WO2022012364A1

WO2022012364A1 - Electromyographic signal processing method and apparatus, and exoskeleton robot control method and apparatus

Info

Publication number: WO2022012364A1
Application number: PCT/CN2021/104565
Authority: WO
Inventors: 李红红; 姚秀军
Original assignee: 京东科技信息技术有限公司
Priority date: 2020-07-15
Filing date: 2021-07-05
Publication date: 2022-01-20
Also published as: CN111844032B; CN111844032A

Abstract

An electromyographic signal processing method and apparatus, and an exoskeleton robot control method and apparatus. The electromyographic signal processing method comprises: acquiring an electromyographic signal; performing feature extraction on an active segment of the electromyographic signal, so as to obtain a first feature of the electromyographic signal; performing dimensionality reduction on the first feature, so as to obtain a second feature corresponding to the first feature; and inputting the second feature into a pre-trained classifier, so as to obtain a type corresponding to the electromyographic signal.

Description

EMG signal processing, exoskeleton robot control method and device

Citations to Related Applications

This disclosure claims the entire rights and interests of the invention patent application with the application number 202010680429.5 and the title "Myoelectric Signal Processing, Exoskeleton Robot Control Method and Device" submitted to the State Intellectual Property Office of the People's Republic of China on July 15, 2020, and approved by It is hereby incorporated by reference in its entirety.

field

The present disclosure generally relates to the field of robotics, and more particularly, to a method for processing myoelectric signals, a method and apparatus for controlling an exoskeleton robot.

background

In recent years, with the development of industry and transportation, the number of amputations caused by human beings due to industrial production, engineering construction, traffic accidents and other reasons is increasing year by year. For disabled persons with missing hands, multi-degree-of-freedom myoelectric prosthetic hands with bionic control can enable them to live a better life and integrate into society to a certain extent, so the need for prosthetics becomes more urgent.

By inputting control signals such as switches, force sensors, and EMG signals, the prosthetic controller can help users drive the prosthesis to achieve different joint actions. An essential component of many modern prostheses is the myoelectric control system, which uses electromyographic signals from human muscles to control the motion of the prosthesis.

In order to realize the control of the prosthetic hand, EMG signal processing is the most commonly used control information extraction method. Surface Electromyography (sEMG) refers to a signal of a certain length collected by electrodes on the surface of the skin during muscle action or static state. It has the characteristics of small amplitude and easy interference.

At present, the processing methods of few-channel EMG signals include threshold switch control, single-degree-of-freedom proportional control, and coding control. Among them, threshold-switch control and single-degree-of-freedom proportional control are usually controlled based on the threshold value of the EMG signal amplitude. Controlling one degree of freedom has fewer controllable degrees of freedom, which cannot adapt to the motor function of the current dexterous prosthesis. Although the coding control can theoretically control any number of degrees of freedom, with the complexity of the motor function, the coding method also becomes more and more It is complicated, which makes the prosthetic hand control unintuitive, and the coding process has a large delay. Due to the weak, aliasing and low signal-to-noise ratio of the EMG signal, it becomes difficult to identify multi-modal actions from the few-channel EMG signal. .

Overview

In a first aspect, the present disclosure relates to an electromyographic signal processing method, comprising:

Obtain EMG signals;

Perform feature extraction from the active segment of the electromyographic signal to obtain a first feature corresponding to the active segment;

Perform dimensionality reduction processing on the first feature to obtain a second feature corresponding to the first feature; and

The second feature is input into a pre-trained classifier, so that the classifier outputs the type corresponding to the electromyographic signal.

In some embodiments, the first feature is subjected to a dimensionality reduction process to obtain a second feature corresponding to the first feature, including:

A principal component analysis algorithm is used to perform dimension reduction processing on the first feature, and the feature obtained after the dimension reduction process is used as the second feature.

In some embodiments, the second feature is input into a pre-trained classifier, so that the classifier outputs the type corresponding to the electromyographic signal, including:

The second feature is input into a pre-trained regularized discriminant analysis classifier, and the type output by the regularized discriminant analysis classifier is used as the type corresponding to the electromyographic signal.

In some embodiments, feature extraction is performed from the active segment of the electromyographic signal to obtain the first feature corresponding to the active segment, including:

Using a sliding window algorithm to divide the active segment into multiple sub-windows according to a preset window length;

Extract the features of the EMG signals in each sub-window respectively;

According to the feature of the EMG signal in each sub-window, the corresponding feature vector of each sub-window is generated respectively; And

A feature matrix composed of feature vectors corresponding to the respective sub-windows is used as the first feature of the active segment.

In certain embodiments, before performing feature extraction from the active segment of the EMG signal, the method further comprises:

extracting the envelope signal of the electromyographic signal;

In the case where the previous one or more envelope signals of the current envelope signal are not greater than the preset threshold, if the current envelope signal is greater than the preset threshold, the EMG signal corresponding to the current envelope signal is determined to be the starting point signal of the active segment ;

In the case where the previous one or more envelope signals of the current envelope signal are greater than the preset threshold, if the current envelope signal is not greater than the preset threshold, the EMG signal corresponding to the current envelope signal is determined as the termination point of the active segment signal; and

The active segment of the electromyographic signal is determined according to the starting point signal and the ending point signal.

In certain embodiments, extracting the envelope signal of the electromyographic signal comprises:

Initialize the kernel function;

According to the sequence of EMG acquisition time from first to last, the EMG signals are imported into the kernel function one by one;

Update the kernel function after each EMG signal is imported, and use the trapezoidal method to calculate the unit isometric integral of the updated kernel function; and

The obtained unit equidistant integral is used as the envelope signal corresponding to the EMG signal introduced into the kernel function.

