A method and device for processing accelerated rehabilitation data based on electrophysiological data

Abstract

The present invention discloses a method and device for processing accelerated rehabilitation data based on electrophysiological data, which relates to the field of medical rehabilitation technology. The method comprises the following steps: calculating the full-modal phase lock value based on a noise reduction matrix, generating a fused feature vector through a gated recurrent network; extracting microscopic entropy, mesoscopic entropy, macroscopic entropy, and psychological entropy from the fused feature vector, and constructing an entropy state vector; generating personalized rehabilitation parameters under safety and exploration constraints based on the entropy state vector; converting the personalized rehabilitation parameters into exoskeleton control instructions and virtual reality scene parameters, and collecting execution feedback data; and dynamically updating the parameters for signal noise reduction, feature fusion, and strategy optimization based on the execution feedback data. The present invention improves the quality and accuracy of multimodal electrophysiological signal processing and realizes the optimization of personalized rehabilitation parameters. These innovations not only solve key problems in the existing technology, but also demonstrate broad application prospects in the intersection of biomedical engineering and artificial intelligence.

Description

Technical Field

[0001] The present invention relates to the field of medical rehabilitation technology, and in particular to a method and device for processing accelerated rehabilitation data based on electrophysiological data. Background Art

[0002] In recent years, with the development of biomedical engineering technology, the application of electrophysiological data in rehabilitation therapy has become increasingly widespread. The collection and analysis of multimodal data such as surface electromyography (sEMG), electrocardiogram (ECG), electrodermal activity (EDA), and acceleration signals has become a research hotspot. These technologies can provide rich information for evaluating and guiding the rehabilitation process by monitoring the physiological status of different parts of the human body. However, traditional electrophysiological data analysis methods mainly rely on a single type of signal and lack effective noise reduction and feature fusion strategies, resulting in the information obtained being incomplete and inaccurate, making it difficult to meet the needs of personalized rehabilitation. In addition, existing technologies often ignore the dynamic correlation between signals when processing cross-modal signals, which limits the ability to understand and respond to complex rehabilitation needs.

[0003] Specifically, the analysis systems based on single-modal or multi-modal electrophysiological signals currently used in the field of rehabilitation have two major shortcomings. First, traditional methods usually use fixed filters or predefined noise reduction algorithms, which are powerless when faced with complex practical application scenarios because the noise patterns vary significantly under different individuals and environmental conditions. Secondly, in the feature extraction process, most existing technologies focus on single-dimensional data analysis and fail to fully utilize the interaction between multiple signals to reveal deep changes in physiological state. In contrast, the present invention proposes an accelerated rehabilitation data processing method and system based on electrophysiological data, which not only uses an adaptive noise reduction algorithm to generate a time-aligned noise reduction matrix, but also effectively solves the above problems by calculating the cross-modal phase locking value and using a gated recurrent network to generate a fusion feature vector. This method can more accurately capture the patient's true physiological state and provide a scientific basis for formulating personalized rehabilitation plans. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an ERAP data processing method based on electrophysiological data to solve the problems in the prior art of being unable to effectively integrate multi-sensor data and lacking a mechanism for dynamically adjusting personalized rehabilitation parameters.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides an accelerated rehabilitation data processing method based on electrophysiological data, which includes: synchronously collecting surface electromyography signals, electrocardiogram signals, skin electrical activity signals and acceleration signals, generating a time-aligned denoising matrix through an adaptive denoising algorithm; calculating the full-modal phase locking value based on the denoising matrix, and generating a fusion feature vector through a gated recurrent network; extracting microscopic entropy, mesoscopic entropy, macroscopic entropy and psychological entropy from the fusion feature vector, and constructing an entropy state vector; generating personalized rehabilitation parameters under safety constraints and exploration constraints based on the entropy state vector; converting the personalized rehabilitation parameters into exoskeleton control instructions and virtual reality scene parameters, and collecting execution feedback data; dynamically updating the parameters of signal denoising, feature fusion and strategy optimization based on the execution feedback data.

[0008] As a preferred solution of the method for processing enhanced rehabilitation data based on electrophysiological data of the present invention, wherein: the time-aligned denoising matrix is ​​generated by the adaptive denoising algorithm, and the specific steps are as follows:

[0009] Perform noise reduction processing on the collected surface electromyography signals, electrocardiogram signals, skin electrical activity signals and acceleration signals;

[0010] Time alignment and noise reduction matrix generation are achieved through hardware-level synchronization, resampling alignment, matrix construction, and data verification.

[0011] As a preferred solution of the method for processing enhanced rehabilitation data based on electrophysiological data of the present invention, wherein: the full-modality phase locking value is calculated based on the noise reduction matrix, and the fusion feature vector is generated through the gated recurrent network. The specific steps are as follows:

[0012] Extract the time-frequency phase of the surface electromyography signal, the piecewise linear phase of the electrocardiogram signal, the exponential decay phase of the skin electrode activity signal, and the phase-locked loop tracking phase of the acceleration signal from the denoising matrix, and align them along the time axis to generate a phase matrix.

[0013] Based on the phase matrix, the full-mode phase locking value is calculated using the multi-mode PLV formula;

[0014] Based on the full modal phase locking value, a node set and a hyperedge set are defined, and a dynamic hypergraph is constructed through the hypergraph adjacency matrix;

[0015] Based on the input gate and forget gate, the input weight of the signal channel is allocated through the full-modal phase locking value, and the forgetting ratio of the historical hidden state is adjusted according to the acceleration phase difference;

[0016] Combine the input gate weights and the historical hidden states adjusted by the forget gate to generate candidate hidden states and update the hidden states;

[0017] The hidden states are aggregated through hypergraph-constrained multi-head attention to generate a fused feature vector.

[0018] As a preferred solution of the enhanced rehabilitation data processing method based on electrophysiological data of the present invention, wherein: the microscopic entropy, mesoscopic entropy, macroscopic entropy and psychological entropy are extracted from the fused feature vector, and the entropy state vector is constructed. The specific steps are as follows:

[0019] Extract entropy features from the fusion feature vector at the microscopic, mesoscopic, macroscopic and psychological levels to generate microscopic entropy, mesoscopic entropy, macroscopic entropy and psychological entropy;

[0020] The entropy state vector is constructed by extracting microscopic entropy, mesoscopic entropy, macroscopic entropy and psychological entropy.

