic envelope sequence of multi-channel physiological electrical signals and digital audio streams; The temporal synchronization characteristics of the physiological electrical signal and the acoustic envelope sequence, the spatial asymmetry characteristics constrained by the temporal synchronization characteristic frequency band, and the event response delay characteristics constrained by the spatial asymmetry characteristic channel weight are calculated sequentially. Based on the feature calculation, the neural coupling weight coefficient and consistency deviation index are used to iteratively correct the features that failed to be verified using a dynamic confidence smoothing factor, and the corrected target feature set is output. Based on the corrected target feature set, spatial azimuth parameters, reverberation wet-dry ratio and low-frequency oscillation modulation parameters are mapped and calculated to generate and output dynamic dual-channel output data sequence to perform acoustic reconstruction. The posterior target feature set after acoustic reconstruction is collected to calculate the temporal physiological state gradient, and the feature extraction parameters and consistency decision threshold are updated in a closed loop based on the temporal physiological state gradient.

[0007] Calculate the temporal synchronization characteristics of physiological electrical signals and acoustic envelope sequences, including: calculating the frequency index of the acoustic envelope sequence and the spatially centered merged physiological electrical signal sequence. With discrete-time index instantaneous phase difference matrix at ; Calculate the cross-phase-locked value vector of the acoustic envelope sequence and the physiological electrical signal sequence on a frequency-by-frequency basis. :

[0008] Extract the maximum peak value from the cross-locked phase-locked value vector as the timing synchronization feature of the current sliding time window. And lock the maximum phase-locked target frequency that produces the maximum phase-locked peak value. :

[0009]

[0010] in, This represents the total number of sampling points within the time window. It is the imaginary unit.

[0011] The calculation process for the spatial asymmetric characteristics constrained by the timing synchronization characteristic frequency band is as follows: using the maximum phase-locked target frequency... Dynamically generate a band filter mask function for the center frequency. :

[0012] Application of frequency band filtering mask function Calculate the frequency band energy characteristic value of the target on the left Energy characteristic value of the target frequency band on the right And calculate spatial asymmetric features. :

[0013] in, This is the preset bandwidth search constant for the system.

[0014] The calculation process for event response delay characteristics constrained by spatial asymmetric feature channel weights is as follows: based on spatial asymmetric features... Dynamically calculate the weight of the left channel Right-side channel weight and the length of the dynamic time search window :

[0015]

[0016]

[0017] Calculate the synthetic target channel power sequence Within the defined forward dynamic time search interval Inside, extract to satisfy Index of first desynchronization response time of constraints Output event response delay characteristics :

[0018] in, The preset weight mapping adjustment factor, Based on the length of the search window over the time period, This is the window expansion factor. This is an index for the event triggering reference time. The preset desynchronization decision threshold is used. and These are the preprocessed left and right physiological electrical signal sequences, respectively.

[0019] The neural coupling weight coefficient and consistency deviation index are calculated based on features, including: for the current... A sliding time window is used to extract the corresponding timing synchronization features. Spatial asymmetry characteristics Event response delay characteristics Calculate the neural coupling weight coefficients of the current window. :

[0020] By using neural coupling weight coefficients to penalize spatial asymmetric features, the consistency deviation index is calculated. :

[0021] in, This is the preset delay decay constant.

[0022] Features that fail validation are iteratively corrected using a dynamic confidence smoothing factor, and the corrected target feature set is output, including: the consistency deviation index. Static consistency decision threshold Perform a comparison; if If the verification fails, the dynamic confidence smoothing factor of the current feature vector is calculated. :

[0023] Using dynamic confidence smoothing factor For the current observed feature vector Compared with the corrected target feature set saved in the previous time window Perform one-dimensional iterative correction and output the corrected target feature set at the current time step. :

[0024] in, Based on the confidence constant, The deviation penalty steepness coefficient.

[0025] Based on the corrected target feature set, the spatial azimuth parameters, reverberation wet-dry ratio, and low-frequency oscillation modulation parameters are calculated, including: extracting the corrected target feature set. Parameters in Mapping to calculate spatial azimuth parameters Reverberation wet-to-dry ratio parameter and physical distance gain parameters :

[0026]

[0027]

[0028] in, The physical angle for maximum sound field deflection. For symbol extraction function, The azimuth smoothing adjustment index, The maximum permissible reverberation ratio limit, is the distance approximation gain constant.

[0029] Calculating the temporal physiological state gradient includes: obtaining the... Posterior target feature set after time correction Calculate the temporal physiological state gradient between two adjacent time windows. :

[0030] in, , , These are the preset weights for timing synchronization features, spatial asymmetry features, and event response delay features, respectively.

[0031] The feature extraction parameters and consistency decision threshold are updated in a closed loop based on the temporal physiological state gradient, including: based on the temporal physiological state gradient. Update bandwidth search width constant Weight mapping adjustment factor :

[0032]

[0033] Based on the temporal physiological state gradient Update dynamic boundary parameters :

[0034] in, For bandwidth update step size constant, To update the step size constant for the weights, This is a truncation function. As a factor of historical amnesia, This is a small relaxation coefficient.

[0035] The present invention also provides an adaptive music-controlled upper limb movement system based on a brain-computer interface, the system comprising: Feature de-module: Acquire the acoustic envelope sequence of multi-channel physiological electrical signals and digital audio streams; sequentially calculate the temporal synchronization features of the physiological electrical signals and acoustic envelope sequences, the spatial asymmetric features constrained by the temporal synchronization feature frequency band, and the event response delay features constrained by the spatial asymmetric feature channel weights; Validation module: Based on the feature calculation, the neural coupling weight coefficient and consistency deviation index are calculated. The features that fail to be validated are iteratively corrected using a dynamic confidence smoothing factor, and the corrected target feature set is output. Audio reconstruction module: Based on the corrected target feature set, it maps and calculates spatial azimuth parameters, reverberation wet-dry ratio and low-frequency oscillation modulation parameters, generates and outputs dynamic two-channel output data sequence to perform acoustic reconstruction; Update module: Collects the posterior target feature set after acoustic reconstruction to calculate the temporal physiological state gradient, and updates the feature extraction parameters and consistency decision threshold in a closed loop based on the temporal physiological state gradient.