In certain embodiments, after acquiring the electromyographic signal, the method further comprises:

Determine whether the value of the electromyographic signal is greater than the baseline threshold;

If not greater than the baseline threshold, adjusting the value of the electromyographic signal; and

If it is greater than the baseline threshold, the value of the EMG signal does not change.

In a second aspect, the present disclosure relates to an exoskeleton robot control method, which includes:

Collect the EMG signal through the EMG signal acquisition module;

determining the type corresponding to the electromyographic signal;

generating control instructions for controlling the exoskeleton robot based on the type of the myoelectric signal; and

The exoskeleton of the exoskeleton robot is controlled to operate based on the control instructions.

In certain embodiments, determining the type corresponding to the electromyographic signal includes:

Obtain EMG signals;

In a third aspect, the present disclosure relates to an electromyographic signal processing device, comprising:

an acquisition module, configured to acquire myoelectric signals;

a feature extraction module, configured to perform feature extraction from the active segment of the electromyographic signal to obtain a first feature corresponding to the active segment;

a dimensionality reduction module, configured to perform dimensionality reduction processing on the first feature to obtain a second feature corresponding to the first feature; and

A classification module, configured to input the second feature into a pre-trained classifier, so that the classifier outputs the type corresponding to the electromyographic signal.

In certain embodiments, the dimensionality reduction module is configured to:

In a fourth aspect, the present disclosure relates to an exoskeleton robot control device comprising:

an acquisition module, configured to acquire the electromyographic signal through the electromyographic signal acquisition module;

a category determination module, configured to determine the type corresponding to the electromyographic signal;

an instruction determination module configured to generate a control instruction for controlling the exoskeleton robot based on the type of the electromyographic signal; and

The control module is configured to control the exoskeleton operation of the exoskeleton robot based on the control instruction.

In certain embodiments, the category determination module is configured to:

Obtain EMG signals;

Perform feature extraction from the active segment of the electromyographic signal to obtain the first feature corresponding to the active segment;

In a fifth aspect, the present disclosure relates to a computer device, comprising: a processor and a memory, the processor is configured to execute a data processing program stored in the memory, so as to implement the electromyographic signal processing method or the first aspect described in the first aspect. The exoskeleton robot control method described in the second aspect.

In a sixth aspect, the present disclosure relates to a storage medium, the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the electromyographic signal described in the first aspect The processing method or the exoskeleton robot control method described in the second aspect.

An EMG signal processing method in some embodiments of the present disclosure includes: acquiring an EMG signal, performing feature extraction on an active segment of the EMG signal to obtain a first feature of the EMG signal, and reducing the dimension of the first feature to obtain a first feature of the EMG signal. The second feature corresponding to the feature, the second feature is input into the pre-trained classifier to obtain the type corresponding to the EMG signal. In some embodiments, by means of feature extraction and dimensionality reduction, the dimension of the feature is reduced, the amount of calculation is reduced, and the calculation cost is reduced, and the EMG signals belonging to different types are classified by the classifier to obtain a higher classification The accuracy rate is high, and the multi-degree-of-freedom control of the exoskeleton robot can be realized according to the classification result of the EMG signal.

Brief Description of Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the disclosure and together with the description serve to explain the principles of the disclosure.

In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the accompanying drawings that are required to be used in the description of the embodiments or the prior art will be briefly introduced below. In other words, on the premise of no creative labor, other drawings can also be obtained from these drawings.

FIG. 1 is a flowchart of an electromyographic signal processing method provided by an embodiment of the present disclosure;

FIG. 2 is a flowchart of a control method for an exoskeleton robot provided by an embodiment of the present disclosure;

FIG. 3 is a block diagram of an EMG signal processing apparatus provided by an embodiment of the present disclosure;

FIG. 4 is a block diagram of an exoskeleton robot control device provided by an embodiment of the present disclosure; and

FIG. 5 is a schematic diagram of an electronic device according to an embodiment of the present disclosure.

detail

In order to make the purposes, technical solutions and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present disclosure. Obviously, the described embodiments These are some, but not all, embodiments of the present disclosure. Based on the embodiments in the present disclosure, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present disclosure.

An exoskeleton robot is a robot controlled by EMG signals. The user can assist the user to move by wearing the exoskeleton robot. The user usually generates EMG signals before exercising. The exoskeleton robot recognizes the user's action intention based on the action advance of the EMG signal, so as to control the movement of the exoskeleton robot.

The exoskeleton robot is usually equipped with an EMG signal acquisition module (such as an EMG sensor), a control module (such as a CPU controller), and an exoskeleton, which is a mechanical structure (such as a robotic arm, a robotic arm, a robotic leg, etc.), and the EMG signal acquisition The module collects EMG signals through electrodes installed on the user's skin, and the control module determines whether an action potential is generated according to the EMG signals collected by the EMG signal acquisition module (ie, detects whether there is an active segment in the EMG signal). The user's action intention is determined according to the action potential, and then the movement of the mechanical structure is controlled according to the action intention.

