A time-frequency-space muscle synergy analysis method based on wavelet and non-negative tensor decomposition

Abstract

The present invention discloses a time-frequency-space muscle synergy analysis method based on wavelet and non-negative tensor decomposition, which belongs to the technical field of neurorehabilitation engineering and movement mechanism. The method first performs synchronous acquisition and preprocessing of multi-channel surface electromyographic signals during upper limb movement; then wavelet decomposition is performed on the preprocessed data, and for the time-frequency electromyographic signals after wavelet transformation, the multi-channel time-frequency signals of the same action are constructed into a wavelet signal data tensor; by presetting the number of decomposition layers, the three decomposed matrices and the core tensor are multiplied to obtain a reconstructed data tensor, and then the fitting value FIT is calculated; then the synergy and non-synergy relationship is judged; finally, the spectral components of the synergistic muscles are judged. From the perspective of movement generation and execution, the present invention explores the functional activation state of synergistic muscles in different frequency bands in the muscle synergy model, which helps to reveal the intrinsic pattern of nervous system function and can provide a physiological basis for the movement rehabilitation process of stroke patients.

Description

Technical Field

[0001] The invention relates to the fields of neurorehabilitation engineering and movement mechanism technology, in particular to a time-frequency-space muscle coordination analysis method based on wavelet and non-negative tensor decomposition. Background Art

[0002] According to the "China Stroke Prevention and Treatment Report", the current incidence and mortality of stroke are high, and after a stroke occurs, there is a 75% chance of causing varying degrees of motor dysfunction. The most common clinical symptom of stroke patients is contralateral upper limb hemiplegia, which manifests as muscle weakness or contracture, joint relaxation, etc. How to promote the recovery of upper limb motor function is a hot topic and difficulty in rehabilitation medicine. The central nervous system (CNS) receives incoming information from all over the body and transmits it to the muscles through nerve excitation impulses, exciting the muscles and generating electromyographic signals (EMG) to produce movement. Electromyographic signals contain movement control information and muscle response function information, and have become a powerful means to analyze the progress of limb movement rehabilitation.

[0003] In recent years, muscle synergy theory has emerged in neuroscience research as a neural control mechanism of movement. Muscle synergy is the form of central nervous system (CNS) controlling limb movement. CNS generates control instructions to activate certain muscle groups to complete the corresponding movements. The change in muscle coordination after stroke may form a new movement pattern. Compared with healthy people, stroke patients have different muscle synergy characteristics for different movements. The analysis of muscle synergy is of great significance for exploring the human movement control mechanism and nerve damage. Muscle synergy analysis is mainly based on non-negative matrix decomposition to obtain muscle activation mode and synergy state, providing an overall relative activation level analysis result for all tested muscles. However, traditional non-negative matrix decomposition can only reflect muscle synergy information on a single time scale, and cannot reflect the synergy characteristics in the frequency domain. Electromyographic signals have multi-domain characteristics, so exploring the synergy characteristics in the time-frequency domain can fully understand the patient's motor function.

[0004] Currently, there is still insufficient evidence regarding the neural control mechanism of intermuscular temporal-frequency coordination characteristics after stroke. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a time-frequency-space muscle synergy analysis method based on wavelet and non-negative tensor decomposition, which uses wavelet and non-negative tensor decomposition to extract potential muscle synergy characteristics, and explores the synergy characteristics between upper limb motor muscles from the perspective of neural-driven muscle synergy, aiming to provide a research method and scientific basis for motor function rehabilitation assessment.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0007] A time-frequency-space muscle synergy analysis method based on wavelet and non-negative tensor decomposition specifically includes the following steps:

[0008] Step 1, synchronously collecting multi-channel electromyographic signals;

[0009] Step 2, preprocessing the collected electromyographic signals using data processing software;

[0010] Step 3, performing wavelet transform on the multi-channel electromyographic signal and constructing a wavelet signal data tensor;

[0011] Step 4, determine the number of collaborations;

[0012] Step 5, perform non-negative Tucker decomposition on the data tensor to obtain the time-frequency-space muscle synergy characteristics;

[0013] Step 6, determine the synergistic and non-synergistic relationship and determine the spectral components of the synergistic muscles.

