Data analysis method combining muscle collaborative analysis and muscle function network

Abstract

The invention relates to the technical field of data analysis, and discloses a muscle collaborative analysis and muscle function network-combined data analysis method, which comprises the following steps of S1, collecting surface electromyogram signals of each muscle during movement; s2, performing muscle cooperation mode analysis, integrating the collected surface electromyogram signal data by adopting non-negative matrix factorization, extracting a muscle cooperation mode, and quantizing the quality of a reconstruction matrix by using a variance interpretation rate; and S3, performing muscle function network analysis, obtaining normalized mutual information among muscle channels, taking the normalized mutual information as nonlinear correlation characteristic parameters of multi-channel surface myoelectricity, and constructing a muscle function network. According to the muscle collaborative analysis and muscle function network data analysis method based on the surface electromyogram signals, muscle collaborative analysis is carried out by using non-negative matrix factorization, the correlation between channels is quantized by using normalized mutual information, and the muscle function network is constructed; potential and valuable structured and modular information in the myoelectricity data set can be found.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a data analysis method combining muscle synergy analysis with muscle function network. Background Art

[0002] Electromyography (EMG) is an important type of biomechanical information generated when muscle activity is stimulated. It can directly reflect the electrical activity of muscles and provide a direct indicator of human movement intention. Among the many EMG technologies, surface electromyography (sEMG) has become an important tool for the study and application of motor nervous system functions due to its non-invasive measurement, simple operation and easy signal acquisition. sEMG signals record the changes in skin potential difference during muscle contraction by placing electrodes on the surface of the skin, and generate signals that can be used for further analysis through signal processing steps such as amplification, filtering, and rectification. These signals are widely used in many fields such as gesture recognition, prosthetic control, rehabilitation equipment design, muscle strength assessment, and remote-controlled robots, demonstrating their core role in modern sports science and engineering applications.

[0003] Although sEMG signals are widely used in motion analysis, traditional analysis methods usually only focus on the electromyographic activity of a single muscle or a single action, especially the change of absolute strength. This limited analysis method ignores the complex synergistic relationship between muscle groups and cannot fully evaluate the interaction and modular synergistic structure between multiple muscles during the movement, which is extremely unfavorable for understanding the integrity and coordination of movement and optimizing movement efficiency. Summary of the invention

[0004] The purpose of the present invention is to solve the problems existing in the prior art and to propose a data analysis method combining muscle synergy analysis and muscle functional network.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: The data analysis method combining muscle synergy analysis and muscle functional network includes the following steps: Step S1: collecting surface electromyographic signals of each muscle during movement; Step S2: muscle synergy pattern analysis, using non-negative matrix factorization to integrate the collected surface electromyographic signal data, extract the muscle synergy pattern, and use the variance explanation rate to quantify the quality of the reconstructed matrix; Step S3: repeating steps S1-S2, for each repeated experiment, calculating the cosine similarity between the extracted muscle synergy patterns, sorting the experiments according to the cosine similarity between the repeated experiments, and calculating the average synergy pattern of the sorted repeated experiments; Step S4: Calculate the cosine similarity between the average synergy pattern and the synergy pattern of each repeated experiment to ensure that the average synergy pattern can represent each repeated experiment; Step S5: Analyze the correlation characteristics between electromyographic channels through normalized mutual information, construct a muscle function network, and obtain a normalized mutual information matrix; Step S6: On the normalized mutual information matrix, a threshold is set to construct a weighted undirected network graph; Step S7: Calculate the characteristics of each node in the weighted undirected network to evaluate the node importance, the clustering coefficient to evaluate the network connectivity, analyze the network characteristics under different motion conditions, and draw a network diagram.

[0006] Preferably, the step S1 also includes preprocessing the collected surface electromyography signals, wherein the preprocessing process includes removing motion artifacts, filtering and denoising, rectifying, normalizing and taking linear envelopes, and then storing the processed data according to the classification of motion types.

[0007] Preferably, in step S2, the variance explanation rate is used to quantify the quality of the reconstructed matrix after non-negative matrix decomposition, so that the product of the two matrices approximates the original matrix, where the rank of one matrix is ​​k, and the k value is iterated until the variance explanation rate is greater than 90%, that is, the reconstructed matrix contains more than 90% of the original matrix information.