[0036] This application provides a brain-computer interface-based adaptive music-controlled upper limb movement method and system. For multimodal physiological electrical signals in audio interaction scenarios, a cascaded feature extraction architecture based on prior state constraints is constructed. The whole-brain rhythm phase-locked state is prioritized to generate a filter mask for the target frequency band, which then guides the directional extraction of spatially lateralized energy features. Finally, the dominant channel weights are assigned based on the amplitude and polarity of the spatial force lateralization, and response delay features are calculated within an adaptive temporal search window. This effectively avoids the background neural noise masking effect caused by traditional full-band and full-channel parallel blind search. By dynamically shrinking the computational search space of subsequent features through prior high-confidence features, the free-state EEG activity of unresponsive audio stimuli can be isolated, greatly improving the confidence and computational accuracy of multi-source feature extraction in complex physical interaction environments.

[0037] To ensure the continuity and stability of the digital audio output at the feedback end over time, this application constructs a consistency verification and adaptive smoothing correction mechanism. During dynamic interaction, a consistency deviation evaluation rule is established. When the system identifies logically contradictory data with rapid local state responses but extreme deviations in spatial energy distribution, an exponential smoothing algorithm based on dynamic confidence intervals is employed. This algorithm uses historically stable data manifolds to perform punitive weighted compensation on the current contradictory features. This avoids abrupt changes in acoustic parameters and audio stuttering at the execution end caused by directly discarding data, and provides robust target feature data for subsequent mapping of multi-dimensional acoustic parameters.

[0038] In the acoustic reconstruction process, the corrected spatial asymmetry features are explicitly mapped to stereo virtual azimuth parameters, actively guiding the reconstructed audio energy to shift to the opposite azimuth in the physical three-dimensional sound field. Cross-neural pathways are used to precisely target and compensate for weak neural conduction regions. Simultaneously, the system converts temporal response hysteresis features into the wet / dry ratio and physical distance depth gain of the digital reverberation matrix. When the target object's response is sluggish, environmental reflected sound is stripped to enhance the transient penetration of direct sound. Furthermore, this application introduces an adaptive evolution mechanism for the underlying data processing logic based on temporal feedback. After executing a single digital audio stream reconstruction output, the system collects physiological electrical signals in the next time window to calculate the physiological state gradient before and after the reconstructed sound field intervention, evaluating the effectiveness of the previous physical sound field control. Based on this temporal gradient verification result, the system reverse-corrects the frequency band search bandwidth, channel weight mapping steepness, and consistency verification tolerance boundaries in the initial extraction architecture. This ensures that the system maintains optimal noise filtering capabilities and extremely high audio dynamic control robustness when dealing with continuous and complex motion intent recognition and acoustic environment feedback tasks. Attached Figure Description

[0039] Figure 1 This is a flowchart of the adaptive music-controlled upper limb movement based on a brain-computer interface according to the present invention. Figure 2 This invention provides a diagram of the synchronous acquisition and recording of multi-channel electroencephalogram signals and digital audio envelopes. Figure 3 This is a diagram illustrating the asymmetric analysis of left and right lateral EEG energy under target frequency band constraints in this invention. Figure 4 This is a diagram showing the consistency deviation verification and abnormal feature smoothing correction in this invention. Detailed Implementation

[0040] This embodiment provides an adaptive music-controlled upper limb movement system based on a brain-computer interface. The system is applied in dynamic interactive scenarios using digital audio streams based on multi-channel physiological electrical signal feedback. Specifically, it is used for recognizing the upper limb movement intentions of a target object under the guidance of audio rhythms, determining the execution of actions, and adaptively reconstructing the closed-loop acoustic parameters of external digital audio sequences.

[0041] The system, in terms of hardware topology, includes an ear-worn EEG audio interaction terminal worn on the head of the target object, and a data processing terminal that establishes a two-way data communication link with the ear-worn EEG audio interaction terminal.

[0042] The ear-worn EEG audio interaction terminal serves as the system's signal acquisition and feedback execution node. Preferably, the ear-worn EEG audio interaction terminal adopts a headband or ear-hook structure, and its internal structure integrates a multi-channel physiological electrode group, a dual-channel audio output component, a wireless communication module, and an independent power supply module.

[0043] Furthermore, the multi-channel physiological electrode assembly employs dry, non-invasive acquisition electrodes that do not require the application of conductive liquid. Its distribution structure is configured to include a frontal reference electrode array and bilateral postauricular acquisition electrode arrays. The frontal reference electrode array is embedded within the frontal extension support of the structural body and adheres to the skin surface of the frontal lobe when worn by the target subject, used to acquire reference baseline signals and electrooculography artifact reference sequences. The bilateral postauricular acquisition electrode arrays are symmetrically distributed on the inner contact surfaces of the left and right support portions of the structural body, respectively. When worn by the target subject, the left and right acquisition electrodes are respectively attached to the left and right mastoid regions of the target subject, used to simultaneously acquire independent physiological electrical signal sequences originating from the left and right hemispheres of the brain, providing a hardware signal source for computationally asymmetric features.

[0044] The dual-channel audio output component includes independently driven high-fidelity miniature speakers for the left and right channels. The digital audio decoding chip of the speakers supports dynamic gain adjustment and phase delay control of multi-track digital audio signals to adapt to the three-dimensional stereo spatial azimuth offset output reconstructed based on the head-related transform function.

[0045] The data processing terminal can be a mobile intelligent device or an edge computing controller with digital signal processing capabilities, and it integrates a central processing unit, a cache memory, and an interactive display interface.

