A wavelet-enhanced collaborative activation method for multi-target temporal modulation in FES

Abstract

This invention discloses a wavelet-enhanced co-activation method for multi-target temporal modulation of functional electrical stimulation (FES), relating to the field of biomedical engineering technology. The model constructs a time-frequency energy salient feature matrix by analyzing multi-channel surface electromyography (SEMG) signals using an overlapping sliding window combined with energy proportions. Simultaneously, it constructs an intermuscular co-activation matrix within the multi-channel SEMG sliding window. Each single-channel time-frequency energy salient feature matrix and the multi-channel intermuscular co-activation matrix are then combined to form a multi-channel high-dimensional matrix. This high-dimensional matrix is ​​then dimensionality-reduced to a multi-channel, multi-dimensional time-frequency energy-intermuscular co-activation matrix. An adaptive threshold logic combination rule is used to filter muscle activation temporal vectors, and finally, the activation temporal sequence is applied to FES experiments. This invention, by analyzing multi-channel SEMG signals and multi-target muscle co-activation characteristics, accurately predicts hand movements in the sagittal plane and dynamically adjusts FES parameters, achieving precise control over muscle activity phases.

Description

Technical Field

[0001] This invention relates to the field of biomedical engineering technology, and in particular to a wavelet-enhanced collaborative activation method for multi-target temporal modulation of FES. Background Technology

[0002] Functional electrical stimulation (FES) has shown great potential as an emerging rehabilitation therapy. FES technology uses a pre-programmed low-frequency pulsed current of a certain intensity to stimulate one or more muscle groups, inducing muscle movement or simulating normal voluntary movement. As a technology, FES can be used as a neuroprosthetic to restore motor function and as a therapeutic intervention in post-stroke rehabilitation. FES can improve motor performance, reduce muscle loss, and overcome the limitations of disabled limbs through stimulation. It can restore motor function in patients with spinal cord injury, Parkinson's disease, cerebral palsy, and stroke. FES can reduce muscle degeneration during inactivity. Repetitive exercises of single-joint movements induced by FES, such as flexion and extension, play a crucial role in preventing degeneration of the muscle fibers supporting the joint and are used in the treatment of spinal cord injury and stroke patients. Furthermore, FES can stimulate damaged nerves and muscles, promoting the recovery of neuromuscular function and improving patients' motor abilities and quality of life.

[0003] By precisely controlling the electrical stimulation parameters, FES can simulate the natural stimulation of the nervous system, promoting muscle contraction and the recovery of neuromuscular function, thereby helping patients regain motor ability and improve their quality of life. As an advanced neuromuscular activation method, FES can indeed effectively stimulate muscles to produce movement by adjusting specific stimulation parameters (such as pulse frequency, pulse width, and current intensity). However, it still faces many challenges in precisely controlling the coordinated movements of various muscles and joints in the hand. The hand, as one of the most delicate and complex organs in the human body, relies on the coordinated work of numerous muscles and nerves for its movements. Every subtle movement, such as finger flexion and extension, wrist rotation, and palm clenching, requires these muscles and nerves to work together at precise times and with specific intensities. This complexity and precision present both a challenge and an opportunity for FES technology. Currently, FES technology stimulates muscles to produce movement by adjusting stimulation parameters (such as pulse frequency, pulse width, and current intensity). However, when simulating natural hand movements, adjusting these parameters often fails to achieve ideal results. This is because the neuromuscular signals of hand movements are complex and variable, while the stimulation patterns provided by FES technology are often relatively singular. This means that when simulating natural hand movements, FES technology often fails to accurately capture all neuromuscular signals, thus failing to achieve precise coordinated movements between various muscles and joints of the hand.

