Multi-channel surface electromyogram signal noise reduction and de-aliasing method

Abstract

The invention relates to the technical field of medical treatment, in particular to a multi-channel surface electromyogram signal noise reduction and de-aliasing method which comprises the steps that S1, a wireless terminal collector is used for collecting multi-channel surface electromyogram signals (sEMG) at the same time; s2, performing noise reduction on the sEMG by using second-generation wavelet transform; s3, carrying out independent component analysis on the collected data; and S4, verifying a de-aliasing effect through a cross correlation coefficient, the signal subjected to noise reduction by adopting the improved threshold function has the advantages of good approximation degree with an original signal, smooth signal and high signal-to-noise ratio, and the noise reduction effect of the method applied to the electromyographic signal is obviously superior to that of a traditional soft and hard threshold function method.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and in particular to a multi-channel surface electromyography signal noise reduction and de-aliasing method. Background Art

[0002] Surface electromyography (sEMG) is a bioelectric signal collected on the surface of the skin and generated by muscle contraction caused by limb movement. sEMG contains rich information about muscle movement and can reflect the movement pattern of the limbs. Currently, sEMG has been widely used in clinical diagnosis, rehabilitation engineering, sports medicine and other fields.

[0003] The human-machine interface based on electromyographic signals can identify different movement patterns of the human body, control the operation of external environment equipment, control bionic prosthetic hands based on electromyographic signals, and identify falls based on electromyographic signals through processing and pattern recognition of surface electromyographic signals. In the research of human-machine interfaces based on electromyographic signals, in order to realize pattern recognition of multiple movements through electromyographic signals, it is often necessary to use multiple electrodes to collect multiple electromyographic signals at the same time. Due to the natural electric current in the human body, the interconnection between neurons, and the coupling crosstalk between adjacent channels of the sampling electrodes, the signals collected by multi-channel electromyographic signals contain mixed components of adjacent electromyographic signals.

[0004] In summary, developing a multi-channel surface electromyography signal noise reduction and de-aliasing method is still a key issue that needs to be urgently solved in the field of medical technology. Summary of the invention

[0005] In order to solve the problems existing in the prior art of human-machine interface based on electromyographic signals, the present invention can identify different movement modes of the human body, control the operation of external environment equipment, control a bionic prosthetic hand based on electromyographic signals, and identify falls based on electromyographic signals through processing and pattern recognition of surface electromyographic signals. In the research of human-machine interface based on electromyographic signals, in order to realize pattern recognition of multiple movements through electromyographic signals, it is often necessary to use multiple electrodes to collect multiple electromyographic signals at the same time. Due to the natural electric current in the human body, the interconnection between neurons, and the coupling crosstalk between adjacent channels of the sampling electrodes, the signals collected by the multi-channel electromyographic signals have the problem of aliasing components of adjacent electromyographic signals. The present invention provides a multi-channel surface electromyographic signal noise reduction and de-aliasing method.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The present invention provides a multi-channel surface electromyography signal noise reduction and de-aliasing method, comprising: S1, using a wireless terminal collector to simultaneously collect multi-channel surface electromyography signals (sEMG);

[0008] S2, using the second generation wavelet transform to reduce the noise of sEMG;

[0009] S3, perform independent component analysis on the collected data;

[0010] S4. Verify the anti-aliasing effect through the cross-correlation coefficient.

[0011] Furthermore, in step S1, the method of simultaneously collecting multiple surface electromyography signals (sEMG) using a wireless terminal collector is as follows:

[0012] Powered by batteries and using voltage-stabilizing circuits to provide different voltage requirements, multiple motion sensors are attached to the muscle surface of the human body to detect sEMG. A synchronous clock is used to provide a time reference to ensure synchronous signal acquisition of each channel. A bandpass filter and an operational amplifier are used to filter out noise in the signal and retain the main frequency components of the electromyographic signal. The collected sEMG signal is amplified and then converted into a digital signal through an ADC module for further processing by the MCU. The signal collected by the motion sensor is transmitted to the main controller (MCU) using wireless transmission. The MCU classifies and manages the electromyographic data of each channel. The processed data can be stored or transmitted to the human-computer interaction interface, which displays the electromyographic signal in graphical or numerical form.