[0008] Preferably, in step S3, each experiment is sorted according to the cosine similarity between repeated experiments, and the average synergy pattern of the sorted repeated experiments is calculated, which specifically includes the following steps: Step S31: Calculate the cosine similarity matrix between multiple matrices in repeated experiments, and extract the similarity results of each pair of matrices; Step S32: aligning the collaborative patterns based on the maximum similarity principle, and rearranging the collaborative pattern matrix according to the alignment order; Step S33: Calculate the cosine similarity between the rearranged collaborative pattern matrices, and average the diagonal elements of the similarity matrix to quantify the collaborative consistency between the matrices; Step S34: Calculate the average value of the rearranged multiple cooperative mode matrices for further analysis and comparison.

[0009] Preferably, in step S6, the specific process of setting the threshold is: generating a series of thresholds, and calculating the average degree of the adjacency matrix under each threshold, and then obtaining a relationship curve between the threshold and the average degree, which can help determine the connectivity and sparsity of the network under different connection strengths, and is suitable for analyzing the changes in key characteristics of the network under different thresholds. According to the changes in key characteristics of the network under different thresholds, a suitable threshold is determined to ensure that the network has the desired average degree.

[0010] Preferably, in step S7, the node feature is the point degree, and the network feature is the clustering coefficient, which are used to evaluate the importance of each node in the network and the network connectivity respectively, so as to find the key muscles in the action process and evaluate the overall coordination connectivity of each muscle group under the action conditions.

[0011] Preferably, in step S5, analyzing the correlation characteristics between electromyographic channels by normalized mutual information further includes the following steps: Step S51: Perform standard check and preprocessing on input data; Step S52: calculating the joint distribution matrix of the signal according to the defined histogram edge; Step S53: Calculate the edge probability distribution of the signal and the corresponding entropy value; Step S54: calculating the mutual information between signals by accumulating the information in the joint probability distribution, and normalizing the mutual information; Step S55: Obtain the mutual information matrix of each repeated experiment and take the mean matrix.

[0012] Compared with the prior art, the present invention has the following beneficial effects: The present invention is based on the data analysis method of muscle synergy analysis and muscle function network of surface electromyographic signals. It uses non-negative matrix decomposition to perform muscle synergy analysis, uses normalized mutual information to quantify the relationship between muscle channels and constructs a muscle function network. It can discover potential and valuable structured and modular information in electromyographic data sets, and highlight the integrity and coordination of muscle groups in motion analysis. It not only helps to improve the accuracy and efficiency of motion analysis, but also optimizes motion patterns and improves motion performance, and has broad application prospects. It provides new perspectives and tools for research in the fields of rehabilitation medicine, sports science, and neural engineering, and also provides theoretical support and technical basis for practical applications such as prosthetic control and the design of intelligent rehabilitation equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a flow chart of the data analysis method combining muscle synergy analysis and muscle functional network proposed by the present invention. DETAILED DESCRIPTION

[0014] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0015] Reference Figure 1, the embodiment of the present application provides a data processing method for muscle synergy analysis and muscle functional network based on surface electromyographic signals, which conducts in-depth analysis of muscle synergy through non-negative matrix decomposition, and uses normalized mutual information to quantify the relationship between different muscle channels, thereby constructing a muscle functional network. This method can reveal the potential structured and modular information in electromyographic data, and emphasize the importance of muscle group integrity and coordination in motion analysis, helping to discover the deep-seated laws of muscle synergy.

[0016] Step S1: collecting surface electromyographic signals of each muscle during movement; Step S2: muscle synergy pattern analysis, using non-negative matrix factorization to integrate the collected surface electromyographic signal data, extract the muscle synergy pattern, and use the variance explanation rate to quantify the quality of the reconstructed matrix; This step introduces the synergistic module, because the synergistic module is the smallest functional unit of the central nervous system to control the musculoskeletal system to complete various complex movements. Muscle synergy theory can identify the key muscle groups that complete specific action tasks, namely synergists, by analyzing the synergistic effects of multiple muscles contracting simultaneously. Studies have shown that human movement can be modeled by the synergy of a small number of muscles with specific biomechanical functions. These synergists construct the basic modules of movement and simplify the complex control of multiple muscles by the nervous system.