[0004] Existing FES systems mostly employ open-loop control. In open-loop control systems, the stimulation parameters of the functional electrical stimulation output (such as stimulation amplitude, frequency, waveform, etc.) are fixed, and these parameters depend on the experience of the rehabilitation instructor. Because a fixed pulse sequence is used for stimulation, excess residual charge may result in muscle fatigue, or the stimulation intensity may be insufficient to induce the muscles to contract accordingly and complete the prescribed training movements. Furthermore, the lack of real-time feedback and precise control of the patient's actual performance during rehabilitation (such as trajectory, angle, angular velocity, joint torque, etc.) in open-loop control systems makes it difficult to achieve ideal rehabilitation results. Summary of the Invention

[0005] The purpose of this invention is to propose a wavelet-enhanced co-activation method for multi-target temporal modulation of FES. By integrating time-frequency analysis and quantification of muscle co-activation characteristics, a comprehensive framework for muscle activity analysis is constructed. Furthermore, by dynamically adjusting the energy threshold index, the activation trend of muscles can be predicted more accurately, thereby enabling precise control of the movement of the upper limb in the sagittal plane.

[0006] To achieve the above objectives, this invention proposes a wavelet-enhanced collaborative activation method for multi-target temporal modulation in FES, with the following specific steps:

[0007] S1. Acquire multi-channel surface electromyography (EMG) signals and bending angles, and preprocess the acquired EMG signals;

[0008] S2. Wavelet denoising and reconstruction are performed on the preprocessed electromyography signals of each channel in S1. Two consecutive wavelet transforms are performed on the reconstructed electromyography signals of each channel, and the time-frequency energy significant feature matrix is ​​adaptively obtained by combining the energy proportion ratio and the overlapping moving average window.

[0009] S3. The preprocessed multi-channel electromyography (EMG) signals in S1 are combined in pairs to construct multiple muscle pairs. The EMG signals of each muscle pair are aligned and segmented using an overlapping sliding average window to obtain multiple EMG signal pairs. Multi-level co-activation feature indexes are calculated for each pair of EMG signals to obtain the intermuscular co-activation distribution feature matrix.

[0010] S4. Construct a multi-channel high-dimensional matrix by combining the salient feature matrix of each single-channel time-frequency energy in S2 with the intermuscular coactivation matrix in S3. At the same time, reduce the dimension by the cumulative variance contribution rate and obtain the multi-dimensional time-frequency energy-intermuscular coactivation matrix.

[0011] S5. For each component of the multi-dimensional time-frequency energy-muscle coactivation matrix of each channel in S4, a threshold is adaptively set according to statistical probability. An adaptive threshold logic combination rule is proposed to optimize and select the corresponding muscle activation state time-series vector.

[0012] S6. Incorporate the activation time vectors of multiple muscles into the design of stimulation parameters for the FES experiment and conduct the corresponding experiments.

[0013] Preferably, in S1, the experiment is designed for sagittal plane motion. During the hand clenching motion, multi-channel surface electromyography (EMG) signals of the forearm muscles and finger bending angles are simultaneously acquired, and the acquired multi-channel EMG signals are preprocessed.

[0014] Specifically, this includes: collecting surface electromyographic signals of the extensor carpi ulnaris, extensor digitorum, flexor digitorum superficialis, palmaris longus, and flexor carpi radialis muscles on the right side of a normal person; simultaneously collecting the flexion of the right index finger; and performing bandpass filtering and notch filtering on the collected EMG signals based on different angle values ​​of the movement to obtain preprocessed electromyographic signals (V1,...,Vi).

[0015] Preferably, in S2, each single-channel electromyography signal after preprocessing in S1 is subjected to fourth-order Daubechies level 6 wavelet decomposition, and a threshold for each signal is set based on the signal length and logarithmic function to remove noise. Then, the wavelet coefficients of each frequency band are reconstructed to obtain a multi-channel reconstructed signal (v1, v2, ..., vi); where vi is the reconstructed signal.