At present, in order to reduce the hardware cost in the existing exoskeleton robots, a few-channel (for example, 2-channel) EMG system is usually used as the EMG signal acquisition module to collect EMG signals. When the EMG signal controls the mechanical structure, the threshold switch control method, the single-degree-of-freedom proportional control method or the programming control method are usually used. Generally, only one degree of freedom can be controlled at a time, and there are few controllable degrees of freedom, which cannot adapt to the motor function of the current dexterous prosthesis. Although the coding control can theoretically control any number of degrees of freedom, with the complexity of the motor function, the coding The method has also become more complex, making the prosthetic hand control unintuitive, and the coding process has a large delay.

Due to the weakness, aliasing and low signal-to-noise ratio of EMG signals, it is difficult to identify multi-modal actions from few-channel EMG signals, so the commercialization of multi-DOF exoskeleton robots with real-time control is not ideal.

Taking the exoskeleton robot as an EMG prosthetic hand as an example, the traditional EMG prosthetic hand control method mainly controls the degree of freedom of opening and closing the palm of the prosthetic hand after inputting the EMG signal, shaping, filtering and signal processing. This type of myoelectric prosthetic hand has only one degree of freedom for opening/closing because it directly collects muscle electrical signals from the hand and controls the motor of the prosthetic hand to drive the mechanical structure of the prosthetic hand.

On the basis of the existing commercial exoskeleton robot, some embodiments of the present disclosure use a small number of channels (2 channels) to collect EMG signals, and through feature extraction and PCA (principal components analysis, principal component analysis) projection methods, Reduce the feature dimension, reduce the amount of calculation, and keep the original information to the maximum extent. Through the regularized discriminant analysis classifier, the EMG signals belonging to different action categories can be classified correctly to the maximum extent, so that the EMG signals belonging to different action categories can be classified correctly. Realize multi-DOF control of exoskeleton robots and improve accuracy.

FIG. 1 is a flowchart of an EMG signal processing method provided by an embodiment of the present disclosure. The method may be applied to a control module in an exoskeleton robot. As shown in FIG. 1 , the method may include S11 to S14.

S11. Acquire an electromyographic signal.

Usually, if the exoskeleton robot needs to be controlled by the classification result of the EMG signal, the EMG signal collected in real time by the EMG acquisition module (such as the EMG sensor) can be obtained. If you just want to classify the EMG signal, you can also obtain the EMG signal. Pre-stored EMG signals.

When acquiring the EMG signal collected by the EMG signal acquisition module in real time, the EMG signal collected by the EMG acquisition module can be acquired in a wired or wireless manner.

In certain embodiments, the acquired EMG signals may be two-way EMG signals of the extensor and flexor muscles of the user's forearm.

S12. Perform feature extraction from the active segment of the electromyographic signal to obtain a first feature corresponding to the active segment.

In certain embodiments, the first characteristic is data that can characterize the intrinsic properties of the myoelectric signal.

Because the purpose of extracting the first feature of the EMG signal in the present disclosure is to determine the type corresponding to the EMG signal according to the first feature, in some implementations, the extracted first feature may include a more effective classification effect on the EMG signal. Large features, such as the wavelength feature of the EMG signal, the number of zero-crossing points, the change in the sign of the slope, the AR parameter model and/or the skewness, etc.

S13. Perform dimensionality reduction processing on the first feature to obtain a second feature corresponding to the first feature.

There are usually not one but multiple types of features that have a greater impact on the classification of EMG signals, that is to say, the first feature usually contains multiple features, which makes it usually require a large amount of time to directly analyze the first feature. The amount of computation increases the computational cost. Therefore, in order to reduce the computational cost and reduce the computational cost, this solution performs dimension reduction processing on the extracted first feature to obtain the second feature corresponding to the first feature.

S14. Input the second feature into a pre-trained classifier, so that the classifier outputs the type corresponding to the electromyographic signal.

The classifier is used to identify the type corresponding to the EMG signal according to the characteristics of the EMG signal, where the type usually refers to the type of action, such as making a fist, bending the wrist, extending the wrist, extending the palm, bending the thumb, bending the index finger, bending the middle finger, bending the ring finger, The little finger is bent, the thumb is bent, the thumb and index finger are combined, etc., the corresponding mechanical structure can be controlled according to the type of EMG signal.

An EMG signal processing method provided by an embodiment of the present disclosure includes: acquiring an EMG signal, performing feature extraction on an active segment of the EMG signal, obtaining a first feature of the EMG signal, and reducing the dimension of the first feature to obtain the first feature Corresponding to the second feature, the second feature is input into the pre-trained classifier to obtain the corresponding type of the EMG signal. In this scheme, feature extraction and dimensionality reduction are used to reduce the dimension of features, reduce the amount of calculation, and reduce the computational cost. The EMG signals belonging to different types are classified by the classifier to obtain a higher classification accuracy. , and the multi-degree-of-freedom control of the exoskeleton robot can be realized according to the classification results of the EMG signals.

In some embodiments, when acquiring the EMG signal in S11, the EMG signal collected by the EMG signal acquisition module can be obtained, for example, the EMG signal of the extensor muscle and the flexor muscle of the forearm can be collected by using the EMG signal acquisition module, The collected EMG signals are transmitted to the control module through wired or wireless transmission.

In some embodiments, the collected EMG signals may be needle electrode EMG signals or surface electrode EMG signals.

In some embodiments, before S12 can also include:

Determining the active segment of the electromyographic signal, which can include the following steps 1 to 4 for determining the active segment of the electromyographic signal.

Step 1: Extract the envelope signal of the EMG signal.

In some embodiments, the following steps 1.1 to 1.4 can be used to extract the envelope signal of the EMG signal.

Step 1.1: Initialize the kernel function.