[0017] At the same time, in muscle synergy analysis, non-negative matrix factorization (NMF) and principal component analysis (PCA) are commonly used dimensionality reduction methods. The NMF algorithm can decompose the high-dimensional electromyographic signal matrix into a set of non-negative basis matrices and coefficient matrices, thereby extracting characteristic patterns representing different synergies, which can reflect the contribution of each muscle in a specific action task. In addition, PCA, as another common data dimensionality reduction method, simplifies the complexity of the data by extracting the principal components of the signal and reveals the most representative change pattern in the data. However, since the principal components of PCA may contain negative values, it is difficult to directly interpret them as actual muscle activity patterns. The matrix factors obtained by NMF decomposition have non-negative restrictions, which makes the matrix results obtained by decomposition more practical electromyographic physiological significance. At the same time, NMF decomposition usually produces sparse results, that is, each gene pattern is mainly contributed by a few muscles. This sparsity is closer to the actual physiological situation, that is, in some actions, usually only a small number of muscles will have the main contribution. Therefore, non-negative matrix decomposition is more suitable for dimensionality reduction analysis of electromyographic data.

[0018] As a nonlinear association measurement method, mutual information does not rely on an established model and can accurately evaluate the linear or nonlinear dependence between two or more random variables. Mutual information has been widely used in the field of EEG, mainly for analyzing the association between different brain regions, information transmission and the construction of functional networks. In recent years, it has been gradually introduced into muscle functional network analysis. Compared with the traditional Pearson correlation coefficient, mutual information has a wider applicability and stronger robustness, and thus has been increasingly used in complex biological signal analysis. Through mutual information analysis, the complex nonlinear relationship between different muscles can be revealed, and a functional network reflecting muscle correlation can be constructed.

[0019] Step S3: repeating steps S1-S2, for each repeated experiment, calculating the cosine similarity between the extracted muscle synergy patterns, sorting the experiments according to the cosine similarity between the repeated experiments, and calculating the average synergy pattern of the sorted repeated experiments; Step S4: Calculate the cosine similarity between the average synergy pattern and the synergy pattern of each repeated experiment to ensure that the average synergy pattern can represent each repeated experiment; Step S5: Analyze the correlation characteristics between electromyographic channels through normalized mutual information, construct a muscle function network, and obtain a normalized mutual information matrix; Step S6: On the normalized mutual information matrix, a threshold is set to construct a weighted undirected network graph; Step S7: Calculate the characteristics of each node in the weighted undirected network to evaluate the node importance, the clustering coefficient to evaluate the network connectivity, analyze the network characteristics under different motion conditions, and draw a network diagram.

[0020] In this embodiment, the step S1 further includes: Preprocessing of surface EMG signals: First, the surface EMG signals of each muscle during movement are collected and preprocessed separately, including removing motion artifacts, filtering and denoising, rectifying, normalizing, and taking linear envelopes. Then the processed data are classified and stored according to the type of movement. Specifically, a fourth-order Butterworth high-pass filter with a cutoff frequency of 20Hz is used to remove motion artifacts and take the absolute value of the EMG signal; the maximum value during the movement is taken for further normalization, and finally the EMG data envelope is extracted for collaborative feature extraction or calculation of functional connections.

[0021] Among them, in step S2, non-negative matrix decomposition is used to integrate the EMG data, and the variance explanation rate is used to quantify the quality of the reconstructed matrix, specifically: It is used to decompose a non-negative matrix H into two non-negative matrices W and H. First, the code checks the input matrix V to ensure that its elements are non-negative and the sum of each row is not zero. Then, the matrices W and H are initialized to random non-negative values, and the values ​​of W and H are updated through an iterative optimization algorithm so that the matrix W×H can approximate V. During the iteration process, the square error (i.e., the loss function) between V and W×H is minimized through the update rule, and the error value of each iteration is recorded until the set maximum number of iterations is reached. The formula for the above calculation can be expressed as: Given a non-negative matrix V, the goal of NMF is to find two non-negative matrices W and H such that: (1) Where: V is the original matrix, W is the feature matrix, and H is the coefficient matrix.

[0022] The formula for updating the matrix W is: (2) Where: ⊙ represents the multiplication of elements one by one, H T is the transposed matrix of H.