[0016] Specifically, this involves using a 4th-order Daubechies wavelet (Db4) to perform a 6-level wavelet decomposition to extract the time-frequency features of the electromyographic signal (V1,..,Vi). This process divides the signal into an approximate part (low frequency) and a detail part (high frequency), and further decomposes the approximate part. After decomposition, the high-frequency detail coefficients are suppressed or removed by calculating a threshold, and then the signal (v1,v2,…,vi) is reconstructed by using wavelet basis functions for deconvolution.

[0017] Preferably, the multi-channel reconstructed signal (v1, v2, ..., vi) is transformed by scales from 1 to 128 to obtain the overall coefficient distribution C of the electromyography signal for each channel. i (a,b), by calculating each C i The square of (a,b) is used to obtain the reconstructed energy, and then the time-frequency scale energy matrix E of the entire signal is obtained. i (a,b); Based on this, calculate the energy percentage, that is, calculate the energy distribution E for each energy distribution. i The proportion of each scale in the total energy of (a,b) is determined, and a proportion exceeding 10% is considered significant, thus achieving m. i An adaptive selection of energy-significant scales; where a is the scale parameter, b is the location parameter, and m is the energy-significant scale. i Select m energy saliency scales for the i-th scale;

[0018] Preferably, an overlapping moving average window is used to divide the electromyographic signal vi of each channel in the multi-channel reconstructed signal (v1, v2, ..., vi) into n electromyographic signal segments, and m i The energy saliency scale is used as the scale selection criterion for the continuous wavelet transform of the electromyographic signal after sliding window n. For each saliency scale energy value after each window, a time-frequency energy saliency feature matrix (m) corresponding to the electromyographic signal vi is constructed. i (×n), where n is an integer.

[0019] Preferably, in S3, multiple muscle pairs are constructed by combining the preprocessed multi-channel reconstructed signals (v1, v2, ..., vi) in pairs according to S2. An overlapping sliding average window n is applied to the electromyographic signals of each muscle pair, and the multi-level co-activation feature index of each pair of electromyographic signals after each window is calculated, including the degree of overlap of the muscle pair combination cci, the time correlation tc, and the degree of information exchange cmi. Thus, a muscle co-activation distribution feature matrix (cci; tc; cmi)×n containing all combinations is obtained.

[0020] Specifically, this includes: after applying an overlapping moving average window n to the preprocessed whole-segment signals (v1, v2, ..., vi) between multiple muscles, calculating the common contraction index cci1, ..., cci between muscles after each window. a Time correlation tc1,...,tc b and cross-information cmi1,…,cmi c These characteristic values ​​quantify the degree of co-activation and information exchange between the electromyographic envelopes of two muscles at a given time point, and are used as the co-activation distribution feature matrix (cci; tc; cmi)×n of vi.

[0021] Preferably, in S4, the significant feature matrix of time-frequency energy of each single channel in S2 (m i The high-dimensional data matrix (cci; tc; cmi; m) for each muscle ri is constructed by combining the intermuscular coactivation matrix (cci; tc; cmi; m) in S3 with the intermuscular coactivation matrix (cci; tc; cmi) × n. i Based on the correlation between variance and eigenvectors, the most important features are extracted. Principal components with a cumulative feature variance contribution rate of 95% are selected, and the high-dimensional data matrix is ​​reduced to a multi-dimensional time-frequency energy-myocoel co-activation matrix k. i ×n, the multidimensional time-frequency energy-myomuscular co-activation matrix will be used as the energy threshold indicator; where k i This is the row number index.

[0022] Specifically, this includes integrating the reconstructed energy values ​​and co-activation characteristic values ​​of the complete signals (v1, v2, ..., vi) of multiple muscles into a multidimensional dataset to obtain a high-dimensional data matrix (cci; tc; cmi; m) for each muscle ri. i )×n, where each row represents a time point and each column represents a feature, including energy value and co-activation characteristic value, the covariance matrix of the standardized data is calculated for the multidimensional data to determine the correlation between variables. The eigenvalues ​​and corresponding eigenvectors of the covariance matrix are solved. Based on the magnitude of the eigenvalues, principal components with a cumulative variance contribution rate exceeding 95% are selected. Using the selected ri and k... i Principal components are used to construct feature vectors to obtain a multi-dimensional time-frequency energy-myo-intermuscular co-activation matrix k. i ×n, where each feature vector corresponds to the muscle activity state at a given time point.