[0046] Based on the hardware architecture, this embodiment elaborates on the initial steps in the system data processing flow, namely step S1: initial multi-channel signal acquisition and preprocessing. This step aims to acquire multimodal raw data from heterogeneous sensor devices, and eliminate scene-specific physiological and physical noise through spatial rereference, feature frequency band extraction, and envelope detection algorithms, ultimately constructing data that is strictly aligned in the time dimension.

[0047] Specifically, step S1 includes the following processing procedures: S1, acquire multi-channel physiological electrical signals.

[0048] When the target subject performs upper limb motor imagery or attempts at movement, facial muscle tension and involuntary blinking are very likely to occur, thus introducing high-amplitude electrooculography (EOG) artifacts and electromyography (EMG) noise into the EEG acquisition terminal. Because the frontal reference electrode array is physically close to the eye interference source, the signal it acquires contains a very high proportion of common-mode environmental noise and artifact components.

[0049] Multi-channel signal acquisition module in discrete time index At this location, the raw left-sided physiological electrical signal sequence output from the left-sided postauricular electrode array is acquired in real time. The original right-sided physiological electrical signal sequence output from the right postauricular electrode array. and the original frontal reference signal sequence output by the frontal reference electrode array. To suppress common-mode interference and highlight subtle brain neural oscillations, the system employs a differential spatial rereference algorithm to calculate the left-side physiological electrical signal sequence after differential rereference. With right-side physiological electrical signal sequence :

[0050]

[0051] in, Represents the index of discrete time series sampling points.

[0052] S2, target frequency band filtering preprocessing of physiological electrical signals.

[0053] The determination of upper limb movement execution is highly dependent on the specific neural rhythmic response of the target subject's sensorimotor cortex, i.e., sensorimotor rhythm. Preserving the full-band signal would introduce a large amount of background EEG noise unrelated to movement. Therefore, a finite impulse response filter with linear phase characteristics is selected to construct an EEG bandpass filter function. Its passband range is strictly set to cover the core frequency band of sensorimotor rhythms. The rereferenced signal is input into this filter to obtain the pre-filtered left-side electrophysiological signal sequence. and right-side physiological electrical signal sequence :

[0054]

[0055] in, This represents the bandpass filtering operator.

[0056] S3, Acoustic envelope extraction of digital audio streams.

[0057] While outputting digital audio, the system simultaneously extracts the original digital audio stream signal sequence from the front-end audio bus of the dual-channel audio output component. In dynamic interactive scenarios of digital audio streams, the core factor that induces the brain to produce a phase-locked follow-along response is not the high-frequency carrier details of the audio, but the physical fluctuations of the audio's temporal rhythm, such as the drum envelope and the energy mutations of voice commands.

[0058] Therefore, the Hilbert transform operator is used. Extract the analytic signal of the audio and calculate its amplitude, then apply it to the audio envelope low-pass filter function. Smooth the ultra-high frequency ripples and extract the acoustic envelope sequence. :

[0059] in, Defined as the process of calculating the Discrete Hilbert Transform after discretization of a continuous-time signal. This indicates a low-pass filter operator whose cutoff frequency is adapted to the low-frequency oscillation response range of EEG.

[0060] S4, time-series resampling and constructing a time window buffer matrix.

[0061] The raw EEG sampling rate of the multi-channel physiological electrical signals was set to The original audio sampling rate of the original digital audio stream signal is set to Because the hardware for acquiring the two physical signals has differences in clock source and digital-to-analog conversion rate (usually...), Much larger To meet the stringent timing alignment requirements of subsequent consistency analysis, the timestamp synchronization subunit within the multi-channel signal acquisition module employs a multi-phase filtering anti-aliasing resampling algorithm. , as well as Unified downsampling or interpolation resampling to the target synchronous sampling rate After resampling, the system extracts a data segment containing a single complete "audio-physical stimulus-brain neural response" interaction cycle, setting the total number of sampling points within the sliding time window to [number missing]. The system uses the current moment as the endpoint and traces back along the historical timeline. Data points are used to construct a time window buffer matrix. Its mathematical structure is represented as:

[0062] in, The constructed time window buffer matrix precisely stores the left physiological electrical signal sequence, the right physiological electrical signal sequence, and the acoustic envelope sequence, which are strictly aligned under the same clock domain, in its first, second, and third rows, respectively. This matrix serves as the standard data input source for subsequent cascaded feature extraction modules.

[0063] This embodiment uses the constructed time window buffer matrix Based on the data, this paper elaborates on the S2 process of the system's data processing flow: the cascaded feature extraction process based on the preceding state constraints.

[0064] In complex digital audio interactions, multi-channel physiological electrical signals from the target object's cerebral cortex are coupled in the time, frequency, and spatial domains. Using independent extraction methods makes it difficult to eliminate background neural activity noise that does not respond to audio rhythms. This embodiment introduces prior state constraint logic, allowing the extracted prior features to directly determine the search boundary of subsequent features, thereby constructing a robust feature extraction method.

[0065] Specifically, S21 calculates the timing synchronization characteristics. .

[0066] When given a digital audio sequence input, the first prerequisite for determining whether the underlying neural activity of the target object has generated a motor intention is to verify whether the auditory and motor cortices of the brain have achieved stable frequency locking or phase following in response to external audio stimuli on a temporal scale. Only lateralization of neural electrical activity based on this synchronization can be determined as a valid intention to execute guided upper limb movements.

[0067] To extract this physical feature, the cascaded feature extraction module first uses a time window buffer matrix. Middle-aligned left-side physiological electrophysiological signal sequence Right-side physiological electrical signal sequence and acoustic envelope sequence To assess the overall phase response of the whole brain to audio rhythms, a spatially centered combined physiological electrical signal sequence was calculated. :

[0068] To extract the instantaneous phase information of non-stationary signals, a continuous wavelet transform algorithm with optimal time-frequency resolution is employed. The complex Morlet wavelet is selected as the mother function. The spatial center was combined with the physiological electrical signal sequence. Harmony and acoustic envelope sequence Perform time-scale transformation and calculate the frequency index. With discrete-time index The complex wavelet coefficient matrices under the joint mapping are denoted as follows: and .