[0023] The formula for updating the matrix H is: (3) Where: ⊙ represents the multiplication of elements one by one, W T is the transposed matrix of W.

[0024] The calculation formula of the loss function loss represents the sum of the squares of the Euclidean distances between the matrix V and its approximation matrix WH: (4) Evaluate the impact of the number of decomposed features on the accuracy of signal reconstruction. The code first divides the signal data into different parts (left and right, top and bottom), applies NMF decomposition to each part, and calculates the total variance and reconstruction error, and then calculates the variance explained (vaf), that is, the accuracy of signal reconstruction. Finally, NMF decomposition is performed on the entire signal data, and vaf is calculated in the same way to quantify the reconstruction quality of the entire signal. The formula for the above calculation can be expressed as: Calculation of total variance (total_variance): (5) Where emg_envelope represents the electromyographic signal envelope.

[0025] Calculation of reconstruction error (reconstruction_error): (6) Substituting equations (5) and (6) into equation (7) yields the variance explained (vaf): (7) In step S3, the experiments are sorted according to the cosine similarity between the synergy patterns of the repeated experiments, and the average synergy pattern of the sorted repeated experiments is calculated, which includes the following steps: Take 3 repeated experiments as an example: Step S31: Calculate the cosine similarity matrix between the three collaborative pattern matrices, and extract the similarity results of each pair of matrices therefrom.

[0026] The calculation formula of cosine similarity is: (8) Where 'i' and 'j' are the indices of the column vectors in matrix1 and matrix2 respectively. For each column vector vec1 from matrix1 and column vector vec2 from matrix2, their cosine similarity is calculated.

[0027] Step S32: aligning the collaborative patterns based on the maximum similarity principle, and rearranging the second and third matrices according to the alignment order.

[0028] Step S33: Calculate the cosine similarity between the rearranged collaborative pattern matrices and average the diagonal elements of the similarity matrix to quantify the collaborative consistency between the matrices.

[0029] Step S34: Calculate the average value of the three aligned collaborative pattern matrices for further analysis and comparison.

[0030] In step S5, the normalized mutual information analysis algorithm includes: It combines the histogram method, joint probability distribution, entropy and mutual information calculation. First, the code calculates the joint distribution matrix of the two input data by the histogram method, and then normalizes it to get the joint probability distribution. Next, the respective marginal entropies and the mutual information between the two are calculated, and finally the mutual information is normalized to get NMI, which is used to quantify the mutual dependence between the two variables. First, the joint distribution matrix of the two input vectors is calculated using the histogram method (histcounts2), the data is divided into specified intervals (bins) and the frequency is counted. Next, the matrix is ​​normalized to calculate the joint probability distribution. Then, the marginal entropy of the two input vectors (H(X) and H(Y)), that is, the individual uncertainty of each variable, is calculated using the formula of information theory. Next, the mutual information (MI) of the two variables is calculated to measure the dependence between them. Finally, the mutual information is divided by the smaller marginal entropy to get the normalized mutual information (NMI), which is used to quantify the relationship and degree of dependence between the two variables. The formula for the above calculation is as follows: Calculate the joint distribution matrix (Cxy), that is, the histogram calculation based on the two-dimensional data (x, y) under the given boundaries eX and eY: (9) Where eX is the boundary of the marginal histogram of x, eY is the boundary of the marginal histogram of y, and Cxy(i, j) is the frequency corresponding to the i-th row and j-th column, indicating the number of samples of x in the i-th interval and y in the j-th interval.

[0031] Calculation of marginal probability distribution, calculate the marginal probability distribution px and py of x and y: (10) (11) To calculate entropy, substitute equation (10) and equation (11) into equation (12) and equation (13) respectively to calculate the entropy H(X) and H(Y) of x and y: (12) (13) where ϵ is a very small constant used to prevent the logarithm from being zero (i.e., to prevent undefined cases in the calculation).