[0021] As a preferred solution of the electrophysiological data-based accelerated rehabilitation data processing method of the present invention, wherein: the personalized rehabilitation parameters are generated under safety constraints and exploration constraints based on the entropy state vector, the specific steps are as follows:

[0022] Setting safety thresholds C1, C3, and C4 and time decay factors based on historical rehabilitation datasets;

[0023] Dynamically limit high-risk actions through safety constraint modeling, including:

[0024] When the microscopic entropy exceeds the safety threshold C1, the training intensity parameter is reduced through the Sigmoid decay function;

[0025] When the macro entropy exceeds the safety threshold C3, the joint range parameters are limited by the exponential truncation function;

[0026] When the psychological entropy exceeds the safety threshold C4, the motion speed parameter is reduced through a piecewise linear zeroing function;

[0027] Encourage diverse action patterns by exploring constraint modeling, including,

[0028] The training intensity exploration weight is generated by the hyperbolic tangent function, the joint range of motion exploration weight is generated by the inverse exponential function, the movement speed exploration weight is generated by the power function, and the time decay factor is superimposed;

[0029] Multiply the training intensity parameter by the training intensity exploration weight element by element to generate a preliminary training intensity parameter;

[0030] Multiplying the joint range of motion safety parameter and the joint range of motion safety exploration weight element by element to generate a preliminary joint range of motion parameter;

[0031] Multiplying the motion speed safety parameter by the motion speed safety exploration weight element by element to generate a preliminary motion speed parameter;

[0032] The exploration weights of training intensity, joint range of motion, and movement speed are superimposed on the corresponding parameters to form enhanced, adaptive, and driven intermediate parameters, which are then superimposed twice to generate the final personalized rehabilitation parameters.

[0033] As a preferred solution of the electrophysiological data-based accelerated rehabilitation data processing method of the present invention, the steps of converting personalized rehabilitation parameters into exoskeleton control instructions and virtual reality scene parameters and collecting execution feedback data are as follows:

[0034] The joint range parameters are converted into a spatial coordinate sequence using a cubic spline interpolation algorithm;

[0035] The training intensity parameter is converted into the motor torque output threshold through the fuzzy PID controller;

[0036] The motion velocity parameters are converted into joint angular velocity constraint values ​​through the Jacobian matrix inverse solution algorithm;

[0037] The spatial coordinate sequence, motor torque output threshold and joint angular velocity constraint value are used as exoskeleton control instructions;

[0038] The joint range of motion parameters are converted into the motion amplitude of the virtual limb model through nonlinear amplification mapping rules;

[0039] The motion speed parameters are converted into the frequency of virtual obstacle generation through the speed-density correlation function;

[0040] The training intensity parameters are converted into a multimodal prompting strategy through intensity grading logic, and the virtual limb model movement amplitude, virtual obstacle generation frequency and multimodal prompting strategy are used as virtual reality scene parameters;

[0041] Exoskeleton drive execution and virtual scene rendering are performed through exoskeleton control instructions and virtual reality scene parameters;

[0042] Based on the six-dimensional force sensor array, inertial measurement unit and eye tracking module, the actual joint torque, torso posture angle and gaze point coordinate data are synchronously collected.

[0043] As a preferred solution of the enhanced rehabilitation data processing method based on electrophysiological data of the present invention, the parameters of signal noise reduction, feature fusion and strategy optimization are updated based on the execution feedback data. The specific steps are as follows:

[0044] The torque error scalar is generated by the root mean square error between the actual torque and the expected torque of the exoskeleton joint. The posture coordination error is calculated based on the dynamic time warping distance of the torso posture angle, and the visual attention error is generated by combining the overlap rate of the virtual task gaze points.

[0045] The wavelet packet sub-band noise threshold of the surface electromyography signal is dynamically adjusted according to the torque error scalar. The median filtering is performed on the R-wave detection noise threshold of the ECG signal based on the posture coordination error. The parameters of signal denoising, feature fusion and strategy optimization are updated in combination with the visual attention error.

[0046] In the second aspect, the present invention provides an accelerated rehabilitation data processing device based on electrophysiological data, including: an acquisition module for synchronously acquiring surface electromyography signals, electrocardiogram signals, skin electrical activity signals and acceleration signals, and generating a time-aligned denoising matrix through an adaptive denoising algorithm; a fusion module for calculating the full-modal phase locking value based on the denoising matrix, and generating a fusion feature vector through a gated recurrent network; a multi-scale entropy state analysis module for extracting microscopic entropy, mesoscopic entropy, macroscopic entropy and psychological entropy from the fused feature vector, and constructing an entropy state vector; a personalized rehabilitation strategy generation module for generating personalized rehabilitation parameters under safety constraints and exploration constraints based on the entropy state vector; an exoskeleton-VR execution control module for converting personalized rehabilitation parameters into exoskeleton control instructions and virtual reality scene parameters, and collecting execution feedback data; a dynamic parameter optimization module for dynamically updating the parameters of signal denoising, feature fusion and strategy optimization based on the execution feedback data.

[0047] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the accelerated rehabilitation data processing method based on electrophysiological data as described in the first aspect of the present invention is implemented.

[0048] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the accelerated rehabilitation data processing method based on electrophysiological data as described in the first aspect of the present invention.

[0049] The beneficial effects of this invention include significantly improving the quality and accuracy of multimodal electrophysiological signal processing through synchronized acquisition and adaptive noise reduction, as well as calculating the full-modality phase-locking value based on the noise reduction matrix and generating a fused eigenvector. The former ensures signal consistency and synchronization through precise time alignment and high-quality noise reduction, while the latter enables personalized optimization of rehabilitation parameters through deep feature extraction and dynamic adjustment mechanisms. These innovations not only address key issues in existing technologies but also demonstrate broad application prospects at the intersection of biomedical engineering and artificial intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0051] Figure 1 Flowchart of the enhanced rehabilitation data processing method based on electrophysiological data in Example 1;

[0052] Figure 2 This is a module diagram of the enhanced rehabilitation data processing device based on electrophysiological data in Example 1. DETAILED DESCRIPTION

[0053] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0054] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0055] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0056] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a method for processing enhanced rehabilitation data based on electrophysiological data, comprising the following steps:

[0057] S1: Synchronously collects surface electromyography, electrocardiogram, skin electrical activity, and acceleration signals, and generates a time-aligned denoising matrix using an adaptive denoising algorithm.

