A method for analyzing muscle dynamic synergy based on dynamic community detection

Abstract

The present invention proposes a muscle dynamic synergy analysis method based on dynamic community detection. First, the surface electromyographic signals of the human body during movement are collected to construct a time-varying electromyographic functional network; then, the muscle community clustering of the single-layer electromyographic functional network is detected to obtain the muscle communities of each layer, and the muscle communities are arranged in the order of corresponding time snapshots to form a community set; a resolution parameter is set to adjust the sensitivity of each layer of the electromyographic functional network to the community, and a coupling parameter is set to adjust the coupling degree between layers; then, the modularity is optimized with maximizing the multi-layer modularity as an indicator; finally, the clusters of strongly interconnected nodes in the electromyographic functional network are divided into different modules, and the node clusters within the same module are regarded as muscle groups with strong synergy at this moment. As the muscle synergy changes dynamically over time, this change process is revealed through dynamic community detection, thereby realizing dynamic decoding of muscle synergy in movement changes.

Description

Technical Field

[0001] The present invention relates to the technical field of bioelectric signal processing, and in particular to a method for analyzing muscle dynamic synergy based on dynamic community detection. Background Art

[0002] During exercise, the desired movement requires the collaboration of muscles and the coordinated activation of numerous muscles. Decoding muscle synergy during movement helps us understand how the central nervous system controls the generation, execution, and coordination of human movement. Analyzing muscle synergy during exercise not only helps athletes develop targeted training programs but also provides rehabilitation strategies for patients with motor dysfunction caused by stroke and spinal cord injury. Therefore, analyzing dynamic muscle synergy during exercise is of great significance for assessment in sports medicine and rehabilitation medicine.

[0003] Currently, commonly used muscle synergy analysis methods include independent component analysis (ICA), principal component analysis (PCA), and non-negative matrix factorization (NMF). Independent component analysis and PCA often impose restrictive conditions such as mutual orthogonality or statistical independence during matrix decomposition, making them ineffective in practical applications. In comparison, the non-negative constraints of NMF easily satisfy the assumptions of muscle synergy analysis, and the decomposition results are more physiologically interpretable, thus attracting widespread attention from scholars. However, muscle synergy often occurs during exercise and is not limited to the transient process of a particular movement. The NMF decomposition algorithm cannot update synergy changes in a timely manner, and is therefore insufficient in analyzing dynamic muscle changes. Therefore, it is necessary to establish an analysis framework that covers the entire exercise process without losing detail to decode the dynamic synergy changes of muscles. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for analyzing muscle dynamic synergy based on dynamic community detection, so as to solve the problem that existing theories cannot analyze the dynamic synergy of muscles during exercise. By constructing a time-varying electromyographic network during exercise, introducing a dynamic community detection algorithm, analyzing the community changes of the electromyographic network with time-varying processes, further extracting muscle group clusters, and explaining the dynamic synergy of muscles from a macro perspective.

[0005] The present invention is achieved through the following technical solutions:

[0006] A method for analyzing muscle dynamic synergy based on dynamic community detection includes the following steps:

[0007] Step 1: Collect surface electromyographic signals of the human body during movement, extract time domain eigenvalues ​​through a sliding time window after preprocessing, and construct a time-varying electromyographic functional network;

[0008] Step 2: Detect muscle community clustering in the single-layer EMG functional network;

[0009] Step 3: Perform step 2 on each layer of the EMG functional network divided by the time window to obtain the muscle community of each layer, and arrange them in the order of the corresponding time snapshots to form a community set;

[0010] Step 4: Set the resolution parameter γ to adjust the sensitivity of each layer of the EMG functional network to the community; add the coupling parameter δ between the layers in the community set to adjust the coupling between the layers;

[0011] Step 5: Optimize the size of modularity by maximizing multi-layer modularity; map the module in one time window to itself in the next time window. The adjacency matrix of the electromyographic functional network in one time window is linked to the adjacency matrix in the adjacent time window through the identification edge. The identification edge in the adjacent time window is connected to itself in the adjacent time window. This identification link executes all nodes, identifies the modules and their time changes by maximizing the multi-layer modularity quality function, and maximizes the multi-layer modularity quality function Q. multslice The expression is as follows:

[0012]