The kernel function can be a data set including the acquired pre-order signal of the EMG signal, and the initial state of the kernel function is 0. For example, the kernel function is ke(jk)={j1, j2, j3, ..., jn}, then the initial state j1,...,jn=0 in the kernel function in the state, where j1,...,jn represent the data points in the kernel function, and the data in the kernel function is gradually enriched in the process of signal processing.

Step 1.2: Import the EMG signals into the kernel function one by one in the order of the EMG signal acquisition time.

For example, after sorting the EMG signals according to the acquisition time of the acquired EMG signals, the order of the EMG signals is: s1, s2, . . . si, si+1, . in the function.

Step 1.3: Update the kernel function after each EMG signal is imported, and use the trapezoidal method to calculate the unit isometric integral of the updated kernel function.

For example, if the EMG signal s1 is imported into the kernel function, the kernel function is updated to kernel={j2,...,jn,s1}, where j2,...,jn=0, and s2 is imported into the kernel function, and the kernel function is updated to kernel ={j3,...,jn,s1,s2}, where j3,j4,...,jn=0, and so on.

Each time an EMG signal is imported, the trapezoidal method is used to calculate the unit isometric integral of the updated kernel function. For example, after importing s1, the trapezoidal method is used to calculate the unit of kernel={j2,...,jn,s1} according to the following formula, etc. Distance integral envelopeSignal:

envelopeSignal=sum{j2,...,jn,s1}÷2;

After importing si+1, the trapezoidal method is used to calculate the unit equidistant integral envelopeSignal of kernel={j3,...,jn,s1,s2}:

envelopeSignal=sum{j3,...,jn,s1,s2}÷2.

Step 1.4: Use the obtained unit equidistant integral as the envelope signal corresponding to the EMG signal imported into the kernel function.

The unit equidistant integral corresponding to the EMG signal s1 is used as the envelope signal y1 corresponding to the EMG signal s1, and the unit equidistant integral corresponding to the EMG signal s2 is used as the envelope signal y2 corresponding to the EMG signal si+1. By analogy, the envelope signals {y1, y2, ..., yi, yi+1, ...} are calculated from the EMG signals {s1, s2, ..., si, si+1, ...}.

Step 2: In the case that the previous one or more envelope signals of the current envelope signal are not greater than the preset threshold value, if the current envelope signal is greater than the preset threshold value, determine that the EMG signal corresponding to the current envelope signal is the active segment starting signal.

Step 3: In the case that the previous one or more envelope signals of the current envelope signal are greater than the preset threshold, if the current envelope signal is not greater than the preset threshold, determine that the EMG signal corresponding to the current envelope signal is the active segment end point signal.

The preset threshold is a preset signal value for judging whether the EMG signal is in the active segment. When the EMG signal is greater than the preset threshold, it can be determined that the EMG signal is in the active segment. When the EMG signal is not greater than the preset value When the threshold value is reached, it is determined that the EMG signal is in the resting segment, and the resting segment is the stable part between the action segments. In some embodiments, the specific value of the preset threshold can be set according to experience, and the preset threshold can be adjusted according to requirements, so as to improve the accuracy of determining the active segment.

Because the signal value of the EMG signal in the action segment will be higher than the signal value of the EMG signal in the resting segment, it is determined by judging the change of the signal value between the current envelope signal and its previous envelope signal whether the current envelope signal is It is the start signal or end signal of the action segment.

Step 4: Determine the active segment of the EMG signal according to the starting point signal and the ending point signal.

When the EMG signal just enters the active segment or just leaves the active segment, the amplitude of the signal data is small, and it is difficult to distinguish the resting segment from the active segment according to the EMG signal value, which is easy to misjudge and has a low accuracy rate. In the example, the EMG signal is first processed based on the kernel function, and the envelope signal corresponding to the EMG signal is obtained. The accuracy of active segment detection is improved.

The above method of determining the active segment is just an example, and other methods may be used to determine the active segment, such as a short-time Fourier method, a self-organizing artificial neural network method, a moving average method, and the like.

In some embodiments, after acquiring the electromyographic signal in S1, in order to ensure the accuracy of subsequent detection of the active segment, before determining the active segment of the electromyographic signal, the electromyographic signal processing method may further include:

To correct the acquired EMG signal, for example, the EMG signal can be corrected in the following ways:

Determine whether the value of the EMG signal is greater than the baseline threshold; if not greater than the baseline threshold, adjust the value of the EMG signal; if greater than the baseline threshold, the value of the EMG signal remains unchanged.

The baseline threshold may be determined according to resting-state EMG data. In some embodiments, the detected user may be asked to be in a resting state and data may be collected. The data is confirmed resting-state EMG data. In an embodiment, the baseline threshold thr can be calculated according to the following formula:

thr=mean{MAV ₁ , MAV ₂ , MAV ₃ ,...MAV _K }+A

Among them, mean is the operator of the average operation, MAV _i is the maximum value of the signal in the sliding window after the collected resting-state EMG data is segmented by the sliding window algorithm, and i is a positive integer between 1 and k , k is the number of sliding windows obtained by dividing the resting state EMG data, A is a preset constant, the constant A can be set based on experience, and the value of A can be adjusted during the calculation process to improve the determination of the active segment accuracy.

After acquiring the EMG signal, the EMG signal is corrected based on the following formula:

Among them, xi is the signal value of the EMG signal si. If the signal value of si is less than the baseline threshold, the signal value of si is adjusted to 0. If the signal value of si is not less than the baseline threshold, the signal value of si is not adjusted.

