A muscle coordination analysis method based on surface electromyography signals
By using a muscle synergy analysis method based on surface electromyography signals, the number of muscle synergies is determined by interpreting variance. Furthermore, sparsity constraints and fitness function optimization are introduced into nonnegative matrix factorization, which solves the problem of balancing reconstruction accuracy and sparsity in muscle synergy analysis and achieves higher data decomposition quality and interpretability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HANGZHOU DIANZI UNIV
- Filing Date
- 2023-09-18
- Publication Date
- 2026-08-04
AI Technical Summary
Existing muscle synergy analysis methods struggle to effectively balance reconstruction accuracy and sparsity, resulting in insufficient accuracy and interpretability in muscle synergy analysis.
A muscle synergy analysis method based on surface electromyography signals is adopted. The number of muscle synergies is determined by interpreting variance. Sparsity constraints are introduced into the non-negative matrix factorization. Combined with the crossover and mutation operations of the fitness function, the decomposition results are optimized to improve the reconstruction accuracy and sparsity.
It improves the reconstruction accuracy and sparsity of muscle synergy analysis, and enhances the quality and interpretability of data decomposition.
Smart Images
Figure CN117204867B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of biosignal processing technology, specifically to a method for muscle synergistic analysis based on surface electromyography signals. Background Technology
[0002] Muscle synergy theory explains the control mechanism of the central nervous system over the musculoskeletal system. A muscle synergy consists of multiple muscles with different activation intensities, and a single muscle can be activated in multiple synergies. Complex human limb movements can be represented by the superposition of a few muscle synergies in the time domain. Therefore, muscle synergy analysis is crucial for the quantitative analysis of human muscle, nerve, and other motor functions. Generally, muscle synergy analysis first requires acquiring the electromyographic signals of the relevant muscles using electromyography (EMG) sensors. After processing the raw EMG signals, matrix factorization algorithms can be used to obtain the structure and recruitment patterns of each synergy.
[0003] Currently, muscle synergy analysis primarily relies on matrix factorization algorithms to analyze multi-channel electromyography (EMG) data, such as independent component analysis, principal component analysis, and nonnegative matrix factorization. By analyzing surface EMG signals and extracting features, muscle synergy can be decoupled. To enhance the sparsity of the decomposed data, some sparse coding-based methods have been employed to obtain sparse muscle activation patterns. Researchers have also proposed methods based on compressed sensing. Compressed sensing theory posits that as long as the signal possesses sparsity, it can be sampled at a rate much smaller than the Nyquist sampling rate and then reconstructed from the original signal. However, these methods suffer from issues with reconstruction accuracy, as they often require a trade-off between sparsity and reconstruction precision. Therefore, balancing sparsity and reconstruction error to improve the accuracy and interpretability of muscle synergy analysis remains a significant challenge in the field. Summary of the Invention
[0004] To address the shortcomings of existing technologies, this invention proposes a muscle synergy analysis method based on surface electromyography signals, which can achieve higher reconstruction accuracy and sparsity in muscle synergy analysis, thereby improving the quality and interpretability of data decomposition.
[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:
[0006] A method for muscle synergy analysis based on surface electromyography signals includes the following steps:
[0007] S1. Acquire multi-channel surface electromyography signals;
[0008] S2. The number of muscle coordinations is determined by interpreting variance as a standard for the number of muscle coordinations.
[0009] The variance expression is explained as follows:
[0010]
[0011] Z i×r H is the envelope signal matrix. i×n For the cooperative structure matrix, Y n×r H is the activation coefficient matrix. i×n Y n×r The reconstructed matrix after matrix decomposition, where n is the number of muscle synergies, i represents the number of muscles, and r represents the length of the data;
[0012] S3. Extracting muscle synergistic features
[0013] According to the envelope signal matrix Z i×r With respect to the number of muscle coordinators n, the envelope signal matrix Z i×r By performing nonnegative matrix factorization and introducing sparsity constraints during the factorization process, the extraction of muscle synergistic features is transformed into an optimization problem as follows:
[0014]
[0015] Among them, ||Z i×r -H i×n Y n×r || F 2 The F-norm represents the difference between the envelope signal matrix and the reconstructed matrix. Parameters used to control the degree of sparsity;
[0016] S4. Based on the optimization problem in step S3, obtain the decomposed reconstruction matrix HY. res That is, the decomposition result;
[0017] S5. The co-structure matrix H in the decomposition results... res Normalization is performed to obtain the muscle activation state under different muscle synergy modules. Based on the analysis of the muscle activation state under different muscle synergy modules, the synergy mode of the muscles during the movement is obtained, and the reconstruction accuracy and sparsity are calculated.