[0069] Based on the real part of the complex wavelet coefficient matrix With the imaginary part The system calculates discrete coordinate points respectively. Instantaneous phase of neural oscillation With audio acoustic instantaneous phase And solve its instantaneous phase difference matrix. :

[0070] To determine the two heterogeneous physical signals within a given time window The system calculates the cross-phase-locked value vector of the acoustic envelope sequence and the physiological electrical signal sequence point by point within the preset effective low-frequency response band to determine the phase-locked tightness. :

[0071] in, It is the imaginary unit. The range of values ​​is The closer the value is to 1, the more constant the instantaneous phase difference between the two at a specific frequency remains throughout the entire time window.

[0072] Extract the maximum peak value from the cross-locked phase-locked value vector as the timing synchronization feature of the current sliding time window. And simultaneously lock onto the maximum phase-locked target frequency that generates the maximum phase-locked peak value. :

[0073]

[0074] S22, Computational space asymmetric features .

[0075] After extracting the time-series synchronization features Subsequently, it was determined that the target's nervous system had become coupled with external physical rhythms. When the target attempted to drive the left or right upper limbs in a motor response, the sensorimotor cortex of the left and right hemispheres of the brain produced an asymmetrical distribution of energy desynchronization within a characteristic frequency band.

[0076] If channel energy is blindly accumulated across the entire frequency band, a large amount of background neural noise in a non-phase-locked state will severely overwhelm this spatial asymmetry. Therefore, the information obtained in this application via S21... As a preceding state constraint parameter, it guides the frequency domain search space of S22.

[0077] Cascaded feature extraction module A frequency band filter mask function with Gaussian smooth attenuation characteristics is dynamically generated for the center frequency. :

[0078] in, The preset bandwidth search constant for the system is used to control the attenuation gradient of the mask on the frequency axis.

[0079] Similarly, using continuous wavelet transform, the physiological electrical signal sequence on the left side of the original input matrix is ​​calculated independently. and right-side physiological electrical signal sequence The complex wavelet coefficient matrices are denoted as follows: and Applying a frequency band filtering mask function As a weighting factor, integration and summation are performed in the two-dimensional plane of frequency and time domains to calculate the left target frequency band energy characteristic value within an extremely narrow frequency band where the target object exhibits phase-locked following of the audio rhythm. Energy characteristic value of the target frequency band on the right :

[0080]

[0081] Based on the high signal-to-noise ratio energy value extracted after the above masking mapping, the normalized spatial asymmetric features are calculated. :

[0082] This space asymmetry feature The range of values ​​is Its magnitude indicates the lateralization depth of electrical activity in the left and right cerebral hemispheres, while its polarity directly indicates the lateral direction (left or right) of the physical force exerted by the target object in driving upper limb movement under the audio rhythm. Through temporal and spatial constraints, the system eliminates the interference of free-state background noise in the scene, achieving the extraction of high-confidence multi-source feature sequences.

[0083] Furthermore, when the system issues audio sequences with significant rhythmic fluctuations or verbal commands, the target subject's brain's execution of upper limb motor imagery induces a decrease in energy response in specific frequency bands of the sensorimotor cortex, a phenomenon known as event-related desynchronization (ERD). The time difference between the occurrence of physical stimuli and the generation of significant ERD in the underlying nervous system is a temporal indicator for assessing the efficiency of neural transmission and motor intention execution.

[0084] If the search is performed independently across all channels to extract this time difference, it is highly likely that spontaneous background EEG fluctuations in the ipsilateral non-motor cortex will be captured, leading to serious false-positive logical errors in response time determination. This application introduces spatial asymmetry features extracted in the preceding sequence. The polarity and amplitude are used as prior guiding parameters to dynamically allocate the channel weight matrix and time search window boundaries, thereby achieving high-confidence directional time feature extraction.

[0085] S23 Calculate event response delay characteristics : Buffer matrix from the constructed time window In this process, the timing of the physical stimulus occurrence in digital audio is extracted. This is based on acoustic envelope sequences. A first-order backward difference operator is used. Calculate the instantaneous gradient sequence of the envelope energy, and search for the discrete time point corresponding to the gradient peak within the current time window as the event trigger reference time index. :

[0086]

[0087] This time index This corresponds to the exact physical moment when a drum beat accent or clear guiding instruction is issued in a digital audio sequence.

[0088] Based on spatial asymmetry Dynamic channel weights and dynamic temporal search windows are constructed. Unilateral upper limb motor intention is controlled by the contralateral cerebral hemisphere. When the target object attempts to drive the right upper limb, the left sensorimotor cortex is in an active inhibitory state, producing the ERD phenomenon, resulting in a decrease in the target frequency band energy characteristic value on the left side. Lower than the target frequency band energy characteristic value on the right Thus, we can conclude Conversely, when driving the left upper limb, .

[0089] To highlight the dominant neural pathway signals that actually generate motor intention, according to Polarity dynamic calculation of left channel weight Weights of the right channel :

[0090]

[0091] in, The preset weight mapping adjustment factor for the system ( ).when At that time, the left hemisphere is the dominant side for mapping output. This ensures that the search weights always adaptively tilt towards the actual activated lateral cortex where ERD occurs.

[0092] Meanwhile, the absolute amplitude of spatial asymmetry features This indicates the intensity of the target object's invocation of the lateral hemisphere motion network. A weaker lateralization ( The closer the value is to 0, the slower the nervous system response, requiring a longer time interval to capture the delayed ERD phenomenon. The system calculates the dynamic time search window length based on this. :

[0093] in, Based on the length of the search window over the time period, The preset window expansion factor ( ).