[0058] Furthermore, an 8-channel flexible electrode array is used to collect electrophysiological signals of target muscle groups, a 3-lead Ag / AgCl electrode is used to synchronously collect ECG signals of leads I, II, and III, a constant voltage (0.5V) sensor is used to collect changes in palmar thenar muscle conductance, and a 3-axis MEMS sensor is used to collect limb movement acceleration;

[0059] Perform noise reduction processing on the collected surface electromyography signals, electrocardiogram signals, skin electrical activity signals and acceleration signals;

[0060] Surface electromyography signals are denoised using an improved wavelet packet decomposition algorithm;

[0061] It should be noted that the power supply interference was eliminated by using a 50 Hz notch filter, and the surface electromyographic signal of each channel was decomposed into 8 sub-bands by wavelet packet transform. The noise threshold was dynamically calculated for each sub-band, and soft threshold processing was performed on each sub-band. The filtered sub-bands were reconstructed into the noise-reduced surface electromyographic signal;

[0062] The ECG signal is denoised by adaptive comb filtering driven by R waves;

[0063] It should be noted that the Pan-Tompkins algorithm is used to locate the R wave in the QRS complex in real time. The specific steps are as follows: enhance the QRS complex by bandpass filtering (5-15 Hz), use differentiation and squaring operations to highlight the R wave characteristics, generate a smooth envelope by sliding window integration (150 ms), and determine the R wave position by adaptive thresholding;

[0064] The adaptive threshold is divided into a detection threshold and a noise threshold. The detection threshold is used to determine the effective R wave, and its initial value is 50% of the average value of the signal peak value in the previous 2 seconds. The noise threshold is used to identify noise interference, and its initial value is 25% of the average value of the signal peak value in the previous 2 seconds.

[0065] When a valid R wave is detected, the new detection threshold inherits 75% of the weight of the historical detection threshold and integrates 25% of the weight of the current peak value to smoothly track the change of signal amplitude. The noise threshold is updated synchronously.

[0066] If there is no valid R wave, the detection threshold will gradually approach the noise threshold to reduce the risk of missed detection, and the noise threshold will be updated synchronously.

[0067] Design a filter based on the current RR interval (interval between adjacent R waves) to perform noise reduction;

[0068] Denoising of the electrodermal activity signal through acceleration-assisted motion artifact restoration;

[0069] It should be noted that baseline drift correction was performed by eliminating low-frequency drift through 5-second sliding window mean filtering, and motion artifact detection and repair were performed through artifact marking and linear interpolation repair;

[0070] The acceleration signal is de-noised through low-pass filtering and gravity compensation;

[0071] It should be noted that high-frequency noise suppression was performed by eliminating vibration noise through a 20 Hz fourth-order Butterworth low-pass filter, and gravity compensation was performed by separating the static gravity component through quaternion attitude solution;

[0072] Time alignment and noise reduction matrix generation are achieved through hardware-level synchronization, resampling alignment, matrix construction, and data verification.

[0073] It should be noted that hardware-level synchronization is based on the IEEE 1588PTP protocol to achieve multi-device clock synchronization with a maximum deviation of <1ms. Resampling alignment is to perform cubic spline interpolation on surface electromyography signals, electrocardiogram signals, skin electrodermal activity signals and acceleration signals (EDA, ECG, acceleration) that are not sampled at 1000Hz, and unify them to a 1000Hz time axis. Matrix construction is to merge the noise-reduced surface electromyography signals, electrocardiogram signals, skin electrodermal activity signals and acceleration signals into a standardized matrix in chronological order. Data verification is to verify the signal alignment accuracy through cross-correlation analysis to ensure that the maximum delay error is <5ms.

[0074] S2: Calculate the full-modal phase locking value based on the denoising matrix and generate a fusion feature vector through a gated recurrent network;

[0075] Furthermore, the time-frequency phase of the surface electromyography signal, the piecewise linear phase of the electrocardiogram signal, the exponential decay phase of the skin electrodermal activity signal, and the phase-locked loop tracking phase of the acceleration signal are extracted from the denoising matrix and aligned along the time axis to generate a phase matrix.

[0076] Based on the phase matrix, the full-mode phase locking value is calculated using the multi-mode PLV formula, which is expressed as:

[0077]

[0078] Among them, PLV k,m,p,q (t) is the full-modal phase locking value of EMG channel k, ECG lead m, EDA signal, and acceleration axis q at time t, L is the sliding window length, τ is the time index variable, j is the imaginary unit, is the time-frequency phase of the kth EMG channel at time point τ, is the piecewise linear phase of the mth ECG lead at time τ, φ EDA (τ) is the exponential decay phase of the EDA signal at time point τ, is the phase-locked loop tracking phase of the qth acceleration axis at time point τ, g(τ; t) is the spatiotemporal joint weight function;

[0079] It should be noted that γ is the exponential decay factor, is the cosine smoothing term;

[0080] Based on the full modal phase locking value, a node set and a hyperedge set are defined and a dynamic hypergraph is constructed through a hypergraph adjacency matrix;

[0081] It should be noted that the node set is defined by treating each signal channel as a hypergraph node (a total of 15 nodes: 8EMG+3ECG+1EDA+3ACC) and forming a node set, and the hyperedge set is defined by each hyperedge corresponding to a PLV quadruple (k, m, p, q) and forming a hyperedge set;

[0082] Based on the input gate and forget gate, the input weight of the signal channel is allocated through the full-modal phase locking value, and the forgetting ratio of the historical hidden state is adjusted according to the acceleration phase difference;

[0083] It should be noted that the full-modal phase-locked value tensor regulates the input gate weight distribution through the weighted summation process of the hypergraph weight matrix. Specifically, the phase synchronization quaternion of the surface electromyography channel, electrocardiogram lead, skin electrodermal activity signal and acceleration axis is multiplied and accumulated with the corresponding hypergraph weight matrix, and then superimposed on the linear transformation result of the input signal. The input gate weight vector is generated by the Sigmoid function, and the high synchronization signal combination (such as PLV>0.8) obtains the priority fusion weight; the acceleration phase difference is mapped to the forget gate bias term through linear transformation, superimposed on the linear transformation result of the input signal, and then the forget gate weight vector is generated by the Sigmoid function. When the motion phase is mismatched (such as the phase difference>π / 2), the forget gate weight approaches zero, forcing the removal of non-coordinated motion noise in the historical hidden state;

[0084] Combine the input gate weights and the historical hidden states adjusted by the forget gate to generate candidate hidden states and update the hidden states;

[0085] It should be noted that the input gate weight vector is element-wise multiplied with the current noise reduction signal to filter out high synchronization components. After linear transformation, it is added to the linear transformation result of the historical hidden state adjusted by the forget gate (the result of element-wise multiplication of the forget gate weight vector and the historical hidden state), and the normalized candidate hidden state is generated through the hyperbolic tangent function. During the hidden state update process, the input gate weight vector controls the mixing ratio of the candidate hidden state and the historical hidden state. The high-weight area (such as the input gate weight > 0.9) prioritizes the fusion of the current signal, and the low-weight area (such as the input gate weight < 0.1) retains the historical information, and finally generates the updated hidden state vector.

