Electromyographic signal denoising method based on RIME-SVMD

Abstract

The invention provides an electromyographic signal denoising method based on RIME-SVMD (Reactive International Mobile Equipment-Space Vector Multiple Data). The method comprises the following steps: S1, acquiring an original electromyographic signal, and adaptively iteratively searching an optimal balance parameter of an SVMD algorithm through an RIME algorithm; s2, carrying out successive variational mode decomposition on the original electromyographic signals by using SVMD to obtain decomposed IMF; S3, calculating NCC corresponding to each IMF, and by taking 0.3 as a standard, determining a part greater than 0.3 as a useful signal, and determining a part less than 0.3 as a useless signal. S4, denoising reconstruction is carried out on the signals with the NCC larger than 0.3, and the electromyographic signals obtained after denoising processing are obtained. According to the RIME-SVMD-based electromyographic signal denoising method disclosed by the invention, the SVMD maximum equilibrium parameter is solved through the RIME, the SVMD is used for decomposing the original electromyographic signals, and the wavelet threshold method is combined for denoising the useful signals, so that the characteristics of the original electromyographic signals are better reserved for subsequent analysis and judgment.

Description

Technical Field

[0001] The invention relates to a method for denoising an electromyographic signal, which is applied to a bionic manipulator and belongs to the field of signal processing. Background Art

[0002] SEMG is a weak electrical signal collected from the skin surface when the human body performs limb movements. It is closely related to the generation of muscle force and can be used to reflect muscle activity. Surface electromyography is widely used in rehabilitation engineering for the design and control of rehabilitation and assistive devices, as well as gesture recognition and other applications. However, since SEMG is a random signal with nonlinear, non-stationary, low amplitude and low frequency characteristics, it is easily interfered by human body and environmental noise, which will affect the accuracy and reliability of detection in severe cases, and even cause medical accidents. Therefore, minimizing harmful noise in electromyography signals is a serious issue that needs to be considered.

[0003] In order to remove noise interference in SEMG, many methods have been proposed and tested at home and abroad, and some results have been achieved. For example, empirical mode decomposition, wavelet threshold method and variational mode decomposition (VMD) are used. Among them, empirical mode decomposition (EMD) is one of the most commonly used denoising strategies. It is the most famous method to decompose a signal into a set of zero mean and almost the same number of maximum and minimum components. It recursively decomposes a non-stationary signal into a data-dependent basis function, called intrinsic mode function (IMF). An important feature of EMD is that it can represent nonlinear and non-stationary signals as the sum of time-frequency components with physical meaning. EMD has been satisfactorily applied in extracting life signals from biological signals, removing interference from signals, and climate analysis. However, the results of EMD are greatly affected by extreme point search and interpolation methods. Therefore, in order to solve the problems of low robustness to noise and lack of mathematical theory, some EMD-like methods have been developed. The VMD method adaptively decomposes SEMG into multiple intrinsic mode components (IMF), converts the variational constraint problem into a non-variational constraint problem, thereby realizing the decomposition of SEMG and overcoming the problem of modal aliasing in the EMD method. Based on the high adaptability of VMD to non-stationary sequences, VMD denoising has been successfully applied to earthquake time-frequency analysis, gear fault diagnosis, biological signal denoising, wind speed prediction and other fields, but VMD decomposition also has the defects of manually setting the penalty factor α and the decomposition layer K parameter. At present, scholars at home and abroad have tried to use optimization algorithms to solve the two parameters in VMD. This method can save time for manual parameter search, but due to different selections of algorithms and fitness functions, the results of the solution vary greatly. The wavelet threshold method has the characteristics of multi-resolution and time-frequency localization, which can better capture the detailed characteristics of the signal. Therefore, it is widely used in the random signal denoising process. Neural electrical signals such as electrocardiogram, electroencephalogram, and electromyography are all suitable for wavelet denoising. However, wavelet transform also has influencing factors such as wavelet basis selection, threshold function selection, and decomposition layer selection, which have a certain impact on the accuracy of signal denoising and the amount of calculation is relatively large. In view of this, this paper proposes a new signal denoising method to solve the above problems. Summary of the invention