[0013] Among them, A ijs is the weight of the edge between node i and node j in time window s; the community assignment of node i in layer s is g is , the community assignment of node j in layer r is g jr , if g is =g jr , then δ(g is ,g jr )=1, otherwise it is 0; the total edge weight is where K js =k js +c js is the strength of node j in layer s, k js is the intra-layer strength of node j in layer s, c js =∑ r C jsr is the inter-layer strength of node j in layer s, m s =Σ j k js ; parameter γ s is the structural resolution parameter of the s layer, which is used to adjust the number of communities identified, δ sr It represents the inter-layer coupling parameter between the s layer and the r layer, which is used to adjust the coupling size between the layers and indirectly affects the size of the community;

[0014] Step 6: Set the parameter domain for the two parameters mentioned in step 4, and perform multiple iterations of step 5 in the parameter domain to prune and prioritize the different community structures identified in multiple runs, and determine the optimal resolution parameter and coupling parameter by adjusting the mutual information;

[0015] Step 7: Through the operations of the first six steps, the dynamic community changes of muscle groups during exercise are determined. Finally, the clusters of strongly interconnected nodes in the electromyographic functional network are divided into different modules. These modules change over time as the movement process changes. The nodes and edges of the same module are colored with different colors. Each module represents a group of nodes that are highly connected to each other and sparsely connected to the nodes of other modules, and their participation is transferred between modules over time.

[0016] The node clusters within the same module are regarded as muscle groups with strong synergy at this moment. As the muscle synergy changes dynamically over time, dynamic community detection is used to reveal this change process and realize dynamic decoding of muscle synergy in movement changes.

[0017] In the above technical solution, in step 1, the collected original signals of multiple channels are subjected to data preprocessing operations such as filtering, rectification, and normalization.

[0018] In the above technical solution, in step 1, four time domain eigenvalues ​​in the continuous time window are extracted by sliding the time window, namely: maximum amplitude, root mean square, integrated electromyography, and absolute value average; the four extracted time domain eigenvalues ​​are constructed into a feature matrix T it , calculate the Pearson correlation between each pair of EMG channels in the feature matrix; then establish an undirected weighted adjacency matrix with muscle areas as nodes and Pearson correlations between EMG channels as edges; divide the EMG signals of the human body movement process into continuous time windows through a sliding time window, extract the adjacency matrix in each time window respectively, and combine them into an overlapping adjacency matrix set in order to obtain a time-varying EMG functional network.

[0019] In the above technical solution, the specific steps of step 2 are as follows:

[0020] 2.1 Extract the undirected weighted adjacency matrix of the single-layer EMG functional network as the initial data set; regard each node in the EMG functional network as a community, and the number of communities is consistent with the number of nodes;

[0021] 2.2 Merge each node with its adjacent nodes in turn and calculate the modularity gain ΔQ;

[0022] 2.3 Determine whether the modularity gain ΔQ is greater than 0. If it is greater than 0, place the node in the community where the adjacent node with the largest ΔQ is located.

[0023] 2.4 Repeat the iterative step 2.3 and perform the same operation on all nodes in the EMG functional network in turn until ΔQ no longer changes;

[0024] 2.5 Merge and compress the existing communities into a new super node. The weights within the original community are converted into the self-loop weights of the new node, and the weights between the original communities are converted into the edge weights of the new node.

[0025] 2.6 Repeat steps 2.2-2.4 until the final stabilization is completed.

[0026] In the above technical solution, in step 2, the modularity gain ΔQ is calculated as follows:

[0027]

[0028] Among them, ∑in is the total edge weight of the community composed of all adjacent nodes, that is, the correlation value of the electromyographic channel. Initially, one node is one community, and its value is the default value 1; ∑tot is the sum of the weights of the links associated with each node community, that is, the node degree; k i is the sum of the weights of the links associated with node i, k i，in is the sum of links from node i to the associated community; m is the total weight of all paths in the network.

[0029] In the above technical solution, in step 6, the calculation expression of mutual information is:

[0030]

[0031] where MI(X,Y) is the mutual information between random variables X and Y, and H(X) is the entropy of random variable X. The expected value E(MI(X,Y)) is computed over random partitions sampled from a hypergeometric null distribution. An AMI of 1 between two partitions indicates perfect agreement, and a value of 0 indicates a partition that is worse than random.

[0032] In the above technical solution, the specific steps of step 6 are:

[0033] 6.1 Given a set of partitions, determine the modular optimization domain for each partition;

[0034] 6.2 Abandon partitions with empty domains to obtain a subset of partitions that maintain potential optimization in the specified parameter domain;

[0035] 6.3 Use mutual information to select the optimized partitions, retain the subset with AMI>0.7, and determine it as the final community structure. The corresponding parameters at this time are the optimal parameters.