[0013] Further, in step S2, the method of using the second generation wavelet transform to reduce the noise of sEMG is: 1. splitting;

[0014] The signal sequence is divided into two non-overlapping subsets, usually divided into two sequences of equal length according to the odd samples Xodd[n] and the even samples Xeven[n]: X[n]=Xodd[n]+Xeven[n], Xeven[n]=X[2n](1), Xodd[n]=X[2n+1].

[0015] Further, in step S2, the method of using the second generation wavelet transform to reduce the noise of sEMG is: 2. prediction;

[0016] A prediction operator P is used that is independent of the data structure, and the even samples are kept unchanged. The odd samples are estimated from the even samples: Xodd[n] = P(Xeven[n])(2), where Xodd[n] is the estimated value of the odd samples. When the signal has a local correlation structure, the even sample set and the odd sample set must be highly correlated. One subset is used to estimate another subset with a certain accuracy. The difference between the original value and the predicted value is used as the wavelet or detail coefficient d[n], which is expressed as:

[0017] d[n]=Xodd[n]-P(Xeven[n]) (3).

[0018] Further, in step S2, the method for denoising sEMG using second-generation wavelet transform is as follows: 3. Correction;

[0019] To preserve certain scale characteristics existing in the original signal, a correction operator U is introduced to obtain the scale coefficients at the next scale: U = Xeven[n] - U(d[n]) (4). The above 1. Splitting, 2. Prediction, and 3. Correction constitute a lifting process. Repeating the above lifting process for the output c[n] can form a complete discrete wavelet transform, and the obtained scale coefficients and wavelet coefficients are cj[n] and dj[n] respectively. From the frequency domain perspective, the wavelet coefficient d represents the high-frequency component of the original data, and the scale coefficient c represents the low-frequency component of the original data.

[0020] Further, in step S2, the method for denoising sEMG using second-generation wavelet transform is as follows: Second-generation wavelet threshold selection;

[0021] The wavelet threshold denoising theory is mainly based on setting thresholds for wavelet coefficients. Achieving an ideal denoising effect by setting thresholds for wavelet coefficients is the main content of the wavelet threshold denoising theory: setting wavelet coefficients with small absolute values to zero, making wavelet coefficients with large absolute values shrink or be retained as much as possible, and then inverse-transforming these wavelet coefficients after threshold processing to reconstruct the signal and complete the denoising process. It is proposed to adopt a parameter-based non-linear improved threshold function on the basis of soft and hard threshold functions, which can make the transition between the effective signal and noise smoother, thereby improving the continuity of the filtered signal: where α ∈ [0, 1] is a variable parameter. When α = 0, the threshold function is equivalent to the original signal.

[0022] Further, in step S3, the method for performing independent component analysis on the collected data is as follows:

[0023] Suppose there is a source signal vector, and these source signals are statistically independent at each discrete time point. When the source signals cannot be directly observed, we can collect linearly mixed observed signals through instruments. Then, the ICA separation mathematical model is expressed as: where n(i) is the noise term, which is usually not considered. A = [a 1 , a 2 ···, a N , is an unknown mixing matrix of M*N order. The number of source signals should not be greater than the number of observed signals, that is, N ≥ M. For the convenience of analysis, usually M = N;

[0024] When using the ICA method to perform ICA separation on the signal, it is necessary to first whiten and remove the mean value of the observed data to make the components of the vector X uncorrelated with each other and have unit variance;

[0025] In short, the ICA method is: under the condition that both S and A are unknown, find a separation matrix W so that the output Y (Y=WX) obtained after X passes through it is the optimal approximation of S.

[0026] Furthermore, in step S3, the method of performing independent component analysis on the collected data is:

[0027] Fast ICA is a fixed-point calculation of the ICA method based on the negative entropy criterion. It takes fast optimization as the principle and batch processing as the method. A large amount of sample data is involved in the calculation in each iterative calculation. Usually, the Fast ICA algorithm has forms based on kurtosis, maximum likelihood, maximum negative entropy, etc. The research experiment in this paper is based on the Fast ICA algorithm based on maximum negative entropy, and uses the maximum negative entropy as the search direction to realize the extraction of independent signals. In addition, since the algorithm uses a fixed-point iteration method, the convergence is faster and more robust.