[0094] After completing the allocation of guidance parameters, the system is based on the preprocessed left-sided physiological electrical signal sequence. and right-side physiological electrical signal sequence Calculate the instantaneous power and synthesize a one-dimensional synthetic target channel power sequence using the assigned weights. :

[0095] To pinpoint the exact time of occurrence of the ERD phenomenon, refer to the event trigger reference time index. Within the previous static, unstimulated interval, the truncated length is... The baseline reference window is used to calculate the baseline power reference mean of the synthesized target channel power sequence within this window. :

[0096] Set desynchronization decision threshold ,in The preset desynchronization drop ratio constant is set within a range of [range]. Within a strictly defined forward dynamic time search interval Within, sequentially detect and extract those that satisfy... The earliest time index of the constraint is denoted as the first desynchronization response time index. .

[0097] By calculating the time difference between this response time and the time of occurrence of the physical stimulus, the final event response delay feature is output. :

[0098] If within the complete forward dynamic time search interval None of them exceeded the synchronization decision threshold. Then force assignment This represents the loss of motion responsiveness of the target object during the current interaction cycle.

[0099] Understandably, in the current field of digital signal processing, feature extraction for multimodal physiological electrical signals generally adopts an independent parallel operation mode, that is, solving for frequency domain energy, time domain delay, and phase features independently. However, the reception, processing, and motor intention response of the target subject's cerebral cortex to external audio rhythm signals are by no means isolated data fragments, but follow a strict time mechanism from sensory cortex locking to motor network lateralization and then to local energy synchronization triggering.

[0100] Therefore, the cascaded extraction logic designed in this application ( This constitutes data filtering and state constraints, where timing synchronization characteristics ( As the first cascaded threshold globally, this verifies the effectiveness of the input source. In dynamic audio interaction, if the underlying neural oscillations of the target object do not achieve stable phase-locking with the acoustic envelope of the external audio, i.e., if effective perceptual tracking is not established, then any spatial energy distribution extracted across the entire frequency band is considered disordered spontaneous background noise. and its corresponding target phase-locked frequency Dynamically generated frequency band filter masks eliminate redundant frequency band signals not modulated by physical audio stimuli, ensuring that subsequent spatial feature calculations strictly converge within the effective data domain that actually had an impact. Secondly, spatial asymmetry features ( )exist Extracted under extremely narrow bandwidth constraints, it serves as a spatial guide for connectivity states and response times. Due to limitations of portable dual-channel acquisition hardware, the system is susceptible to spontaneous fluctuations in ipsilateral brain regions not involved in motor networks, resulting in false energy drops, i.e., the false-positive ERD phenomenon. The lateralization depth and physical force direction dominated by upper limb movement intention were quantified and directly converted into channel weight coefficients for subsequent time feature extraction. The directional search based on spatial asymmetry avoids the feature space masking effect caused by traditional full-channel blind search. Event response delay features ( The extraction directly received Prior guidance on amplitude magnitude. In real-world interactions, due to the dynamic fluctuations in the cognitive state of the target individual, the time it takes for the cerebral cortex to generate a desynchronized response exhibits significant non-stationarity. This system will... The absolute value of the lateralization is mapped to the boundary of the dynamic temporal search window: the weaker the lateralization, the more sluggish the neural response, and the system adaptively widens the search window to capture the lagging signal; the extremely strong lateralization narrows the search window to save computational power. The originally resource-intensive global temporal traversal is transformed into a targeted local temporal search for a specific high-confidence dominant channel. This not only maximizes the elimination of interference from detached environment artifacts and significantly reduces redundant computational overhead, but also ensures that the extracted multidimensional feature set possesses extremely high confidence in the data manifold.

[0101] Based on the data from the cascaded feature extraction, this embodiment details S3 in the system data processing flow: consistency verification and smoothing correction of multidimensional features.

[0102] In normal data processing, time synchronization characteristics Spatial asymmetry characteristics Event response delay characteristics There are constraints between them. Specifically, when the digital audio stream is continuously output, if the system extracts an extremely high cross-locked phase value (i.e., ... Approaching 1 indicates that the brain has an excellent ability to follow musical rhythms, and it extracts an extremely short response delay time (i.e., If the value is extremely small (indicating smooth execution of the motor neural pathway), then under long-term continuous motor response, the energy scheduling of the sensorimotor cortex in both hemispheres should be in a stable and coordinated state, and the energy distribution of the left and right channels should tend to be in physical equilibrium (i.e., It should approach 0). If the system extracts Extremely high Extremely short, but However, the presence of extreme abnormal deviations (absolute values ​​approaching 1) indicates a causal contradiction in the current data. This contradiction clearly suggests that the spatial features are highly likely contaminated by transient physical loosening of a unilateral electrode or a unilateral localized electromyographic burst, such as a sudden chewing by the target object.

[0103] S31, Multidimensional feature space mapping and consistency logic verification.

[0104] The current number The three-dimensional feature vector extracted within each sliding time window is mapped to the currently observed feature vector. :

[0105] Among them, superscript This represents the matrix transpose operation.

[0106] Based on the above causal constraints, a consistency deviation index calculation function is constructed.

[0107] Calculate the neural coupling weight coefficients of the current window. This coefficient measures whether the brain is currently in a state of high-frequency lock-on and rapid response to audio rhythms.

[0108] in, The system's preset delay decay constant ( ), This is the length of the dynamic time search window. When... Approaching 1 and When it approaches 0, the neural coupling weight coefficient Reaching the maximum value; however, when the target object becomes distracted, resulting in an unlocked frequency ( Extremely low) or extremely slow response ( Approaching the upper limit )hour, It will decrease significantly.