In this embodiment, the influence of individual differences on the signal is weakened by calibrating the EMG signal, and then the envelope signal is extracted from the EMG signal after correction, which is equivalent to performing secondary conversion on the obtained EMG signal, increasing the The difference between the resting segment potential and the active segment potential is increased, and the accuracy of subsequent active segment detection is improved.

In some embodiments, performing feature extraction on the active segment in S12 to obtain the first feature may include the following steps I to IV.

Step 1: using a sliding window algorithm to divide the active segment into multiple sub-windows according to a preset window length.

The window length can be set according to requirements or experience, and is not specifically limited.

Because a too long sequence is processed accordingly, such as extracting features, it will bring a lot of inconvenience, and the information description of the sequence is inaccurate, so the sliding window is used to segment the active segment for more convenient and accurate extraction. Characteristics of the active segment.

Step II: Extract the features of the EMG signals in each sub-window respectively.

Which features to extract can be set according to requirements, for example, features such as wavelength features, number of zero-crossing points, number of slope symbol changes, AR model, and skewness can be extracted.

The wavelength can be calculated using the following formula:

Among them, WL represents the wavelength, k represents the number of EMG signals in the sub-window, and the length of the wavelength is the simple accumulation of the lengths of the k EMG signals, which reflects the complexity of the EMG signal waveform and also reflects the EMG signal amplitude. , frequency, and duration.

Zeros crossing (ZC) is a simple frequency statistic that counts the number of times a signal waveform crosses the time axis (ie, zero) over a period of time. Given two adjacent samples x _i , x _i+1 satisfies the following conditions, and the value of the zero-crossing point is increased by one:

x _i x _i+1 ≤0,|x _i -x _i+1 |≥ε.

Slope sign changes (SSC), this statistic is another feature that describes signal frequency information. Given three consecutive sample values of the signal, x _i-1 , x _i , x _i+1 satisfying the following conditions, the value of the number of slope sign changes is incremented by one:

(x _i+1 -x _i )*(x _i -x _i-1 )≤0,|x _i -x _i+1 |≥ε,|x _i -x _i-1 |≥ε.

AR model is a commonly used time series model.

where s(n) represents the EMG signal, a _i (i=1, 2, 3...p) represents the AR model coefficient, p represents the model order, and w(n) represents random white noise.

Skewness is a eigenvector that measures the direction and degree of skewness of data. Assuming that it is a sample x of the sensor data of the motion sensor, the skewness of the data x can be estimated as:

In the formula, SK is the skewness, x _i is the sample observation value, n is the number of samples, x _mean is the mean value of n observations of the sample, and sd is the sample standard deviation.

Step III: According to the characteristics of the electromyographic signals in each sub-window, a feature vector corresponding to each sub-window is generated respectively.

During feature extraction, a sub-window may extract multiple features, and these features can form a feature vector corresponding to the sub-window.

Step IV: Use a feature matrix composed of feature vectors corresponding to the respective sub-windows as the first feature of the active segment.

One sub-window corresponds to one feature vector, and multiple sub-windows can form a feature matrix, and the feature matrix is used as the first feature of the active segment, which is convenient for analysis and calculation.

In some embodiments, S13 may use the principal component analysis algorithm PCA to perform dimension reduction processing on the first feature, and use the feature obtained after the dimension reduction process as the second feature.

Feature dimensionality reduction is the projection transformation of features. Feature projection transformation is an important tool in pattern recognition. It is often used to extract important information (such as discrimination information, variance information, etc.) in redundant features to improve classification. generalization performance.

PCA is a vector dimensionality reduction method based on linear transformation. Its core idea is to project the data onto several coordinate axes that maximize the variance of the data through coordinate rotation (that is, to find a new orthonormal basis), and get the data in The representation of the new space can eliminate the multicollinearity of the original data space, so as to achieve the purpose of data dimensionality reduction.

Usually to obtain a low-dimensional subspace, the simplest is to linearly change the original high-dimensional space, given a sample set X{x _i ,x ₂ ,…x _m }∈R ^d*m in the d-dimensional space, where x _i , x ₂ ,...x _m are all d-dimensional vectors, the samples in the n-dimensional space Z obtained after transformation, where n<<d.

Z = U ^T * X,

In the formula, U∈R ^d*m is the transformation matrix (that is, the new orthonormal basis found), U is the orthonormal matrix composed of the eigenvectors corresponding to the first n items with the largest covariance feature of X; Z is The projection of X{x _i ,x ₂ ,…x _m }∈R ^d*m in the new space is the low-dimensional data after the high-dimensional X dimension reduction.

If the first feature is a feature matrix X with d rows and m columns, where m represents the number of sub-windows, and d represents the number of features extracted by each sub-window (for example, if the features that can be extracted are wavelength features, zero-crossing points, The number of slope sign changes, AR model and skewness, the value of n is 5). The process of using PCA to reduce the dimension of the first feature includes:

Zero-mean each row of X (representing an attribute field), that is, subtract the mean of this row, obtain the covariance matrix, obtain the eigenvalues of the covariance matrix and the corresponding eigenvectors, and divide the eigenvectors according to the corresponding characteristics. The values are arranged in a matrix from top to bottom, and the first n rows are taken to form a matrix U. U is an orthogonal matrix composed of the eigenvectors corresponding to the first n items with the largest covariance feature of X, Z=U ^T * X is the data after dimensionality reduction to n dimensions.