[0028] Furthermore, in step S3, the method of performing independent component analysis on the collected data is:

[0029] 1. De-mean and whiten the observed signal X to make it a vector with zero mean and unit variance and independent components. This process can simplify the ICA algorithm.

[0030] 2. Find the unmixing matrix W based on the negative entropy criterion;

[0031] 3. Separation of independent components is achieved by formula (7);

[0032] The ICA algorithm of Newton's iteration theorem is: w(n+1)=E{X(w(n) T X) 3}-3w(n)(7), where w(n) is obtained by Newton iteration;

[0033] 4. Normalize according to formula (8);

[0034]

[0035] When iterating formula (7), we hope that the value of the objective function will become smaller and smaller as the iteration progresses until it converges. Usually, the separation result of the Fast ICA algorithm is closely related to the starting point, and it cannot be guaranteed that the iteration result will converge. Through experiments, it is found that a relaxation factor E{XXT} related to the input observation signal and the number of sampling points is introduced into the iteration formula, where γ is a variable parameter, so that the iteration formula (7) becomes: w(n+1)=E{X(w(n) T X) 3}-3γE{XX T}w(n)(9), when w(n+1) TWhen w(n) converges to 1, the iteration stops and w(n+1) is the output result. If it does not converge, the iteration algorithm of step (3) continues until |w(n+1) T w(n)|converges.

[0036] Further, in step S4, the method of verifying the anti-aliasing effect through the cross-correlation coefficient is:

[0037] After collecting sEMG signal data, adjacent channel signals are selected for verification, the cross-correlation coefficient between the original signal and the de-aliased signal is calculated, and then the cross-correlation coefficients between adjacent channels are compared to verify the effectiveness of the de-aliasing method.

[0038] Beneficial Effects

[0039] Compared with the known public technology, the technical solution provided by the present invention has the following advantages:

[0040] Beneficial effects:

[0041] The signal after denoising by using the improved threshold function in the present invention has the advantages of good approximation to the original signal, smooth signal and high signal-to-noise ratio. The denoising effect thereof applied to electromyographic signals is obviously better than the denoising results of the traditional soft and hard threshold function methods.

[0042] The electromyographic signal subjected to noise reduction and de-aliasing processing of the present invention has a part of interference caused by the environment and the acquisition instrument removed, which facilitates the feature extraction of the electromyographic signal when subsequent actions occur and provides convenience for research such as action recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A flowchart of a multi-channel surface electromyography signal noise reduction and de-aliasing method of the present invention;

[0044] Figure 2 A second generation wavelet denoising flow chart of a multi-channel surface electromyography signal denoising and de-aliasing method of the present invention;

[0045] Figure 3 A flowchart of the problem solving method of the ICA method of the multi-channel surface electromyography signal noise reduction and anti-aliasing method of the present invention;

[0046] Figure 4 A system framework flow chart of a multi-channel surface electromyography signal noise reduction and de-aliasing method of the present invention;

[0047] Figure 5 It is a flow chart of a data receiving box framework of a multi-channel surface electromyography signal noise reduction and de-aliasing method of the present invention;

[0048] Figure 6It is a flow chart of the hardware framework of a data receiving box of a multi-channel surface electromyography signal noise reduction and de-aliasing method of the present invention;

[0049] Figure 7 The present invention is a flow chart of a collector functional module for a multi-channel surface electromyography signal noise reduction and de-aliasing method. DETAILED DESCRIPTION

[0050] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0051] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0052] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0053] Example:

[0054] like Figure 1 - Figure 7 As shown, the present invention provides a multi-channel surface electromyography signal noise reduction and de-aliasing method, including: S1, using a wireless terminal collector to simultaneously collect multiple surface electromyography signals (sEMG).

[0055] Furthermore, in step S1, the method of simultaneously collecting multiple surface electromyography signals (sEMG) using a wireless terminal collector is as follows:

[0056] Powered by batteries and using voltage stabilization circuits to provide different voltage requirements, motion sensors of multiple acquisition terminals are attached to the muscle surface of the human body to detect sEMG. A synchronous clock is used to provide a time reference to ensure the synchronous acquisition of signals from each channel. A bandpass filter and an operational amplifier are used to filter out noise in the signal and retain the main frequency components of the electromyographic signal. The acquired sEMG signal is amplified and then converted into a digital signal through an ADC module for further processing by the MCU. The signal acquired by the motion sensor is transmitted to the main controller (MCU) through wireless transmission. The MCU classifies and manages the electromyographic data of each channel. The processed data can be stored or transmitted to a data receiving box, which sends it to the computer via USB.