[0023] Preferably, in S5, the multi-dimensional time-frequency energy-myomyelin co-activation matrix k for each channel in S4 is... i Each component k of ×n i The threshold is adaptively set based on the maximum probability density. To provide additional principal components related to muscle activation, considering the simultaneous influence of multiple principal components, a threshold logic combination rule is proposed to determine the activation status of n points in the multi-dimensional time-frequency energy-muscle co-activation matrix. Only when the value of each principal component simultaneously exceeds its respective threshold is the muscle considered activated within a specific time period, thus obtaining the activation sequence g1,...,g of different channels. i Among them, g i Let be the activation timing vector of the i-th channel.

[0024] Specifically, this includes: Principal components are linear combinations of the original dataset, representing the main directions of variation in the data. The first principal component usually contains the most variation information, thus serving as a powerful indicator for monitoring changes in muscle activation state. Besides the first principal component, other principal components can also be used to set thresholds, especially when they provide additional information related to muscle activation. The multi-dimensional matrix k for each channel... i Each component k of ×n i Thresholds are adaptively set based on the maximum probability density value. Each principal component can have its own threshold, and these thresholds work together to determine the muscle activation state. An adaptive threshold logic combination rule is proposed to consider multiple principal components simultaneously. The logic rule states that the muscle is considered activated only when the values ​​of multiple principal components simultaneously exceed their respective thresholds, resulting in the activation time sequence n. i The activation timing g1,...,g of different channels is obtained by reconstructing the signal (v1,v2,...,vi). i .

[0025] Preferably, in step S6, the electromyographic electrodes are fixedly attached to the relevant muscle area of ​​the patient's left arm, and the electrical stimulation electrodes are held in a set position. The electromyographic acquisition device collects surface electromyographic signals for subsequent analysis, and the activation sequence g1,...,g of multiple muscles is determined. i The stimulation parameters designed for the FES experiment include the frequency, amplitude, and pulse width of the current, which stimulate the muscles on the affected side, enabling the patient's affected hand to complete the fist-clenching action.

[0026] Therefore, this invention proposes a wavelet-enhanced collaborative activation method for multi-target temporal modulation in FES, with the following beneficial effects:

[0027] (1) The present invention provides a wavelet-enhanced co-activation method for multi-target temporal modulation of FES. By integrating time-frequency analysis and muscle co-activation characteristic quantification, a comprehensive muscle activity analysis framework is constructed. The energy threshold index is dynamically adjusted, which can flexibly adapt to changes in muscle activity, thereby more accurately predicting the activation trend of muscles.

[0028] (2) The present invention provides a wavelet-enhanced co-activation method for multi-target temporal modulation of FES. Based on the activation timing of different channels, it deepens the understanding of muscle co-activation mode, thereby enabling accurate judgment of the co-activation state of muscles in a specific time period, providing more comprehensive and richer information on muscle activity, improving the accuracy of muscle activation judgment, and providing a new perspective for the quantitative analysis of muscle co-activation.

[0029] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0030] Figure 1 This is a flowchart of a preferred embodiment of the functional electrostimulation control strategy based on multi-target temporal modulation of the present invention;

[0031] Figure 2 This is a schematic diagram of the forearm sampling site of the conventional hand according to the present invention;

[0032] Figure 3 This is a schematic diagram of the finger curvature acquisition module of the present invention; wherein, a is a schematic diagram of the finger curvature acquisition device, and b is a schematic diagram of finger curvature data acquisition;

[0033] Figure 4 This is a visualization interface for collecting finger curvature according to the present invention;

[0034] Figure 5 This is a schematic diagram of the muscle temporal activation segment of the present invention. Detailed Implementation

[0035] To make the technical solutions, advantages, and objectives of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below. The described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the protection scope of this application.