The goal of feature dimensionality reduction is to extract the information with the best discrimination from the input feature space. PCA is an unsupervised dimensionality reduction method (without classification labels) capable of identifying features whose linear projections are consistent with major changes in the data. Compared with LDA, PCA does not require label information, and can maintain the original information to the greatest extent while reducing the dimension. PCA only needs to keep the eigenvector matrix and the mean vector of the samples, and then the new samples can be projected into the low-dimensional space through simple vector subtraction and matrix-vector multiplication, which is simple and convenient.

Although the low-dimensional space is different from the original high-dimensional space, it is often necessary to discard some information in the high-dimensional information in practice. The data is easier to use and the computational overhead of many algorithms is reduced, which is also an important motivation for dimensionality reduction; on the other hand, when the data is affected by noise, the eigenvectors corresponding to the smallest eigenvalues are often related to noise, and they are discarded. To a certain extent, it has the effect of denoising.

In some embodiments, the regularized discriminant analysis classifier RDA can be used for classification in S14, and RDA provides a continuous relationship between linear discriminant analysis LDA and quadratic discriminant analysis QDA by fitting a class-specific covariance matrix. By providing a continuum between LDA and QDA, RDA is guaranteed to achieve at least the same performance as the other two methods, and employing RDA also avoids the problem of overfitting.

RDA is trained before using RDA for classification to ensure the accuracy of the classification results.

The regularized discriminant analysis (RDA) model has two parameters (gamma and lambda), and the values of the two parameters are between 0 and 1. To train RDA, you need to set the regularization hyperparameters gamma and lambda first. For example, you can use In a linear search with a step size of 0.025 in the range of [0, 1], the parameter value with the smallest crossover loss in the validation set is selected as the set hyperparameter value.

After the hyperparameters are optimized, the training samples are obtained. The training samples are the eigenvectors of the corresponding EMG signals with known types. The training samples are input into the RDA classifier after dimensionality reduction through PCA, and the RDA is adjusted according to the classification results. The trained RDA is verified through the validation set, and 10-fold cross-validation can be used to evaluate the performance of the trained RDA classifier, and the RDA classifier with the best performance is selected as the classifier used in S14.

In some embodiments, training samples can be obtained in the following manner:

Wear an EMG signal acquisition device (such as armband, EMG acquisition sensor, EMG acquisition electrode, etc.) on the relevant position of the tested user (such as flexor and extensor muscles, etc.) Wrist, palm extension, thumb bending, index finger bending, middle finger bending, ring finger bending, little finger bending, thumb bending, thumb and index finger combination bending and other gestures, the EMG signals of repeated movements. In order to reduce the error brought by the acquisition process, each gesture was measured 6 times, with a 5-second rest between each test, and a 30-second rest between each type of gesture to avoid fatigue. The features of the collected EMG activity segments are extracted as training samples.

In this embodiment, by training the classifier, the accuracy of the classification result is guaranteed.

The present disclosure also provides an embodiment of a control method for an exoskeleton robot, which is applied to an exoskeleton robot. The exoskeleton robot may include an EMG signal acquisition module, a control module and a mechanical structure. As shown in FIG. 2 , the method may include the following: Steps S21 to S24.

S21. Collect the electromyography signal through the electromyography signal acquisition module.

S22. Determine the type corresponding to the electromyographic signal.

In certain embodiments, determining the type corresponding to the electromyographic signal may include:

Obtain an electromyographic signal, perform feature extraction from the active segment of the electromyographic signal, obtain a first feature corresponding to the active segment, perform dimensionality reduction processing on the first feature, and obtain a first feature corresponding to the first feature. Second feature, the second feature is input into the pre-trained classifier, so that the classifier outputs the type corresponding to the electromyographic signal.

In some embodiments, a principal component analysis algorithm can be used to perform dimensionality reduction processing on the first feature.

S23. Based on the type of the electromyographic signal, generate a control instruction for controlling the exoskeleton robot.

In some embodiments, a correspondence table between the electromyographic signal types and control instructions may be preset, and the correspondence table may be pre-stored in the local memory of the exoskeleton robot or in a data node that can be accessed by the control module of the exoskeleton robot , so that after determining the type corresponding to the electromyographic signal, the corresponding control instruction can be determined by looking up the table.

S24. Control the exoskeleton operation of the exoskeleton robot based on the control instruction.

In some embodiments, after the electromyographic signal is collected, the type of the electromyographic signal is determined, and the corresponding control instruction is determined according to the type of the electromyographic signal, and the exoskeleton of the exoskeleton robot, that is, the mechanical structure is controlled according to the control instruction, So that the mechanical structure performs the corresponding action. When determining the type of EMG signal, the EMG signal with fewer channels (for example, 2 channels) can be used to ensure maximum separation between finger movements through feature set extraction and PCA projection. Types of EMG signals are classified, and then the multi-degree-of-freedom control of the mechanical structure is realized according to the categories.

