Motion intention prediction method, system and equipment based on multi-muscle-group surface electromyogram signals and medium

Abstract

The invention relates to a motion intention prediction method, system and device based on multi-muscle-group surface electromyogram signals and a medium. The method comprises the steps that S1, the multi-muscle-group surface electromyogram signals are collected; s2, performing action sequence detection and segmentation on the multi-muscle group surface electromyographic signals to obtain segmented electromyographic sequences; s3, multi-dimensional myoelectricity characteristic value extraction is carried out on the segmented myoelectricity sequence, and multi-muscle-group multi-modal characteristic parameters are obtained; s4, performing dimension reduction processing on the multi-muscle-group multi-modal characteristic parameters by adopting a binary particle swarm optimization algorithm, and determining a myoelectricity optimal characteristic matrix; and S5, inputting the myoelectricity optimal feature matrix into a pre-trained MobileNet V2 neural network model, and outputting a motion intention prediction result. Compared with the prior art, the method has the advantages of high prediction accuracy, good real-time performance and the like.

Description

Technical Field

[0001] The present invention relates to the field of motion recognition technology, and in particular to a method, system, device and medium for predicting motion intention based on surface electromyography signals of multiple muscle groups. Background Art

[0002] In today's society, exoskeleton robots have been promoted in the field of rehabilitation as rehabilitation assistive devices to help patients who have temporarily lost the ability to walk for some reason to undergo rehabilitation treatment. This treatment method has greatly reduced the excessive interference of human factors in the traditional rehabilitation treatment process and helped to restore their walking ability.

[0003] At present, with the research and development in recent years, lower limb exoskeleton robots are mainly aimed at two types of people. One is for healthy people, as a wearable power device, used in military, industrial, emergency and other fields. The robot users can use it to improve their own functions to achieve the purpose of reducing weight and physical energy consumption. The second is people who have impaired lower limb motor function and need rehabilitation treatment. In this case, the research is mainly reflected in the simulation of normal human movement, using fixed movement trajectories to assist patients in exercising leg muscles and help restore motor function.

[0004] In medical rehabilitation, robots can be further divided into exoskeleton rehabilitation robots and lower limb exoskeleton walking robots. Lower limb walking robots provide support for patients with lower limb function loss, allowing them to stand on two feet, thus getting rid of the limitations of wheelchairs and barrier-free facilities. This type of exoskeleton walking robot can also help patients exercise their lower limb muscles and bones to prevent muscle atrophy. Wearable lower limb walking robots can help patients with lower limb disabilities stand up again and regain confidence in life, which has positive significance for physical and mental health. Due to the low accuracy of existing motion intention recognition methods, the exoskeleton robots developed today have strong human-machine resistance and cannot achieve human-machine intelligent collaboration. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a method, system, device and medium for predicting movement intention based on surface electromyography signals of multiple muscle groups with high prediction accuracy and real-time performance.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] According to a first aspect of the present invention, a method for predicting movement intention based on surface electromyographic signals of multiple muscle groups is provided, comprising:

[0008] S1, collect surface electromyographic signals of multiple muscle groups;

[0009] S2, performing motion sequence detection and segmentation on surface electromyographic signals of multiple muscle groups to obtain segmented electromyographic sequences;

[0010] S3, extracting multi-dimensional electromyographic feature values ​​from the segmented electromyographic sequences to obtain multi-muscle group multi-modal feature parameters;

[0011] S4, using binary particle swarm optimization algorithm to reduce the dimension of multi-muscle group multi-modal feature parameters and determine the optimal electromyographic feature matrix;

[0012] S5. Input the optimal electromyography feature matrix into the pre-trained MobileNet V2 neural network model and output the motion intention prediction result.

[0013] Preferably, the multi-muscle group surface EMG signals include the rectus femoris surface EMG signal, the vastus medialis surface EMG signal, the vastus lateralis surface EMG signal, the semitendinosus surface EMG signal and the biceps femoris surface EMG signal.

[0014] Preferably, after collecting the surface electromyography signals of multiple muscle groups, the method also includes denoising the surface electromyography signals of multiple muscle groups, specifically: using a 50Hz power frequency notch filter to remove power frequency interference in the surface electromyography signals of multiple muscle groups, and using a 20Hz~500Hz Butterworth bandpass filter to filter out high-frequency noise in the surface electromyography signals of multiple muscle groups.

