A Real-Time Decomposition Method for Separated Vector Alternating Electromyography Based on Muscle Force Level

Abstract

This application relates to the field of surface electromyography (SEMG) decomposition technology, specifically, to a real-time SEMG decomposition method based on alternating vector separation of muscle force levels. This method can, to some extent, address the problem in existing technologies where the accuracy of pre-extracted muscle function (MU) features decreases over time, leading to insufficient accuracy in real-time decoding of motor intent using neural information. The method includes: offline initialization at different neural drive levels, constructing a separation vector group of motor units by extracting motor unit feature information at different neural drive levels; selecting the separation vector group, estimating the neural drive level level from the SEMG signal using a neural drive level classifier, and matching the estimated neural drive level with the separation vector group to obtain a separation vector group B to be detected; and real-time detection of motor unit discharge events, extracting a sequence of motor unit discharge events from the separation vector group B to be detected in real time.

Description

Technical Field

[0001] This application relates to the field of surface electromyography technology, and more specifically, to a real-time electromyography decomposition method based on alternating vector separation of muscle force levels. Background Technology

[0002] The motor unit (MU) discharge information obtained from electromyography (EMG) decomposition is generally considered to be more reflective of muscle activity than traditional surface electromyography information, has a clear physiological basis, and has been shown to have significant advantages in predicting finger kinematics and dynamics.

[0003] Decoding motor intent using neuronal activity information is the most direct method to achieve active conscious control of the human body. However, surface electromyography (EMG) signals are highly mixed signals, which are a temporal and spatial superposition of the action waveform potentials of motor units in muscle fibers. Therefore, it is crucial to accurately extract neuronal activity from surface EMG signals. Currently, offline EMG decomposition methods based on template matching and latent component analysis are widely used. These methods provide a way to obtain neural activity information, but they are still time-consuming and cannot meet the need for real-time acquisition of neural activity information, thus making them unsuitable for real-time control requirements of external devices.

[0004] To achieve real-time extraction of motion unit activity, the offline decomposition method has been improved. Existing real-time decomposition methods still retain the feature information of motion units in the offline stage and directly apply the feature information of the offline stage to the new EMG signal to detect the discharge activity of motion units in real time. For latent component analysis methods, the feature information is the separation vector of motion units.

[0005] However, the feature information of motor units is not static. Therefore, as the level of muscle activation and the recruitment mode of motor units change, the feature information of motor units also changes. The accuracy of the pre-extracted MU features will also decrease over time, which will lead to a significant decrease in the accuracy of motor unit discharge event detection, thereby affecting the accurate estimation of muscle contraction intensity and the accuracy of decoding motor intention in real time through neural information. Summary of the Invention

[0006] To address the issue that the accuracy of pre-extracted MU features in existing technologies decreases over time, leading to insufficient accuracy in real-time decoding of motion intent using neural information, this application provides a real-time decomposition method for alternating vector separation based on muscle strength levels.

[0007] The embodiments of this application are implemented as follows:

[0008] In a first aspect, this application provides a real-time decomposition method for electromyography based on alternating vector separation at muscle strength levels, comprising:

[0009] Offline initialization at different neural drive levels involves extracting the feature information of the motor units at different neural drive levels to form a set of separate vectors for the motor units.

[0010] Select a separation vector group, use a neural drive level classification classifier to estimate the neural drive level from surface electromyography signals, and match the estimated neural drive level with the separation vector group to obtain the separation vector group to be detected.

[0011] Real-time detection of discharge events in the motion unit, and separation of the vector group to be detected. Real-time extraction of the discharge event sequence of the moving unit.

[0012] In one possible implementation, the offline initialization at different neural drive levels, which involves extracting motor unit feature information at different neural drive levels to construct a motor unit separation vector group, further includes:

[0013] Measure the maximum voluntary contraction force (MVC) and complete voluntary contractions at three force levels: high, medium, and low.

[0014] Train a neural drive level classifier that takes EMG amplitude information as input;

[0015] For all neural drive levels, motor units are identified from the electromyographic signals of different neural drives, and feature information of the motor units is extracted. The separation matrices obtained at low, medium, and high drive levels are denoted as B, respectively. l B m and B h ;

[0016] This forms a set of separate vectors.