In some embodiments, the second feature is input into a pre-trained classifier, so that the classifier outputs the type corresponding to the electromyographic signal, which may include:

In some embodiments, feature extraction is performed from the active segment of the electromyographic signal to obtain the first feature corresponding to the active segment, which may include:

Extract the features of the EMG signals in each sub-window respectively;

According to the characteristics of the electromyographic signals in each sub-window, the corresponding feature vectors of each sub-window are respectively generated; and

In some embodiments, before the feature extraction is performed from the active segment of the electromyographic signal, the method may further include:

extracting the envelope signal of the electromyographic signal;

In the case that the previous one or more envelope signals of the current envelope signal are greater than the preset threshold, if the current envelope signal is not greater than the preset threshold, the EMG signal corresponding to the current envelope signal is determined as the termination point of the active segment signal; and

In certain embodiments, extracting the envelope signal of the electromyographic signal may include:

Initialize the kernel function;

In certain embodiments, after acquiring the electromyographic signal, it can also include:

In some embodiments, after the active segment of the electromyographic signal is determined, and before the feature extraction is performed on the active segment, the electromyographic signal processing method may further include: preprocessing the electromyographic signal in the active segment, in a certain In some embodiments, pretreatment can be performed in the following manner:

Use a notch filter to remove 50Hz power interference, and then perform 20-450Hz bandpass filtering.

By preprocessing the active segment, the accuracy of subsequent feature extraction is improved.

The present disclosure also provides an embodiment of an electromyographic signal processing device, as shown in FIG. 3 , the device may include:

an acquisition module 301, configured to acquire myoelectric signals;

The feature extraction module 302 is configured to perform feature extraction from the active segment of the electromyographic signal to obtain a first feature corresponding to the active segment;

A dimension reduction module 303, configured to perform dimension reduction processing on the first feature to obtain a second feature corresponding to the first feature; and

The classification module 304 is configured to input the second feature into a pre-trained classifier, so that the classifier outputs the type corresponding to the electromyographic signal.

In certain embodiments, the dimensionality reduction module 303 is configured to:

The principal component analysis algorithm is used to perform dimension reduction processing on the first feature, and the feature obtained after the dimension reduction process is used as the second feature.

In certain embodiments, classification module 304 is configured to:

In certain embodiments, feature extraction module 302 is configured to:

Extract the features of the EMG signals in each sub-window respectively;

In certain embodiments, the apparatus may further include an active segment detection module configured to:

extracting the envelope signal of the electromyographic signal;

In some embodiments, extracting the envelope signal of the electromyographic signal may include:

Initialize the kernel function;

In certain embodiments, the apparatus may further include a correction module configured to:

The present disclosure also provides an exoskeleton robot control device, as shown in FIG. 4 , the device may include:

The collection module 401 is configured to collect the EMG signal through the EMG signal collection module;

a category determination module 402, configured to determine the type corresponding to the electromyographic signal;

an instruction determination module 403, configured to generate a control instruction for controlling the exoskeleton robot based on the type of the electromyographic signal; and

The control module 404 is configured to control the exoskeleton of the exoskeleton robot to operate based on the control instruction.

In certain embodiments, category determination module 402 is configured to:

Obtain EMG signals;

In another embodiment of the present disclosure, an electronic device is also provided, as shown in FIG. 5 , including a processor 501 , a communication interface 502 , a memory 503 and a communication bus 504 , wherein the processor 501 , the communication interface 502 , and the memory 503 pass through The communication bus 504 completes mutual communication;

memory 503 configured to store computer programs; and

When the processor 501 is configured to execute the program stored in the memory 503, it realizes:

Acquire an electromyographic signal, perform feature extraction from the active segment of the electromyographic signal, obtain a first feature corresponding to the active segment, perform dimensionality reduction processing on the first feature, and obtain a first feature corresponding to the first feature. Second feature, the second feature is input into the pre-trained classifier, so that the classifier outputs the type corresponding to the electromyographic signal. or

The EMG signal is collected by the EMG signal acquisition module, the type corresponding to the EMG signal is determined, the control instruction for controlling the exoskeleton robot is generated based on the type of the EMG signal, and the exoskeleton robot is controlled based on the control instruction. The exoskeleton of the skeletal robot runs.

The communication bus 504 mentioned by the above electronic device may be a Peripheral Component Interconnect (PCI for short) bus or an Extended Industry Standard Architecture (EISA for short) bus or the like. The communication bus 504 can be divided into an address bus, a data bus, a control bus, and the like. For ease of presentation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

The communication interface 502 is used for communication between the above-mentioned electronic device and other devices.

The memory 503 may include random access memory (Random Access Memory, RAM for short), or may include non-volatile memory (non-volatile memory), such as at least one disk memory. In certain embodiments, the memory may also be at least one storage device located remotely from the aforementioned processor.

The above-mentioned processor 501 may be a general-purpose processor, including a central processing unit (Central Processing Unit, referred to as CPU), a network processor (Network Processor, referred to as NP), etc.; may also be a digital signal processor (Digital Signal Processor, referred to as DSP) ), Application Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components.

In another embodiment of the present disclosure, a computer-readable storage medium is also provided, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, any one of the aforementioned electromyography is implemented A signal processing method or any one of the control methods of the exoskeleton robot.

When implementing the embodiments of the present disclosure, reference may be made to the above-mentioned embodiments, which have corresponding technical effects.

It should be noted that, in this document, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply these Any such actual relationship or sequence exists between entities or operations. Moreover, the terms "comprising", "comprising" or any other variation thereof are intended to encompass non-exclusive inclusion such that a process, method, article or device comprising a list of elements includes not only those elements, but also includes not explicitly listed or other elements inherent to such a process, method, article or apparatus. Without further limitation, an element qualified by the phrase "comprising a..." does not preclude the presence of additional identical elements in a process, method, article or apparatus that includes the element.