[0014] The further improvement of the technical solution of the present invention is that: in step 1, when collecting multi-channel electromyographic signals related to upper limb wrist flexion and extension movements, the resolution of the acquisition device is set to 16bit and the sampling rate is 2000Hz; before collecting the signal, the subject sits in front of the experimental computer, the subject's shoulder and elbow are supported by the upper limb bracket, the forearm is in a horizontal position, and the forearm is 90 degrees to the body; all subjects complete the wrist movement with the maximum voluntary contraction force of the movement under the guidance of the target picture displayed on the computer screen; and at the same time, the electromyographic signals of multiple muscles under the wrist flexion and extension movement are collected.

[0015] A further improvement of the technical solution of the present invention is that in step 2, preprocessing the electromyographic signal data collected in step 1 specifically includes:

[0016] 2.1 Delete data segments with severe handshake or delayed response;

[0017] 2.2 Remove the mean value and baseline drift, and use an adaptive 50Hz power frequency notch filter to suppress power frequency interference; and downsample all surface electromyography signals to 500Hz;

[0018] 2.3 The signal was filtered with a 150 Hz low-pass filter and full-wave rectified to obtain the surface electromyographic signal envelope for muscle coordination analysis.

[0019] The further improvement of the technical solution of the present invention is that: in step 3, the measured data pre-processed in step 2 is used to construct a multi-channel electromyographic signal x(t) time series; the electromyographic signal x(t) is subjected to wavelet transformation, and the time domain and frequency domain series are constructed into a two-dimensional array EMG frequency×time ; The specific method is as follows:

[0020] 3.1 First construct the wavelet function

[0021] ψ j,k (t) = 2 j / 2 ψ(2 j tk)

[0022] Where ψ(t) is the basic wavelet, k is the translation of ψ(t) in the ordinate direction, and j represents the number of signal layers; j,k∈Z, Z is an integer set, 2 j is the scale parameter, 2 j / 2 is the zoom factor, 2 j t is the translation parameter; t is the time index;

[0023] 3.2 Then decompose the electromyographic signal x(t) to obtain the wavelet transform coefficients:

[0024]

[0025] In the formula, is j,k (t) is the complex conjugate form; the wavelet coefficient C j,k Reconstruct the time-frequency electromyographic signal EMG with frequency domain characteristics frequency×time ;

[0026] Reconstructed time-frequency electromyographic signal EMG based on wavelet transform frequency×time Time series; construct the multi-channel time-frequency signals of the same action into a wavelet signal data tensor X(EMG frequency×time×channel ).

[0027] The further improvement of the technical solution of the present invention is that the specific method of step 4 is as follows:

[0028] By presetting the number of decomposition layers, the three decomposed matrices and the core tensor are multiplied to obtain the reconstructed data tensor, and the fitting value FIT is calculated. The formula is as follows:

[0029]

[0030]

[0031] In the formula, X represents the original tensor constructed, and Y represents the reconstructed data tensor after decomposition by the non-negative tensor decomposition algorithm;

[0032] The optimal decomposition level of the synergy matrix is ​​defined as the minimum synergy number R that achieves an average FIT>90% and an increase in the synergy number by 1 results in an increase in the average FIT of less than or equal to 2%.