[0015] Preferably, the method of performing action sequence detection and segmentation on surface electromyographic signals of multiple muscle groups to obtain the segmented electromyographic sequence is specifically: using sliding window signal energy threshold detection and segmenting the starting position of the electromyographic action to obtain the segmented electromyographic sequence.

[0016] Preferably, the multi-dimensional electromyographic characteristic values ​​include electromyographic integral value, variance, root mean square value, wavelength, average power frequency, median frequency, wavelet packet energy, approximate entropy, sample entropy and fuzzy entropy.

[0017] Preferably, the binary particle swarm optimization algorithm is used to perform dimension reduction processing on the characteristic parameters to determine the optimal electromyographic characteristic matrix, which specifically includes:

[0018] S401, initializing the particle population, including taking the number of multi-muscle group multi-modal characteristic parameters as the particle space dimension d, setting the number of particles n, setting the maximum number of iterations kmax, randomly generating the particle initial position information vector X and the flight speed vector V, recording the initial state position and speed as the current local optimal solution, and importing the multi-muscle group multi-modal characteristic parameter data set;

[0019] S402, using the Sigmoid function to perform binary conversion on the particles, using the conversion result to form a new feature parameter subset, substituting the new feature parameter subset into the MobileNet V2 neural network for classification and recognition, and using the negative value of the F1 score obtained by the MobileNet V2 neural network as the fitness function value;

[0020] S403, based on the fitness function value and current position of each particle, update the individual best position, and record the individual best position of the i-th particle in the k-th iteration

[0021] S404, update the global best position according to the individual historical best solution and the current individual best position, and record the individual best position of the i-th particle in the k-th iteration

[0022] S405, update the speed and position of all particles, where the speed of the i-th particle in the k+1th iteration and location They are:

[0023]

[0024] Where: and They represent the current position and velocity of the i-th particle at the k-th iteration respectively; is the global optimal position of the particle swarm at the kth iteration, the subscript g is the index of the optimal particle position; c1 and c2 represent the acceleration factors of the particles; r1 and r2 are random numbers in the interval [0,1]; ω represents the inertia weight that is linearly related to the number of iterations;

[0025] S406, determine whether the number of iterations reaches the maximum number of iterations kmax, if yes, go to S407, otherwise go to S402;

[0026] S407, saving the global optimal result and obtaining the electromechanical optimal characteristic matrix.

[0027] Preferably, in S4, a Bayesian optimization algorithm is used to pre-train the MobileNet V2 neural network model, wherein all hyperparameters of the MobileNet V2 convolutional neural network are used as inputs of the objective function, and the average AUC of multiple cross-validations is used as the optimization objective function.

[0028] According to a second aspect of the present invention, a system for predicting movement intention based on the surface electromyography signals of multiple muscle groups is provided, the system comprising:

[0029] Myoelectric signal acquisition module, used to collect surface myoelectric signals of multiple muscle groups;

[0030] The electromyographic sequence segmentation module is used to detect and segment the action sequence of surface electromyographic signals of multiple muscle groups and obtain the segmented electromyographic sequence;

[0031] A multi-muscle group multi-modal feature parameter acquisition module is used to extract multi-dimensional electromyographic feature values ​​from the segmented electromyographic sequences to obtain multi-muscle group multi-modal feature parameters;

[0032] The module for obtaining the optimal electromyographic feature matrix is ​​used to reduce the dimension of multi-modal feature parameters of multiple muscle groups using a binary particle swarm optimization algorithm to determine the optimal electromyographic feature matrix. The MobileNet V2 neural network is used to classify and identify the binary converted particle data set, and the negative value of the F1 score obtained by the MobileNet V2 neural network is used as the fitness function value.

[0033] The recognition prediction module is used to input the optimal electromyography feature matrix into the pre-trained MobileNet V2 neural network model and output the motion intention prediction result.

[0034] According to a third aspect of the present invention, there is provided an electronic device, comprising a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the program, any one of the methods described above is implemented.

[0035] According to a fourth aspect of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, any one of the methods described above is implemented.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] (1) The present invention adopts a binary particle swarm optimization algorithm to reduce the dimensionality of multi-dimensional electromyographic features and determine the optimal electromyographic feature matrix, thereby improving the real-time performance of motion intention recognition while improving recognition accuracy.