[0086] The hidden states are aggregated through hypergraph-constrained multi-head attention to generate a fused feature vector.

[0087] It should be noted that the hypergraph adjacency matrix constructs a binary mask matrix to mark the effective interaction relationship between the internal nodes of the hyperedge (such as EMG channel 1-ECG lead I-EDA-ACC X-axis); after the hidden state vector is linearly transformed to generate the query, key, and value vectors of each attention head, the attention weights of invalid node pairs are reset to zero; the effective attention weights are calculated by query-key dot product scaling, and the normalized weights are generated by mask filtering and normalized exponential function. After the weighted sum value vector is added, the multi-head output is spliced ​​and finally mapped to a fusion feature vector through a linear projection matrix, which only aggregates the feature information of the physiological correlation signal combination (such as muscle activation and heart rate synchronization).

[0088] S3: Extract microscopic entropy, mesoscopic entropy, macroscopic entropy and psychological entropy from the fused feature vector and construct the entropy state vector;

[0089] Furthermore, entropy features are extracted from the fusion feature vector at the microscopic, mesoscopic, macroscopic and psychological levels to generate microscopic entropy, mesoscopic entropy, macroscopic entropy and psychological entropy.

[0090] The microscopic entropy is extracted by sub-band segmentation, complex Morlet wavelet transform and permutation entropy calculation, and the expression is:

[0091]

[0092] Among them, S1 is the microscopic entropy, is the u-th sub-band of the fused feature vector, is the complex Morlet wavelet transform, PE is the permutation entropy, log2(3!) is the most likely value of the permutation entropy, u is the sub-band index, f u is the center frequency, σ u is the bandwidth factor, o is the embedding dimension, and δ is the delay parameter;

[0093] It should be noted that sub-band segmentation is to divide the fused feature vector into four sub-bands (20-100 Hz, 100-200 Hz, 200-300 Hz, and 300-500 Hz), complex Morlet wavelet transform is to perform time-frequency decomposition of each sub-band signal and extract the instantaneous energy distribution, and permutation entropy calculation is to calculate the permutation entropy of each sub-band to measure the signal complexity;

[0094] The mesoscopic entropy is extracted by Hurst exponent calculation, phase synchronization modulation and detrended fluctuation analysis, and the expression is:

[0095]

[0096] Among them, S2 is the mesoscopic entropy, is the Hurst index based on the RR interval, PS EMG-ECGis the myoelectric-ECG phase synchronization index, λ=5 is the phase synchronization sensitivity adjustment parameter, A α is the detrended volatility analysis scaling index, α max =1.2 is the maximum physiological scaling index;

[0097] It should be noted that the Hurst index calculation is to calculate the long-term memory of the heart rate signal based on the RR interval, the phase synchronization modulation is to dynamically adjust the Hurst index weight using the electromyographic-electrocardiographic phase synchronization index, and the detrended fluctuation analysis is to calculate the scaling index of the 500 ms window;

[0098] Macroscopic entropy is extracted by multi-scale entropy calculation and acceleration gradient detection, and the expression is:

[0099]

[0100] Among them, S3 is the macro entropy, is the 5-second window acceleration signal of the d-th axis, MSE is the multi-scale entropy, s=5 is the scale factor, r=0.15 is the tolerance threshold, J d is the acceleration gradient, ∥·∥2 is the L2 norm;

[0101] It should be noted that multi-scale entropy calculation is a multi-scale complexity analysis of acceleration signals, and acceleration gradient detection is the calculation of acceleration mutation detection indicators;

[0102] Psychological entropy is extracted by SCR event counting and phase synchronization analysis, and the expression is:

[0103]

[0104] Among them, S4 is psychological entropy, is the number of skin conductance response (SCR) events in the past 10 seconds, is the exponential decay phase of the EDA signal, is the R wave trigger phase of the ECG signal;

[0105] It should be noted that SCR event counting was the counting of the number of skin conductance responses within a 10-s window, and phase synchrony analysis was the calculation of the phase difference between EDA and ECG;

[0106] The entropy state vector is constructed by extracting microscopic entropy, mesoscopic entropy, macroscopic entropy and psychological entropy.

[0107] It should be noted that based on the fused feature vector, microscopic entropy was extracted by complex Morlet wavelet frequency division and permutation entropy to quantify the chaos degree of high-frequency electromyographic signals within a 50ms window; mesoscopic entropy was extracted by the Hurst exponent and the dynamic weight of electromyographic-electrocardiographic phase synchronization to evaluate the autonomic sympathetic-vagal balance; macroscopic entropy was extracted by multi-scale entropy and acceleration signal gradient analysis to quantify the movement coordination and joint mutation characteristics within a 5-second window; psychological entropy was extracted by skin conductance response (SCR) event density and skin electrodermal activity-electrocardiographic phase difference power fusion to characterize the psychological stress level; and finally, they were sequentially combined into four-dimensional entropy state vectors S1, S2, S3, and S4, which comprehensively covered the multi-scale physiological states of muscle activation, autonomic nervous regulation, movement control, and psychological load.

[0108] S4: Generate personalized rehabilitation parameters based on the entropy state vector under safety constraints and exploration constraints;

[0109] Setting safety thresholds C1, C3, and C4 and time decay factors based on historical rehabilitation datasets;

[0110] It should be noted that the safety threshold is determined by analyzing the correlation between multimodal physiological signals (surface electromyography, electrocardiography, skin electrodermal activity, acceleration) and clinical safety events (muscle strain, joint compensation, and anxiety triggering) in historical rehabilitation data sets:

[0111] Micro-entropy safety threshold C1 = 0.7: When the micro-entropy (muscle chaos) exceeds 0.7, the risk probability of muscle cramps or fatigue is ≥ 75% (Logistic regression analysis results);

[0112] Macro entropy safety threshold C3 = 0.6: When macro entropy (movement coordination) exceeds 0.6, the incidence of joint compensatory movements increases significantly (p < 0.01, based on motion capture data statistics);