[0033] The further improvement of the technical solution of the present invention is that in step 5, the specific method of time-frequency-space muscle synergy characteristic analysis is as follows:

[0034] For the wavelet signal data tensor X(EMG frequency×time×channel ) Perform non-negative tensor decomposition to obtain time-frequency-space muscle synergy characteristics;

[0035] The specific application of non-negative tensor decomposition method to muscle synergy analysis is as follows: According to muscle synergy theory, at any given non-negative tensor X ijn , the non-negative tensor decomposition method can find three non-negative matrices and With a core tensor G, such that

[0036]

[0037] In the formula, X ijn Represents the sEMG raw data tensor of size i×j×n, Y ijn represents the reconstructed data tensor of size i×j×n, where i=8 is the number of recorded muscles, j is the number of sampling points, n is the number of spectrum points, and μ is the number of decomposition levels; As the core tensor, the decomposition yields three non-negative matrices, namely the synergy matrix, the relative contribution weight matrix, and the spectrum component matrix; It reflects the degree of participation of each muscle in a certain movement. reflects the activation degree of the collaborative module, reflects the spectral information of the cooperative module, and × is the modulo-m matrix product operator of the tensor.

[0038] A further improvement of the technical solution of the present invention is that in step 6, based on the characteristic matrix decomposed in step 5 and Determine the synergistic and non-synergistic relationship and the spectral components of the synergistic muscles; specifically:

[0039] 6.1 Determine the relationship between synergy and non-synergy, and the muscle synergy vector Perform maximum value normalization processing, and take the value in the range of [0-1];

[0040] 6.2 Determine the synergistic and non-synergistic relationship between muscles by setting a threshold h; if the normalized W value is greater than h, it is considered that the muscle is significantly activated in the synergistic effect and has a synergistic relationship with the most active muscle, otherwise it does not;

[0041] 6.3 Determine the spectral components of the synergistic muscle and determine the main spectral components of the synergistic muscle by setting a threshold value o; if the P value obtained by decomposition is greater than o, the main spectral components of the corresponding muscle synergistic module are obtained.

[0042] A further improvement of the technical solution of the present invention is that the threshold h is selected as 0.50; the threshold o is selected as 0.10.

[0043] Due to the adoption of the above technical solution, the technical progress achieved by the present invention is:

[0044] 1. Compared with the traditional muscle synergy analysis method, the present invention extracts the time-frequency-spatial characteristics of synergistic modulation in the motion control system, explores the functional activation state of synergistic muscles in different frequency bands in the muscle synergy model from the perspective of motion generation and execution, and can analyze the synergistic characteristics between the time-frequency scales of electromyographic signals, which helps to reveal the intrinsic pattern of nervous system function and provide a physiological basis for the motor rehabilitation process of stroke patients.

[0045] 2. The present invention utilizes wavelet-nonnegative tensor decomposition to extract potential muscle synergy effects, and quantitatively describes the muscle time-frequency-space synergy and information transmission characteristics based on the spectral characteristics of the significantly activated muscle-to-synergistic muscle; from the perspective of neural-driven muscle synergy, it helps to explore and study the motor control feedback mechanism and the pathological mechanism of movement disorders, and establish a rehabilitation status evaluation index based on electromyographic signals, which can provide a physiological basis for the motor rehabilitation process of stroke patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a work flow chart of the present invention;

[0047] Figure 2 It is a diagram showing the number of coordinated wrist flexion and muscle time-frequency-space coordinated characteristics in a healthy person in the present invention;

[0048] Figure 3 It is a diagram of the coordinated number of wrist extensions of healthy people and the time-frequency-space coordinated characteristics of muscles in the present invention. DETAILED DESCRIPTION

[0049] The present invention is further described in detail below with reference to the accompanying drawings and embodiments:

[0050] The electromyographic signal is very weak, and has the characteristics of nonlinearity, non-stationarity, and prominent frequency domain characteristics. During the movement process, the interactive control mechanism between the nervous system and muscles can be reflected through muscle synergy analysis. Wavelet decomposition can extract the time-frequency data segments of the electromyographic signal, and non-negative tensor decomposition can characterize the multi-domain synergy characteristics and information transfer characteristics between signals. The present invention explores the time-frequency synergy mechanism by studying the wavelet-non-negative tensor analysis between electromyography. It is of great significance to analyze how the central nervous system relies on muscle synergy to activate different motor functions of muscles in the time-frequency domain. Then study the physiological mechanism of motor dysfunction.

