Fast self-adaptive sEMG signal denoising method and system based on VMD decoupling optimization strategy

Abstract

The invention provides a rapid self-adaptive sEMG signal denoising method and system based on a VMD decoupling optimization strategy. The specific method comprises the following steps: S1, obtaining an original electromyographic signal; s2, constructing a Hankel matrix by using the preprocessed sEMG signal, determining a signal principal component through an SVD difference spectrum, and determining a modal number K of VMD; s3, determining an optimal penalty factor alpha of the VMD through an HSO algorithm; s4, performing variational mode decomposition on the original electromyographic signals by adopting the VMD of the optimal parameter combination to obtain K intrinsic mode components (IMF), performing size sorting according to the NCC, and partitioning the IMF into a high-correlation region, a middle-correlation region and a low-correlation region; and S5, according to respective characteristics of the three correlation areas, different denoising processing modes are applied, reconstruction is carried out, and a denoised sEMG signal is obtained. According to the method, VMD parameter adaptive optimization is achieved through decoupling optimization, denoising processing is conducted on the noisy IMF in combination with the NCC-based three-area adaptive reconstruction strategy, efficient and high-fidelity denoising of the sEMG signals is achieved, and the complete features of the electromyographic signals are better reserved.

Description

Technical Field

[0001] This invention belongs to the field of signal processing technology, specifically relating to a fast adaptive sEMG signal denoising method and system based on VMD decoupling optimization strategy. Background Technology

[0002] sEMG, as a non-invasive window for detecting muscle electrophysiological activity, has become a core technology for decoding human motor intentions. Its widespread application in rehabilitation medicine, human-computer interaction, and biomechanics, such as driving intelligent prostheses and manipulating exoskeleton robots, all rely on high-precision real-time analysis of sEMG signals. However, sEMG signals are inherently nonlinear and non-stationary random signals, and their effective energy is easily contaminated by various noise sources after passing through skin tissue, primarily including baseline drift, electric field interference, and Gaussian white noise. These noises severely reduce the signal-to-noise ratio of sEMG, leading to distortion in subsequent feature extraction and potentially causing catastrophic application failures, such as prosthetic control delays or human-computer interaction misjudgments. Therefore, efficient and high-fidelity denoising preprocessing of the raw sEMG signal is a prerequisite and key to unlocking its application potential.

[0003] To address these issues, Huang and his team proposed Empirical Mode Decomposition (EMD) in 1998 in the *Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences*. This method decomposes complex signals into several Intrinsic Mode Components (IMFs), each representing a local characteristic oscillation at different time scales within the signal. This decomposition method effectively reveals the instantaneous frequency characteristics of a signal, laying the foundation for subsequent Hilbert-Huang transforms. However, while this method can handle nonlinear and non-stationary signals and possesses high decomposition accuracy, flexibility, and adaptability, its mode aliasing and endpoint effects can lead to confusion in the time-frequency distribution of the IMFs, failing to reflect the physical characteristics of the signal and thus affecting the recovery of the original signal.

[0004] In contrast, Variational Mode Decomposition (VMD), proposed by Konstantin Dragomiretskiy et al. in the top international journal *IEEE Transactions on Signal Processing* in 2014, overcomes the mode aliasing defect of EMD effectively by solving constraint problems within a variational framework, demonstrating stronger robustness due to its solid mathematical foundation. Nevertheless, the performance of VMD is highly dependent on the prior settings of two key parameters: the mode decomposition number K and the quadratic penalty factor α. Inappropriate parameter selection can lead to under- or over-decomposition of the signal, greatly limiting its adaptive capability in practical applications. While existing research has attempted to use intelligent optimization algorithms to optimize parameters so that VMD can effectively separate noise and signal components, it typically faces the enormous computational overhead of joint optimization in a two-dimensional parameter space, and the optimization efficiency still needs improvement.

[0005] CN 118680584 discloses a machine learning-based method for denoising electromyographic (EMG) signals. It constructs a first processing module including singular value decomposition; an optimization processing network including an input layer, intermediate layers, and an output layer; a second processing module including a loss function and backpropagation; and an EMG signal data denoising model including input, the first processing module, the optimization processing network, the second processing module, and output. The EMG signal data denoising model is trained and tested using an EMG dataset. After training, the EMG signal requiring noise removal is obtained and input into the EMG signal data denoising model to obtain the denoised EMG signal. However, the method utilizes an SGD optimizer for denoising. The static processing logic of static supervision signals and fixed filtering cannot fully adapt to the dynamic changes in EMG signals, leading to a decrease in the physiological integrity of the denoised EMG signal and potentially causing severe signal distortion.

[0006] In view of this situation, the present invention proposes a new signal noise reduction method to solve the above problems. Summary of the Invention

[0007] The purpose of this invention is to provide a fast adaptive sEMG signal denoising method and system based on VMD decoupling optimization strategy. The method uses VMD to decompose the electromyography signal, and optimizes the maximum balance parameter of VMD and improves the decomposition accuracy through SVD and HSO. Then, the NCC of each IMF after decomposition is calculated, and the three-zone adaptive modal denoising strategy is used for reconstruction to better preserve the complete features of the original electromyography signal.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] In a first aspect, the present invention provides a fast adaptive sEMG signal denoising method based on a VMD decoupling optimization strategy, comprising:

[0010] S1: Acquire the raw sEMG signal and perform preprocessing to obtain the preprocessed sEMG signal;

[0011] S2: Construct the Hankel matrix using the preprocessed sEMG signal, determine its principal components through the SVD difference spectrum, and then determine the number of modes K of VMD;

[0012] S3: The fitness value of VMD is adaptively and iteratively solved using the HSO algorithm with PE as the objective function to determine the optimal penalty factor α of VMD and obtain the optimal parameter combination of VMD.