[0017] In one possible implementation, training the neural drive level classifier, which takes EMG amplitude information as input, further includes:

[0018] Calculate the root mean square of each channel of the electromyography signal within 1 second, and form a dataset labeled with the current neural drive level;

[0019] The training and test sets are divided in a 4:1 ratio to train the classification network until the accuracy requirements are met.

[0020] In one possible implementation, the step of identifying motor units from electromyographic signals driven by different nerves and extracting feature information of the motor units further includes:

[0021] The FastICA algorithm is used to obtain feature information of a specified number of motion units.

[0022] Repetitive motion units are removed by synchronous discharge.

[0023] We obtain n motion units, and their separation vectors are denoted as: s1, s2...s n ;

[0024] The separation matrix is ​​obtained by merging the separation vectors of all motion units.

[0025] In one possible implementation, the formation of the separated vector group further includes:

[0026] B l B m and B h Applying this to the same EMG segment, we obtain the discharge event sequence of all motion units within that EMG segment.

[0027] Identify B l B m and B h The repetitive motion unit in;

[0028] Generate a set of separate vectors for all appearing motion units. and

[0029] In one possible implementation, the B l B m and B h Applied to the same EMG segment, the discharge event sequence of all motion units in that EMG segment is obtained, further including:

[0030] Extend and whiten the 20s EMG signal;

[0031] The extended whitened EMG signal and the separation vector s of motion unit i are then compared. i Multiply to obtain the source signal ss i ;

[0032] After peak detection and K-means clustering, the discharge time series sp is obtained. i .

[0033] In one possible implementation, the identification B l B m and B h The repetitive motion unit in the figure further includes:

[0034] Given the discharge event sequences of motion unit i and motion unit j, respectively spi and sp j sp i and sp j If the synchronous discharge rate exceeds 85%, it is considered to be the same motion unit.

[0035] In one possible implementation, the separation vector set is generated for all occurring motion units. and In this step, when the motor unit fails to extract feature information at a certain neural drive level, feature information from the neighboring level is used to supplement it.

[0036] In one possible implementation, the selection of the separation vector group involves estimating the neural drive level level from surface electromyography signals using a neural drive level level classifier, and matching the estimated neural drive level with the separation vector group to obtain the separation vector group to be detected. Further includes:

[0037] The electromyographic data with a window length of 1 second is input into a trained neural drive level classifier to obtain the level results.

[0038] In one possible implementation, the discharge event of the motion unit is detected in real time, and the detection separation vector group is... Real-time extraction of the firing event sequence of the motion unit, further including:

[0039] The electromyographic data with a window length of 1 second were expanded and whitened, and denoted as Z;

[0040] Multiplying the selected set of separation vectors by the extended whitened electromyographic data Z yields the source signals SS for all motor units, i.e.

[0041] Peak detection and K-means clustering are performed on each row of the source signal, and the weighted average silhouette distance within each row is calculated. When it is less than a certain threshold, the detected discharge is discarded, and the discharge time series SP is finally obtained.

[0042] The technical solution provided in this application can achieve at least the following beneficial effects:

[0043] The real-time electromyography decomposition method based on alternating separation vectors at muscle strength levels provided in this application assigns different separation vectors to motor units by estimating the level of neural drive, thereby making the decomposition results more accurate. It realizes the replacement of feature information of motor units at different levels of neural drive, achieving more accurate extraction of motor unit activity information and thus improving the accuracy of muscle activity state estimation. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a schematic flowchart illustrating an exemplary embodiment of the present application of a method for real-time decomposition of electromyography based on muscle strength level using alternating vector separation.

[0046] Figure 2 This is a flowchart illustrating a specific implementation of a real-time electromyography decomposition method based on muscle strength level using alternating vector separation, as shown in another exemplary embodiment of this application.