[0032] Import equations (10) and (11) into equation (14) to calculate the mutual information MI(X;Y) of x and y: (14) By introducing equation (14), equation (12) and equation (13) into equation (15), we can obtain NMI(X;Y): (15) Wherein in step S6, the topological structure of the network is analyzed by systematically adjusting the threshold and calculating the corresponding average degree. Specifically, it generates a series of thresholds, and calculates the average degree of the adjacency matrix under each threshold, and then depicts the relationship curve between the threshold and the average degree. It can help determine the connectivity and sparsity of the network under different connection strengths, and is particularly suitable for analyzing the changes in key characteristics of the network under different thresholds, such as determining a suitable threshold to ensure that the network has the desired average degree. When calculating the average point degree of each node, by traversing each threshold, the elements in the matrix are compared with the current threshold, a binary matrix is ​​generated, and then the degree of each row (corresponding to each node) is calculated, and the average value is taken as the average degree of the node under the threshold.

[0033] The formula for the above calculation is as follows: Calculate the point degree: (16) In the formula, the point degree k i Represents the number of edges associated with node i.

[0034] Substitute equation (16) into equation (17) to calculate the average degree of all nodes and record it as the average degree of the entire network: (17) Wherein in step S7, the importance of each node in the network and the network connectivity are evaluated respectively by point degree and clustering coefficient, so as to find out the key muscles in the action process and evaluate the overall network coordination connectivity under the action condition. The calculation of point degree can be seen in the specific description of step S6. The calculation of clustering coefficient is the average clustering coefficient of the undirected binary graph. By traversing each node, identifying its neighboring nodes, calculating the actual number of edges and the number of edges that may exist between neighboring nodes, and then calculating the clustering coefficient of each node. Finally, by averaging the clustering coefficients of all nodes, the average clustering coefficient of the entire graph is obtained, the local density of the nodes in the graph is quantified, and the network graph is finally drawn.

[0035] Step S7 is specifically as follows: Step S71: Thresholding the adjacency matrix: setting the elements in the adjacency matrix avg_mi_matrix that are less than the specified threshold to zero, removing the connections below the threshold, and setting the diagonal elements to zero to exclude self-connections; Step S72: constructing a network graph: using the processed adjacency matrix to create an undirected graph G; Step S73: Use the force layout algorithm to visualize the network graph, set the color according to the node degree (number of connections), the higher the degree, the darker the color, set the size of the node according to the node degree, and set the width and color of the edge according to the weight of the edge, where the weight of the edge is the mutual information correlation index between two nodes, and the edge width and color are associated with the weight through linear mapping; Step S74: Calculate the degree of each node in the network. The degree is calculated as shown in formula (16) to measure its importance in the network. Step S75: Calculate the average clustering coefficient of the entire network to evaluate the local density of nodes in the network, observe the network connectivity under the action condition, and divide the muscle group modules according to the local network.

[0036] The formula for the above calculation is as follows: Calculate the clustering coefficient of a single node (c i ): (18) In the formula, E i is the connection ratio between neighboring nodes of node i, k i (k i -1) / 2 is the E in the undirected network i The maximum value of .

[0037] Substitute the result of formula (18) into formula (19) to calculate the network clustering coefficient (c): (19) In the formula, n is the number of nodes.

[0038] Therefore, the present invention provides a data processing method for muscle synergy analysis and muscle functional network based on surface electromyographic signals, which deeply analyzes muscle synergy through non-negative matrix decomposition, and uses normalized mutual information to quantify the relationship between different muscle channels, thereby constructing a muscle functional network. This method can reveal the potential structured and modular information in electromyographic data, emphasize the importance of muscle group integrity and coordination in motion analysis, and help discover the deep-seated laws of muscle synergy.

[0039] The present invention provides a data processing method for muscle synergy analysis and muscle function network based on surface electromyographic signals. Combined with muscle synergy analysis and muscle function network analysis, it can comprehensively quantify the interaction between key muscles and muscle channels under a certain action condition, and provide an in-depth understanding of the overall characteristics of the action. It can be used not only for the analysis of simple movements such as walking, standing up, sitting down, etc., but also for running, dance movements, and qigong movements to explore the laws of synergy and improve exercise efficiency. At the same time, in the field of rehabilitation medicine, it is helpful for the exploration of the therapeutic mechanism of certain exercise therapies, interpreting the therapeutic effects of the exercise therapy on patients from the perspective of biomechanical mechanisms, simplifying and optimizing the therapy, and analyzing the muscle coordination patterns of patients during daily life activities according to the method provided by the present invention, so as to carry out personalized training and treatment.