The foregoing are merely some embodiments of the present disclosure to enable those skilled in the art to understand or implement the present disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure is not to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

An electromyographic signal processing method, comprising:

Obtain EMG signals;

Perform feature extraction from the active segment of the electromyographic signal to obtain a first feature corresponding to the active segment;

Perform dimensionality reduction processing on the first feature to obtain a second feature corresponding to the first feature; and

The second feature is input into a pre-trained classifier, so that the classifier outputs the type corresponding to the electromyographic signal.
The method of claim 1, wherein performing dimension reduction processing on the first feature to obtain a second feature corresponding to the first feature, comprising:

A principal component analysis algorithm is used to perform dimension reduction processing on the first feature, and the feature obtained after the dimension reduction process is used as the second feature.
The method according to claim 1 or 2, wherein inputting the second feature into a pre-trained classifier, so that the classifier outputs the type corresponding to the electromyographic signal, comprising:

The second feature is input into a pre-trained regularized discriminant analysis classifier, and the type output by the regularized discriminant analysis classifier is used as the type corresponding to the electromyographic signal.
The method according to any one of claims 1 to 3, wherein, performing feature extraction from the active segment of the electromyographic signal to obtain the first feature corresponding to the active segment, comprising:

Using a sliding window algorithm to divide the active segment into multiple sub-windows according to a preset window length;

Extract the features of the EMG signals in each sub-window respectively;

According to the characteristics of the electromyographic signals in each sub-window, the corresponding feature vectors of each sub-window are respectively generated; and

A feature matrix composed of feature vectors corresponding to the respective sub-windows is used as the first feature of the active segment.
The method according to any one of claims 1 to 4, wherein, before performing feature extraction from the active segment of the electromyographic signal, the method further comprises:

extracting the envelope signal of the electromyographic signal;

In the case where the previous one or more envelope signals of the current envelope signal are not greater than the preset threshold, if the current envelope signal is greater than the preset threshold, the EMG signal corresponding to the current envelope signal is determined to be the starting point signal of the active segment ;

In the case where the previous one or more envelope signals of the current envelope signal are greater than the preset threshold, if the current envelope signal is not greater than the preset threshold, the EMG signal corresponding to the current envelope signal is determined as the termination point of the active segment signal; and

The active segment of the electromyographic signal is determined according to the starting point signal and the ending point signal.
The method of claim 5, wherein extracting the envelope signal of the EMG signal comprises:

Initialize the kernel function;

According to the sequence of EMG acquisition time from first to last, the EMG signals are imported into the kernel function one by one;

Update the kernel function after each EMG signal is imported, and use the trapezoidal method to calculate the unit isometric integral of the updated kernel function; and

The obtained unit equidistant integral is used as the envelope signal corresponding to the EMG signal introduced into the kernel function.
The method according to any one of claims 1 to 6, wherein after acquiring the electromyographic signal, the method further comprises:

Judging whether the value of the electromyographic signal is greater than the line-based threshold;

If not greater than the baseline threshold, adjusting the value of the electromyographic signal; and

If it is greater than the baseline threshold, the value of the EMG signal does not change.
A control method for an exoskeleton robot, comprising:

Collect the EMG signal through the EMG signal acquisition module;

determining the type corresponding to the electromyographic signal;

generating control instructions for controlling the exoskeleton robot based on the type of the myoelectric signal; and

The exoskeleton of the exoskeleton robot is controlled to operate based on the control instructions.
The method of claim 8, wherein determining the type corresponding to the electromyographic signal comprises:

Obtain EMG signals;

Perform feature extraction from the active segment of the electromyographic signal to obtain a first feature corresponding to the active segment;

Perform dimensionality reduction processing on the first feature to obtain a second feature corresponding to the first feature; and

The second feature is input into a pre-trained classifier, so that the classifier outputs the type corresponding to the electromyographic signal.
An electromyographic signal processing device, comprising:

an acquisition module, configured to acquire myoelectric signals;

a feature extraction module, configured to perform feature extraction from the active segment of the electromyographic signal to obtain a first feature corresponding to the active segment;

a dimensionality reduction module, configured to perform dimensionality reduction processing on the first feature to obtain a second feature corresponding to the first feature; and

A classification module, configured to input the second feature into a pre-trained classifier, so that the classifier outputs the type corresponding to the electromyographic signal.
The apparatus of claim 10, wherein the dimensionality reduction module is configured to:

A principal component analysis algorithm is used to perform dimension reduction processing on the first feature, and the feature obtained after the dimension reduction process is used as the second feature.
Exoskeleton robot control device, which includes:

an acquisition module, configured to acquire the electromyographic signal through the electromyographic signal acquisition module;

a category determination module, configured to determine the type corresponding to the electromyographic signal;

an instruction determination module configured to generate a control instruction for controlling the exoskeleton robot based on the type of the electromyographic signal; and

The control module is configured to control the exoskeleton operation of the exoskeleton robot based on the control instruction.
The apparatus of claim 12, wherein the category determination module is configured to:

Obtain EMG signals;

Perform feature extraction from the active segment of the electromyographic signal to obtain a first feature corresponding to the active segment;

Perform dimensionality reduction processing on the first feature to obtain a second feature corresponding to the first feature; and

The second feature is input into a pre-trained classifier, so that the classifier outputs the type corresponding to the electromyographic signal.
An electronic device, comprising: a processor and a memory, the processor is configured to execute a data processing program stored in the memory, so as to implement the electromyographic signal processing method or the rights described in any one of claims 1 to 7 The control method of an exoskeleton robot according to any one of claims 8 to 9.
A storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement the electromyographic signal processing method according to any one of claims 1 to 7 or The control method of an exoskeleton robot according to any one of claims 8 to 9.