[0004] The purpose of the present invention is to provide an electromyographic signal denoising processing method based on RIME optimized SVMD, by using SVMD to decompose the electromyographic signal, and at the same time optimizing the maximum balance parameter of SVMD through RIME and improving the decomposition accuracy, calculating the NCC of each IMF after decomposition, directly removing the component with NCC less than 0.3, and performing wavelet threshold denoising processing on the component with NCC greater than 0.3, so as to better retain the complete characteristics of the original electromyographic signal for subsequent research.

[0005] In order to solve the above technical problems, the present invention proposes the following technical solutions:

[0006] A method for denoising an electromyographic signal based on RIME optimized SVMD, comprising the following steps:

[0007] S1, obtain the original electromyographic signal, and adaptively iterate and search for the optimal balance parameters of the SVMD algorithm through the RIME algorithm;

[0008] S2, using SVMD to perform successive variational modal decomposition on the original electromyographic signal to obtain the decomposed IMF;

[0009] S3. Calculate the NCC corresponding to each IMF, taking 0.3 as the standard. The part greater than 0.3 is considered as a useful signal, and the part less than 0.3 is considered as a useless signal;

[0010] S4. De-noise and reconstruct the signal with NCC greater than 0.3 to obtain the electromyographic signal after noise reduction.

[0011] Optionally, the step S1 specifically includes:

[0012] S11 collects the original electromyographic signal of the subject through the electromyographic signal sensor, amplifies the collected original electromyographic signal, and performs fitting and normalization processing;

[0013] S12 sets the fitness function of the RIME algorithm;

[0014] S13 initializes the frost particle population of the RIME algorithm, sets the range of the maximum equilibrium parameter α, and calculates the fitness value of the particles;

[0015] S14 uses a soft frost search strategy to simulate the condensation process of particles and calculate the position of particles;

[0016] S15 uses the hard frost puncture mechanism to update the position of particles;

[0017] S16 performs a greedy selection mechanism to compare the fitness values ​​of the frost crystals after and before the update and take the optimal value;

[0018] S17 determines whether the termination condition is met, if so, outputs α, if not, returns to S14;

[0019] Specifically, the minimum envelope spectrum peak factor of the original electromyographic signal is used as the fitness function of the RIME algorithm, and the minimum value of the envelope spectrum peak factor of the original electromyographic signal is used as the fitness of the individual particle.

[0020] Optionally, the envelope spectrum peak factor of the original electromyographic signal is calculated by the following formula:

[0021]

[0022] Where X(z) represents the envelope spectrum amplitude sequence, X(z) (z=1,2,3...Z); Ec is a dimensionless index.

[0023] Optionally, the S14 specifically calculates the position of the particle by the following formula:

[0024]

[0025] Among them, R ij new is the new position of the updated particle, i and j represent the jth particle in the i-th frost crystal; R best.j is the jth particle of the best frost crystal in the frost population R; parameter r1 is a random number in the range of (-1, 1); cosθ changes with the number of iterations, and θ controls the direction of movement of the particle; h represents the adhesion, which is a random number in the range of (0, 1); β is the environmental factor, where the mathematical model of β is a step function, ω1 represents the number of segments of the step function and defaults to 5, and [] represents rounding operation; Ub ij and Lb ij are the upper and lower bounds of the frost space, which limit the effective area of ​​particle movement; t1 represents the current number of iterations, T2 represents the maximum number of iterations; r2 is a random number in the range of (0,1), and E is the adhesion coefficient. The expression of E is:

[0026]

[0027] r2 and E are random values ​​in the interval (0,1), which determine whether the particles are condensed or not, that is, whether the particle positions are updated.