[0057] In this embodiment, the skin is kept clean before use to avoid interference signals. The motion sensor is placed on the surface of the target muscle group. The placement of the motion sensor is adjusted according to different training needs to ensure accurate collection of telecommunication signals of each group of muscles. Using wireless transmission technology, electromyographic data can be transmitted to the terminal device in real time, reducing the constraints of data cables. The data receiving box includes a USB interface board, a data receiving board 1, a data receiving board 2, a medical power supply and a power filter. There are two USB-A female ports on the USB interface board, which are connected to the two data receiving boards respectively. The power supply AC220V is connected from a pin-shaped plug with a power filter. There is a boat-shaped switch on it to control the 220V switch. The medical power supply is divided into two routes and connected to the data receiving board 1 and the data receiving board 2 to provide power for the data receiving board 1 and the data receiving board 2.

[0058] S2. Use the second-generation wavelet transform to reduce the noise of sEMG.

[0059] Further, in step S2, the method of using the second generation wavelet transform to reduce the noise of sEMG is as follows: Figure 2 As shown, 1. split;

[0060] The signal sequence is divided into two non-overlapping subsets, usually divided into two sequences of equal length according to the odd samples Xodd[n] and the even samples Xeven[n]: X[n]=Xodd[n]+Xeven[n], Xeven[n]=X[2n](1), Xodd[n]=X[2n+1].

[0061] Further, in step S2, the method of using the second generation wavelet transform to reduce the noise of sEMG is as follows: Figure 2 As shown, 2. prediction;

[0062] A prediction operator P is used that is independent of the data structure, and the even samples are kept unchanged. The odd samples are estimated from the even samples: Xodd[n] = P(Xeven[n])(2), where Xodd[n] is the estimated value of the odd samples. When the signal has a local correlation structure, the even sample set and the odd sample set must be highly correlated. One subset is used to estimate another subset with a certain accuracy. The difference between the original value and the predicted value is used as the wavelet or detail coefficient d[n], which is expressed as:

[0063] d[n]=Xodd[n]-P(Xeven[n]) (3).

[0064] Further, in step S2, the method of using the second generation wavelet transform to reduce the noise of sEMG is as follows: Figure 2 As shown, 3. Correction;

[0065] In order to maintain certain scale characteristics existing in the original signal, the correction operator U is introduced to obtain the scale coefficient at the next scale: U = Xeven[n]-U(d[n])(4). The above 1. splitting, 2. prediction, 3. correction constitute a lifting process. Repeating the above lifting process on the output c[n] can constitute a complete discrete wavelet transform. The obtained scale coefficient and wavelet coefficient are cj[n] and dj[n] respectively. From the frequency domain perspective, the wavelet coefficient d represents the high-frequency component of the original data, and the scale coefficient and c represent the low-frequency component of the original data.

[0066] Further, in step S2, the method of using the second generation wavelet transform to reduce sEMG noise is as follows: second generation wavelet threshold selection;

[0067] The wavelet threshold denoising theory is mainly based on setting thresholds for wavelet coefficients. The main content of the wavelet threshold denoising theory is to achieve the ideal denoising effect by setting thresholds for wavelet coefficients: set the wavelet coefficients with small absolute values ​​to zero, shrink or retain the wavelet coefficients with large absolute values ​​as much as possible, and then inversely transform these wavelet coefficients that have been processed by thresholds to reconstruct the signal and complete the denoising process. It is proposed to use a parameter-based nonlinear improved threshold function based on the soft and hard threshold functions, which can make the transition between the effective signal and the noise smoother, thereby improving the continuity of the filtered signal: Where α∈[0,1] is a variable parameter. When α=0, the threshold function is equivalent to the original signal.