[0113] Psychological entropy safety threshold C4 = 0.8: When the psychological entropy (psychological stress level) exceeds 0.8, the probability of the patient having moderate or above anxiety is ≥ 90% (correlation verification with the anxiety scale GAD-7);

[0114] Time decay factor z = 0.95: Determined through reinforcement learning optimization, it controls the exploration intensity to decay by 5% every 10 minutes, balancing the needs of exploration and stability;

[0115] Dynamically limit high-risk actions through safety constraint modeling, including:

[0116] When the microscopic entropy exceeds the safety threshold C1, the training intensity parameter is reduced through the Sigmoid decay function;

[0117] When the macro entropy exceeds the safety threshold C3, the joint range parameters are limited by the exponential truncation function;

[0118] When the psychological entropy exceeds the safety threshold C4, the motion speed parameter is reduced through a piecewise linear zeroing function;

[0119] It should be noted that when the microscopic entropy (muscle chaos) exceeds the safety threshold C1 = 0.7, the training intensity parameter is dynamically decayed from the current value through the Sigmoid decay function. Specifically, the input of the Sigmoid function is the difference between the microscopic entropy and the safety threshold (S1-C1), and the slope parameter is set to 10, so that the output value of the function decays rapidly from 1 to close to 0 when S1>C1. For example, when S1 = 0.8, the training intensity is reduced to about 30%, and when S1 = 0.9, it is further reduced to 5%, thereby avoiding muscle spasms or fatigue injuries.

[0120] When the macro entropy (S3, movement coordination) exceeds the safety threshold C3 = 0.6, the exponential truncation function is used to limit the joint range parameters. Specifically, the function form is: When S3 = 0.7, the range of joint motion is reduced to approximately 37%, and when S3 = 0.8, it is reduced to 13.5%, suppressing the risk of sports injuries caused by joint compensation or imbalance;

[0121] When the psychological entropy (S4, psychological stress level) exceeds the safety threshold C4 = 0.8, the movement speed parameter is reduced through a piecewise linear zeroing function. Specifically, in the interval 0.8≤S4<0.9, the speed parameter is reduced from 1.0 to 0.5 according to the linear slope 5; when S4≥0.9, the speed parameter is forced to zero and triggers an emergency pause to prevent anxiety or pain from causing loss of control of movement.

[0122] Encourage diverse action patterns by exploring constraint modeling, including,

[0123] The training intensity exploration weight is generated by the hyperbolic tangent function, the joint range of motion exploration weight is generated by the inverse exponential function, the movement speed exploration weight is generated by the power function, and the time decay factor is superimposed;

[0124] Multiply the training intensity parameter by the training intensity exploration weight element by element to generate a preliminary training intensity parameter;

[0125] Multiplying the joint range of motion safety parameter and the joint range of motion safety exploration weight element by element to generate a preliminary joint range of motion parameter;

[0126] Multiplying the motion speed safety parameter by the motion speed safety exploration weight element by element to generate a preliminary motion speed parameter;

[0127] It should be noted that based on the mesoscopic entropy (S2, autonomic nervous balance), the hyperbolic tangent function Generate training intensity exploration weights. When the vagus nerve is active (S2 approaches 1), the weight is close to 1.0, and when the sympathetic nerve is active (S2 approaches 0), the weight is close to 0. The superimposed time decay factor z = 0.95 causes the weight to decay by 5% every 10 minutes.

[0128] Based on macro entropy (S3, movement coordination), through the inverse exponential function Generate joint range of motion exploration weights. When coordination is good (S3 approaches 0), the weight is close to 1.0, and when coordination is poor (S3 approaches 1), the weight is close to 0. A superimposed time decay factor z = 0.95 gradually reduces the exploration intensity.

[0129] Based on psychological entropy (S4, psychological stress level), through power function (1-S4) 2 Generate movement speed exploration weights. When psychological stress is low (S4 approaches 0), the weight is close to 1.0, and when stress is high (S4 approaches 1), the weight is close to 0. A time decay factor z = 0.95 is added to suppress long-term exploration risks.

[0130] The exploration weights of training intensity, joint range of motion, and movement speed are superimposed on the corresponding parameters to form enhanced, adaptive, and driven intermediate parameters, which are then superimposed twice to generate the final personalized rehabilitation parameters.

[0131] It should be noted that the initial training intensity parameter is added to the 30% training intensity exploration weight to generate the final training intensity parameter; the initial range of motion parameter is added to the 30% joint range of motion exploration weight to generate the final joint range of motion parameter; and the initial speed parameter is added to the 30% speed exploration weight to generate the final speed parameter. For example, if the training intensity exploration weight is 0.7, the independent superposition portion is 0.7 × 30% = 0.21. If the initial training intensity parameter is 0.3, the final parameter is 0.3 + 0.21 = 0.51.

[0132] S5: Convert personalized rehabilitation parameters into exoskeleton control instructions and virtual reality scene parameters, and collect execution feedback data;

[0133] Furthermore, the joint range parameters are converted into a spatial coordinate sequence using a cubic spline interpolation algorithm;

[0134] The training intensity parameter is converted into the motor torque output threshold through the fuzzy PID controller;

[0135] The motion velocity parameters are converted into joint angular velocity constraint values ​​through the Jacobian matrix inverse solution algorithm;

[0136] The spatial coordinate sequence, motor torque output threshold and joint angular velocity constraint value are used as exoskeleton control instructions;

[0137] It should be noted that the interpolation nodes are defined based on the joint range parameters, and the cubic spline interpolation algorithm is used to construct a continuous and differentiable joint angle-space coordinate mapping function, and C is applied between adjacent nodes. 2 Continuity constraints are used to generate a smooth trajectory spatial coordinate sequence. A fuzzy PID controller is designed based on the training intensity parameter. The intensity error and its rate of change are quantified through a membership function. The dynamic torque output threshold is generated by combining the fuzzy self-tuning rule of the proportional-integral-differential parameters, and a saturation function is used to safely limit the output torque. A kinematic Jacobian matrix is ​​constructed based on the exoskeleton link parameters, and the generalized inverse of the Jacobian matrix is ​​solved for the motion velocity parameters. The damped least squares method is used to handle singular configurations, and the Cartesian space velocity is mapped to the joint angular velocity constraint value. A velocity threshold comparator is used to limit the angular velocity fluctuation range to within ±3 rad / s.