[0028] Optionally, the S15 calculates and updates the particle position of the hard frost by the following formula:

[0029] In the stage of implementing the hard frost ice puncture mechanism, the replacement formula between particles is as follows:

[0030] R ij new =R best.j ,r3 <F normr (Si );

[0031] Where r3 is a random number in the range of (-1,1); S i F stands for frost crystals. normr (S i ) represents the normalized value of the current crystal fitness value; finally, the positive greedy selection mechanism stage is carried out to compare the fitness value of the frost crystal after the update with that before the update. If the fitness after the update is better than the value before the update, the updated value replaces the value before the update.

[0032] Optionally, the step S2 specifically includes:

[0033] S21, the original electromyographic signal is defined to be decomposed into L orders, and the mode corresponding to each order is u L (t), then each mode u L (t) constructing the Lth-order mode minimization criterion J1 compactly around its center frequency and minimizing the minimization criterion J1;

[0034] S22, constructing a minimization criterion J2 so as to minimize the frequency spectrum overlap between the residual signal and the L-th order mode;

[0035] S23, updating the impulse frequency response of the filter by the minimization criterion J1 and the minimization criterion J2, and establishing an additional criterion J3;

[0036] S24, solving the variational problem based on the minimization criterion J1, the minimization criterion J2 and the additional criterion J3, and constructing an augmented Lagrangian formula;

[0037] S25, using a multiplier algorithm to perform an iterative update optimization on the mode u(t), the center frequency ω in the SVMD decomposition, and the input signal f(t), and correct the Lagrange multiplier λ by a cross multiplier method.

[0038] Optionally, the signal decomposition in step S21 is expressed as:

[0039] f(t)=u L (t)+f r (t);

[0040] The residual signal f r (t) is except u L (t), it consists of two parts: the L-1 order mode and the unprocessed part of the signal f u (t), that is

[0041]

[0042] The minimization criterion J1 is expressed as:

[0043]

[0044] ω L is the center frequency of the Lth mode; * is the convolution operator, is the partial derivative with respect to time t, δ(t) is the Dirac function, t represents the time series of sampling, and j is an imaginary number and has no practical meaning.

[0045] Optionally, the step S22 minimization criterion J2 is expressed as: in order to make the residual signal f r (t) and u L (t) by minimizing the spectral overlap between them, which is achieved by using appropriate filters, The expression for the frequency response is:

[0046]

[0047] In order to make β L (ω) filter out f r (t) energy is minimized, so the J2 criterion is established to minimize the spectrum overlap between the residual signal and the Lth mode:

[0048]

[0049] where β L (t) represents the impulse response of the Lth-order mode;

[0050] The additional criterion of step S23 is:

[0051]

[0052] The construction formula of step S24 is:

[0053]

[0054] When L-1 constraints are known, the Lth mode can be transformed into a constraint minimization problem. According to the constraints in the above formula, the combination of J1, J2 and J3 is minimized, and the minimization constraint model is established:

[0055]

[0056] Where α is the maximum equilibrium parameter, which can be solved using the Lagrange multiplier method;

[0057]

[0058] Where λ is the Lagrange multiplier. According to Parseval's theorem, the above formula can be transformed and solved using the cross multiplication method to finally obtain the decomposed component IMF.

[0059] Optionally, the step S3 specifically includes:

[0060] S31 calculates the correlation coefficient NCC between each IMF and the original signal using the following formula:

[0061]

[0062] In the above formula, f(t) represents the original signal, Represented as the signal of the IMF, represents the mean of the original signal, Represents the mean of the IMF signal.

[0063] After calculating the NCC of each IMF, S32 retains the noise component greater than 0.3 and removes the component less than 0.3.

[0064] Optionally, the step S4 specifically includes:

[0065] S41 performs wavelet threshold processing on the retained electromyographic signal;

[0066] S42 reconstructs the denoised signal to obtain a denoised signal.