[0068] In this embodiment, the second-generation wavelet transform is used to denoise the sEMG signal. The lifting wavelet transform is performed on the original sEMG signal through splitting, prediction, and correction steps, and then high-frequency noise is removed through threshold processing. Different from the traditional wavelet transform, the second-generation wavelet transform provides a more flexible framework, which can better eliminate noise and interference while maintaining the effective components of the signal, is beneficial to improving the signal-to-noise ratio of the signal, and provides high-quality input data for subsequent analysis and processing.

[0069] S3. Perform independent component analysis on the collected data.

[0070] Further, in step S3, the method for performing independent component analysis on the collected data is as follows:

[0071] Suppose there is a source signal vector, and these source signals are statistically independent at each discrete time point. When the source signals cannot be directly observed, we can collect linearly mixed observed signals through an instrument. Then, the ICA separation mathematical model is expressed as: where n(i) is the noise term, which is usually not considered. A = [a 1 , a 2 ···, a N , is an unknown mixing matrix of M*N order. The number of source signals should not be greater than the number of observed signals, that is, N ≥ M. For the convenience of analysis, usually M = N;

[0072] When using the ICA method to perform ICA separation on the signal, it is necessary to whiten and remove the mean value of the observed data first, so that the components of the vector X are uncorrelated with each other and have unit variance. The process of solving the problem by the ICA method can be described by the Figure 3 model shown;

[0073] In short, the ICA method is: under the condition that both S and A are unknown, find a separation matrix W such that the output Y (Y = WX) obtained after X passes through it is the best approximation of S.

[0074] Further, in step S3, the method for performing independent component analysis on the collected data is as follows:

[0075] Fast ICA is a fixed-point algorithm of the ICA method based on the negative entropy criterion. Based on the principle of fast optimization and in a batch processing manner, a large number of sample data participate in the calculation in each iteration. Usually, the Fast ICA algorithm has forms based on kurtosis, maximum likelihood, maximum negative entropy, etc. The research experiment in this paper is based on the Fast ICA algorithm based on maximum negative entropy, and the maximum negative entropy is used as the search direction to extract independent signals. In addition, because this algorithm uses the fixed-point iteration method, the convergence is faster and more stable.

[0076] Furthermore, in step S3, the method of performing independent component analysis on the collected data is:

[0077] 1. De-mean and whiten the observed signal X to make it a vector with zero mean and unit variance and independent components. This process can simplify the ICA algorithm.

[0078] 2. Find the unmixing matrix W based on the negative entropy criterion;

[0079] 3. Separation of independent components is achieved by formula (7);

[0080] The ICA algorithm of Newton's iteration theorem is: w(n+1)=E{X(w(n) T X) 3}-3w(n)(7), where w(n) is obtained by Newton iteration;

[0081] 4. Normalize according to formula (8);

[0082]

[0083] When iterating formula (7), we hope that the value of the objective function will become smaller and smaller as the iteration progresses until it converges. Usually, the separation result of the Fast ICA algorithm is closely related to the starting point, and it cannot be guaranteed that the iteration result will converge. Through experiments, it is found that a relaxation factor E{XXT} related to the input observation signal and the number of sampling points is introduced into the iteration formula, where γ is a variable parameter, so that the iteration formula (7) becomes: w(n+1)=E{X(w(n) T X) 3}-3γE{XX T}w(n)(9), when w(n+1) T When w(n) converges to 1, the iteration stops and w(n+1) is the output result. If it does not converge, the iteration algorithm of step (3) continues until |w(n+1) T w(n)|converges.

[0084] In this embodiment, through the fast ICA algorithm, the aliasing components in the sEMG signals of multiple channels can be separated to obtain independent electromyographic signals and avoid mutual interference between signals. Through the negative entropy criterion and the Newton iterative optimization algorithm, the method can quickly find the optimal separation matrix, and after the introduction of the relaxation factor, the convergence speed is further accelerated and the stability of the algorithm is enhanced. The method does not need to know the specific form or characteristics of the source signal in advance, and can perform separation when the signal is complex and the noise is large. It is suitable for real-time processing of multi-channel sEMG signals. Combined with the second-generation wavelet transform denoising, the ICA algorithm can effectively remove the aliasing between signals while reducing noise, thereby ensuring the validity and accuracy of the signal.