[0036] The advantages and beneficial effects of the present invention are:

[0037] The application scenarios of this invention are not limited to the dynamic coordination between upper and lower limb muscle groups; it is universally applicable to multi-muscle sports. Compared with existing non-negative matrix factorization algorithms for extracting muscle synergy, this invention takes a different approach, explaining dynamic muscle synergy by constructing a muscle functional network. This method can not only detect real-time changes in muscle synergy, but also identify the most contributing muscles in different motion states. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of a muscle dynamic synergy analysis method based on dynamic community detection in the present invention.

[0039] Figure 2 This is a time-varying network community connection diagram in an embodiment of the present invention.

[0040] Figure 3 This is a dynamic muscle coordination community change diagram in an embodiment of the present invention.

[0041] For ordinary technicians in this field, other relevant drawings can be obtained based on the above drawings without any creative work. DETAILED DESCRIPTION

[0042] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention are further described below with reference to specific embodiments.

[0043] A method for analyzing muscle dynamic synergy based on dynamic community detection includes the following steps:

[0044] Step 1: Collect the surface electromyographic signals of the lower limbs of the human body, extract the time domain eigenvalues ​​after preprocessing, and construct a time-varying electromyographic functional network.

[0045] 1.1 Surface electromyographic signals (EMGs) were collected from subjects during lower limb flexion and extension movements using a lower limb rehabilitation robot. Subjects completed four repeated experiments in active (with the rehabilitation robot providing resistance) and passive (with the rehabilitation robot providing assistance) modes. During exercise, 32 channels of EMG information were collected in real time from the lower limb region. Because muscles such as the rectus femoris, vastus lateralis, quadriceps femoris, and triceps surae are longer and larger in area, a slightly larger number of EMG sensors were placed in these locations than in other areas. The raw signals from these 32 channels were filtered, rectified, and normalized using a 20-450 Hz bandpass filter for data preprocessing.

[0046] 1.2 After data preprocessing, a sliding time window was used to extract four time-domain eigenvalues ​​within consecutive time windows: maximum amplitude, root mean square (RMS), integrated EMG, and absolute value average. To balance information integrity and the time-varying nature of the process, a sliding window length of h = 800 ms and a step size of S = 160 ms were selected to ensure 80% overlap between consecutive time windows.

[0047] Maximum amplitude: max{x it (1),x it (2),...,x it (h)}

[0048] RMS:

[0049] Integrated EMG:

[0050] Absolute value average:

[0051] Here, i represents the EMG channel and h represents the amount of EMG signal data.

[0052] 1.3 The four eigenvalues ​​extracted in step 1.2 are constructed into a feature matrix T it , calculate the Pearson correlation between each pair of EMG channels in the feature matrix as the edge weight of the EMG functional network. No threshold division is performed on the weights, that is, no binarization is performed.

[0053] T it =(Mav it ,RMS it ,IEMG it ,MA it )

[0054] Where i = (1, 2, 3…, n), n = 32, representing 32 channels, and t represents the time window sequence.

[0055] Normalize the feature matrix to get:

[0056] T it '=(mav it ,rms it ,IEMG it ,ma it )

[0057] The feature matrix T i ' t The eigenvector of P (P=1,2,3,4), H1=mav, H2=rms, H3=IEMG, H4=mav. Calculate H pThe Pearson correlation coefficient r between any two column vectors pij :

[0058]

[0059] Where k represents the amount of data in the time window t, H pi (t) represents the p-th eigenvalue of the i-th node in the t-th time window, <H pi > represents the average value of k elements. Finally, the Pearson correlation between EMG channel i and channel j is:

[0060]

[0061] 1.4 An undirected weighted adjacency matrix was established with muscle regions as nodes and Pearson correlations between EMG channels as edges. The adjacency matrices within all continuous time windows were extracted and stacked in chronological order to construct a time-varying EMG functional network for the lower limbs.

[0062] Step 2: Detect muscle community clustering in the single-layer EMG functional network. The specific steps are as follows:

[0063] 2.1 Extract the undirected weighted adjacency matrix of the single-layer electromyographic functional network as the initial data set A; regard each node in the electromyographic functional network as a community, and the number of communities is consistent with the number of nodes.