[0067] Compared with the existing technology, it has the following beneficial effects:

[0068] The present invention decomposes noisy electromyographic signals through the SVMD algorithm. The optimization in SVMD can be approximately regarded as K one-dimensional optimization problems. Compared with the existing VMD, it has lower computational complexity and can well avoid the trouble of determining the number of IMFs related to VMD. At the same time, it also avoids the problem that the initial center frequency is difficult to determine in VMD, that is, it can improve the efficiency and effect of electromyographic signal decomposition, thereby improving the accuracy of useful signal feature extraction.

[0069] On the basis of signal decomposition through SVMD algorithm, the present invention optimizes the balance parameter selection of SVMD through RIME (Frost Ice Optimization Algorithm) to address the problem that SVMD algorithm needs to pre-set balance parameters and different balance parameter values ​​determine the SVMD decomposition accuracy. The RIME algorithm is simple to implement, has strong global search capability, fast solution accuracy and convergence speed, and efficiently and accurately selects the optimal balance parameters for the SVMD algorithm to give full play to its signal decomposition performance, that is, it can achieve high-precision signal decomposition to obtain several IMF components, and denoise the useful signal to obtain useful information. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present invention but do not constitute a limitation of the present invention. In the accompanying drawings:

[0071] Figure 1 This is the overall denoising flow chart of the patent of this invention.

[0072] Figure 2 This is the normalized electromyographic signal diagram of the patent of this invention.

[0073] Figure 3 This is the fitness function curve of the patent of the present invention.

[0074] Figure 4 This is the result diagram of SVMD decomposition of electromyographic signals according to the patent of this invention.

[0075] Figure 5 This is the NCC calculation result diagram of the patent of this invention.

[0076] Figure 6 This is the phase diagram of the chaotic attractor trajectory before and after the electromyographic signal denoising of the patented invention. DETAILED DESCRIPTION

[0077] The present invention is further described below in conjunction with the accompanying drawings:

[0078] 1. Explanation of abbreviations in this article:

[0079] RIME: Rime optimization algorithm (frost ice optimization algorithm);

[0080] SVMD: Successive variational mode decomposition;

[0081] IMF: Intrinsic Mode Function (intrinsic mode component);

[0082] NCC: Normalized Cross Correlation.

[0083] See also Figure 1 As shown, the electromyographic signal denoising method based on RIME-SVMD provided by the present invention comprises the following steps:

[0084] S1, obtain the original electromyographic signal, and adaptively iterate and search for the optimal balance parameters of the SVMD algorithm through the RIME algorithm;

[0085] S2, using SVMD to perform successive variational modal decomposition on the original electromyographic signal to obtain the decomposed IMF;

[0086] S3. Calculate the NCC corresponding to each IMF, taking 0.3 as the standard. The part greater than 0.3 is considered as a useful signal, and the part less than 0.3 is considered as a useless signal;

[0087] S4. De-noise and reconstruct the signal with NCC greater than 0.3 to obtain the electromyographic signal after noise reduction.

[0088] The following description will provide detailed descriptions of steps S1 to S4.

[0089] Step S1 specifically includes:

[0090] S11, collect the original electromyographic signal information of the subject through the electromyographic signal sensor, amplify the collected signal, and perform fitting and normalization processing, such as Figure 2 As shown;

[0091] S12 sets the fitness function of the RIME algorithm;

[0092] S13 initializes the frost particle population of the RIME algorithm, sets the range of the maximum balance parameter α, and calculates the fitness value of the particles. The results are as follows Figure 3 As shown;

[0093] S14 uses a soft frost search strategy to simulate the condensation process of particles and calculate the position of particles;

[0094] S15 uses the hard frost puncture mechanism to update the position of particles;

[0095] S16 performs a greedy selection mechanism to compare the fitness values ​​of the frost crystals after and before the update and take the optimal value;

[0096] S17 determines whether the termination condition is met, if so, outputs α, if not, returns to S14.