[0040] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A data analysis method combining muscle synergy analysis and muscle functional network, characterized in that: The steps include: Step S1: collecting surface electromyographic signals of each muscle during movement; Step S2: muscle synergy pattern analysis, using non-negative matrix factorization to integrate the collected surface electromyographic signal data, extract the muscle synergy pattern, and use the variance explanation rate to quantify the quality of the reconstructed matrix; Step S3: repeating steps S1-S2, for each repeated experiment, calculating the cosine similarity between the extracted muscle synergy patterns, sorting the experiments according to the cosine similarity between the repeated experiments, and calculating the average synergy pattern of the sorted repeated experiments; Step S4: Calculate the cosine similarity between the average synergy pattern and the synergy pattern of each repeated experiment to ensure that the average synergy pattern can represent each repeated experiment; Step S5: analyzing the correlation characteristics between electromyographic channels by normalized mutual information to obtain a normalized mutual information matrix to construct a muscle function network; Step S6: On the normalized mutual information matrix, a threshold is set to construct a weighted undirected network graph; Step S7: Calculate the characteristics of each node in the weighted undirected network to evaluate the node importance, the clustering coefficient to evaluate the network connectivity, analyze the network characteristics under different motion conditions, and draw a network diagram.

2. The data analysis method combining muscle synergy analysis and muscle functional network according to claim 1, characterized in that: The step S1 also includes preprocessing the collected surface electromyography signals, wherein the preprocessing process includes removing motion artifacts, filtering and denoising, rectifying, normalizing and taking linear envelopes, and then storing the processed data according to the classification of motion types.

3. The data analysis method combining muscle synergy analysis and muscle functional network according to claim 1, characterized in that: In step S2, the variance explanation rate is used to quantify the quality of the reconstructed matrix after non-negative matrix decomposition, so that the product of the two matrices approximates the original matrix, where the rank of one matrix is ​​k, and the k value is iterated until the variance explanation rate is greater than 90%, that is, the reconstructed matrix contains more than 90% of the original matrix information.

4. The data analysis method combining muscle synergy analysis and muscle functional network according to claim 2, characterized in that: In step S3, each experiment is sorted according to the cosine similarity between repeated experiments, and the average synergy pattern of the sorted repeated experiments is calculated, which specifically includes the following steps: Step S31: Calculate the cosine similarity between multiple matrices in repeated experiments, and extract the similarity results of each pair of matrices; Step S32: aligning the collaborative patterns based on the maximum similarity principle, and rearranging the collaborative pattern matrix according to the alignment order; Step S33: Calculate the cosine similarity between the rearranged collaborative pattern matrices, and average the diagonal elements of the similarity matrix to quantify the collaborative consistency between the matrices; Step S34: Calculate the average value of the rearranged multiple cooperative mode matrices for further analysis and comparison.

5. The data analysis method combining muscle synergy analysis and muscle functional network according to claim 1, characterized in that: In step S6, the specific process of setting the threshold is: generating a series of thresholds, and calculating the average degree of the adjacency matrix under each threshold, and then obtaining a relationship curve between the threshold and the average degree, which can help determine the connectivity and sparsity of the network under different connection strengths, and is suitable for analyzing the changes in key characteristics of the network under different thresholds. According to the changes in key characteristics of the network under different thresholds, a suitable threshold is determined to ensure that the network has the desired average degree.

6. The data analysis method combining muscle synergy analysis and muscle functional network according to claim 1, characterized in that: In step S7, the node feature is the point degree, and the network feature is the clustering coefficient, which are used to evaluate the importance of each node in the network and the network connectivity respectively, so as to find the key muscles in the action process and evaluate the overall coordination connectivity of each muscle group under the action conditions.

7. The data analysis method combining muscle synergy analysis and muscle functional network according to claim 1, characterized in that: In the step S5, the correlation characteristics between the electromyographic channels are analyzed by normalized mutual information, and the following steps are also included: Step S51: Perform standard check and preprocessing on input data; Step S52: calculating the joint distribution matrix of the signal according to the defined histogram edge; Step S53: Calculate the edge probability distribution of the signal and the corresponding entropy value; Step S54: calculating the mutual information between signals by accumulating the information in the joint probability distribution, and normalizing the mutual information; Step S55: Obtain the mutual information matrix of each repeated experiment and take the mean matrix.

Images