[0138] The joint range of motion parameters are converted into the motion amplitude of the virtual limb model through nonlinear amplification mapping rules;

[0139] The motion speed parameters are converted into the frequency of virtual obstacle generation through the speed-density correlation function;

[0140] Convert training intensity parameters into multimodal prompting strategies through intensity grading logic;

[0141] The virtual limb model's motion amplitude, virtual obstacle generation frequency, and multimodal prompt strategy are used as virtual reality scene parameters;

[0142] It should be noted that a nonlinear amplification mapping rule is defined based on the joint range of motion parameters. A piecewise exponential-linear composite function is used to map the joint range of motion to the virtual limb model's motion amplitude. When the joint range of motion parameter exceeds the preset joint angular velocity threshold, an exponential amplification mode is activated. When it falls below the threshold, a linear amplification factor is used. The amplification curve is dynamically calibrated using the joint range of motion-virtual model proportional coefficient. A velocity-density correlation function is constructed for the motion velocity parameter. A mathematical relationship between the velocity parameter and the spatial density of virtual obstacles is established based on the cubic power function. The normalized velocity parameter is input into a Poisson distribution model to generate an obstacle probability distribution field, and the virtual obstacle generation frequency is calculated in real time based on the density gradient field. The intensity grading logic is set based on the training intensity parameter. The intensity parameter is divided into low, medium, and high levels using a double-threshold segmentation method. The hybrid weight of the multimodal cueing strategy is determined by combining a fuzzy membership function. Low-intensity levels activate visual cues (color gradients and flashing icons), medium-intensity levels superimpose auditory cues (adjustable frequency beeps), and high-intensity levels introduce tactile cues (vibration intensity and rhythm encoding). The synergistic activation coefficient of the multimodal cues is dynamically adjusted based on the real-time intensity data stream.

[0143] Exoskeleton drive execution and virtual scene rendering are performed through exoskeleton control instructions and virtual reality scene parameters, and the actual joint torque, torso posture angle and gaze point coordinate data are synchronously collected based on the six-dimensional force sensor array, inertial measurement unit and eye tracking module.

[0144] It should be noted that the strain gauge measurement unit based on the six-dimensional force sensor array of the exoskeleton joint collects the force components of the X, Y, and Z axes and the torque components around the axis in real time in each joint coordinate system, quantizes the original signal at a sampling rate of 1000Hz through a 24-bit analog-to-digital converter, and uses a second-order Butterworth low-pass filter to eliminate high-frequency noise. Then, the voltage signal is converted into an actual torque vector in Newton-meter units based on the pre-calibrated sensor sensitivity matrix; an inertial measurement unit integrating a three-axis gyroscope, accelerometer, and magnetometer is used to fuse angular velocity, linear acceleration, and geomagnetic data through the quaternion extended Kalman filter algorithm to solve The pitch, roll, and yaw angles of the patient's torso posture are sampled at 200Hz and output as Euler angle data. Sensor drift errors are eliminated through a temperature compensation algorithm. The eye tracking module, based on the principle of infrared corneal reflection, uses a 940nm non-invasive light source and a high-speed CMOS image sensor to capture eye movement characteristics. The pupil-corneal reflection vector method is used to calculate the three-dimensional spatial coordinates of the gaze point. The raw eye movement data is mapped to the normalized screen coordinate system of the virtual reality scene through a nine-point calibration algorithm. Adaptive threshold filtering is used to eliminate blink artifact interference, and the XY two-dimensional time series of the gaze point coordinates is output.

[0145] S6: Dynamically update parameters for signal noise reduction, feature fusion, and strategy optimization based on execution feedback data;

[0146] Furthermore, the torque error scalar is generated by the root mean square error between the actual torque and the expected torque of the exoskeleton joint, the posture coordination error is calculated based on the dynamic time warping distance of the torso posture angle, and the visual attention error is generated by combining the overlap rate of the virtual task gaze points;

[0147] It should be noted that 1000 sampling points of the six-dimensional torque of each joint are collected within a 500ms time window, the square mean of the Euclidean distance between the actual torque vector and the target torque vector is calculated, and the square root is taken and normalized to the interval [0,1]. When the scalar value exceeds 0.5, the noise reduction module parameter update is triggered; for the trunk posture angle data, the dynamic time warping algorithm is used to calculate the posture coordination error, and the pitch angle and roll angle time series collected by the inertial measurement unit are non-rigidly aligned with the ideal rehabilitation trajectory, and the bending window is limited to 20% of the trajectory length to construct a The optimal regularized path is calculated, and the cumulative distance value is divided by the regularized path length to generate a standardized error index. When the error exceeds 10 degrees, the acceleration signal noise reduction is determined to be invalid. Based on the virtual task gaze point coordinate data, the visual attention error is generated by calculating the Jaccard overlap rate between the gaze point heat map and the target area. The error coefficient is defined as 1 minus the ratio of the overlapping area to the target area. After smoothing the raw eye movement data with the Kalman filter, when the error coefficient exceeds 0.3 for 3 seconds, the parameter correction of the psychological entropy estimation module is triggered.

[0148] The surface electromyography signal wavelet packet sub-band noise threshold is dynamically adjusted according to the torque error scalar. The ECG signal R wave detection noise threshold is corrected by median filtering based on the posture coordination error. The parameters of signal denoising, feature fusion and strategy optimization are updated in combination with the visual attention error.

[0149] It should be noted that when the torque error exceeds the torque error threshold, the response strength of the hyperbolic tangent function is enhanced by a gain factor, causing the noise threshold to increase in a step-wise manner as the error increases, suppressing high-frequency motion artifacts. For error signals with amplitudes below the torque error threshold, an exponential decay mode is used to gradually restore the baseline noise suppression level. To correct the noise threshold for ECG signal R-wave detection, a dynamic weighted fusion is performed by combining posture coordination error and median filtering results: a sliding window median filter extracts the statistical characteristics of the ECG signal baseline amplitude, and a linear compensation term for posture error is superimposed. When posture instability causes a sharp increase in error, an emergency threshold increase mechanism is activated to suppress false detection of abnormal R waves. During the optimization of acceleration signal gravity compensation, an adaptive learning rate adjustment strategy is constructed based on visual attention error: an exponential decay function is used to map the visual attention error into a learning rate correction coefficient for the gradient descent algorithm. This accelerates the iterative convergence of the gravity component in high-error conditions and maintains steady-state optimization to prevent overfitting in low-error conditions. A chain derivation is performed to link the rotation matrix parameters with the gravity vector estimation residual, achieving dynamic accuracy improvement in spatial posture solution.