[0097] The minimum envelope spectrum peak factor of the original signal is used as the fitness function of the RIME algorithm, and the minimum value of the envelope spectrum peak factor is used as the fitness of the individual particle.

[0098] The envelope spectrum peak factor of the signal is calculated using the following formula:

[0099]

[0100] Where X(z) represents the envelope spectrum amplitude sequence X(z) (z=1,2,3…Z); Ec is a dimensionless index that takes into account the periodicity and intensity of the signal impact component. The smaller Ec is, the stronger the periodic impact characteristics are and the more obvious the signal characteristics are.

[0101] The particle position of the updated soft cream is calculated by the following formula:

[0102] R ij new =R best.j +r1 cosθ×β×(h×(Ub ij -Lb ij )+Lb ij ),r2 <E;

[0103]

[0104] Among them, R ij new is the new position of the updated particle, i and j represent the jth particle in the i-th frost crystal; R best,j is the jth particle of the best frost crystal in the frost population R; parameter r1 is a random number in the range of (-1,1); cosθ changes with the number of iterations, and θ controls the direction of movement of the particle; h represents the adhesion, which is a random number in the range of (0,1); β is the environmental factor, where the mathematical model of β is a step function, ω1 represents the number of segments of the step function and defaults to 5, and [] represents rounding operation; Ub ij and Lb ij are the upper and lower bounds of the frost space, which limit the effective area of ​​particle movement; t1 represents the current number of iterations, T2 represents the maximum number of iterations; r2 is a random number in the range of (0,1); E is the adhesion coefficient, and the expression of E is:

[0105]

[0106] r2 and E are random values ​​in the interval (0,1), which determine whether the particles are condensed or not, that is, whether the particle positions are updated.

[0107] The particle position of the updated hard frost is calculated by the following formula:

[0108] In the stage of realizing the hard frost ice puncture mechanism, the replacement formula between particles is as follows:

[0109]

[0110] Where r3 is a random number in the range of (-1,1); S i F stands for frost crystals. normr (S i ) represents the normalized value of the current crystal fitness value. Finally, the positive greedy selection mechanism stage is carried out to compare the fitness value of the frost crystal after the update with that before the update. If the fitness value after the update is better than the value before the update, the updated value replaces the value before the update.

[0111] Step S2 specifically includes:

[0112] S21, the original electromyographic signal input is defined to be decomposed into L orders, and the mode corresponding to each order is u L (t), then each mode u L (t) constructing the Lth-order mode minimization criterion J1 compactly around its center frequency and minimizing the minimization criterion J1;

[0113] S22, constructing a minimization criterion J2 so as to minimize the frequency spectrum overlap between the residual signal and the L-th order mode;

[0114] S23, updating the impulse frequency response of the filter by the minimization criterion J1 and the minimization criterion J2, and establishing an additional criterion J3;

[0115] S24, solving the variational problem based on the minimization criterion J1, the minimization criterion J2 and the additional criterion J3, and constructing an augmented Lagrangian formula;

[0116] S25, using a multiplier algorithm to perform an iterative update optimization on the mode u(t), the center frequency ω, and the input signal f(t), and correcting the Lagrange multiplier λ by a cross multiplier method.

[0117] As a further improvement of the present invention, the signal decomposition in step S21 is expressed as:

[0118] f(t)=u L (t)+f r (t);

[0119] The residual signal f r (t) is except u L (t), it consists of two parts: the L-1 order mode and the unprocessed part of the signal f u (t), that is:

[0120]

[0121] The minimization criterion J1 is expressed as:

[0122]

[0123] ω L is the center frequency of the Lth mode; * is the convolution operator, is the partial derivative with respect to time t, δ(t) is the Dirac function, t represents the time series, and j is an imaginary number and has no practical meaning.