[0109] The consistency deviation index is calculated by using the neural coupling weight coefficient to perform a nonlinear penalty mapping on spatial asymmetric features. :

[0110] When the system is in a high-lock, fast-response state ( When the spatial lateralization is maximized, this application has a strict tolerance for spatial lateralization. Due to noise, there is a drastic shift, for example This will lead to The output shows a large deviation value; conversely, if the system itself is in an unlocked, ionized state... At this point, the brain is in a state of disordered spontaneous thought activity, and an energy imbalance between the left and right hemispheres is a normal physiological phenomenon. Here, at this very small... Automatic suppression The magnification makes It remains at a low level.

[0111] The calculated consistency deviation index With respect to the system's preset static consistency decision threshold Perform a comparison. If... If the consistency verification of the currently observed feature vector is passed, then it is determined that the current observation feature vector has passed; if If a physical contradiction occurs in the extraction logic, the verification fails, and the feature correction and compensation procedure is triggered.

[0112] S32, adaptive smoothing correction compensation based on dynamic confidence interval.

[0113] In continuous digital audio stream mapping, directly discarding contradictory data can cause abrupt changes and stuttering in the acoustic parameters of the audio output. Therefore, this application constructs an adaptive exponential smoothing equation based on dynamic confidence factors, using historical stable manifolds to perform weighted correction on the current contradictory data.

[0114] Calculate the dynamic confidence smoothing factor of the current feature vector. :

[0115] in, The system's preset basic confidence constant has a value range of [value range missing]. For example, setting it to 0.9 indicates a high degree of trust in the current observation when there is no contradiction. The preset deviation penalty steepness coefficient ( ).

[0116] When verification fails, deviations exceeding the threshold are reduced by a negative exponent to decrease the current confidence level. The more extreme the deviation, the lower the confidence level. The closer it gets to 0.

[0117] Using the acquired dynamic confidence smoothing factor, the current observed feature vector is... Compared with the corrected target feature set saved in the previous time window Perform one-dimensional iterative correction and output the corrected target feature set at the current time step. :

[0118] Next, in this embodiment, the corrected target feature set is output. Based on the data, this paper details S4 in the system data processing flow: stereo stream reconstruction calculation based on the target feature set.

[0119] S41, calculate spatial acoustic phase shift and azimuth parameters.

[0120] The human primary auditory cortex has deep synaptic connections with the ipsilateral sensorimotor network, and the binaural auditory conduction pathway exhibits significant contralateral contralateral control. When the motor network on one side of the target object is damaged or inactive, it manifests as a significant bias in spatial asymmetry. The system needs to actively shift the energy of digital audio to the contralateral direction in a three-dimensional virtual sound field, utilizing the contralateral neural pathway to provide targeted compensatory stimulation to the weaker brain region.

[0121] The stereo stream reconstruction module extracts the corrected spatial asymmetry features at the current moment. A nonlinear target azimuth mapping function is constructed to calculate the spatial azimuth parameters of the current digital audio stream. :

[0122] in, The preset maximum sound field deflection physical angle is preferably set to [value]. Indicates the extreme left or extreme right direction; For symbol extraction functions; The preset azimuth smoothing adjustment index, This is used to avoid sudden changes in acoustic phase.

[0123] when When this occurs, it indicates that the energy in the left sensorimotor cortex is weaker, meaning the target object may have a transduction disorder of motor intentions in the right limbs. The calculated [result] at this time... This involves controlling the sound source to shift towards the right ear in three-dimensional space. This acoustic spatial traction forces the target audience to focus their auditory attention to the right, thereby spontaneously activating the cortical network in the left hemisphere of the brain, forming a spatial sound-brain interaction.

[0124] S42, calculate the reverberation wet-to-dry ratio.

[0125] Corrected event response delay characteristics It characterizes the excitation level and conduction efficiency of the target's nervous system. When the response is severely delayed, the target is in a state of cognitive dissipation or deep fatigue, in which case it is necessary to enhance the transient impact of digital audio; while when the response is rapid, a sense of spatial envelopment should be provided to consolidate a stable emotion.

[0126] Based on this, the wet / dry ratio parameter of the digital reverberation matrix is ​​calculated. That is, the proportion of reverberant reflected sound signal to the total audio signal energy:

[0127] in, The preset maximum allowable reverberation ratio limit has a value range of [value range missing]. , The length of the dynamic time search window.

[0128] Simultaneously, to further enhance the stimulation, the physical distance gain parameter for compensating for the direct sound is calculated. :

[0129] in, This is the preset distance approximation gain constant.

[0130] when Approaching When the response is extremely sluggish, it leads to an excessively low wet-to-reverberation ratio. Approaching 0, and the distance gain Reaching the maximum means stripping away all virtual spatial reflections, making the sound extremely singular and amplifying the absolute loudness, resulting in a strong sense of closeness and penetration in the drumbeats or guiding commands, thereby forcibly stimulating the nerve excitation of the target audience.

[0131] S43, calculate the low-frequency oscillation modulation parameters of the target frequency-locked resonance.

[0132] To solidify the brain's established rhythmic following state, the target object's spontaneous neurophysiological frequencies are re-injected into the acoustic system. This is based on the modified temporal synchronization characteristics. and the extracted optimal target phase-locked frequency Constructing a digital low-frequency oscillator carrier :

[0133] in, The preset maximum amplitude modulation depth coefficient. The sampling time for continuous digital audio. This is the initial phase.

[0134] The tighter the brain's phase lock on rhythms ( The larger the amplitude of the audio stream, the more it will generate a frequency based on the target object's own brainwaves. It is a periodic, strong, in-sync vibrato (resonance reward); if the frequency lock is lost, the vibrato depth automatically decays to zero.

[0135] S44, dual-channel reconstructed output.

[0136] After completing the above physical acoustic parameter mapping, extract the target reconstructed audio segment to be output within the current interaction cycle. It calls the system's built-in HRTF database. Based on the currently dynamically calculated azimuth parameters... The system extracts the corresponding spatial impulse response filter function for the left ear. With right ear spatial impulse response filter function .