[0047] Figure 3 This is a schematic diagram showing the results of a simulation experiment verifying an exemplary embodiment of this application. Detailed Implementation

[0048] To make the objectives, implementation methods and advantages of this application clearer, the exemplary implementation methods of this application will be clearly and completely described below with reference to the accompanying drawings of the exemplary embodiments of this application. Obviously, the exemplary embodiments described are only some embodiments of this application, and not all embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0049] It should be noted that the brief descriptions of terms in this application are only for the convenience of understanding the embodiments described below, and are not intended to limit the embodiments of this application. Unless otherwise stated, these terms should be understood in their ordinary and common meaning.

[0050] The terms "first," "second," "third," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar or related objects or entities, and do not necessarily imply a specific order or sequence, unless otherwise specified. It should be understood that such terms are interchangeable where appropriate.

[0051] The terms “comprising” and “having”, and any variations thereof, are intended to cover but not exclude inclusion, for example, a product or device that includes a range of components is not necessarily limited to all of the components that are clearly listed, but may include other components that are not clearly listed or that are inherent to such product or device.

[0052] Before explaining the real-time decomposition method of alternating vector separation based on muscle strength level provided in the embodiments of this application, the application scenarios and implementation environment of the embodiments of this application will be introduced first.

[0053] The motor unit (MU) discharge information obtained from electromyography (EMG) decomposition is generally considered to be more reflective of muscle activity than traditional surface electromyography information, has a clear physiological basis, and has been shown to have significant advantages in predicting finger kinematics and dynamics.

[0054] Decoding motor intent using neuronal activity information is the most direct method to achieve active conscious control of the human body. However, surface electromyography (EMG) signals are highly mixed signals, which are a temporal and spatial superposition of the action waveform potentials of motor units in muscle fibers. Therefore, it is crucial to accurately extract neuronal activity from surface EMG signals. Currently, offline EMG decomposition methods based on template matching and latent component analysis are widely used. These methods provide a way to obtain neural activity information, but they are still time-consuming and cannot meet the need for real-time acquisition of neural activity information, thus making them unsuitable for real-time control requirements of external devices.

[0055] To achieve real-time extraction of motion unit activity, the offline decomposition method has been improved. Existing real-time decomposition methods still retain the feature information of motion units in the offline stage and directly apply the feature information of the offline stage to the new EMG signal to detect the discharge activity of motion units in real time. For latent component analysis methods, the feature information is the separation vector of motion units.

[0056] However, the feature information of motor units is not static. Therefore, as the level of muscle activation and the recruitment mode of motor units change, the feature information of motor units also changes. The accuracy of the pre-extracted MU features will also decrease over time, which will lead to a significant decrease in the accuracy of motor unit discharge event detection, thereby affecting the accurate estimation of muscle contraction intensity and the accuracy of decoding motor intention in real time through neural information.

[0057] Based on this, this application provides a real-time electromyography decomposition method based on muscle strength level and alternating vector separation. This method enables the replacement of feature information of motor units at different neural drive levels, thereby achieving more accurate extraction of motor unit activity information and improving the accuracy of muscle activity state estimation.

[0058] Next, the technical solutions of this application and how they solve the aforementioned technical problems will be described in detail through embodiments and in conjunction with the accompanying drawings. The embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this application.

[0059] Figure 1 This is a schematic flowchart illustrating an exemplary embodiment of the present application of a method for real-time decomposition of electromyography based on muscle strength level using alternating vector separation.

[0060] In one exemplary embodiment, such as Figure 1 As shown, a real-time decomposition method for alternating vector electromyography based on muscle strength level is provided. In this embodiment, the method may include the following steps:

[0061] Step 100: Offline initialization at different neural drive levels, by extracting the feature information of the motor units at different neural drive levels to form a set of separation vectors for the motor units.

[0062] Step 200: Select the separation vector group, estimate the neural drive level level from the surface electromyography signal using a neural drive level classifier, and match the estimated neural drive level with the separation vector group to obtain the separation vector group to be detected.

[0063] Step 300: Real-time detection of discharge events in the motion unit, and separation of the vector group to be detected. Real-time extraction of the discharge event sequence of the moving unit.