[0124] As a further improvement of the present invention, the minimization criterion J2 in step S22 is expressed as:

[0125]

[0126] As a further improvement of the present invention, the additional criterion in step S23 is:

[0127]

[0128] As a further improvement of the present invention, the construction formula of step S24 is:

[0129]

[0130] When L-1 constraints are known, the Lth mode can be transformed into a constrained minimization problem. The constraints minimize the combination of J1, j2, and J3, and establish a minimization constraint model:

[0131]

[0132] Where α is the maximum equilibrium parameter, which can be solved using the Lagrange multiplier method.

[0133]

[0134] Where λ is the Lagrange multiplier. According to Parseval's theorem, the above formula can be transformed and solved using the cross multiplication method to finally obtain the decomposed component IMF. The decomposition result is as follows: Figure 4 shown.

[0135] Step S3 specifically includes:

[0136] S31 calculates the correlation coefficient NCC between each IMF and the original signal using the following formula:

[0137]

[0138] In the above formula, f(t) represents the original signal, Represented as the signal of the IMF, represents the mean of the original signal, Represents the mean of the IMF signal.

[0139] After calculating the NCC of each IMF, S32 retains the noise components greater than 0.3 and removes the components less than 0.3. Figure 5 shown.

[0140] Step S4 specifically includes:

[0141] S41 performs wavelet threshold processing on the retained electromyographic signal.

[0142] S42 reconstructs the denoised signal to obtain a denoised signal.

[0143] Specifically, the RIME-SVMD-based electromyographic signal denoising method of the present invention first solves α in SVMD by the RIME algorithm, and decomposes the electromyographic signal by SVMD to obtain component IMF. The NCC of each component is calculated to determine the usefulness of the signal, and the useful signal is reconstructed by wavelet threshold denoising. The denoised signal is obtained for subsequent research.

[0144] Compared with the traditional denoising method, the present invention does not need to manually set parameters, and adaptively decomposes the signal. From the result diagram analysis, this method has a better effect on the denoising of non-stationary signals. It only needs to manually input the signal to be denoised. At the same time, the electromyographic signal with obvious characteristics has an important influence on the accurate identification of the bionic manipulator. Therefore, this method has certain significance for the field of mechanical arms and information processing, and ensures the accuracy of mechanical automation.

[0145] In order to verify the practicality of the denoising method proposed in this paper, this paper uses the electromyographic signal in the Ninapro dataset DB2 library for verification, where the sampling frequency is 2000HZ and the number of sampling points is 2000. The normalized waveform of the electromyographic signal is as follows Figure 2 shown.

[0146] RIME is used to optimize the parameter α of SVMD. The population size is set to 20, the maximum number of iterations is set to 10, the iteration range of α is (0, 2000), and the iteration curve of the minimum envelope peak factor is as follows: Figure 3 As shown. Figure 3 It can be seen that the minimum envelope peak factor of the electromyographic signal reaches the minimum value of 2.5225 in the 7th generation, and the optimal α of the electromyographic signal is 399 at this time.

[0147] After setting the maximum balance parameter α of SVMD, the decomposition result of SVMD on SEMG is as follows: Figure 4 As shown. Figure 4 It can be seen that the main features of the signal exist in IMF1-IMF2. As the frequency increases, the noise of the signal becomes larger and larger, and a large amount of noise mainly exists after IMF3. The NCC of each IMF of the decomposed signal and the original signal is calculated separately, and the calculation results are as follows: Figure 5 As shown in the previous text, we know that the signal with NCC less than 0.3 is directly removed, and the signal with NCC greater than 0.3 is subjected to wavelet threshold denoising. Figure 5 It can be seen that the useful part of the electromyographic signal is IM1-IMF3, and the useless signal is IMF4-IMF5;