[0051] Example:

[0052] like Figure 1 As shown, a time-frequency-space muscle synergy analysis method based on wavelet and non-negative tensor decomposition includes the following steps:

[0053] Step 1, using the Wireless EMG wireless synchronous acquisition system to synchronously acquire 8 channels of electromyographic signals;

[0054] Synchronous acquisition of multi-channel surface electromyography signals during upper limb movement:

[0055] Use with electrode patch sensor to collect the superficial surface muscles of right upper limb: biceps brachii (BB), brachioradialis (B), flexor carpi radialis (FCR), palmaris longus (PL), extensor carpi radialis (ECR), extensor digitorum (ED), extensor carpi ulnaris (ECU) and flexor digitorum superficialis (FDS) as research object, it is considered to be the main muscle group involved in wrist movement, meet research needs. First, wipe the skin surface of the measured part with alcohol, remove skin surface grease and dandruff, then paste the electromyographic electrode on the muscle belly position along the muscle fiber direction. In the present invention, the resolution of the acquisition device is set to 16bit, and the sampling frequency is 2000Hz.

[0056] Step 2: Use Matlab data processing software to pre-process the collected EMG signal data. First, delete the data segments with severe handshakes or delayed responses. Then remove the mean and baseline drift, and use an adaptive 50Hz power frequency notch filter to suppress power frequency interference. And downsample all surface EMG signals to 500Hz. Finally, filter the signal with a 150Hz low-pass filter, and full-wave rectify to obtain the surface EMG signal envelope for muscle coordination analysis.

[0057] Step 3, based on the measured data preprocessed in step 2, construct the time series of the electromyographic signal x(t); use the complex Morlet wavelet CMOR6-1 with bandwidth parameters to perform wavelet transform on the electromyographic signal x(t) to obtain the signal's one-dimensional time domain feature t and one-dimensional frequency domain feature f, and construct the time domain and frequency domain series into a two-dimensional array EMG frequency×time ;

[0058] First, construct the wavelet function

[0059] ψ j,k (t) = 2 j / 2 ψ(2 j tk)

[0060] Where ψ(t) is the basic wavelet, k is the translation of ψ(t) in the ordinate direction, j represents the number of signal layers, j,k∈Z, Z is an integer set, 2 j is the scale parameter, 2 j / 2 is the zoom factor, 2 j t is the translation parameter, t is the time index;

[0061] Then the electromyographic signal x(t) is decomposed to obtain the wavelet transform coefficients;

[0062]

[0063] In the formula, is j,k (t) is the complex conjugate form; the wavelet coefficient C j,k Reconstruct the EMG signal with frequency domain characteristics frequency×time , and construct the multi-channel time-frequency signals of the same action into a wavelet signal data tensor X(EMG frequency ×time×channel );Construct the multi-channel time-frequency signals of the same action into a wavelet signal data tensor X(EMG frequency ×time×channel ).

[0064] Step 4, determine the number of collaborations;

[0065] Specifically, by presetting the number of decomposition layers, the three decomposed matrices and the core tensor are multiplied to obtain the reconstructed data tensor, and the fitting value FIT is calculated:

[0066]

[0067]

[0068] In the formula, X represents the original tensor constructed, and Y represents the reconstructed data tensor after decomposition by the non-negative tensor decomposition (NTD) algorithm;

[0069] The optimal decomposition level of the synergy matrix is ​​defined as the minimum synergy number R that achieves an average FIT>90% and an increase in the synergy number by 1 results in an increase in the average FIT of less than or equal to 2%.

[0070] Step 5, perform non-negative Tucker decomposition on the data tensor to obtain the time-frequency-space muscle synergy characteristics;

[0071] Non-negative Tucker decomposition is a method in non-negative tensor decomposition;

[0072] Muscle synergy, a fixed activation pattern of a group of muscles, is a simplified strategy of the central nervous system to overcome the problem of multiple degrees of freedom in motor control.