[0036] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0037] like Figure 1 As shown, this invention provides a wavelet-enhanced collaborative activation method for multi-target temporal modulation in FES, with the following specific steps:

[0038] S1. Obtain the dataset. The electromyography (EMG) acquisition device is Delsys, and its sampling frequency is 2000 Hz. Figure 3 As shown in Figure a, the bending angle is acquired using a Flex sensor. Before the experiment, the skin surface is wiped with alcohol to increase conductivity and reduce interference. Figure 2 As shown, the electrodes of the electromyography (EMG) acquisition device and the functional electrical stimulation (fEP) electrodes are attached to the extensor carpi ulnaris, extensor digitorum, flexor digitorum superficialis, palmaris longus, and flexor carpi radialis muscles on the right side of a normal individual. Figure 4 As shown, at the start of the experiment, electromyography (EMG) and muscle contraction measurements were simultaneously acquired and recorded, including the start and maintenance points of the movement. Figure 3 As shown in b, the experimental subject first wore a flexible glove with his arm hanging naturally and his palm relaxed. Surface electromyography (EMG) signals were collected from the user at rest for 20 seconds. Then, the experimental subject began to grasp with slight force. The finger flexion was collected and displayed on a visualization interface. The flexion was divided into three groups: 0 to 30 degrees, 0 to 60 degrees, and 0 to 90 degrees. Each group performed 40 movements. After reaching the desired angle, the experimental subject maintained the current angle for a period of time. Based on the different angle values ​​of the movement, the collected EMG signals were bandpass filtered (cutoff frequency 15Hz~250Hz) and notch filtered (frequency 50Hz) to obtain preprocessed EMG signals (V1,..,Vi).

[0039] S2. The preprocessed electromyographic signal (V1,...,Vi) is decomposed using the 4th-order Daubechies (6-level decomposition) as the mother wavelet to extract time-frequency features. The wavelet decomposition process begins by convolving the original electromyographic signal with the Db4 wavelet basis function to extract the approximate and detail components of the signal. In each level of decomposition, the signal is divided into an approximate part (containing low-frequency information) and a detail part (containing high-frequency information). The approximate part is further decomposed. After 6 levels of decomposition, a series of wavelet coefficients in different frequency bands are obtained. At the same time, the threshold of each signal is calculated based on the signal length and logarithmic function. The high-frequency detail coefficients are removed, and then the inverse process of wavelet decomposition is reconstructed. For the wavelet coefficients of each frequency band, the corresponding wavelet basis function is used for inverse convolution to reconstruct the signal (v1,v2,...,vi).

[0040] S3. Perform a continuous wavelet transform on the reconstructed signal (v1, v2, ..., vi), with a scale ranging from 1 to 128, to capture the time variations of different frequency components. This provides the coefficient distribution at different scales of the signal, which is typically represented as a three-dimensional image. Two dimensions are time t and frequency 1 / a, and the third dimension is the amplitude C of the coefficients. i (a,b), by calculating each C i The reconstructed energy is obtained by squaring (a,b), and then the time-frequency scale energy matrix E of the entire signal is obtained. i (a,b); where the formula for continuous wavelet transform is as follows:

[0041]

[0042] Among them, W f f(x) is the result of the wavelet transform; it is two-dimensional and depends on the scale a and the location b. f(x) is the original signal; ψ(x) is the wavelet function, usually called the mother wavelet. * τ is the complex conjugate of the wavelet function; a is the scaling parameter, which controls the extent of the wavelet's expansion and is inversely proportional to the frequency; b is the position parameter, representing the wavelet's position in the signal; τ is the center position of the wavelet function.