[0013] S4: The original sEMG signal is subjected to variational mode decomposition using VMD with optimal parameter combination to obtain K intrinsic mode components (IMFs). The IMFs are sorted and partitioned according to the NCC corresponding to the K IMFs. Among them, NCC>0.6 is the high correlation region, 0.3≤NCC≤0.6 is the medium correlation region, and NCC<0.3 is the low correlation region.

[0014] S5: Apply different denoising methods to the three relevant regions identified in step S4, reconstruct the denoised intrinsic mode components, and obtain the denoised sEMG signal.

[0015] Optionally, step S1 specifically involves: acquiring the original sEMG signal, performing fitting and normalization processing on the original sEMG signal, and preprocessing it using a 50Hz notch filter and a bandpass filter to obtain the preprocessed sEMG signal.

[0016] Optionally, step S2 specifically includes:

[0017] S21: Construct the Hankel matrix H using the preprocessed sEMG signal;

[0018] S22: Perform singular value decomposition on the Hankel matrix H to obtain H = UΣV T Σ is a diagonal matrix whose diagonal elements are singular values.

[0019] Find the point r with the largest abrupt change in the difference spectrum as the boundary between the principal components of the signal and the noise;

[0020] S23: Based on the characteristics of electromyographic signals, the maximum mutation point r is mapped to the number of VMD modalities K, thereby determining the number of VMD modalities K.

[0021] Optionally, the mapping rule for the mutation point r in step S23 to the number of modes K of VMD is as follows:

[0022]

[0023] Specifically, when r < 2, two modes are added; when r > 7, K = 7 is forcibly set.

[0024] Optionally, step S3 specifically includes:

[0025] S31: Input the preprocessed sEMG signal, set the VMD balance parameters, and generate a random solution in the search space using the parameter range;

[0026] S32: Obtain the HSO algorithm, select permutation entropy PE as the objective function, use it as the evaluation criterion for VMD decomposition effect, calculate the PE of each random solution, and use it as the evaluation basis for subsequent searches;

[0027]

[0028] Where, p i This represents the probability of the i-th permutation pattern occurring.

[0029] S33: Calculate the root mean square (RMS) and displacement coefficient c of the random solution. i To achieve global guidance for individual movement direction:

[0030]

[0031] Define the difference d between an individual and the root mean square (RMS). i And normalize to obtain the displacement coefficient:

[0032]

[0033] Where n represents the number of solutions generated; f(x) i ) represents the fitness value corresponding to each solution; sign(d i ) is the sign function, which determines the direction of population movement, ensuring that individuals move to better areas; x i This indicates the current position of the i-th solution;

[0034] S34: Integrate the information of the entire population with random perturbations to update the position of random solutions, and realize the individual's utilization of global information through the weighted sum of differences of the entire population:

[0035]

[0036] Where θ is the step size coefficient, controlling the movement amplitude; r ij It is a random number; c j x represents the difference between the fitness and RMS of the j-th solution; j Indicates the current position of the j-th solution; x iThis indicates the current position of the i-th solution;

[0037] S35: Introduce the simulated annealing (SA) mechanism to dynamically accept new solutions, expressed as follows:

[0038] T = T initial ·CR iter

[0039] Among them, T initial The initial temperature is CR, the cooling rate is CR, and iter represents the current iteration number.

[0040] If the fitness value of the new solution is better than that of the previous solution, the new solution is used instead of the old solution for computation; otherwise, the new solution is selected with probability P = exp(-Δf / P), where Δf is the fitness deterioration factor, representing the difference in fitness between the new and old solutions.

[0041] S36: Using the mutation operator algorithm, the random solution is dynamically adjusted based on the mutation rate and the variable asynchronous length, i.e.

[0042]

[0043] in, This represents the mutation rate in the i-th iteration. λ represents the initial rate of variation. final It represents the final mutation rate, where iter is the current iteration number. max δ is the maximum number of iterations. init It is the initial variable asynchronous length, δ final It is ultimately a variable asynchronous length;

[0044] S37: Calculate the fitness value of the random solution and determine whether the fitness value corresponding to each random solution is the minimum value. If not, return to S32; if yes, output the optimal penalty factor α to obtain the optimal parameter combination of VMD.

[0045] Optionally, step S4 specifically includes:

[0046] S41: Perform Hilbert transform on the original sEMG signal to obtain the one-sided spectrum;

[0047] S42: The spectrum of the initial intrinsic mode components (IMFs) is transferred to the baseband region by multiplying them with complex exponential signals tuned to their respective estimated center frequencies;

[0048] S43: By estimating the gradient of the demodulated signal using Gaussian smoothing through L2 regularization, an optimization model and constraints for VMD are constructed.

[0049] S44: Solve the VMD optimization model using the augmented Lagrange function to transform the constrained variational problem into an unconstrained variational problem;

[0050] S45: The K intrinsic mode components (IMFs) are obtained by optimizing each component and its corresponding center frequency using the alternating direction multiplier method.