[0064] 2.2 Each node is merged with its adjacent nodes in turn to calculate the modularity gain ΔQ. The modularity gain ΔQ is calculated as follows:

[0065] Among them, ∑in is the total edge weight of the community composed of all adjacent nodes, that is, the correlation value of the electromyographic channel. Initially, one node is one community, and its value is the default value 1; ∑tot is the sum of the weights of the links associated with each node community, that is, the node degree; k i is the sum of the weights of the links associated with node i, k i , in is the sum of links from node i to the associated community; m is the total weight of all paths in the network.

[0066] 2.3 Determine whether the modularity gain ΔQ is greater than 0. If it is greater than 0, place the node in the community where the adjacent node with the largest ΔQ is located.

[0067] 2.4 Repeat the iterative step 2.3 and perform the same operation on all nodes in the electromyographic functional network in turn until ΔQ no longer changes.

[0068] 2.5 Merge and compress the already formed communities into a new super node. The weights within the original community are converted into the self-loop weights of the new node, and the weights between the original communities are converted into the edge weights of the new node.

[0069] 2.6 Repeat steps 2.2-2.4 until the final stabilization is completed.

[0070] Step 3: Perform step 2 on each layer of the electromyographic functional network divided by the time window to obtain the muscle community C of each layer. i , i represents the number of network layers, and the community set is formed according to the order of the corresponding time snapshots, that is, C={C0,C1,C2,…,C N}.

[0071] Step 4: In C i The resolution parameter γ is set in to adjust the sensitivity of each layer of the electromyographic functional network to the community, and the coupling parameter δ is added between the layers in C to adjust the coupling between the layers.

[0072] Step 5: Integrate the detection process from step 2 to step 4. Based on step 2, optimize the size of the modularity to maximize the multi-layer modularity as the indicator. The steps are as follows:

[0073] 5.1 Identify modules in each time window.

[0074] 5.2 Map the module in a time window to itself in the next time window. The adjacency matrix of the electromyographic functional network in a time window is linked to the adjacency matrix in the adjacent time window through the identification edge. The identification edge in the adjacent time window is connected to itself in the adjacent time window. This identification link executes all nodes and identifies the modules and their time changes by maximizing the multi-layer modular quality function, maximizing the multi-layer modular quality function Q. multslice The expression is as follows:

[0075]

[0076] Among them, A ijs is the weight of the edge between node i and node j in time window s; the community assignment of node i in layer s is g is , the community assignment of node j in layer r is g jr , if g is =g jr , then δ(g is ,g jr )=1, otherwise it is 0; the total edge weight is where K js =k js +c js is the strength of node j in layer s, k jsis the intra-layer strength of node j in layer s, c js =∑ r C jsr is the inter-layer strength of node j in layer s, m s =∑ j k js ; Parameter γs is the structural resolution parameter of the s layer, which is used to adjust the number of communities identified, δ sr It represents the inter-layer coupling parameter between the s layer and the r layer, which is used to adjust the coupling size between layers and indirectly affects the size of the community.

[0077] Step 6: Set the parameter domain for the two parameters mentioned in step 4, and perform multiple iterations of step 5 in the parameter domain to prune and prioritize the different community structures identified in multiple runs, and determine the optimal resolution parameter γ and coupling parameter δ by adjusting the mutual information AMI. The calculation expression of mutual information AMI is:

[0078]

[0079] where MI(X,Y) is the mutual information between random variables X and Y, and H(X) is the entropy of random variable X. The expected value E(MI(X,Y)) is computed over random partitions sampled from a hypergeometric null distribution. An AMI of 1 between two partitions indicates perfect agreement, and a value of 0 indicates a partition that is worse than random.

[0080] The specific steps of step 6 are:

[0081] 6.1 Given a set of partitions, determine the modularity optimization domain for each partition.

[0082] 6.2 Discard partitions with empty domains to obtain a subset of partitions that maintain potential optimization in the specified parameter domain.

[0083] 6.3 Use mutual information to select the optimized partitions, retain the subset whose AMI meets the requirements, and determine it as the final community structure. The corresponding parameters at this time are the optimal parameters.

[0084] In this embodiment, the resolution parameter and the coupling parameter are optimized by setting a parameter domain. After 2000 iterative calculations, the parameter range when AMI>0.7 is defined, and the optimal resolution parameter value is determined to be 0.68, and the coupling parameter value is determined to be the default value of 1.

[0085] This approach has the following benefits:

[0086] (1) It solves the matching problem under adjacent functional EMG network slices and defines which module in one time window is the same as which module in another time window.

[0087] (2) Providing adjustment parameters can obtain information about the fine and coarse topological scales of network reconstruction, as well as the fine and coarse time scales of network reconstruction, and can artificially interfere with the division of dynamic communities.