[0073] Specifically: According to the muscle synergy theory, the activity state of the muscle can be expressed as a linear combination of the muscle synergy module, the activation scale coefficient and the spectral component of the muscle synergy module:

[0074]

[0075] In the formula, X ijn Represents the sEMG raw data tensor of size i×j×n, Y ijn represents the reconstructed data tensor of size i×j×n, where i=8 is the number of recorded muscles, j is the number of sampling points, n is the number of spectrum points, and μ is the number of decomposition levels; As the core tensor, three non-negative matrices are decomposed. They are the synergy matrix, the relative contribution weight matrix and the spectrum component matrix. Among them, It reflects the degree of participation of each muscle in a certain movement. reflects the activation degree of the collaborative module, reflects the spectral information of the cooperative module. Where × is the modulo-m matrix product operator of the tensor.

[0076] Step 6, determine the synergistic and non-synergistic relationship and determine the spectral components of the synergistic muscles;

[0077] Specifically: for muscle synergy vector Perform maximum value normalization and take the value in the range of [0-1]. Determine the synergistic and non-synergistic relationship between muscles by setting the threshold h. If the normalized W value is greater than h, it is considered that the muscle is significantly activated in the synergistic effect and has a synergistic relationship with the most active muscle, otherwise it does not. h is selected as 0.50.

[0078] The main spectral components of the synergistic muscle are determined by setting a threshold value o. If the P value obtained by decomposition is greater than o, the main spectral components of the corresponding muscle synergistic module are obtained. o is selected as 0.10.

[0079] In order to verify the feasibility and effectiveness of the wavelet and non-negative tensor time-frequency-space muscle synergy analysis method described in the present invention, the experimental part recruited 10 healthy subjects (7 males and 3 females. The average age was 24.0±3.0 years old) to conduct upper limb wrist flexion and extension movement experiments. All subjects signed informed consent. The experiment was conducted in a quiet shielded room. The subjects sat upright in comfortable chairs. To avoid the influence of fatigue, all subjects were in good condition and had not done strenuous exercise in the last 24 hours. The subjects were able to complete the corresponding movements with maximum voluntary contraction (MVC). During the experiment, the subjects sat in front of the experimental computer. The subjects' shoulders and elbows were supported by upper limb brackets, which could help the forearms to be in a horizontal position with the forearms at 90 degrees to the body. All subjects completed wrist movements with MVC under the guidance of the target picture displayed on the computer screen.

[0080] Before the experiment, the skin was wiped with 75% medical alcohol to remove dandruff, and then the electrodes were pasted on the corresponding muscle positions along the direction of the muscle fibers according to anatomical knowledge. According to the myoelectric acquisition and analysis process of the present invention, the myoelectric signals of the subjects under wrist flexion and extension movements were synchronously acquired, and the muscle time-frequency coordination characteristics and information transmission mechanism during the subjects' movements were analyzed and studied.

[0081] Under the monitoring of the wireless surface electromyography synchronous acquisition device, 8 muscles (biceps brachii (BB), brachioradialis (B), flexor carpi radialis (FCR), palmaris longus (PL), extensor carpi radialis (ECR), extensor digitorum (ED), extensor carpi ulnaris (ECU) and flexor digitorum superficialis (FDs)) were synchronously acquired. Its sensor can acquire EMG signals with a bandwidth of 0.5-450Hz and a sampling frequency of 2000Hz.

[0082] Average muscle temporal-frequency-spatial characteristics of 10 healthy subjects during wrist flexion and extension Figure 2 As shown in Figure 3.