[0137] Direct sound components are generated by finite-length convolution operations, preserving accurate binaural time difference (ITD) and binaural sound level difference (ILD). and :

[0138]

[0139] in, This is the index for the discrete delay tap of the filter. The length of the impact response function.

[0140] In parallel, the target audio segment is reconstructed. Generate reverberation components for simulating diffuse sound fields. and By integrating the aforementioned physical distance depth, reverberation wet / dry ratio, and low-frequency oscillation carrier, the final stereo stream synthesis equation is executed to generate a dynamic two-channel output data sequence. and :

[0141]

[0142] After generating the above reconstruction sequence, the system will and Encapsulated as digital audio data packets, the data is sent to the decoding chip of the ear-worn EEG audio interaction terminal to drive the left and right channel high-fidelity micro-speakers to perform acoustic reproduction.

[0143] Next, after the stereo stream reconstruction calculation and acoustic output, this embodiment describes S5 in the system data processing flow: verification and extraction logic update based on temporal feedback. After the system outputs the reconstructed stereo stream, i.e., after generating spatial phase shift, wet / dry ratio changes, and resonance modulation, the target object's nervous system will generate subsequent physiological potential responses upon receiving this targeted acoustic stimulus. This application evaluates the effectiveness of the previous audio intervention by acquiring the physiological electrical signal at the next moment after playing the reconstructed audio, and adaptively adjusts the feature extraction boundary parameters in S2 and the consistency decision threshold in S3.

[0144] S51, collect posterior data and calculate temporal physiological state gradients.

[0145] In the dynamic dual-channel output data sequence of S4 and After physical playback via the speaker, the multi-channel signal acquisition module slides a time window to capture this moment (denoted as the [number]th moment). The system retrieves the physiological electrophysiological signal sequence and acoustic envelope sequence (using a sliding window). The system sequentially calls operations S2 and S3 to obtain the sequence. Posterior target feature set after time correction .

[0146] To evaluate whether the reconstruction of the digital audio stream in physical space positively guides the motion intention of the target object, the temporal state gradient between two adjacent time windows is calculated. :

[0147] in, , , These are the system's preset weights for timing synchronization, spatial asymmetry, and event response delay, respectively, and all are greater than 0.

[0148] If the spatial acoustic traction and direct acoustic impact in the previous moment successfully activated the weak lateral brain region of the target and improved attention, then the posterior data should show: an increase in rhythmic phase lock ( The energy of the left and right hemispheres tends to be in a coordinated balance, that is, lateralization is reduced. ), and reduced response latency ( Any of the above positive feedbacks will make It generates a positive increment. Conversely, if This indicates that the extraction rules set by the current system are not performing well under the current complex free-state noise, and have failed to effectively capture or guide neural responses.

[0149] S52 updates cascaded feature extraction parameters based on temporal state gradient.

[0150] Based on the calculated temporal physiological state gradient The bandwidth search constant for generating the band filter mask. And the weight mapping adjustment factor for allocating dynamic channel weights. Perform discrete time-series step updates.

[0151] For the bandwidth search constant Its update equation is:

[0152] in, The preset bandwidth update step size constant; The cutoff function ensures that the constant is constrained to a reasonable physical frequency boundary. Inside.

[0153] when This indicates that the interaction is going well and the system is reducing [the impact of the system's performance]. Because the target object's phase-locking to audio is extremely stable when the state is good, the system can narrow the frequency mask bandwidth to filter out background EEG noise in adjacent frequency bands with stricter conditions; conversely, if the state deteriorates, the system adaptively widens the bandwidth. This is to prevent over-filtering from causing the loss of effective signals.

[0154] For weight mapping adjustment factor Its update equation is:

[0155] in, The preset weight update step size constant.

[0156] when When the system increases Because the better the state, the more spatial asymmetric features are extracted. The higher the confidence level, the more it should be increased. This allows for a more decisive shift of the search weight of the time window to the activated side channel, thereby further improving the extraction efficiency of time features.

[0157] S53, update the consistency decision threshold based on state performance.

[0158] In addition to feature extraction parameters, the static consistency decision threshold also needs to be adaptively adjusted based on the stability of the interaction. Upgrade it to dynamic boundary parameters Its update rules are as follows: when At this point, it indicates that the system is not only effectively guiding neural activity, but also that the input data source is relatively pure. The system then employs a forgetting factor smoothing model to actively lower the threshold to approximate the consistency deviation index actually measured at the current moment. This tightens consistency verification:

[0159] in, This is a pre-set historical forgetting factor.

[0160] when This indicates that the target object may be experiencing a period of sharp cognitive fluctuation or being subjected to continuous and significant physical disturbances. Maintaining a strict threshold would cause the system to frequently trigger corrections, resulting in data distortion. In this case, the system performs threshold relaxation compensation:

[0161] in, The preset small relaxation coefficient ( This allows the system to have a higher tolerance for transient deviations in harsh interactive environments.

[0162] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0163] In this specification, the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the descriptions of the embodiments described later are relatively simple, and relevant parts can be referred to the descriptions of the foregoing embodiments.

[0164] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for regulating upper limb movement based on brain-computer interface adaptive music, characterized in that, The method includes: Acquire the acoustic envelope sequence of multi-channel physiological electrical signals and digital audio streams; The temporal synchronization characteristics of the physiological electrical signal and the acoustic envelope sequence, the spatial asymmetric characteristics constrained by the frequency band of the temporal synchronization characteristics, and the event response delay characteristics constrained by the channel weights of the spatial asymmetric characteristics are calculated sequentially. Among them, a frequency band filtering mask function is dynamically generated based on the phase-locked target frequency corresponding to the temporal synchronization characteristics to extract the spatial asymmetric characteristics, and the event response delay characteristics are extracted based on the dynamically allocated dual-side channel weights and time search window length according to the spatial asymmetric characteristics. Based on the feature calculation, the neural coupling weight coefficient and consistency deviation index are used to iteratively correct the features that failed to be verified using a dynamic confidence smoothing factor, and the corrected target feature set is output. Based on the corrected target feature set, spatial azimuth parameters, reverberation wet-dry ratio and low-frequency oscillation modulation parameters are mapped and calculated to generate and output dynamic dual-channel output data sequence to perform acoustic reconstruction. The posterior target feature set after acoustic reconstruction is collected to calculate the temporal physiological state gradient, and the feature extraction parameters and consistency decision threshold are updated in a closed loop based on the temporal physiological state gradient.