[0018] Preferably, the specific method of step S1 is as follows: surface electromyography (EMG) data is collected during human movement; after collection, the surface EMG signals are denoised; and the envelope signal matrix Z during movement is obtained using the maximum value normalization method. i×r .
[0019] Preferably, the specific method for determining the number of muscle synergies n in step S2 is as follows:
[0020] S2-1. Set the initial number of muscle synergies n = 1, and calculate the explained variance of n;
[0021] S2-2, n:=n+1, calculate the explained variance of n in this case;
[0022] S2-3. When VAF(n) > 0.95 and VAF(n+1) - VAF(n) < 0.01, then n is the number of muscle synergies; otherwise, proceed to S2-2 to continue iteration.
[0023] Preferably, the optimization method for step S4 is as follows:
[0024] S4-1. Randomly initialize a set of matrix decompositions HY1, HY2, K, HY m ;
[0025] S4-2. Calculate the adaptation function of reconstruction error and sparsity for each matrix decomposition, and obtain F1, F2, K, F m ;
[0026] S4-3. Based on the fitness function value F of each matrix decomposition m ,choose The matrix decomposition of quantities undergoes crossover and mutation operations to generate new matrix decomposition individuals;
[0027] S4-4. Repeat the above steps until the predetermined number of iterations is reached. Through multiple iterations, a set of high-quality solutions that balance reconstruction error and sparsity will be obtained. Select the solution with the largest fitness function value from these solutions as the final decomposition result HY. res .
[0028] Preferably, in step S4-2, the adaptation function expression for the reconstruction error and sparsity is as follows:
[0029] Fitness = w1RE + w2SP (3)
[0030]
[0031]
[0032] Where w1 is the weight of the reconstruction error, RE is the reconstruction error of the decomposed matrix, w2 is the weight of the sparsity of the decomposed matrix, SP is the sparsity of the decomposed matrix, and |||Z i×r -H i×n Y n×r || F 2 The F-norm of the difference between the original matrix and the reconstructed matrix is ||H. i×n ||1 is the L1 norm of the activation coefficient matrix, S H For matrix H i×n The number of elements.
[0033] Preferably, in step S4-3, a probability selection based on the fitness function is introduced in the crossover and mutation operation, as shown in the following formula:
[0034]
[0035]
[0036] Where w1 is the weight of the reconstruction error, RE min The minimum reconstruction error of the decomposed matrix is given by w2, where w2 is the weight of the sparsity of the decomposed matrix, and SP is the minimum reconstruction error of the decomposed matrix. max The maximum sparsity of the decomposed matrix is max(F). a ,F b ) indicates selecting the larger fitness function value from the two cross matrix decompositions, F m For the fitness function value of the mutation matrix decomposition, F avg To fit the average value of the function, q1 and q2 are both constants.
[0037] This invention has the following characteristics and beneficial effects:
[0038] By employing the above technical solution, a sparsity regularization term is introduced into the non-negative matrix factorization to make the generated factorization matrix more sparsity, thereby better reflecting the true characteristics of muscle activity. Simultaneously, by searching for a superior solution set based on the selection of the fitness function, crossover, and mutation probabilities, a balance is effectively struck between reconstruction error and sparsity, while avoiding the selection of sparse parameters, resulting in a more optimized decomposition result. By introducing sparsity constraints and probabilistic selection based on the fitness function's crossover and mutation probabilities, the method of this invention can achieve higher reconstruction accuracy and sparsity in muscle synergistic analysis, thereby improving the quality and interpretability of data decomposition. Attached Figure Description
[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0040] Figure 1 This is a flowchart of the overall method of the present invention.
[0041] Figure 2 This is a bar chart showing the reconstruction accuracy and sparsity of this invention compared to other methods.
[0042] Figure 3 This is a diagram showing the muscle synergy analysis results of the present invention. Detailed Implementation
[0043] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.
[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0045] This invention provides a method for muscle synergy analysis based on surface electromyography signals, such as... Figure 1 As shown, the method includes the following steps:
[0046] Step 1: Acquire multi-channel surface electromyography signals
[0047] In this embodiment, the subject remained relaxed, looking straight ahead, with arms extended forward and palms facing each other. The subject squatted down until their thighs were nearly parallel to the ground, then stood up. Throughout the experiment, the subject was required to keep their back straight. The subject remained still for the first 3 seconds, then began squatting and standing up, continuing this deep squat for 200 seconds. After each data collection session, the subject rested for one hour, and three data sets were collected from each subject. The electromyography (EMG) sampling frequency was set to 2000Hz. EMG sensors were attached to the tibialis anterior, lateral gastrocnemius, vastus lateralis, biceps femoris, rectus femoris, and vastus medialis muscles of the subject's lower limbs. After data collection, the surface EMG signals were denoised, and the envelope signal matrix Z during movement was obtained using the maximum value normalization method. 6×280 (6 represents the number of muscles, and 280 represents the length of the data).