[0051] S46: Determine the NCC corresponding to the K intrinsic mode components (IMFs), sort them by size and partition them according to the NCC corresponding to the K IMFs, where NCC>0.6 is the high correlation region, 0.3≤NCC≤0.6 is the medium correlation region, and NCC<0.3 is the low correlation region.

[0052] Optionally, the optimization model and constraints of VMD in step S43 are as follows:

[0053]

[0054] In the formula: f(t) is the original signal; k is the total number of components; u k For the k modal components of the original signal f(t), w k Let be the set of center frequencies of all sub-signals; j is the imaginary part of the complex term; The partial derivative operator is δ(t); δ(t) is the Diclave function. The Hilbert transform converts each modal function into an analytic signal, transforming the real-valued signal into a complex-valued form to obtain the single-sided spectrum.

[0055] The transformation formula used in step S44 to convert the constrained variational problem into an unconstrained variational problem is as follows:

[0056]

[0057] Where α is the penalty factor; λ is the Lagrange multiplier;

[0058] The specific expression of the alternating direction multiplier method in step S45 is as follows:

[0059]

[0060] In the formula: and These are the Wiener filters and frequency centers for each component, respectively; and ω represents the Fourier transforms of f(t), μ(t), and λ(t), respectively; n represents the number of iterations; and dw represents the derivative with respect to w.

[0061] Optionally, step S5 specifically includes:

[0062] S51: The wden function is used in conjunction with the minimaxi threshold selection rule to denoise the intrinsic mode components (IMFs) in the high-correlation region using the minimaxi threshold selection rule, and a nonparametric estimation method that minimizes the maximum risk is employed for noise reduction.

[0063] S52: The threshold for the intrinsic mode components (IMF) of the mid-correlation region is determined by the heuristic SureShrink rule. By adaptively selecting between the Stein unbiased risk estimation threshold and the general threshold, a balance is achieved between suppressing residual noise and preserving effective signal details.

[0064] S53: The intrinsic mode components (IMF components) in the low correlation region are directly discarded;

[0065] S54: Reconstruct the intrinsic mode components (IMFs) of the high-correlation region and the medium-correlation region after denoising to obtain the denoised sEMG signal.

[0066] In a second aspect, the present invention provides an sEMG signal denoising processing system, including a signal extraction module, a VMD optimization module, a variational mode decomposition module, a denoising processing module, and a reconstruction module;

[0067] The signal extraction module is used to extract the raw sEMG signal and perform preprocessing to obtain the preprocessed sEMG signal.

[0068] The VMD optimization module is used to construct a Hankel matrix using the preprocessed sEMG signal, determine the principal components of the signal through the SVD difference spectrum, and then determine the number of modes K of VMD; the HSO algorithm is used for adaptive iterative search to determine the optimal penalty factor α of VMD, and obtain the optimal combination of parameters of VMD.

[0069] The variational mode decomposition module is used to perform variational mode decomposition on the original sEMG signal using VMD with optimal parameter combination, to obtain K intrinsic mode components (IMFs). The IMFs are sorted and partitioned according to the NCC corresponding to the K IMFs, where NCC>0.6 is the high correlation region, 0.3≤NCC≤0.6 is the medium correlation region, and NCC<0.3 is the low correlation region.

[0070] The noise reduction module applies different noise reduction methods to the high correlation region, medium correlation region and low correlation region according to their respective characteristics.

[0071] The reconstruction module reconstructs the denoised intrinsic mode components and outputs the denoised sEMG signal.

[0072] Thirdly, the present invention provides a computer device, the computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the computer program to implement the fast adaptive sEMG signal denoising method based on VMD decoupling optimization strategy described above.

[0073] The beneficial effects of this invention are:

[0074] (1) This invention decomposes noisy electromyography (EMG) signals using the VMD algorithm. The dual-parameter optimization of the intelligent optimization algorithm in VMD is reduced to K one-dimensional optimization problems by SVD. Compared with the two-dimensional optimization problem of the traditional intelligent optimization algorithm, it has lower computational complexity and scientific interpretability. It can also solve the prior setting problem of the two key parameters of VMD, that is, it can improve the efficiency and effect of EMG signal decomposition, thereby improving the accuracy of useful signal feature extraction.

[0075] (2) Based on the signal decomposition through the VMD algorithm, this invention addresses the problem that the prior setting of the balance parameters of the VMD algorithm and the fact that different balance parameter values ​​determine the accuracy of VMD decomposition. It adaptively optimizes the selection of balance parameters of VMD by using SVD (Singular Value Decomposition) and HSO (Global Cluster Optimization Algorithm). It utilizes the advantages of SVD in revealing the intrinsic structure and key features of the matrix and the strong global search capability and fast convergence speed of the HSO algorithm to efficiently and accurately select the optimal balance parameters for the VMD algorithm to give full play to its signal decomposition performance, that is, to achieve high-precision signal decomposition to obtain several IMF components.

[0076] (3) The present invention designs a three-zone adaptive denoising and reconstruction strategy based on NCC to supplement each modal component after decomposition with a refined post-processing strategy, which can achieve excellent denoising effect at a relatively fast speed, realize efficient and high-fidelity denoising of sEMG signal, so as to facilitate subsequent analysis and discrimination. Attached Figure Description

[0077] Figure 1 This is the overall noise reduction flowchart of the present invention.

[0078] Figure 2 This is the normalized electromyography signal diagram of the present invention.