[0150] The root mean square error between the actual torque and the target torque of the exoskeleton joint is used as the loss function. The partial derivative of the error with respect to the input gate weight matrix is ​​calculated through chain derivation. The momentum optimizer is used to fuse the historical gradient direction and the current gradient value to generate the weight update. When the torque error exceeds the safety threshold for three consecutive frames, the gradient clipping mechanism is activated to limit the update amplitude. The dynamic pruning of the hypergraph adjacency matrix and the hyperedge weight attenuation are guided by the visual attention error. The exponential moving average of the attention error is set as the pruning trigger condition. When the error mean exceeds 0.25, the hyperedge set is traversed to synchronize the phases of the associated nodes. Hyperedges with values ​​lower than 0.4 are subjected to weight coefficient decay operations, and the decay factor decreases piecewise linearly with the error intensity. Hyperedge connections with weights continuously lower than 0.1 are removed simultaneously. L1 regularization constraints are introduced in the multi-head attention mask sparsification process, and an absolute value penalty term is imposed on the dot product result of the query vector and the key vector. The proximal gradient descent algorithm is used to separate the significant components and noise components in the attention weight, retain the high-weight connections of the nodes associated with the head gaze area, and force the attention weights of non-related node pairs to zero, ultimately forming a sparse attention mask pattern that only retains the top 30% significant connections.

[0151] Using the number of safety incidents per hour as input, the threshold adjustment coefficient is generated using piecewise linear interpolation. When the incident frequency exceeds 3 events per hour, the threshold increment mode is triggered, gradually increasing the upper safety threshold in steps of 0.05. When the incident frequency falls below 1 event per hour, the threshold recovery mechanism is activated, gradually returning to the baseline threshold using an exponential decay function. While optimizing the decay factor using a policy gradient algorithm, a two-dimensional reward function is constructed, including the proportion of new action patterns and a safety indicator. Monte Carlo sampling is used to evaluate policy improvements, and the Adam optimizer is used to calculate the gradient update of the exploration decay factor. The learning rate is set to 0.01. When the reward growth rate is less than 2% over five consecutive training cycles, the learning rate cosine annealing mechanism is activated to prevent local optimality. When adjusting the inertia factor superposition ratio in stages, the rehabilitation training cycle is divided into three stages: early, middle, and late. In the early stage (0-14 days), a fixed ratio of 0.3 is maintained. In the middle stage (15-42 days), an exponential decay function is used to dynamically adjust the superposition weight, and the decay time constant is set to 30 days. In the late stage (more than 43 days), the inertia factor ratio is locked at 0.05. Each time the stage switches, the sensitivity of the proportional parameter to the torque error is verified using the gradient descent algorithm. When the error change rate exceeds 10%, the proportional parameter freezing protection mechanism is triggered.

[0152] This embodiment also provides an accelerated rehabilitation data processing device based on electrophysiological data, including an acquisition module for synchronously acquiring surface electromyographic signals, electrocardiographic signals, skin electrical activity signals and acceleration signals, and generating a time-aligned denoising matrix through an adaptive denoising algorithm; a fusion module for calculating the full-modal phase locking value based on the denoising matrix, and generating a fusion feature vector through a gated recurrent network; a multi-scale entropy state analysis module for extracting microscopic entropy, mesoscopic entropy, macroscopic entropy and psychological entropy from the fused feature vector, and constructing an entropy state vector; a personalized rehabilitation strategy generation module for generating personalized rehabilitation parameters under safety constraints and exploration constraints based on the entropy state vector; an exoskeleton-VR execution control module for converting personalized rehabilitation parameters into exoskeleton control instructions and virtual reality scene parameters, and collecting execution feedback data; and a dynamic parameter optimization module for dynamically updating the parameters of the signal denoising, feature fusion and strategy optimization modules based on the execution feedback data.

[0153] This embodiment also provides a computer device suitable for the accelerated rehabilitation data processing method based on electrophysiological data, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the accelerated rehabilitation data processing method based on electrophysiological data proposed in the above embodiment.

[0154] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0155] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for processing enhanced rehabilitation data based on electrophysiological data as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0156] In summary, this invention significantly improves the quality and accuracy of multimodal electrophysiological signal processing through: synchronous acquisition and adaptive noise reduction, as well as calculation of all-modal phase-locking values ​​based on a noise reduction matrix and generation of fused eigenvectors. The former ensures signal consistency and synchronization through precise time alignment and high-quality noise reduction; the latter enables personalized optimization of rehabilitation parameters through deep feature extraction and dynamic adjustment mechanisms. These innovations not only address key issues in existing technologies but also demonstrate broad application prospects at the intersection of biomedical engineering and artificial intelligence.

[0157] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for processing enhanced rehabilitation data based on electrophysiological data, characterized by: include, Synchronously collect surface electromyography signals, electrocardiogram signals, skin electrical activity signals and acceleration signals, and generate a time-aligned noise reduction matrix through an adaptive noise reduction algorithm; The full-modal phase locking value is calculated based on the denoising matrix, and the fusion feature vector is generated through the gated recurrent network. The specific steps are as follows: Extract the time-frequency phase of the surface electromyography signal, the piecewise linear phase of the electrocardiogram signal, the exponential decay phase of the skin electrode activity signal, and the phase-locked loop tracking phase of the acceleration signal from the denoising matrix, and align them along the time axis to generate a phase matrix. Based on the phase matrix, the full-mode phase locking value is calculated using the multi-mode PLV formula, which is expressed as: Among them, PLV k,m,p,q (t) is the full-modal phase locking value of EMG channel k, ECG lead m, EDA signal, and acceleration axis q at time t, L is the sliding window length, τ is the time index variable, j is the imaginary unit, is the time-frequency phase of the kth EMG channel at time point τ, is the piecewise linear phase of the mth ECG lead at time τ, φ EDA (τ) is the exponential decay phase of the EDA signal at time point τ, is the phase-locked loop tracking phase of the qth acceleration axis at time point τ, g(τ; t) is the spatiotemporal joint weight function; Based on the full modal phase locking value, a node set and a hyperedge set are defined, and a dynamic hypergraph is constructed through the hypergraph adjacency matrix; Based on the input gate and forget gate, the input weight of the signal channel is allocated through the full-modal phase locking value, and the forgetting ratio of the historical hidden state is adjusted according to the acceleration phase difference; Combine the input gate weights and the historical hidden states adjusted by the forget gate to generate candidate hidden states and update the hidden states; Aggregate the hidden states through multi-head attention constrained by hypergraph to generate a fused feature vector; Extract microscopic entropy, mesoscopic entropy, macroscopic entropy and psychological entropy from the fused feature vector and construct the entropy state vector; According to the entropy state vector, personalized rehabilitation parameters are generated under safety constraints and exploration constraints; Convert personalized rehabilitation parameters into exoskeleton control instructions and virtual reality scene parameters, and collect execution feedback data; Dynamically update parameters for signal denoising, feature fusion, and strategy optimization based on execution feedback data.