[0064] Figure 2 This is a flowchart illustrating a specific implementation of a real-time electromyography decomposition method based on muscle strength level using alternating vector separation, as shown in another exemplary embodiment of this application.

[0065] In one possible implementation, such as Figure 2 As shown, its specific implementation methods include:

[0066] Step 1: Offline initialization at different neural drive levels, i.e., constructing a set of motor unit separation vectors by extracting motor unit feature information at different neural drive levels.

[0067] Step 2: Select the separation vector group, that is, use the neural drive level classifier to estimate the neural drive level from the surface electromyography signal, and match one of the separation vector groups obtained in step 1 according to the estimated neural drive level. By inputting the electromyography data with a window length of 1 second into the trained neural drive level classifier, the level result is obtained.

[0068] Step 3: Real-time detection of motion unit discharge events, i.e., based on the separation vector group selected in Step 2. Real-time extraction of the discharge event sequence of the moving unit.

[0069] In step 1, motor unit feature information groups are constructed by extracting motor unit feature information at different neural drive levels. This mainly includes the following steps:

[0070] Step 101: Measure the maximum voluntary contraction force (MVC) and complete the voluntary contraction at three force levels (in this embodiment, based on experience, it is set to 20% MVC, 30% MVC, and 40% MVC).

[0071] Step 102: Train a neural drive level classifier with EMG amplitude information as input (in this embodiment, a convolutional neural network classifier is selected).

[0072] Step 103: Identify motor units in electromyographic signals driven by different nerves and extract feature information of motor units.

[0073] Step 104: For all neural drive levels, repeat step 103. Denote the separation matrices obtained at low, medium, and high drive levels as B, respectively. l B m and B h ;

[0074] Step 105: Construct a set of separate vectors.

[0075] In step 102 above, the process of training the neural drive hierarchy classifier is as follows:

[0076] (1) Calculate the root mean square of each channel of the electromyography signal within 1 second, and reorder them according to the arrangement of the electrode pads. For example, when an 8x8 electrode array is used, the root mean square of the electrode pads is finally sorted into 8x8 format data and input into the network, and the current neural drive level is used as the label to form a dataset.

[0077] (2) The network was built, which included convolutional layers, smoothing layers and fully connected layers. Except for the output layer which used Softmax, the ReLU activation function was selected, and other parameters were configured.

[0078] (3) Train the classification network until the accuracy requirements are met.

[0079] In step (3) above, the steps for training the classification network are as follows:

[0080] (a) An early-stopping strategy is adopted, with the data being the validation set accuracy and the tolerance number being 50 epochs. That is, if the validation set accuracy does not improve significantly after 50 epochs, training is terminated early.

[0081] (b) The training set and validation set are randomly divided in a 4:1 ratio, with a maximum of 500 training iterations.

[0082] (c) After training, save the classifier network model.

[0083] In step 103 above, the process of identifying motor units from electromyographic signals driven by different nerves and extracting feature information of motor units is as follows:

[0084] The FastICA algorithm is used to obtain feature information of a specified number of motion units (80 in this embodiment), and duplicate motion units are removed by synchronous discharge, resulting in n motion units. Their separation vectors are denoted as s1, s2...s n The separation matrix is ​​obtained by merging the separation vectors of all motion units.

[0085] In step 105 above, the process of constructing the separated vector group is as follows:

[0086] (1) B l B m and B h The data is applied to the same EMG segment (20 seconds of EMG data in this embodiment) to obtain the discharge event sequence of all motion units in that EMG segment.

[0087] (2) Identify B l B m and B h The repetitive motion unit in the middle.

[0088] (3) Finally, generate a set of separation vectors for all appearing motion units. and

[0089] In step (1), the process of calculating the discharge sequence of the motion unit is as follows:

[0090] The 20s EMG signal is expanded and whitened, and then compared with the separation vector s of motion unit i. i Multiply to obtain the source signal ss i After peak detection and K-means clustering, the discharge time series sp was obtained. i .

[0091] In step (2), the process of identifying the same motion unit is as follows:

[0092] Given the discharge event sequences of motion unit i and motion unit j, respectively sp i and sp j .