[0079] Figure 3 This is the difference spectrum curve of the present invention.

[0080] Figure 4 This is a fitness curve diagram of the HSO algorithm of the present invention.

[0081] Figure 5 This is a diagram showing the VMD decomposition of electromyographic signals according to the present invention.

[0082] Figure 6 This is a graph showing the NCC calculation results of the present invention.

[0083] Figure 7 This is a comparison diagram of the time-frequency domain before and after noise reduction in this invention.

[0084] Figure 8 This is a phase diagram of the chaotic attractor trajectory before and after denoising of the electromyographic signal in this invention.

[0085] Figure 9 Block diagram of the sEMG signal denoising system for invention. Detailed Implementation

[0086] 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.

[0087] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the claims.

[0088] The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular forms “a,” “the,” and “the” used in this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0089] In the description of this application, it should be understood that the numerical labels before the steps do not indicate the order in which the steps are performed, but are only used to facilitate the description of this application and to distinguish each step, and therefore should not be construed as a limitation of this application.

[0090] The present invention patent will be further described below with reference to the accompanying drawings:

[0091] Regarding the abbreviations used in this article:

[0092] sEMG: Surface Electromyographic Signal;

[0093] VMD: Variational mode decomposition;

[0094] SVD: Singular Value Decomposition

[0095] HSO: Holistic swarm optimization algorithm

[0096] IMF: Intrinsic Mode Function;

[0097] NCC: Normalized Cross Correlation.

[0098] Exemplary implementations of this application are provided below.

[0099] See Figure 1 The diagram illustrates a fast adaptive sEMG signal denoising method based on VMD decoupling optimization strategy according to the present invention, including:

[0100] S1: Acquire the raw sEMG signal and perform preprocessing to obtain the preprocessed sEMG signal;

[0101] S2: Construct the Hankel matrix using the preprocessed sEMG signal, determine its principal components through the SVD difference spectrum, and then determine the number of modes K of VMD;

[0102] S3: The fitness value of VMD is adaptively and iteratively solved using the HSO algorithm with PE as the objective function to determine the optimal penalty factor α of VMD and obtain the optimal parameter combination of VMD.

[0103] S4: The original sEMG signal is subjected to variational mode decomposition using VMD with optimal parameter combination to obtain K intrinsic mode components (IMFs). The IMFs are sorted and partitioned according to the NCC corresponding to the K IMFs. Among them, NCC>0.6 is the high correlation region, 0.3≤NCC≤0.6 is the medium correlation region, and NCC<0.3 is the low correlation region.

[0104] S5: Apply different denoising methods to the three relevant regions identified in step S4, reconstruct the denoised intrinsic mode components, and obtain the denoised sEMG signal.

[0105] The following provides a detailed explanation of the specific operating steps for steps S1 to S5.

[0106] Step S1 in this invention specifically involves: acquiring the original sEMG signal, performing fitting and normalization processing on the original sEMG signal, and preprocessing it using a 50Hz notch filter and a bandpass filter to obtain the preprocessed sEMG signal.

[0107] Step S2 in this invention specifically includes:

[0108] S21: Construct the Hankel matrix H using the preprocessed sEMG signal, as follows:

[0109]

[0110] Where m = [N / 2], n = N - m + 1.

[0111] S22: Perform singular value decomposition on the Hankel matrix H to obtain H = UΣV T Σ is a diagonal matrix whose diagonal elements are singular values. Find the point r with the largest abrupt change in the difference spectrum as the boundary between the principal components of the signal and the noise;

[0112] Specifically, calculate the difference sequence B = [b1, b2, b3, ... b] between adjacent singular values. r-1 ],in:

[0113]

[0114] Find the point r with the largest abrupt change in the difference spectrum as the boundary between the principal components of the signal and the noise, where;

[0115] r = argmax(b i )

[0116] Where b i The singular values ​​are the points of difference. The singular values ​​before r correspond to the principal components of the signal, and the singular values ​​after r correspond to noise. The effective dimension of the signal can be determined by r.

[0117] S23: Based on the characteristics of electromyographic signals, the maximum mutation point r is mapped to the number of VMD modalities K, thereby determining the magnitude of the number of VMD modalities K. The specific mapping rule is as follows:

[0118]

[0119] When r < 2, two modes are added to adopt a conservative strategy to avoid under-decomposition; when r > 10, K = 7 is forcibly set to limit the number of modes, which can avoid the error of manually preset K and adaptively match the signal complexity.

[0120] Step S3 in this invention specifically includes:

[0121] S31: Input the preprocessed sEMG signal, set the VMD balance parameters, and generate a random solution in the search space using the parameter range;

[0122] S32: Obtain the HSO algorithm, select permutation entropy PE as the objective function, use it as the evaluation criterion for VMD decomposition effect, calculate the PE of each random solution, and use it as the evaluation basis for subsequent searches;

[0123]

[0124] Where, p i This represents the probability of the i-th permutation pattern occurring.

[0125] This invention selects permutation entropy (PE) as a measure of decomposition performance. It has the characteristics of simple and fast calculation and strong noise resistance. The larger the permutation entropy value, the more complex the signal; the smaller the value, the more obvious the characteristics of the signal.