[0083] Figure 2 In the figure, under wrist flexion, when the number of collaborative decomposition layers R = 3, the average FIT of the reconstructed data capability is 91.01%>90%, and with the increase of K, the increase of FIT is <2%. In the three-layer collaborative units W1, W2, and W3 (the leftmost column), the threshold h = 0.5 is used to divide the muscle pairs to represent the collaborative relationship in the collaborative module. In W1, the "synergistic" muscle groups can be divided into (PL, FDS), in W2, the "synergistic" muscle groups can be divided into (BB), and in W3, the "synergistic" muscle groups can be divided into (ECR, FCR). In the three-layer collaborative units W1, W2, and W3 (the middle column), the threshold o = 0.1 is used to divide the main spectral components of the corresponding muscle collaborative modules. In W1, the main spectral components corresponding to the synergistic muscles are 0-20 Hz, in W2, the main spectral components corresponding to the synergistic muscles are 5-30Hz, and in W3, the main spectral components corresponding to the synergistic muscles are 25-50Hz.

[0084] Figure 3In the figure, under wrist extension, when the number of collaborative decomposition layers R = 3, the average FIT of the reconstructed data capability is 90.22%>90%, and with the increase of K, the increase of FIT is <2%. Similarly, in W1, the "synergistic" muscle groups can be divided into (ED, ECU), in W2, the "synergistic" muscle groups can be divided into (BB), and in W3, the "synergistic" muscle groups can be divided into (ECR, B). In the three-layer collaborative units W1, W2, and W3 (middle column), similarly, in W1, the main spectral components corresponding to the synergistic muscles are 0-20Hz, in W2, the main spectral components corresponding to the synergistic muscles are 15-30Hz, and in W3, the main spectral components corresponding to the synergistic muscles are 7-50Hz.

[0085] The above results show that in the synergistic unit, some muscle pairs with a high synergistic relationship have a strong coupling strength in a certain characteristic functional frequency band. In addition, the spectral components of different synergistic muscles under the same action are different, and the spectral components of synergistic muscles between different actions contain similarities. This provides a theoretical research method for exploring the synergistic mechanism of neuromuscular function.

[0086] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A time-frequency-space muscle synergy analysis method based on wavelet and non-negative tensor decomposition, characterized by: The specific steps include: Step 1, synchronously collecting multi-channel electromyographic signals; Step 2, preprocessing the collected electromyographic signals using data processing software; Step 3, performing wavelet transform on the multi-channel electromyographic signal and constructing a wavelet signal data tensor; The measured data preprocessed in step 2 is used to construct a multi-channel EMG signal x(t) time series; the EMG signal x(t) is subjected to wavelet transform using the complex Morlet wavelet CMOR6-1 with bandwidth parameters to obtain the signal's one-dimensional time domain characteristics and one-dimensional frequency domain characteristics, and the time domain and frequency domain series are constructed into a two-dimensional array EMG frequency×time ; The specific method is as follows: 3.1 First construct the wavelet function ψ j,k (t)=2 j2 ψ(2 j tk) Where ψ(t) is the basic wavelet, k is the translation of ψ(t) in the ordinate direction, and j represents the number of signal layers; j,k∈Z, Z is an integer set, 2 j is the scale parameter, 2 j / 2 is the zoom factor, 2 j t is the translation parameter; t is the time index; 3.2 Then decompose the electromyographic signal x(t) to obtain the wavelet transform coefficients: In the formula, is j,k (t) is the complex conjugate form; the wavelet coefficient C j,k Reconstruct the time-frequency electromyographic signal EMG with frequency domain characteristics frequency×time ; Reconstructed time-frequency electromyographic signal EMG based on wavelet transform frequency×time Time series; construct the multi-channel time-frequency signals of the same action into a wavelet signal data tensor X(EMG frequency×time×channel ); Step 4, determine the number of collaborations; Step 5, perform non-negative Tucker decomposition on the data tensor to obtain the time-frequency-space muscle synergy characteristics; The specific method for analyzing the time-frequency-space muscle synergy characteristics is as follows: For the wavelet signal data tensor X(EMG frequency×time×channel ) Perform non-negative Tucker decomposition to obtain the time-frequency-space muscle synergy characteristics; The specific application of non-negative tensor decomposition method to muscle synergy analysis is as follows: According to muscle synergy theory, at any given non-negative tensor X ijn , the non-negative tensor decomposition method can find three non-negative matrices and With a core tensor G, such that Where, X ijn Represents the sEMG raw data tensor of size i×j×n, Y ijn represents the reconstructed data tensor of size i×j×n, where i=8 is the number of recorded muscles, j is the number of sampling points, n is the number of spectrum points, and μ is the number of decomposition levels; As the core tensor, the decomposition yields three non-negative matrices, namely the synergy matrix, the relative contribution weight matrix, and the spectrum component matrix; It reflects the degree of participation of each muscle in a certain movement. reflects the activation degree of the collaborative module, reflects the spectral information of the cooperative module, and × is the modulo-m matrix product operator of the tensor; Step 6, determine the synergistic and non-synergistic relationship and determine the spectral components of the synergistic muscles; Based on the feature matrix decomposed in step 5 and Determine the synergistic and non-synergistic relationship and the spectral components of the synergistic muscles; specifically: 6.1 Determine the relationship between synergy and non-synergy, and the muscle synergy vector Perform maximum value normalization processing and take the value in the range of [0-1]; 6.2 Determine the synergistic and non-synergistic relationship between muscles by setting the threshold h; If the normalized W value is greater than h, it is considered that the muscle is significantly activated in the synergistic effect and has a synergistic relationship with the most active muscle, otherwise it does not; 6.3 Determine the spectral components of the synergistic muscle and determine the main spectral components of the synergistic muscle by setting a threshold value o; if the P value obtained by decomposition is greater than o, the main spectral components of the corresponding muscle synergistic module are obtained; The threshold h is selected as 0.50; the threshold o is selected as 0.