[0093] sp i and sp j If the synchronous discharge rate exceeds 85%, it is considered to be a single moving unit. The synchronicity can be calculated as 2N. com / (N1+N2), where Ncom represents the number of discharges synchronized between the two discharge event sequences, and N1 and N2 represent the number of discharges in the two discharge sequences, respectively.

[0094] In step (3), the process of generating the separated vector group is as follows:

[0095] If the motor unit fails to extract feature information at a certain neural drive level, feature information from the neighboring level is used to supplement it.

[0096] In step 3 above, the process of extracting the real-time discharge event sequence of the motion unit is as follows:

[0097] Step 301: Expand and whiten the electromyographic data with a window length of 1 second, and record the processed signal as Z.

[0098] Step 302: Multiply the expanded whitened electromyographic data Z by the selected separation vector set to obtain the source signals SS of all motor units, i.e.

[0099] Step 303: Perform peak detection and K-means clustering on each row of the source signal, and calculate the weighted average silhouette distance within each row. When it is less than a certain threshold (0.70 in this embodiment), discard the detected discharge, and finally obtain the discharge time series SP.

[0100] As can be seen, some embodiments of this application assign different separation vectors to motor units by estimating the neural drive level, thereby making the decomposition results more accurate.

[0101] The offline acquisition of motor unit discharge events is mainly based on the FastICA method. First, the electromyography at different neural drive levels is expanded and whitened. Then, the fixed-point algorithm is used to obtain the local optimum, thereby obtaining the separation vector of the motor unit.

[0102] When generating the set of separate vectors, a certain motor unit may appear in all three neural drive levels, or it may appear only in one drive level. When the motor unit is not found in the current neural drive level, separate vectors from other levels are used to complete it.

[0103] When training a classifier, there are several options for the classifier, such as choosing a convolutional neural network, which can achieve good results. The input of the classifier is the root mean square of the electromyographic signal.

[0104] In real-time decomposition, a classifier is used to estimate the current neural drive level in real time and select the separation vector group. Similarly, in online decomposition, electromyography (EMG) is expanded and whitened, then multiplied by the separation matrix. The separation vectors of motor units are called the feature information of the motor units. The separation vectors of all identified motor units together constitute the separation matrix.

[0105] Figure 3 This is a schematic diagram showing the results of a simulation experiment verifying an exemplary embodiment of this application.

[0106] To verify the effectiveness of the method proposed in this embodiment, simulation data was used to verify the proposed method and its ability to improve the real-time detection accuracy of discharge time by alternating separation vectors. The simulation data used in this example was 30 minutes long, with a neural drive range of 0-45% MVC, generating six data segments. Both traditional methods and the method proposed in this invention were used for real-time extraction of motor unit discharge events. The first 60 seconds of data were used for offline initialization, and each 20 seconds corresponded to low, medium, and high neural drive levels, respectively. The average results of the six simulation data segments are shown below. Figure 3 As shown.

[0107] in, Figure 3 Each point in the figure represents the average decomposition accuracy per minute. The dashed circle marks the discharge event extraction accuracy of the traditional method, while the square marks represent the discharge event extraction accuracy of the method proposed in this invention. It can be seen that the method proposed in this embodiment improves the discharge event extraction accuracy and slows down the decrease in discharge accuracy over time. In the 30-minute real-time extraction process, the accuracy can still be maintained at no less than 85%.

[0108] This invention can automatically adjust the characteristic information of the motor unit, namely the separation vector, according to the level of neural drive, so that the real-time electromyography decomposition method can more accurately detect the discharge activity of the motor unit and improve the estimation ability and stability of muscle contraction intensity.

[0109] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially as indicated, these steps are not necessarily executed in the indicated order. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps.

[0110] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0111] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A real-time decomposition method of sEMG based on the level of muscle strength of the separation vector alternation, characterized in that, The method comprises the following steps: Offline initialization under different neural drive levels, by extracting the feature information of motor units under different neural drive levels to form a separation vector group of motor units; The selection vector group is matched with the estimated neural drive level grade to obtain a to-be-detected separation vector group Real-time detection of motion unit discharge events, the to-be-detected separation vector group Real-time extraction of motion unit discharge event sequences.