[0126] S33: Calculate the root mean square (RMS) and displacement coefficient c of the random solution. i To achieve global guidance for individual movement direction:

[0127]

[0128] Define the difference d between an individual and the root mean square (RMS). i And normalize to obtain the displacement coefficient:

[0129]

[0130] Where n represents the number of solutions generated; f(x) i ) represents the fitness value corresponding to each solution; sign(d i ) is the sign function, which determines the direction of population movement, ensuring that individuals move to better areas; x i This indicates the current position of the i-th solution;

[0131] S34: Integrate the information of the entire population with random perturbations to update the position of random solutions, and realize the individual's utilization of global information through the weighted sum of differences of the entire population:

[0132]

[0133] Where θ is the step size coefficient, controlling the movement amplitude; r ij It is a random number; c j x represents the difference between the fitness and RMS of the j-th solution; j Indicates the current position of the j-th solution; x i This indicates the current position of the i-th solution;

[0134] S35: Introduce the simulated annealing (SA) mechanism to dynamically accept new solutions, expressed as follows:

[0135] T = T initial ·CR iter

[0136] Among them, T initial The initial temperature is CR, the cooling rate is CR, and iter represents the current iteration number.

[0137] If the fitness value of the new solution is better than that of the previous solution, the new solution is used instead of the old solution for computation; otherwise, the new solution is selected with probability P = exp(-Δf / P), where Δf is the fitness deterioration factor, representing the difference in fitness between the new and old solutions.

[0138] S36: Using the mutation operator algorithm, the random solution is dynamically adjusted based on the mutation rate and the variable asynchronous length, i.e.

[0139]

[0140] in, This represents the mutation rate in the i-th iteration. λ represents the initial rate of variation. final It represents the final mutation rate, where iter is the current iteration number. max δ is the maximum number of iterations. init It is the initial variable asynchronous length, δ final It is ultimately a variable asynchronous length;

[0141] S37: Calculate the fitness value of the random solution and determine whether the fitness value corresponding to each random solution is the minimum value. If not, return to S32; if yes, output the optimal penalty factor α to obtain the optimal parameter combination of VMD.

[0142] Step S4 in this invention specifically includes:

[0143] S41: Perform Hilbert transform on the original sEMG signal to obtain the one-sided spectrum;

[0144] S42: The spectrum of the initial intrinsic mode components (IMFs) is transferred to the baseband region by multiplying them with complex exponential signals tuned to their respective estimated center frequencies;

[0145] S43: By Gaussian smoothing the gradient of the demodulated signal through L2 regularization, an optimization model and constraints for VMD are constructed. The specific optimization model and constraints are as follows:

[0146]

[0147] In the formula: f(t) is the original signal; k is the total number of components; u kFor the k modal components of the original signal f(t), w k Let be the set of center frequencies of all sub-signals; j is the imaginary part of the complex term; The partial derivative operator is δ(t); δ(t) is the Diclave function. The Hilbert transform converts each modal function into an analytic signal, transforming the real-valued signal into a complex-valued form to obtain the single-sided spectrum.

[0148] S44: Solve the VMD optimization model using the augmented Lagrange function, transforming the constrained variational problem into an unconstrained variational problem. The specific transformation formula is as follows:

[0149]

[0150] Where α is the penalty factor; λ is the Lagrange multiplier;

[0151] S45: Each component and its corresponding center frequency can be optimized using the alternating direction multiplier method to obtain K intrinsic mode components (IMFs). The specific expression of the alternating direction multiplier method is as follows:

[0152]

[0153] In the formula: and These are the Wiener filters and frequency centers for each component, respectively; and denoted as f(t), μ(t), and λ(t), respectively; ω represents the frequency; n is the number of iterations; dw represents the derivative with respect to w;

[0154] S46: Reference Figure 5 and Figure 6 Determine the NCC corresponding to the K intrinsic mode components (IMFs). Sort and partition the regions according to the NCC corresponding to the K IMFs. NCC > 0.6 is the high correlation region, 0.3 ≤ NCC ≤ 0.6 is the medium correlation region, and NCC < 0.3 is the low correlation region.

[0155] Step S5 in this invention specifically includes:

[0156] S51: The wden function is used in conjunction with the minimaxi threshold selection rule to denoise the intrinsic mode components (IMFs) in the high-correlation region using the minimaxi threshold selection rule, and a nonparametric estimation method that minimizes the maximum risk is employed for noise reduction.

[0157] S52: The threshold for the intrinsic mode components (IMF) of the mid-correlation region is determined by the heuristic SureShrink rule. By adaptively selecting between the Stein unbiased risk estimation threshold and the general threshold, a balance is achieved between suppressing residual noise and preserving effective signal details.

[0158] S53: The intrinsic mode components (IMF components) in the low correlation region are directly discarded;

[0159] S54: Reconstruct the intrinsic mode components (IMFs) of the high-correlation region and the medium-correlation region after denoising to obtain the denoised sEMG signal.

[0160] Specifically, the fast adaptive sEMG signal denoising method based on VMD decoupling optimization strategy of the present invention first uses SVD to reduce the dimensionality of parameter optimization into K one-dimensional optimization problems, then solves α in VMD using the HSO algorithm, and simultaneously performs VMD decomposition on the electromyography signal to obtain component IMFs. The NCC of each IMF component is calculated to determine the usefulness of the signal, and a three-region adaptive reconstruction strategy is applied to each IMF component to obtain the denoised signal for subsequent research.