2. The method according to claim 1, characterized in that, Calculate the temporal synchronization characteristics of physiological electrical signals and acoustic envelope sequences, including: calculating the frequency index of the acoustic envelope sequence and the spatially centered merged physiological electrical signal sequence. With discrete-time index instantaneous phase difference matrix at ; Calculate the cross-phase-locked value vector of the acoustic envelope sequence and the physiological electrical signal sequence on a frequency-by-frequency basis. : ； Extract the maximum peak value from the cross-locked phase-locked value vector as the timing synchronization feature of the current sliding time window. And lock the maximum phase-locked target frequency that produces the maximum phase-locked peak value. : ； ； in, This represents the total number of sampling points within the time window. It is the imaginary unit.

3. The method according to claim 2, characterized in that, The calculation process for the spatial asymmetric characteristics constrained by the timing synchronization characteristic frequency band is as follows: using the maximum phase-locked target frequency... Dynamically generate a band filter mask function for the center frequency. : ； Application of frequency band filtering mask function Calculate the frequency band energy characteristic value of the target on the left energy characteristic value of the target frequency band on the right And calculate spatial asymmetric features : ； in, This is the preset bandwidth search constant for the system.

4. The method according to claim 3, characterized in that, The calculation process for event response delay characteristics constrained by spatial asymmetric feature channel weights is as follows: based on spatial asymmetric features... Dynamically calculate the weight of the left channel Right-side channel weight and the length of the dynamic time search window : ； ； ； Calculate the synthetic target channel power sequence Within the defined forward dynamic time search interval Inside, extract to satisfy Index of first desynchronization response time of constraints Output event response delay characteristics : ； in, The preset weight mapping adjustment factor, Based on the length of the search window over the time period, This is the window expansion factor. This is an index for the reference time when the event was triggered. The preset desynchronization decision threshold is used. and These are the preprocessed left and right physiological electrical signal sequences, respectively.

5. The method according to claim 4, characterized in that, The neural coupling weight coefficient and consistency deviation index are calculated based on features, including: for the current... A sliding time window is used to extract the corresponding timing synchronization features. Spatial asymmetry characteristics Event response delay characteristics Calculate the neural coupling weight coefficients of the current window. : ； By using neural coupling weight coefficients to penalize spatial asymmetric features, the consistency deviation index is calculated. : ； in, This is the preset delay decay constant.

6. The method according to claim 5, characterized in that, Features that fail validation are iteratively corrected using a dynamic confidence smoothing factor, and the corrected target feature set is output, including: the consistency deviation index. Static consistency decision threshold Perform a comparison; if If the verification fails, the dynamic confidence smoothing factor of the current feature vector is calculated. : ； Using dynamic confidence smoothing factor For the current observed feature vector Compared with the corrected target feature set saved in the previous time window Perform one-dimensional iterative correction and output the corrected target feature set at the current time step. : ； in, Based on the confidence constant, The deviation penalty steepness coefficient.

7. The method according to claim 6, characterized in that, Based on the corrected target feature set, the spatial azimuth parameters, reverberation wet-dry ratio, and low-frequency oscillation modulation parameters are calculated, including: extracting the corrected target feature set. Parameters in Mapping calculation of spatial azimuth parameters Reverberation wet-to-dry ratio parameter and physical distance gain parameters : ； ； ； in, The physical angle for maximum sound field deflection. For symbol extraction function, The azimuth smoothing adjustment index, The maximum permissible reverberation ratio limit, is the distance approximation gain constant.

8. The method according to claim 7, characterized in that, Calculating the temporal physiological state gradient includes: obtaining the... Posterior target feature set after time correction Calculate the temporal physiological state gradient between two adjacent time windows. : ； in, , , These are the preset weights for timing synchronization features, spatial asymmetry features, and event response delay features, respectively.

9. The method according to claim 8, characterized in that, The feature extraction parameters and consistency decision threshold are updated in a closed loop based on the temporal physiological state gradient, including: based on the temporal physiological state gradient. Update bandwidth search width constant Weight mapping adjustment factor : ； ； Based on the temporal physiological state gradient Update dynamic boundary parameters : ； in, For bandwidth update step size constant, To update the step size constant for the weights, This is a truncation function. As a factor of historical amnesia, This is a small relaxation coefficient.

10. An adaptive music-controlled upper limb movement system based on a brain-computer interface, characterized in that, The system includes: Feature de-module: Acquire the acoustic envelope sequence of multi-channel physiological electrical signals and digital audio streams; sequentially calculate the temporal synchronization features of the physiological electrical signals and acoustic envelope sequences, the spatial asymmetric features constrained by the temporal synchronization feature frequency band, and the event response delay features constrained by the spatial asymmetric feature channel weights; Validation module: Based on the feature calculation, the neural coupling weight coefficient and consistency deviation index are calculated. The features that fail to be validated are iteratively corrected using a dynamic confidence smoothing factor, and the corrected target feature set is output. Audio reconstruction module: Based on the corrected target feature set, it maps and calculates spatial azimuth parameters, reverberation wet-dry ratio and low-frequency oscillation modulation parameters, generates and outputs dynamic two-channel output data sequence to perform acoustic reconstruction; Update module: Collects the posterior target feature set after acoustic reconstruction to calculate the temporal physiological state gradient, and updates the feature extraction parameters and consistency decision threshold in a closed loop based on the temporal physiological state gradient.

Images