[0085] S4. Verify the anti-aliasing effect through the cross-correlation coefficient.

[0086] Further, in step S4, the method of verifying the anti-aliasing effect through the cross-correlation coefficient is:

[0087] After collecting sEMG signal data, adjacent channel signals are selected for verification, the cross-correlation coefficient between the original signal and the de-aliased signal is calculated, and then the cross-correlation coefficients between adjacent channels are compared to verify the effectiveness of the de-aliasing method.

[0088] In this embodiment, the cross-correlation coefficient is calculated to quantitatively evaluate the effectiveness of the anti-aliasing method, rather than relying solely on subjective judgment. When the cross-correlation coefficient is close to zero, it indicates that the signal independence is enhanced and the aliasing is effectively removed.

[0089] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-channel surface electromyography signal noise reduction and de-aliasing method, characterized in that: include: S1. Use a wireless terminal collector to collect multiple surface electromyography signals (sEMG) at the same time; S2, using the second generation wavelet transform to reduce the noise of sEMG; S3, perform independent component analysis on the collected data; S4. Verify the anti-aliasing effect through the mutual correlation coefficient.

2. A multi-channel surface electromyography signal noise reduction and de-aliasing method according to claim 1, characterized in that: In step S1, the method of simultaneously collecting multiple surface electromyography signals (sEMG) using a wireless terminal collector is as follows: Powered by batteries and using voltage-stabilizing circuits to provide different voltage requirements, multiple motion sensors are attached to the muscle surface of the human body to detect sEMG. A synchronous clock is used to provide a time reference to ensure synchronous signal acquisition of each channel. A bandpass filter and an operational amplifier are used to filter out noise in the signal and retain the main frequency components of the electromyographic signal. The collected sEMG signal is amplified and then converted into a digital signal through an ADC module for further processing by the MCU. The signal collected by the motion sensor is transmitted to the main controller (MCU) using wireless transmission. The MCU classifies and manages the electromyographic data of each channel. The processed data can be stored or transmitted to the human-computer interaction interface, which displays the electromyographic signal in graphical or numerical form.

3. A multi-channel surface electromyography signal noise reduction and de-aliasing method according to claim 2, characterized in that: In step S2, the method of using the second generation wavelet transform to reduce the noise of sEMG is:

1. splitting; The signal sequence is divided into two non-overlapping subsets, usually divided into two sequences of equal length according to the odd samples Xodd[n] and the even samples Xeven[n]: X[n]=Xodd[n]+Xeven[n], Xeven[n]=X[2n](1), Xodd[n]=X[2n+1].

4. A multi-channel surface electromyography signal noise reduction and de-aliasing method according to claim 3, characterized in that: In step S2, the method of using the second generation wavelet transform to reduce the noise of sEMG is:

2. prediction; A prediction operator P is used that is independent of the data structure, and the even samples are kept unchanged. The odd samples are estimated from the even samples: Xodd[n] = P(Xeven[n])(2), where Xodd[n] is the estimated value of the odd samples. When the signal has a local correlation structure, the even sample set and the odd sample set must be highly correlated. One subset is used to estimate another subset with a certain accuracy. The difference between the original value and the predicted value is used as the wavelet or detail coefficient d[n], which is expressed as: d[n]=Xodd[n]-P(Xeven[n]) (3).

5. A multi-channel surface electromyography signal noise reduction and de-aliasing method according to claim 4, characterized in that: In step S2, the method of using the second generation wavelet transform to reduce the noise of sEMG is:

3. Correction; In order to maintain certain scale characteristics existing in the original signal, the correction operator U is introduced to obtain the scale coefficient at the next scale: U = Xeven[n]-U(d[n])(4). The above 1. splitting, 2. prediction, 3. correction constitute a lifting process. Repeating the above lifting process on the output c[n] can constitute a complete discrete wavelet transform. The obtained scale coefficient and wavelet coefficient are cj[n] and dj[n] respectively. From the frequency domain perspective, the wavelet coefficient d represents the high-frequency component of the original data, and the scale coefficient and c represent the low-frequency component of the original data.