[0161] Compared to traditional denoising methods, this invention eliminates the need for manual parameter setting, adaptively decomposing the signal. Analysis of the results shows that this method is effective for denoising non-stationary signals. Only the signal to be denoised needs to be manually input. Furthermore, the distinctive electromyographic signals significantly impact the accurate recognition of bionic robotic hands. Therefore, this method has implications for the fields of robotic arms and information processing, ensuring the accuracy of mechanical automation.

[0162] To verify the practicality of the denoising method proposed in this invention, electromyography (EMG) signals from the Ninapro dataset DB4 were used for validation, with a sampling frequency of 2000 Hz and 2000 sampling points. The normalized waveform of the EMG signal is shown below. Figure 2 As shown.

[0163] In this experiment, the VMD parameters of the electromyography signal were optimized using the SVD and HSO algorithms. The population size was set to 20, the maximum number of iterations was set to 20, and the search range of α was (500, 3000). The iterative process of SVD-HSO optimization of VMD parameters is as follows: Figure 3 , Figure 4 As shown, the value of K is first determined using the SVD difference spectrum, and then the HSO algorithm is used to optimize α. When the iterative process converges and stabilizes, it indicates that the optimal solution has been found. Specifically, the iteration results of the HSO algorithm tend to stabilize in the 4th generation, at which point the optimal parameter values ​​of K and α are (4, 2992), and the optimal fitness value is 0.937625.

[0164] The electromyographic signal was decomposed using its optimal parameter combination [K=4, α=2992], yielding four intrinsic mode components (IMFs). From a frequency domain perspective, the center frequencies of IMF1 to IMF4 increase sequentially, clearly deconstructing the original signal from low to high frequencies. IMF1 and IMF2 exhibit the highest correlation coefficients, while IMF3 and IMF4 show lower correlation coefficients. Based on the proposed three-zone adaptive reconstruction strategy, IMF1 and IMF2 are identified as highly correlated modes and undergo gentle denoising. IMF3 and IMF4 are identified as mixed modes and undergo general denoising.

[0165] To comprehensively evaluate the performance of the denoising method of this invention, a comparative analysis was conducted on the denoised results of real sEMG signals from three dimensions: time-domain waveform, frequency-domain structure, and nonlinear dynamics. Figure 7 Time-domain waveform analysis before and after denoising reveals that almost all subtle, high-frequency random jitter waveforms in the original signal are completely filtered out, effectively removing the high-frequency disturbances attached to it. Through analysis... Figure 7 The frequency domain structures of the two signal segments show that the spectral structure in the core signal region (0-300Hz) maintains a high degree of consistency with the spectral shape before denoising. However, in the high-frequency region above 300Hz, the spectral amplitude of the denoised signal is attenuated, indicating the denoising stability of the algorithm in terms of signal distortion. Furthermore, as... Figure 8 As shown, the denoising effect of sEMG is evaluated by attractor traject ...

[0166] This invention also provides an sEMG signal denoising processing system, such as... Figure 9 Specifically, it includes a signal extraction module, a VMD optimization module, a variational mode decomposition module, a denoising module, and a reconstruction module; wherein,

[0167] The signal extraction module is used to extract the raw sEMG signal and perform preprocessing to obtain the preprocessed sEMG signal.

[0168] The VMD optimization module is used to construct a Hankel matrix using the preprocessed sEMG signal, determine the principal components of the signal through the SVD difference spectrum, and then determine the number of modes K of VMD; the HSO algorithm is used for adaptive iterative search to determine the optimal penalty factor α of VMD, and obtain the optimal combination of parameters of VMD.

[0169] The variational mode decomposition module is used to perform variational mode decomposition on the original sEMG signal using VMD with optimal parameter combination, to obtain K intrinsic mode components (IMFs). The IMFs are sorted and partitioned according to the NCC corresponding to the K IMFs, where NCC>0.6 is the high correlation region, 0.3≤NCC≤0.6 is the medium correlation region, and NCC<0.3 is the low correlation region.

[0170] The noise reduction module applies different noise reduction methods to the high correlation region, medium correlation region and low correlation region according to their respective characteristics.

[0171] The reconstruction module reconstructs the denoised intrinsic mode components and outputs the denoised sEMG signal.

[0172] It should be noted that the fast adaptive sEMG signal denoising method based on VMD decoupling optimization strategy provided by this invention can be executed by a server, by a terminal device, or by a combination of both. Correspondingly, this invention also provides an sEMG signal denoising processing system, whose modules can be all located in the server, all in the terminal device, or separately in the server and terminal device.

[0173] The present invention also provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described fast adaptive sEMG signal denoising method based on VMD decoupling optimization strategy.

[0174] In this embodiment, the computer device is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions. The computer device includes, but is not limited to, devices that can communicate with each other via a system bus, such as a memory, a processor, and a network interface.

[0175] The memory includes at least one type of computer-readable storage medium, which can be volatile or non-volatile. Specifically, the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory can be an internal storage module of a computer device, such as the hard disk or RAM of the computer device. In other embodiments, the memory can also be an external storage device of the computer device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. Of course, the memory can also include both internal storage modules and external storage devices of the computer device. In this embodiment, the memory is typically used to store the operating system and various application software installed on the computer device, such as the program code for sEMG signal denoising methods. In addition, memory can also be used to temporarily store various types of data that have been output or will be output.

[0176] In some embodiments, a processor may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other noise reduction processing chip. This processor is typically used to control the overall operation of a computer device, such as performing control and processing related to data interaction or communication with the computer device. In this embodiment, the processor is used to run program code stored in memory or process data.