10.

2. The time-frequency-space muscle synergy analysis method based on wavelet and non-negative tensor decomposition according to claim 1 is characterized by: In step 1, when collecting multi-channel electromyographic signals related to upper limb wrist flexion and extension movements, the resolution of the acquisition device was set to 16 bits and the sampling rate was 2000 Hz; before collecting the signals, the subjects sat in front of the experimental computer, with their shoulders and elbows supported by upper limb brackets, their forearms in a horizontal position, and their forearms at 90 degrees to the body; all subjects completed wrist movements with the maximum voluntary contraction force of the movement under the guidance of the target picture displayed on the computer screen; and at the same time, the electromyographic signals of multiple muscles under wrist flexion and extension movements were collected.

3. The time-frequency-space muscle synergy analysis method based on wavelet and non-negative tensor decomposition according to claim 1 is characterized by: In step 2, preprocessing the electromyographic signal data collected in step 1 specifically includes: 2.1 Delete data segments with severe handshake or delayed response; 2.2 Remove the mean value and baseline drift, and use an adaptive 50Hz power frequency notch filter to suppress power frequency interference; and downsample all surface electromyography signals to 500Hz; 2.3 The signal was filtered with a 150 Hz low-pass filter and full-wave rectified to obtain the surface electromyographic signal envelope for muscle coordination analysis.

4. The time-frequency-space muscle synergy analysis method based on wavelet and non-negative tensor decomposition according to claim 1 is characterized by: The specific method of step 4 is as follows: By presetting the number of decomposition layers, the three decomposed matrices and the core tensor are multiplied to obtain the reconstructed data tensor, and the fitting value FIT is calculated. The formula is as follows: In the formula, X represents the original tensor constructed, and Y represents the reconstructed data tensor after decomposition by the non-negative tensor decomposition algorithm; The optimal decomposition level of the synergy matrix is ​​defined as the minimum synergy number R that achieves an average FIT>90% and an increase in the synergy number by 1 results in an increase in the average FIT of less than or equal to 2%.

Images