[0088] (3) Unlike statistically driven methods based on principal component analysis or independent component analysis, modules do not need to be completely independent of each other. Instead, there can be edges with non-zero weights between the nodes of one module and the nodes of another module, which simplifies the operation process.

[0089] Step 7: Through the operations of the first six steps, the dynamic community changes of muscle groups during exercise are determined. Finally, the clusters of strongly interconnected nodes in the electromyographic functional network are divided into different modules. These modules change over time as the movement process changes. The nodes and edges of the same module are colored with different colors. Each module represents a group of nodes that are highly connected to each other and sparsely connected to the nodes of other modules, and their participation is transferred between modules over time.

[0090] The node clusters within the same module are regarded as muscle groups with strong synergy at this moment. As the muscle synergy changes dynamically over time, dynamic community detection is used to reveal this change process and realize dynamic decoding of muscle synergy in movement changes.

[0091] In this embodiment, if Figure 2 Each layer of the functional network contains 32 nodes, and each node is distinguished by a different color. The horizontal axis represents time, and the layers are arranged in chronological order, which corresponds to the lower limb movement process. The vertical axis represents 1-32 nodes, and the lines between the nodes represent belonging to a community. If there are lines between the nodes, a community relationship is formed. Otherwise, there is no community affiliation. In functional networks 10 and 18, the communities are closely connected, and the number of communities is significantly increased compared to other slices. Combined with the specific movement process, it is not difficult to conclude that functional network 10 represents the community division in the straight state, and functional network 18 represents the electromyographic functional network community division in the bent state.

[0092] Depend on Figure 2 It can be seen that there are more communities in the straightened state than in the bent state, and they are mainly concentrated between the community connections of nodes 18-32 (corresponding to the calf muscle group). From the perspective of sports physiology, when the legs are straight, the thigh and calf muscles are both activated, and the electromyographic signals are strong. When the legs are bent, the force-generating muscles are mainly concentrated in the thigh muscles, supplemented by some of the calf muscles.

[0093] In this embodiment, 20 layers of electromyographic functional network slices were extracted, and further, the dynamic community changes of the 20 time-varying adjacency matrices were analyzed and detected, such as Figure 3 . Figure 3Different colors represent different communities, showing how each node's community affiliation changes over time throughout the entire process. Nodes 1, 8, 9, 12, and 25 remained unchanged from beginning to end, indicating that the muscles they represent are dominant, stable, and less susceptible to transfer during the collaborative process.

[0094] In this embodiment, the dynamic coordination relationship of the lower limb muscles during movement is fully explained by detecting the dynamic coordination of the lower limb muscles during movement and combining the analysis of static and dynamic processes.

[0095] This invention creatively explains the complex dynamic behavior of the human body from the perspective of complex networks, overcomes the limitations of traditional matrix decomposition methods, and models a theoretical framework for muscle synergy analysis, which is conducive to the subsequent dynamic synergy analysis of different types of movements and different muscles.

[0096] The above-described embodiment provides a detailed description of the technical solution of the present invention. It is worth noting that this embodiment is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications and improvements made within the scope of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for analyzing muscle dynamic synergy based on dynamic community detection, characterized in that: The following steps are involved: Step 1: Collect surface electromyographic signals of the human body during movement, extract time domain eigenvalues ​​through a sliding time window after preprocessing, and construct a time-varying electromyographic functional network; Step 2: Detect muscle community clustering in the single-layer EMG functional network; Step 3: Perform step 2 on each layer of the EMG functional network divided by the time window to obtain the muscle community of each layer, and arrange them in the order of the corresponding time snapshots to form a community set; Step 4: Set the resolution parameter γ to adjust the sensitivity of each layer of the EMG functional network to the community; add the coupling parameter δ between the layers in the community set to adjust the coupling between the layers; Step 5: Optimize the size of modularity by maximizing multi-layer modularity; map the module in one time window to itself in the next time window. The adjacency matrix of the electromyographic functional network in one time window is linked to the adjacency matrix in the adjacent time window through the identification edge. The identification edge in the adjacent time window is connected to itself in the adjacent time window. This identification link executes all nodes, identifies the modules and their time changes by maximizing the multi-layer modularity quality function, and maximizes the multi-layer modularity quality function Q. multslice The expression is as follows: Among them, A ijs is the weight of the edge between node i and node j in time window s; the community assignment of node i in layer s is g is , the community assignment of node j in layer r is g jr , if g is =g jr , then δ(g is ,g jr )=1, otherwise it is 0; the total edge weight is where K js =k js +c js is the strength of node j in layer s, k js is the intra-layer strength of node j in layer s, c js =∑ r C jsr is the inter-layer strength of node j in layer s, m s =∑ j k js ; parameter γ s is the structural resolution parameter of the s layer, which is used to adjust the number of communities identified, δ sr It represents the inter-layer coupling parameter between the s layer and the r layer, which is used to adjust the coupling size between the layers and indirectly affects the size of the community; Step 6: Set the parameter domain for the two parameters mentioned in step 4, and perform multiple iterations of step 5 in the parameter domain to prune and prioritize the different community structures identified in multiple runs, and determine the optimal resolution parameter and coupling parameter by adjusting the mutual information; Step 7: Divide clusters of strongly interconnected nodes in the electromyographic functional network into different modules. Node clusters within the same module are considered to be muscle groups with strong synergy at this moment. Muscle synergy changes dynamically over time. Dynamic community detection is used to reveal this change process and achieve dynamic decoding of muscle synergy during movement changes.