[0043] S4. Divide the time dimension t in the two dimensions into time windows and use the overlapping moving average window method to provide temporal continuity, which helps to capture signals that cross the window boundaries. The time window size is set to 50 sampling points and the moving step size is set to 25 sampling points. The sliding window can move on the time axis to cover the calculation of feature values ​​for each time window, and the sliding window takes into account the influence of signal continuity on feature value calculation.

[0044] S5. Based on the energy E of the reconstructed signal vi in ​​S3 i (a,b), for each scale a, calculate the energy E over the entire time series.i (a) Based on the calculated E i (a) Based on this, the energy proportion is calculated to identify the energy salient scale in the energy distribution, according to each energy distribution E i The proportion of total energy at each scale of (a,b) was determined, and significance was defined as an energy proportion exceeding 10%. m was then selected. i A significant energy scale is defined, and then, based on the time window divided by S4, an overlapping moving average window is used with a window size of 50 sampling points and a moving step size of 25 sampling points. For the selected scale m... i Starting from the time dimension t, apply the first window (1-50), calculate the energy distribution of all sampling points within the window at each selected scale, move the window forward by 25 sampling points (26-75), calculate the energy values ​​within the window again, and repeat this process until the entire time dimension is covered, obtaining E(b1) to E(bm) for each selected scale. i ).

[0045] S6. Based on the overlapping moving average window n of the preprocessed whole-segment signals (v1, v2, ..., vi) between multiple muscles in S5, calculate the co-activation characteristic value for each time window, and calculate the co-activation characteristic value between the five muscles, considering each pair of muscles, using the combination formula. Given , represents the number of ways to select 2 muscles from e muscles, meaning there are . The muscle combination needs to be calculated; the specific calculation process is as follows:

[0046] S61. Calculate the common contraction index. For the EMG signal envelopes of the two electromyographic signals within the sliding window, calculate their normalized envelopes, and then calculate the sum of the point-by-point minimum values ​​to obtain cci1,…,cci. a The calculation formula is as follows:

[0047]

[0048] Where, x norm and y norm The envelope is normalized by the average of 98-99% of the respective antagonistic electromyographic signal envelopes; CCI quantifies the degree of overlap between the two muscle envelopes, and the higher the value, the greater the degree of co-contraction;

[0049] S62. Calculate the time correlation: For the EMG signal envelopes of two electromyographic (EMG) signals within the sliding window, calculate the average of their products at all sample points to obtain tc1,...,tc b The calculation formula is as follows:

[0050]

[0051] Among them, x and y are enveloped by their respective antagonistic electromyographic signals;

[0052] S63. Calculate cross-mutual information to quantify the degree of information exchange between the electromyographic envelopes of two different muscles, and obtain cmi1,…,cmi. c The calculation formula is as follows:

[0053] I(x,y,τ)=H(x)+H(y)-H(x,y,τ);

[0054] Where H represents the Shannon entropy, H(x) and H(y) come from the discrete probability distributions of x(n) and y(n), and H(x,y,τ) is the entropy function based on the joint probability distribution with time lag τ.

[0055] These characteristic values ​​quantify the degree of coordinated activation and information exchange between the electromyographic envelopes of two muscles at a given time point.

[0056] S7. Construct a multidimensional dataset and build feature vectors. The specific steps are as follows:

[0057] S71. Calculate the energy value E(b1) to E(bm) of each selected scale from the reconstructed signal vi in ​​S5. i The co-activation characteristic values ​​calculated in S6 are integrated into a multidimensional dataset, where each row represents a time point and each column represents a feature, including energy value and co-activation characteristic value, resulting in a high-dimensional data matrix (cci; tc; cmi; m) for each muscle ri. i )×n;

[0058] S72. Standardize each feature in the multidimensional dataset to ensure that each feature has zero mean and unit variance;

[0059] S73. Calculate the covariance matrix of the standardized data from S72 to determine the correlation between the variables.