2. The real-time decomposition method of sEMG based on isolation vectors alternation according to muscle force level as claimed in claim 1, wherein, The offline initialization under different neural drive levels, by extracting the feature information of motor units under different neural drive levels to form a separation vector group of motor units, further comprises: Measuring the maximum voluntary contraction force (MVC) to complete the voluntary contraction under three force levels; Training a neural drive level classifier with EMG amplitude information as input; For all neural drive levels, motor units are identified from the electromyographic signals of different neural drives, and feature information of the motor units is extracted. The separation matrices obtained at low, medium, and high drive levels are denoted as B, respectively. l B m and B h ; Forming a separation vector group.

3. The real-time decomposition method of sEMG based on isolation vectors alternation according to muscle force level as claimed in claim 2, wherein, The training of the neural drive level classifier with EMG amplitude information as input further comprises: Calculate the root mean square of each channel of the electromyogram signal within 1 second, and take the current neural drive level as the label to form a data set; Divide the training set and test set in the ratio of 4:1, train the classification network until the accuracy requirement is met.

4. The real-time decomposition method of sEMG based on isolation vector alternation according to muscle force level as claimed in claim 2, wherein, The method further comprises the following steps: Obtain the feature information of a specified number of motor units by FastICA algorithm; Remove duplicate motor units by synchronous discharge; n motion units are obtained, and their separation vectors are denoted as s1, s2,..., sn n ; Combine the separation vectors of all motor units to obtain a separation matrix.

5. The real-time decomposition method of sEMG based on isolation vectors alternation according to muscle force level as claimed in claim 2, wherein, The method further comprises the following steps: B l , B m and B h are applied to the same segment of EMG, the sequence of discharge events of all motor units in the segment of EMG is obtained; Identifying B l , B m and B h repeat motion units; generating a set of separation vectors for all occurring motion units and 6. The real-time decomposition method of sEMG based on isolation vectors alternation according to muscle force level as claimed in claim 5, wherein, The B l B m and B h Applied to the same EMG segment, the discharge event sequence of all motion units in that EMG segment is obtained, further including: Expand and whiten the 20s EMG signal; The EMG signal after the expansion whitening is multiplied with the separation vector s of the motor unit i i to obtain the source signal ss i ; After peak detection and K-means clustering, the discharge time series sp is obtained i .

7. The real-time decomposition method of sEMG based on isolation vectors alternation according to muscle force level as claimed in claim 5, wherein, The recognition B l , B m and B h in the repetitive motion unit, further comprising: Given the discharge event sequences of motion unit i and motion unit j, respectively sp i and sp j sp i and sp j If the synchronous discharge rate exceeds 85%, it is considered to be the same motion unit.

8. The real-time decomposition method of sEMG based on isolation vector alternation according to muscle force level of claim 5, wherein, generating a separate set of vectors for all occurring motion units and In the step, when the motion unit does not extract the feature information at a certain level of the neural drive level, the feature information at the adjacent level is supplemented.

9. The real-time decomposition method of sEMG based on isolation vectors alternation according to muscle force level as claimed in claim 1, wherein, The selection separation vector group is matched with the estimated neural drive level grade and the separation vector group according to the estimated neural drive level grade, so as to obtain a to-be-detected separation vector group Further comprising: Input the 1s electromyogram data into the trained neural drive level classifier to obtain the level result.

10. The real-time decomposition method of sEMG based on isolation vectors alternation according to muscle force level as claimed in claim 1, wherein, The motion unit discharge event is detected in real time, and the separated vector group to be detected The motion unit discharge event sequence is extracted in real time, and further comprises: Expand and whiten the 1s electromyogram data, denoted as Z; Using the selected set of separation vectors to left-multiply the expanded whitened EMG data Z, the source signals SS of all motor units are obtained, i.e. Perform peak detection and K-means clustering on each row of the source signal, and calculate the weighted average silhouette distance within each row. When it is less than a certain threshold, discard the detected discharge, and finally obtain the discharge time sequence SP.

Images