[0048] Step 2: Determine the number of synergistic muscles, n.
[0049] The explained variance (VAF) was used as the standard for determining the number of muscle synergies. The specific steps are as follows:
[0050] S2-1. Set the initial number of muscle synergies n = 1, and calculate the explained variance of n;
[0051] S2-2, n = n + 1, calculate the explained variance of n in this case;
[0052] S2-3. When VAF(n) > 0.95 and VAF(n+1) - VAF(n) < 0.01, then n is the number of muscle synergies; otherwise, proceed to step 2 to continue the iteration.
[0053] Specifically, the formula for calculating variance is explained below:
[0054]
[0055] Z i×r H is the original matrix. i×n Y n×r The reconstructed matrix after matrix decomposition.
[0056] Calculations show that when n=4, VAF(4)=0.9931, and when n=5, VAF(5)=0.9942, with an increment of 0.0011. When n is 6, 7, or 8, the VAF value remains constant at 0.9991. Therefore, the number of muscle synergies is determined to be n=4.
[0057] Step 3: Extract muscle synergy features
[0058] Based on the obtained envelope signal matrix Z 6×280 With muscle synergy number n=4, for envelope signal matrix Z 6×280 By performing nonnegative matrix factorization and introducing sparsity constraints during the factorization process, the extraction of muscle synergistic features is transformed into an optimization problem as follows:
[0059]
[0060] Among them, ||Z i×r -H i×n Y n×r || F 2 Z represents the F-norm of the difference between the envelope signal matrix and the reconstruction matrix. i×r H is the envelope matrix of the original sEMG signal to be decomposed. i×n For the cooperative structure matrix, Y n×r The activation coefficient matrix, Parameters used to control the degree of sparsity.
[0061] Step 4: Optimize the muscle synergy feature extraction process
[0062] To address the optimization issue in step three, the following optimization steps are adopted:
[0063] S4-1. Randomly initialize 150 matrix decompositions: HY1, HY2, K, HY 150
[0064] S4-2, Calculate each matrix decomposition HY 150 The reconstruction error and the adaptation function of sparsity are obtained as F1, F2, K, F 150 The formula is as follows:
[0065] Fitness = w1RE + w2SP (10)
[0066]
[0067]
[0068] S4-3, where w1 is the weight of the reconstruction error, RE is the reconstruction error of the decomposed matrix, w2 is the weight of the sparsity of the decomposed matrix, SP is the sparsity of the decomposed matrix, |||Z i×r -H i×n Y n×r || F 2 The F-norm of the difference between the original matrix and the reconstructed matrix, ||H i×n ||1 is the L1 norm of the activation coefficient matrix, S H For matrix H i×n The number of elements.
[0069] S4-4. Based on the fitness function value F of each matrix decomposition m Seventy-five matrix factorizations were selected and subjected to crossover and mutation operations to generate new matrix factorization individuals. The crossover and mutation operations incorporated a probability selection based on the fitness function, as shown in the following formula:
[0070]
[0071]
[0072] Where w1 is the weight of the reconstruction error, RE min The minimum reconstruction error of the decomposed matrix is given by w2, where w2 is the weight of the sparsity of the decomposed matrix, and SP is the minimum reconstruction error of the decomposed matrix. max The maximum sparsity of the decomposed matrix is max(F). a ,F b ) indicates selecting the larger fitness function value from the two cross matrix decompositions, F m For the fitness function value of the mutation matrix decomposition, F avg To fit the average value of the function, q1 and q2 are both constants.
[0073] S4-5. Repeat the above steps until the predetermined number of iterations is reached. Through multiple iterations, a set of high-quality solutions that balance reconstruction error and sparsity will be obtained. The solution with the largest fitness function value is selected as the final decomposition result.
[0074]
[0075] Step 5: Obtain the results of muscle synergy analysis.
[0076] For the above cooperative structure matrix H 6×4 Normalization was performed to obtain the muscle activation states under different muscle synergistic modules, such as Figure 3 Based on the analysis of muscle activation states under different muscle synergy modules, the synergy pattern of muscles during the movement was obtained. Through the reconstruction accuracy and sparsity calculation formula, RE = 92.3728% and SP = 16.2130% were obtained.