[0060] S74. Solve for the eigenvalues ​​and corresponding eigenvectors of the covariance matrix in S73. The eigenvalues ​​represent the variance contribution of each principal component, while the eigenvectors represent the direction of the principal components. Based on the magnitude of the eigenvalues, select the principal components whose cumulative contribution rate reaches 95%. Then, use the selected ri's k... i Each principal component constructs a feature vector, and each feature vector corresponds to the muscle activity state V = [v1, v2, ..., vk] at a given time point. i ].

[0061] S8. Draw the eigenvalue histogram for each principal component. The eigenvalues ​​of the principal components follow a standard normal distribution. Figure 5As shown, the threshold can be defined as the interval with the highest frequency of data occurrence. Once the threshold for each principal component is determined, the following logical rule is proposed to determine whether the muscle is activated at a specific time point, with the selected k... i If all principal component values ​​exceed their thresholds, the muscle is considered activated at that time point, resulting in the activation time sequence n. i The activation timing g1,...,g of different channels is obtained by reconstructing the signal (v1,v2,...,vi). i .

[0062] S9. Electromyographic electrodes are fixedly attached to the relevant muscle areas of the patient's left arm, and the electrical stimulation electrodes are held in the set positions. The electromyographic acquisition device collects surface electromyographic signals for subsequent analysis, and the activation sequence g1,...,g of multiple muscles in S8 is determined. i The stimulation parameters designed for the FES experiment include the frequency, amplitude, and pulse width of the current, to stimulate the muscles on the affected side and enable the patient's affected hand to complete the fist-clenching action.

[0063] Therefore, this invention provides a wavelet-enhanced co-activation method for multi-target temporal modulation of FES. By integrating time-frequency analysis and quantification of muscle co-activation characteristics, a comprehensive framework for muscle activity analysis is constructed. This not only deepens our understanding of muscle co-activation patterns, but also allows for flexible adaptation to changes in muscle activity by dynamically adjusting the energy threshold index, thereby more accurately predicting muscle activation trends.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A wavelet-enhanced synergistic activation method for FES multi-target time sequence regulation, characterized in that, The specific steps are as follows: S1. Acquire multi-channel surface electromyography (EMG) signals and bending angles, and preprocess the acquired EMG signals; S2. Perform wavelet denoising and reconstruction on the preprocessed electromyographic signals of each channel in S1; Two consecutive wavelet transforms were performed on each channel of the reconstructed electromyographic signal, and the time-frequency energy salient feature matrix was adaptively obtained by combining energy proportion and overlapping moving average window. The specific operation is as follows: The pre-processed single-channel electromyogram signal in S1 is decomposed by a fourth-order Daubechies 6-level wavelet, and a threshold of each signal is set based on a signal length and a logarithmic function to remove noise, and then wavelet coefficients of each frequency band are reconstructed to obtain a multi-channel reconstructed signal ; wherein, is the reconstructed signal Multi-channel reconstructed signal Perform a scale transformation from 1 to 128 to obtain the overall coefficient distribution of the electromyography signal for each channel. By calculating each The square of the value is used to obtain the reconstructed energy, and then the time-frequency scale energy matrix of the entire signal is obtained. Based on this, the energy percentage is calculated, that is, the percentage of each energy distribution is calculated. The proportion of each scale in the total energy was determined, and a proportion exceeding 10% was considered significant, thus achieving... An adaptive selection of energy-significant scales; among which... As a scale, For location, Select the i-th scale A significant energy scale; Multi-channel reconstructed signal using an overlapped moving average window Electromyography signals in each channel Divide the time scale and obtain Each electromyographic signal segment will Energy saliency scale as a sliding window The scale selection criteria after continuous wavelet transform of the post-electromyography (EMG) signal, the energy value of each salient scale after each window, and the corresponding EMG signal are used to construct the EMG signal. Time-frequency energy salient feature matrix ),in n It is an integer; S3. The preprocessed multi-channel electromyography (EMG) signals in S1 are combined in pairs to construct multiple muscle pairs. The EMG signals of each muscle pair are aligned and segmented using an overlapping sliding average window to obtain multiple EMG signal pairs. Multi-level co-activation feature indexes are calculated for each pair of EMG signals to obtain the intermuscular co-activation distribution feature matrix. S4. Construct a multi-channel high-dimensional matrix by combining the salient feature matrix of each single-channel time-frequency energy in S2 with the intermuscular coactivation matrix in S3. At the same time, reduce the dimension by the cumulative variance contribution rate and obtain the multi-dimensional time-frequency energy-intermuscular coactivation matrix. S5. For each component of the multi-dimensional time-frequency energy-muscle coactivation matrix of each channel in S4, a threshold is adaptively set according to statistical probability. An adaptive threshold logic combination rule is proposed to optimize and select the corresponding muscle activation state time-series vector. S6. Incorporate the activation time vectors of multiple muscles into the design of stimulation parameters for the FES experiment and conduct the corresponding experiments.