[0148] This paper uses the wavelet soft threshold method to denoise the noisy IMF, and reconstructs the denoised IMF to obtain the denoised electromyographic signal. In addition, we evaluate the denoising effect of SEMG by attractor trajectories, where the attractor trajectory is the final state set of the trajectory described in the phase diagram by the solution of the dynamic equation; in terms of the signal, the attractor trajectory of the noise-free signal is usually very smooth and regular, while the attractor trajectory of the noisy signal is usually very disordered and irregular. The attractor trajectories before and after SEMG denoising are shown in Figure 2. Figure 6 shown. Figure 6 (a) and (b) represent the attractor phase diagrams before and after EMG signal denoising. Figure 6 From the analysis, we can see that the attractor trajectory after denoising is more regular and smoother than that before denoising. Therefore, it is believed that the denoising effect is obvious, which also verifies the rationality of the method in this paper.

[0149] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for denoising electromyographic signals based on RIME optimized SVMD, characterized in that: The following steps are involved: S1, obtain the original electromyographic signal, and adaptively iterate and search for the optimal balance parameters of the SVMD algorithm through the RIME algorithm; S2, using SVMD to perform successive variational modal decomposition on the original electromyographic signal to obtain the decomposed IMF; S3. Calculate the NCC corresponding to each IMF, taking 0.3 as the standard. The part greater than 0.3 is considered as a useful signal, and the part less than 0.3 is considered as a useless signal; S4. De-noise and reconstruct the signal with NCC greater than 0.3 to obtain the electromyographic signal after noise reduction.

2. The electromyographic signal denoising method based on RIME optimized SVMD according to claim 1, characterized in that: The step S1 specifically includes: S11 collects the original electromyographic signal of the subject through the electromyographic signal sensor, amplifies the collected original electromyographic signal, and performs fitting and normalization processing; S12 sets the fitness function of the RIME algorithm; S13 initializes the frost particle population of the RIME algorithm, sets the range of the maximum equilibrium parameter α, and calculates the fitness value of the particles; S14 uses a soft frost search strategy to simulate the condensation process of particles and calculate the position of particles; S15 uses the hard frost puncture mechanism to update the position of particles; S16 performs a greedy selection mechanism to compare the fitness values ​​of the frost crystals after and before the update and take the optimal value; S17 determines whether the termination condition is met, if so, outputs α, if not, returns to S14; Specifically, the minimum envelope spectrum peak factor of the original electromyographic signal is used as the fitness function of the RIME algorithm, and the minimum value of the envelope spectrum peak factor of the original electromyographic signal is used as the fitness of the individual particle.

3. The electromyographic signal denoising method based on RIME optimized SVMD according to claim 2, characterized in that: The envelope spectrum peak factor of the original electromyographic signal is calculated by the following formula: Where X(z) represents the envelope spectrum amplitude sequence, X(z) (z=1,2,3...Z); Ec is a dimensionless index.

4. The electromyographic signal denoising method based on RIME optimized SVMD according to claim 2 or 3, characterized in that: The S14 specifically calculates the position of the particle by the following formula: R ij new =R best.j +r1 cosθ×β×(h×(Ub ij -Lb ij )+Lb ij ),r2<E; Among them, R ij new is the new position of the updated particle, i and j represent the jth particle in the i-th frost crystal; R best.j is the jth particle of the best frost crystal in the frost population R; parameter r1 is a random number in the range of (-1, 1); cosθ changes with the number of iterations, and θ controls the direction of movement of the particle; h represents the adhesion, which is a random number in the range of (0, 1); β is the environmental factor, where the mathematical model of β is a step function, ω1 represents the number of segments of the step function and defaults to 5, and [] represents rounding operation; Ub ij and Lb ij are the upper and lower bounds of the frost space, which limit the effective area of ​​particle movement; t1 represents the current number of iterations, T2 represents the maximum number of iterations; r2 is a random number in the range of (0,1), and E is the adhesion coefficient. The expression of E is: r2 and E are random values ​​in the interval (0,1), which determine whether the particles are condensed or not, that is, whether the particle positions are updated.