2. The muscle dynamic synergy analysis method based on dynamic community detection according to claim 1 is characterized by: In step 1, the collected original signals of multiple channels are subjected to data preprocessing operations such as filtering, rectification, and normalization.

3. The muscle dynamic synergy analysis method based on dynamic community detection according to claim 1 is characterized by: In step 1, the four time domain eigenvalues ​​in the continuous time window are extracted by sliding the time window, namely: maximum amplitude, root mean square, integrated electromyography, and absolute value average; the four extracted time domain eigenvalues ​​are constructed into a feature matrix T it , calculate the Pearson correlation between each pair of EMG channels in the feature matrix; then establish an undirected weighted adjacency matrix with muscle areas as nodes and Pearson correlations between EMG channels as edges; divide the EMG signals of the human body movement process into continuous time windows through a sliding time window, extract the adjacency matrix in each time window respectively, and combine them into an overlapping adjacency matrix set in order to obtain a time-varying EMG functional network.

4. The muscle dynamic synergy analysis method based on dynamic community detection according to claim 1 is characterized in that: Step 2: 2.1 Extract the undirected weighted adjacency matrix of the single-layer EMG functional network as the initial data set; regard each node in the EMG functional network as a community, and the number of communities is consistent with the number of nodes; 2.2 Merge each node with its adjacent nodes in turn and calculate the modularity gain ΔQ; 2.3 Determine whether the modularity gain ΔQ is greater than 0. If it is greater than 0, place the node in the community where the adjacent node with the largest ΔQ is located. 2.4 Repeat the iterative step 2.3 and perform the same operation on all nodes in the EMG functional network in turn until ΔQ no longer changes; 2.5 Merge and compress the existing communities into a new super node. The weights within the original community are converted into the self-loop weights of the new node, and the weights between the original communities are converted into the edge weights of the new node. 2.6 Repeat steps 2.2-2.4 until the final stabilization is completed.

5. The muscle dynamic synergy analysis method based on dynamic community detection according to claim 4 is characterized in that: In step 2, the modularity gain ΔQ is calculated as: Among them, ∑in is the total edge weight of the community composed of all adjacent nodes, that is, the correlation value of the electromyographic channel. Initially, one node is one community, and its value is the default value 1; ∑tot is the sum of the weights of the links associated with each node community, that is, the node degree; k i is the sum of the weights of the links associated with node i, k i，in is the sum of links from node i to the associated community; m is the total weight of all paths in the network.

6. The muscle dynamic synergy analysis method based on dynamic community detection according to claim 1 is characterized by: In step 6, the calculation expression of mutual information is: where MI(X,Y) is the mutual information between random variables X and Y, and H(X) is the entropy of random variable X. The expected value E(MI(X,Y)) is computed over random partitions sampled from a hypergeometric null distribution. An AMI of 1 between two partitions indicates perfect agreement, and a value of 0 indicates a partition that is worse than random.

7. The muscle dynamic synergy analysis method based on dynamic community detection according to claim 6 is characterized by: The specific steps of step 6 are: 6.1 Given a set of partitions, determine the modular optimization domain for each partition; 6.2 Abandon partitions with empty domains to obtain a subset of partitions that maintain potential optimization in the specified parameter domain; 6.3 Use mutual information to select the optimized partitions, retain the subset with AMI>0.7, and determine it as the final community structure. The corresponding parameters at this time are the optimal parameters.

Images