6. A multi-channel surface electromyography signal noise reduction and de-aliasing method according to claim 5, characterized in that: In step S2, the method of using the second generation wavelet transform to reduce sEMG noise is as follows: second generation wavelet threshold selection; The wavelet threshold denoising theory is mainly based on setting thresholds for wavelet coefficients. The main content of the wavelet threshold denoising theory is to achieve the ideal denoising effect by setting thresholds for wavelet coefficients: set the wavelet coefficients with small absolute values ​​to zero, shrink or retain the wavelet coefficients with large absolute values ​​as much as possible, and then inversely transform these wavelet coefficients that have been processed by thresholds to reconstruct the signal and complete the denoising process. It is proposed to use a parameter-based nonlinear improved threshold function based on the soft and hard threshold functions, which can make the transition between the effective signal and the noise smoother, thereby improving the continuity of the filtered signal: Where α∈[0,1] is a variable parameter. When α=0, the threshold function is equivalent to the original signal.

7. A multi-channel surface electromyography signal noise reduction and de-aliasing method according to claim 6, characterized in that: In step S3, the method of performing independent component analysis on the collected data is: There is a source signal vector, and these source signals are statistically independent at each discrete time point. When the source signals cannot be directly observed, we can collect linearly mixed observed signals through instruments. Then, the ICA separation mathematical model is expressed as: where n(i) is the noise term, which is usually not considered. A = [a1, a2 ···, a N , is an unknown mixing matrix of M*N order. The number of source signals should not be greater than the number of observed signals, that is, N ≥ M. For the convenience of analysis, usually M = N; When using the ICA method to separate the signal ICA, the observed data must first be whitened and de-meaned so that the components of vector X are uncorrelated and have unit variance; In short, the ICA method is: under the condition that both S and A are unknown, find a separation matrix W so that the output Y (Y=WX) obtained after X passes through it is the optimal approximation of S.

8. A multi-channel surface electromyography signal noise reduction and de-aliasing method according to claim 7, characterized in that: In step S3, the method of performing independent component analysis on the collected data is: FastICA is a fixed-point calculation of the ICA method based on the negative entropy criterion. It takes fast optimization as the principle and batch processing as the method. A large amount of sample data is involved in the calculation in each iterative calculation. Usually, the FastICA algorithm has forms based on kurtosis, maximum likelihood, maximum negative entropy, etc. The research experiment in this paper is based on the FastICA algorithm based on maximum negative entropy, and uses the maximum negative entropy as the search direction to realize the extraction of independent signals. In addition, since the algorithm uses a fixed-point iteration method, the convergence is faster and more robust.

9. A multi-channel surface electromyography signal noise reduction and de-aliasing method according to claim 8, characterized in that: In step S3, the method of performing independent component analysis on the collected data is:

1. De-mean and whiten the observed signal X to make it a vector with zero mean and unit variance and independent components. This process can simplify the ICA algorithm.

2. Find the unmixing matrix W based on the negative entropy criterion; 3. Separation of independent components is achieved by formula (7); The ICA algorithm of Newton's iteration theorem is: w(n+1)=E{X(w(n) T X) 3 }-3w(n)(7), where w(n) is obtained by Newton iteration; 4. Normalize according to formula (8); When iterating formula (7), we hope that the value of the objective function will become smaller and smaller as the iteration progresses until it converges. Usually, the separation result of the FastICA algorithm is closely related to the starting point, and it cannot be guaranteed that the iteration result will converge. Through experiments, it is found that a relaxation factor E{XXT} related to the input observation signal and the number of sampling points is introduced into the iteration formula, where γ is a variable parameter, so that the iteration formula (7) becomes = w(n+1) = E{X(w(n) T X) 3 }-3γE{XX T }w(n)(9), when w(n+1) T When w(n) converges to 1, the iteration stops and w(n+1) is the output result. If it does not converge, the iteration algorithm of step (3) is continued until w(n+1) T w(n) converges.

10. A multi-channel surface electromyography signal noise reduction and anti-aliasing method according to claim 9, characterized in that: In step S4, the method for verifying the anti-aliasing effect through the cross-correlation coefficient is: After collecting sEMG signal data, adjacent channel signals are selected for verification, the cross-correlation coefficient between the original signal and the de-aliased signal is calculated, and then the cross-correlation coefficients between adjacent channels are compared to verify the effectiveness of the de-aliasing method.

Images