2. The enhanced rehabilitation data processing method based on electrophysiological data according to claim 1, characterized in that: The time-aligned denoising matrix is ​​generated by the adaptive denoising algorithm. The specific steps are as follows: Perform noise reduction processing on the collected surface electromyography signals, electrocardiogram signals, skin electrical activity signals and acceleration signals; Time alignment and noise reduction matrix generation are achieved through hardware-level synchronization, resampling alignment, matrix construction, and data verification.

3. The method for processing enhanced rehabilitation data based on electrophysiological data according to claim 1, characterized in that: The specific steps of extracting microscopic entropy, mesoscopic entropy, macroscopic entropy and psychological entropy from the fusion feature vector and constructing the entropy state vector are as follows: Extract entropy features from the fusion feature vector at the microscopic, mesoscopic, macroscopic and psychological levels to generate microscopic entropy, mesoscopic entropy, macroscopic entropy and psychological entropy; The entropy state vector is constructed by extracting microscopic entropy, mesoscopic entropy, macroscopic entropy and psychological entropy.

4. The method for processing enhanced rehabilitation data based on electrophysiological data according to claim 3, characterized in that: The specific steps of generating personalized rehabilitation parameters under safety constraints and exploration constraints based on the entropy state vector are as follows: Setting safety thresholds C1, C3, and C4 and time decay factors based on historical rehabilitation datasets; Dynamically limit high-risk actions through safety constraint modeling, including: When the microscopic entropy exceeds the safety threshold C1, the training intensity parameter is reduced through the Sigmoid decay function; When the macro entropy exceeds the safety threshold C3, the joint range parameters are limited by the exponential truncation function; When the psychological entropy exceeds the safety threshold C4, the motion speed parameter is reduced through a piecewise linear zeroing function; Encourage diverse action patterns by exploring constraint modeling, including, The training intensity exploration weight is generated by the hyperbolic tangent function, the joint range of motion exploration weight is generated by the inverse exponential function, the movement speed exploration weight is generated by the power function, and the time decay factor is superimposed; Multiply the training intensity parameter by the training intensity exploration weight element by element to generate a preliminary training intensity parameter; Multiplying the joint range of motion safety parameter and the joint range of motion safety exploration weight element by element to generate a preliminary joint range of motion parameter; Multiplying the motion speed safety parameter by the motion speed safety exploration weight element by element to generate a preliminary motion speed parameter; The exploration weights of training intensity, joint range of motion, and movement speed are superimposed on the corresponding parameters to form enhanced, adaptive, and driven intermediate parameters, which are then superimposed twice to generate the final personalized rehabilitation parameters.

5. The enhanced rehabilitation data processing method based on electrophysiological data according to claim 4, characterized in that: The specific steps of converting personalized rehabilitation parameters into exoskeleton control instructions and virtual reality scene parameters and collecting execution feedback data are as follows: The joint range parameters are converted into a spatial coordinate sequence using a cubic spline interpolation algorithm; The training intensity parameter is converted into the motor torque output threshold through the fuzzy PID controller; The motion velocity parameters are converted into joint angular velocity constraint values ​​through the Jacobian matrix inverse solution algorithm; The spatial coordinate sequence, motor torque output threshold and joint angular velocity constraint value are used as exoskeleton control instructions; The joint range of motion parameters are converted into the motion amplitude of the virtual limb model through nonlinear amplification mapping rules; The motion speed parameters are converted into the frequency of virtual obstacle generation through the speed-density correlation function; The training intensity parameters are converted into a multimodal prompting strategy through intensity grading logic, and the virtual limb model movement amplitude, virtual obstacle generation frequency and multimodal prompting strategy are used as virtual reality scene parameters; Exoskeleton drive execution and virtual scene rendering are performed through exoskeleton control instructions and virtual reality scene parameters; Based on the six-dimensional force sensor array, inertial measurement unit and eye tracking module, the actual joint torque, torso posture angle and gaze point coordinate data are synchronously collected.

6. The enhanced rehabilitation data processing method based on electrophysiological data according to claim 5, characterized in that: The parameters of signal denoising, feature fusion and strategy optimization are updated based on the execution feedback data. The specific steps are as follows: The torque error scalar is generated by the root mean square error between the actual torque and the expected torque of the exoskeleton joint. The posture coordination error is calculated based on the dynamic time warping distance of the torso posture angle, and the visual attention error is generated by combining the overlap rate of the virtual task gaze points. The wavelet packet sub-band noise threshold of the surface electromyography signal is dynamically adjusted according to the torque error scalar. The median filtering is performed on the R-wave detection noise threshold of the ECG signal based on the posture coordination error. The parameters of signal denoising, feature fusion and strategy optimization are updated in combination with the visual attention error.

7. An enhanced rehabilitation data processing device based on electrophysiological data, executing the enhanced rehabilitation data processing method based on electrophysiological data according to any one of claims 1 to 6, characterized in that: Including acquisition module, fusion module, multi-scale entropy state analysis module, personalized rehabilitation strategy generation module, exoskeleton-VR execution control module and dynamic parameter optimization module, The acquisition module is used to synchronously collect surface electromyography signals, electrocardiogram signals, skin electrical activity signals and acceleration signals, and generate a time-aligned noise reduction matrix through an adaptive noise reduction algorithm; The fusion module is used to calculate the full-modal phase locking value based on the noise reduction matrix and generate the fusion feature vector through the gated recurrent network; Multi-scale entropy state analysis module, used to extract microscopic entropy, mesoscopic entropy, macroscopic entropy and psychological entropy from the fused feature vector and construct the entropy state vector; A personalized rehabilitation strategy generation module is used to generate personalized rehabilitation parameters under safety constraints and exploration constraints based on the entropy state vector; Exoskeleton-VR execution control module, used to convert personalized rehabilitation parameters into exoskeleton control instructions and virtual reality scene parameters, and collect execution feedback data; The dynamic parameter optimization module is used to dynamically update the parameters of the signal noise reduction, feature fusion and strategy optimization modules based on the execution feedback data.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for processing enhanced rehabilitation data based on electrophysiological data according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for processing enhanced rehabilitation data based on electrophysiological data according to any one of claims 1 to 6 are implemented.

Images