[0177] A network interface, which can include a wireless network interface or a wired network interface, is typically used to establish communication links between computer devices and other computer devices. For example, a network interface is used to connect a computer device to an external terminal via a network, establishing data transmission channels and communication connections between the computer device and the external terminal. Networks can be intranets, the Internet, Global System for Mobile Communications (GSM), Wideband Code Division Multiple Access (WCDMA), 4G networks, 5G networks, Bluetooth, WiFi, and other wireless or wired networks.

[0178] In this embodiment, the sEMG signal denoising stored in the memory can be divided into one or more program modules and executed by one or more processors (in this embodiment, a processor) to complete this application.

[0179] This application provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the steps of the fast adaptive sEMG signal denoising method based on VMD decoupling optimization strategy in the embodiment.

[0180] In this embodiment, the computer-readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the computer-readable storage medium may be an internal storage unit of a computer device, such as the hard disk or memory of the computer device. In other embodiments, the computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device. Of course, the computer-readable storage medium may include both the internal storage unit and the external storage device of the computer device. In this embodiment, the computer-readable storage medium is typically used to store the operating system and various application software installed on the computer device, such as the program code of the fast adaptive sEMG signal denoising method based on VMD decoupling optimization strategy in this embodiment. In addition, computer-readable storage media can also be used to temporarily store various types of data that have been output or will be output.

[0181] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across at least two network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments of this application. Those skilled in the art can understand and implement this without any creative effort.

Claims

1. A fast adaptive sEMG signal denoising method based on VMD decoupling optimization strategy, characterized in that, Includes the following steps: S1: Acquire the raw sEMG signal and perform preprocessing to obtain the preprocessed sEMG signal; S2: Construct the Hankel matrix using the preprocessed sEMG signal, determine its principal components through the SVD difference spectrum, and then determine the number of modes K of VMD; S3: The fitness value of VMD is adaptively and iteratively solved using the HSO algorithm with PE as the objective function to determine the optimal penalty factor α of VMD and obtain the optimal parameter combination of VMD. S4: The original sEMG signal is subjected to variational mode decomposition using VMD with optimal parameter combination to obtain K intrinsic mode components (IMFs). The IMFs are sorted and partitioned according to the NCC corresponding to the K IMFs. Among them, NCC>0.6 is the high correlation region, 0.3≤NCC≤0.6 is the medium correlation region, and NCC<0.3 is the low correlation region. S5: Apply different denoising methods to the three relevant regions identified in step S4, reconstruct the denoised intrinsic mode components, and obtain the denoised sEMG signal.

2. The fast adaptive sEMG signal denoising method based on VMD decoupling optimization strategy according to claim 1, characterized in that, Step S1 specifically involves: acquiring the original sEMG signal, performing fitting and normalization processing on the original sEMG signal, and preprocessing it using a 50Hz notch filter and a bandpass filter to obtain the preprocessed sEMG signal.

3. The fast adaptive sEMG signal denoising method based on VMD decoupling optimization strategy according to claim 1, characterized in that, Step S2 specifically involves: S21: Construct the Hankel matrix H using the preprocessed sEMG signal; S22: Perform singular value decomposition on the Hankel matrix H to obtain H = UΣV T Σ is a diagonal matrix whose diagonal elements are singular values. Find the point r with the largest abrupt change in the difference spectrum as the boundary between the principal components of the signal and the noise; S23: Based on the characteristics of electromyographic signals, the maximum mutation point r is mapped to the number of VMD modalities K, thereby determining the number of VMD modalities K.

4. The fast adaptive sEMG signal denoising method based on VMD decoupling optimization strategy according to claim 3, characterized in that, The mapping rule for the mutation point r in S23 to the number of modes K of VMD is as follows: Specifically, when r < 2, two modes are added; when r > 7, K = 7 is forcibly set.

5. The fast adaptive sEMG signal denoising method based on VMD decoupling optimization strategy according to any one of claims 1 to 4, characterized in that, Step S3 specifically includes: S31: Input the preprocessed sEMG signal, set the VMD balance parameters, and generate a random solution in the search space using the parameter range; S32: Obtain the HSO algorithm, select permutation entropy PE as the objective function, use it as the evaluation criterion for VMD decomposition effect, calculate the PE of each random solution, and use it as the evaluation basis for subsequent searches; Where, p i This represents the probability of the i-th permutation pattern occurring. S33: Calculate the root mean square (RMS) and displacement coefficient c of the random solution. i To achieve global guidance for individual movement direction: Define the difference d between an individual and the root mean square (RMS). i And normalize to obtain the displacement coefficient: Where n represents the number of solutions generated; f(x) i ) represents the fitness value corresponding to each solution; sign(d i ) is the sign function, which determines the direction of population movement, ensuring that individuals move to better areas; x i This indicates the current position of the i-th solution; S34: Integrate the information of the entire population with random perturbations to update the position of random solutions, and realize the individual's utilization of global information through the weighted sum of differences of the entire population: Where θ is the step size coefficient, controlling the movement amplitude; r ij It is a random number; c j x represents the difference between the fitness and RMS of the j-th solution; j Indicates the current position of the j-th solution; x i This indicates the current position of the i-th solution; S35: Introduce the simulated annealing (SA) mechanism to dynamically accept new solutions, expressed as follows: T=T initial ·CR iter Among them, T initial The initial temperature is CR, the cooling rate is CR, and iter represents the current iteration number. If the fitness value of the new solution is better than that of the previous solution, the new solution is used instead of the old solution for computation; otherwise, the new solution is selected with probability P = exp(-Δf / P), where Δf is the fitness deterioration factor, representing the difference in fitness between the new and old solutions. S36: Using the mutation operator algorithm, the random solution is dynamically adjusted based on the mutation rate and the variable asynchronous length, i.e. in, This represents the mutation rate in the i-th iteration. λ represents the initial rate of variation. final It represents the final mutation rate, where iter is the current iteration number. max δ is the maximum number of iterations. init It is the initial variable asynchronous length, δ final It is ultimately a variable asynchronous length; S37: Calculate the fitness value of the random solution and determine whether the fitness value corresponding to each random solution is the minimum value. If not, return to S32; if yes, output the optimal penalty factor α to obtain the optimal parameter combination of VMD.