[0077] like Figure 2 As shown, the method of this invention (RSSNMF) achieves a reconstruction accuracy of 92.3728% and a sparsity of 16.2130%; K-Singular Value Decomposition (K-SVD) achieves a reconstruction accuracy of 88.441% and a sparsity of 15.2782%; Orthogonal Matching Pursuit (OMP) achieves a reconstruction accuracy of 88.8663% and a sparsity of 13.4479%; and Compressed Sensing (CS) achieves a reconstruction accuracy of 90.2198% and a sparsity of 10.3410%. The method of this invention improves upon other methods in both reconstruction accuracy and sparsity. This further demonstrates that the method of this invention can achieve higher reconstruction accuracy and sparsity in muscle synergistic analysis, thereby improving the quality and interpretability of data decomposition.
[0078] like Figure 3 As shown, the synergistic effects are most pronounced in modules W1, W2, and W4 during the squat exercise. In module W1, the lateral gastrocnemius and rectus femoris are activated and contribute significantly, while the biceps femoris remains inactive. In module W2, the lateral gastrocnemius, biceps femoris, and rectus femoris are all activated. In module W4, the rectus femoris and vastus medialis exhibit significant synergistic effects and are activated.
[0079] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions, and variations can be made to these embodiments, including components, without departing from the principles and spirit of the present invention, and these variations still fall within the protection scope of the present invention.
Claims
1. A muscle synergy analysis method based on surface electromyography signals, characterized by, Includes the following steps: S1. Acquire multi-channel surface electromyography signals; S2. The number of muscle coordinations is determined by interpreting variance as a standard for the number of muscle coordinations. The variance expression is explained as follows: （1） wherein is an envelope signal matrix, is a synergy structure matrix, is an activation coefficient matrix, is a reconstruction matrix after matrix decomposition, is a number of muscle synergies, denotes a number of muscles, denotes a length of data; Determining the number of muscle synergies The specific method is: S2-1, set initial number of muscle synergies , compute explained variance of the solution; S2-2、 , compute the explained variance at this time of interpretation; S2-3, when ,and ,but The number of muscle synergies is determined; otherwise, proceed to S2-2 for further iteration. S3. Extracting muscle synergistic features Based on the envelope signal matrix Number of muscles working together For the envelope signal matrix By performing nonnegative matrix factorization and introducing sparsity constraints during the factorization process, the extraction of muscle synergistic features is transformed into an optimization problem as follows: （2） wherein, denotes the F-norm of the difference between the envelope signal matrix and the reconstruction matrix, is a parameter controlling the sparsity degree; S4. Obtain the decomposed reconstruction matrix from the optimization problem of step S3 i.e. the decomposition result; The optimization method for step S4 is as follows: S4-1. Randomly initialize a set of matrix factorizations ; S4-2, calculate the reconstruction error and the sparsity degree of each matrix decomposition, and obtain the adaptive function ; In step S4-2, the reconstruction error and sparsity adaptation function expression for each matrix decomposition are as follows: （3） （4） （5） in For the weights of the reconstruction error, The reconstruction error of the decomposed matrix, The weights for the sparsity of the decomposed matrix are... To determine the sparsity of the decomposed matrix, The F-norm represents the difference between the original matrix and the reconstructed matrix. Let L1 norm be the activation coefficient matrix. For matrix The number of elements; S4-3. Selecting a matrix decomposition from each matrix decomposition based on the fitness value of the matrix decomposition The number of matrix decompositions is increased by performing crossover and mutation operations on the matrix decompositions to produce new matrix decomposition individuals. In step S4-3, the crossover and mutation operation introduces a probability selection based on the fitness function, as shown in the following formula: （6） （7） in To reconstruct the weights of the error, To minimize the reconstruction error of the decomposed matrix, The weights for the sparsity of the decomposed matrix are... To represent the maximum sparsity of the decomposed matrix, This indicates selecting the value with the larger fitness function value from the two cross matrix decompositions. To obtain the fitness function value for mutation matrix decomposition, To adapt the average value of the function, All are constants; S4-4. Repeat the above steps until the predetermined number of iterations is reached. Through multiple iterations, a set of high-quality solutions that balance reconstruction error and sparsity will be obtained. Select the solution with the largest fitness function value from these solutions as the final decomposition result. ; S5. Regarding the co-structure matrix in the decomposition results Normalization is performed to obtain the muscle activation state under different muscle synergy modules. Based on the analysis of the muscle activation state under different muscle synergy modules, the synergy pattern of muscles during exercise is obtained, and the reconstruction accuracy and sparsity are calculated.
2. The muscle synergy analysis method based on surface electromyography signal according to claim 1, characterized in that, The specific method of the step S1 is: collecting surface electromyography data in the human motion process, performing noise removal processing on the surface electromyography signal after the collection is completed, and obtaining an envelope signal matrix in the motion process by using a maximum value normalization method .