5. The electromyographic signal denoising method based on RIME optimized SVMD according to claim 2 or 3, characterized in that: The S15 updates the particle position of the hard frost by calculating the following formula: In the stage of implementing the hard frost ice puncture mechanism, the replacement formula between particles is as follows: R ij new =R best.j ,r3<F normr (S i ); Where r3 is a random number in the range of (-1,1); S i F stands for frost crystals. normr (S i ) represents the normalized value of the current crystal fitness value; finally, the positive greedy selection mechanism stage is carried out to compare the fitness value of the frost crystal after the update with that before the update. If the fitness after the update is better than the value before the update, the updated value replaces the value before the update.

6. The electromyographic signal denoising method based on RIME optimized SVMD according to claim 1, 2 or 3, characterized in that: The step S2 specifically includes: S21, the original electromyographic signal is defined to be decomposed into L orders, and the mode corresponding to each order is u L (t), then each mode u L (t) constructing the Lth-order mode minimization criterion J1 compactly around its center frequency and minimizing the minimization criterion J1; S22, constructing a minimization criterion J2 so as to minimize the frequency spectrum overlap between the residual signal and the L-th order mode; S23, updating the impulse frequency response of the filter by the minimization criterion J1 and the minimization criterion J2, and establishing an additional criterion J3; S24, solving the variational problem based on the minimization criterion J1, the minimization criterion J2 and the additional criterion J3, and constructing an augmented Lagrangian formula; S25, using a multiplier algorithm to perform an iterative update optimization on the mode u(t), the center frequency ω in the SVMD decomposition, and the input signal f(t), and correct the Lagrange multiplier λ by a cross multiplier method.

7. The method for denoising electromyographic signals based on RIME optimized SVMD according to claim 6, characterized in that: The signal decomposition in step S21 is expressed as: f(t)=u L (t)+f r (t); The residual signal f r (t) is the number of cells except u L (t), it consists of two parts: the L-1 order mode and the unprocessed part of the signal f u (t), that is The minimization criterion J1 is expressed as: ω L is the center frequency of the Lth mode; * is the convolution operator, is the partial derivative with respect to time t, δ(t) is the Dirac function, t represents the time series of sampling, and j is an imaginary number and has no practical meaning.

8. The method for denoising electromyographic signals based on RIME optimized SVMD according to claim 6, characterized in that: The step S22 minimization criterion J2 is expressed as: r (t) and u L (t) by minimizing the spectral overlap between them, which is achieved by using appropriate filters, The expression for the frequency response is: In order to make β L (ω) filter out f r (t) energy is minimized, so the J2 criterion is established to minimize the spectrum overlap between the residual signal and the Lth mode: where β L (t) represents the impulse response of the Lth-order mode; The additional criterion of step S23 is: The construction formula of step S24 is: When L-1 constraints are known, the Lth mode can be transformed into a constraint minimization problem. According to the constraints in the above formula, the combination of J1, J2 and J3 is minimized, and the minimization constraint model is established: Where α is the maximum equilibrium parameter, which can be solved using the Lagrange multiplier method; Where λ is the Lagrange multiplier. According to Parseval's theorem, the above formula can be transformed and solved using the cross multiplication method to finally obtain the decomposed component IMF.

9. The method for denoising electromyographic signals based on RIME optimized SVMD according to claim 1, 2 or 3, characterized in that: The step S3 specifically includes: S31 calculates the correlation coefficient NCC between each IMF and the original signal using the following formula: In the above formula, f(t) represents the original signal, Represented as the signal of the IMF, represents the mean of the original signal, Represents the mean of the IMF signal. After calculating the NCC of each IMF, S32 retains the noise component greater than 0.3 and removes the component less than 0.

3.

10. The method for denoising electromyographic signals based on RIME optimized SVMD according to claim 1, 2 or 3, characterized in that: The step S4 specifically includes: S41 performs wavelet threshold processing on the retained electromyographic signal; S42 reconstructs the denoised signal to obtain a denoised signal.

Images