6. The fast adaptive sEMG signal denoising method based on VMD decoupling optimization strategy according to claim 1, characterized in that, Step S4 specifically includes: S41: Perform Hilbert transform on the original sEMG signal to obtain the one-sided spectrum; S42: The spectrum of each initial intrinsic mode component (IMF) is transferred to the baseband region by multiplying it with the complex exponential signal tuned to its respective estimated center frequency; S43: By Gaussian smoothing estimation of the demodulated signal gradient through L2 regularization, an optimization model and constraints for VMD are constructed. S44: Solve the VMD optimization model using the augmented Lagrange function to transform the constrained variational problem into an unconstrained variational problem; S45: The K intrinsic mode components (IMFs) are obtained by optimizing each component and its corresponding center frequency using the alternating direction multiplier method. S46: Determine the NCC corresponding to the K intrinsic mode components (IMFs), sort them by size and partition them according to the NCC corresponding to the K IMFs, where NCC>0.6 is the high correlation region, 0.3≤NCC≤0.6 is the medium correlation region, and NCC<0.3 is the low correlation region.

7. The fast adaptive sEMG signal denoising method based on VMD decoupling optimization strategy according to claim 6, characterized in that, The optimization model and constraints of VMD in step S43 are as follows: In the formula: f(t) is the original signal; k is the total number of components; u k For the k modal components of the original signal f(t), w k Let be the set of center frequencies of all sub-signals; j is the imaginary part of the complex term; The partial derivative operator is δ(t); δ(t) is the Diclave function. The Hilbert transform converts each modal function into an analytic signal, transforming the real-valued signal into a complex-valued form to obtain the single-sided spectrum. The transformation formula used in step S44 to convert the constrained variational problem into an unconstrained variational problem is as follows: Where α is the penalty factor; λ is the Lagrange multiplier; The specific expression of the alternating direction multiplier method in step S45 is as follows: In the formula: and These are the Wiener filters and frequency centers for each component, respectively; and ω represents the Fourier transforms of f(t), μ(t), and λ(t), respectively; n represents the number of iterations; and dw represents the derivative with respect to w.

8. The fast adaptive sEMG signal denoising method based on VMD decoupling optimization strategy according to claim 1, characterized in that, Step S5 specifically includes: S51: The wden function is used in conjunction with the minimaxi threshold selection rule to denoise the intrinsic mode components (IMFs) in the high-correlation region using the minimaxi threshold selection rule, and a nonparametric estimation method that minimizes the maximum risk is employed for noise reduction. S52: The threshold for the intrinsic mode components (IMF) of the mid-correlation region is determined by the heuristic SureShrink rule. By adaptively selecting between the Stein unbiased risk estimation threshold and the general threshold, a balance is achieved between suppressing residual noise and preserving effective signal details. S53: The intrinsic mode components (IMF components) in the low correlation region are directly discarded; S54: Reconstruct the intrinsic mode components (IMFs) of the high-correlation region and the medium-correlation region after denoising to obtain the denoised sEMG signal.

9. An sEMG signal denoising processing system, characterized in that, It includes a signal extraction module, a VMD optimization module, a variational mode decomposition module, a denoising module, and a reconstruction module; The signal extraction module is used to extract the raw sEMG signal and perform preprocessing to obtain the preprocessed sEMG signal. The VMD optimization module is used to construct a Hankel matrix using the preprocessed sEMG signal, determine the principal components of the signal through the SVD difference spectrum, and then determine the number of modes K of VMD; the HSO algorithm is used for adaptive iterative search to determine the optimal penalty factor α of VMD, and obtain the optimal combination of parameters of VMD. The variational mode decomposition module is used to perform variational mode decomposition on the original sEMG signal using VMD with optimal parameter combination, to obtain K intrinsic mode components (IMFs). The IMFs are sorted and partitioned according to the NCC corresponding to the K IMFs, where NCC>0.6 is the high correlation region, 0.3≤NCC≤0.6 is the medium correlation region, and NCC<0.3 is the low correlation region. The noise reduction module applies different noise reduction methods to the high correlation region, medium correlation region and low correlation region according to their respective characteristics. The reconstruction module reconstructs the denoised intrinsic mode components and outputs the denoised sEMG signal.

10. A computer device, characterized in that, The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the fast adaptive sEMG signal denoising method based on the VMD decoupling optimization strategy as described in any one of claims 1 to 8.

Images