2. The wavelet-enhanced collaborative activation method for multi-target temporal modulation of FES according to claim 1, characterized in that, In S1, the experiment was designed to be conducted in the sagittal plane. During the hand clenching motion, multi-channel surface electromyography (EMG) signals of the forearm muscles and finger flexion angles were simultaneously acquired, and the acquired multi-channel EMG signals were preprocessed.

3. The wavelet-enhanced collaborative activation method for multi-target temporal modulation of FES according to claim 1, characterized in that, In S3, based on the preprocessed multi-channel reconstructed signal in S2 Multiple muscle pairs were constructed by pairwise combinations, and an overlapping sliding average window was applied based on the electromyographic signals of each muscle pair. Calculate the multi-layered co-activation feature index for each pair of electromyographic signals after each windowing, including the degree of overlap of muscle pair combinations. Time point correlation and the degree of information exchange This leads to a feature matrix of intermuscular coactivation distribution that includes all combinations. .

4. The wavelet-enhanced collaborative activation method for multi-target temporal modulation of FES according to claim 1, characterized in that, In S4, the time-frequency energy salient feature matrix of each single channel in S2 is... ) and the intermuscular coactivation matrix in S3 Build each muscle High-dimensional data matrix The most important features were extracted based on the correlation between variance and eigenvectors. Principal components with a cumulative feature variance contribution rate of 95% were selected, and the high-dimensional data matrix was reduced to a multi-dimensional time-frequency energy-myo-muscle co-activation matrix. The multi-dimensional time-frequency energy-myocardial co-activation matrix will serve as an energy threshold indicator; among which, This is the row number index.

5. The wavelet-enhanced collaborative activation method for multi-target temporal modulation of FES according to claim 1, characterized in that, In S5, the multi-dimensional time-frequency energy-myo-intermuscular co-activation matrix of each channel in S4 is used. Each component A threshold is adaptively set based on the maximum probability density; a combination rule is formulated for principal components that provide additional information related to muscle activation, considering the simultaneous influence of multiple principal components, and a threshold logic combination rule is proposed for the multi-dimensional time-frequency energy-muscle co-activation matrix. n Activation is determined at each point to assess muscle activation status. Muscles are considered activated only when the value of each principal component simultaneously exceeds its respective threshold, thus obtaining the activation sequence of different channels. ;in, Let be the activation timing vector of the i-th channel.

6. The wavelet-enhanced collaborative activation method for multi-target temporal modulation of FES according to claim 1, characterized in that, In S6, the activation sequence of multiple muscles is determined. The stimulation parameters designed for the FES experiment include the frequency, amplitude, and pulse width of the current, which stimulate the muscles on the affected side, enabling the patient's affected hand to complete the fist-clenching action.

Images