[0038] (2) The present invention introduces the extraction of multi-dimensional electromyographic feature parameters of multiple muscle groups, and selects time domain, frequency domain, time-frequency domain and nonlinear indicators as electromyographic feature values. The indicators include electromyographic integral value, variance, root mean square value, wavelength, average power frequency, median frequency, wavelet packet energy, approximate entropy, sample entropy and fuzzy entropy, covering the multi-dimensional electromyographic feature values ​​of the main muscle groups related to lower limb walking, further improving the accuracy of movement intention recognition.

[0039] (3) The MobileNet V2 neural network optimized by the Bayesian optimization algorithm of the present invention is used to recognize the motion intention of electromyographic signals. Compared with the traditional machine learning algorithm, the MobileNet V2 neural network has a stronger learning ability and the accuracy obtained by training is higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0041] Figure 2 This is a flow chart of the present invention using binary particle swarm optimization algorithm to reduce the dimension of feature parameters;

[0042] Figure 3 Schematic diagram of the MobileNet V2 neural network model training process. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0044] Example

[0045] This embodiment provides a method for predicting movement intention based on surface electromyographic signals of multiple muscle groups. Figure 1 As shown, the method comprises the following steps:

[0046] S1. Collect surface electromyographic signals of multiple muscle groups, including surface electromyographic signals of rectus femoris, vastus medialis, vastus lateralis, semitendinosus and biceps femoris, and perform denoising on the surface electromyographic signals of multiple muscle groups.

[0047] In this embodiment, a 50 Hz power frequency notch filter is designed using a Butterworth filter to remove power frequency interference, and a 20 Hz to 500 Hz Butterworth bandpass filter is designed to filter out high frequency noise to obtain a denoised surface electromyography signal.

[0048] S2. Perform motion sequence detection and segmentation on surface electromyographic signals of multiple muscle groups to obtain segmented electromyographic sequences.

[0049] The length of a single action in the electromyographic signal of different subjects is different, and the length of the electromyographic signal sequence of the same subject at different times is also different. Therefore, this embodiment uses unequal length segmentation to process the experimental data. Specifically, according to the characteristics of dynamic electromyographic signals, a sliding window signal energy threshold is used to detect and segment the starting position of the electromyographic action.

[0050] Step S3: extract multi-dimensional electromyographic feature values ​​from the segmented electromyographic sequence to obtain multi-muscle group multi-modal feature parameters.

[0051] In this embodiment, the multi-dimensional electromyographic characteristic values ​​include electromyographic integral value, variance, root mean square value, wavelength, average power frequency, median frequency, wavelet packet energy, approximate entropy, sample entropy and fuzzy entropy. For the electromyographic time series X={x1,x1,…,x2, N}, the specific calculation process of each electromyographic characteristic value is as follows:

[0052] (1) Electromyography integral value iEMG:

[0053]

[0054] (2) Variance VAR:

[0055]

[0056] (3) Root mean square value RMS:

[0057]

[0058] (4) Wavelength WL:

[0059]

[0060] (5) Median frequency MF:

[0061]

[0062] Where P(f) represents the power spectral density of the surface electromyographic signal in the frequency domain, and f is the frequency of the signal.

[0063] (6) Mean power frequency MPF:

[0064]

[0065] (7) Approximate entropy ApEn, the calculation process includes:

[0066] 1) For a surface electromyography signal time series X(N) = {x(1), x(2), …, x(N)} consisting of N data, reconstruct an m-dimensional vector sequence X m (1),…,X m (N-m+1), where X m (i)={x(i),x(i+1),…,x(i+m-1)},1≤i≤N-m+1,X m (i) represents the m consecutive values ​​of x starting from the i-th point.

[0067] 2) For a given X m (i), statistics X m (i) With X mThe number of j (1≤j≤Nm,j≠i) whose distance between them is less than or equal to r i , for 1≤i≤N-m+1, define:

[0068]

[0069] Definition C m (r) is:

[0070]

[0071] 3) The approximate entropy calculation formula is:

[0072] ApEn=C m (r)-C m+1 (r) (9)

[0073] The value of ApEn is related to the selection of parameters r and m, usually m=2 or m=3, r=0.1~0.25STD, and STD (standard deviation) is the standard deviation of the sequence X(N) of the surface electromyography signal sEMG.

[0074] (8) Sample entropy SampEn, the calculation process includes:

[0075] 1) For a surface electromyography signal time series X(N) = {x(1), x(2), …, x(N)} consisting of N data, reconstruct an m-dimensional vector sequence X m (1),…,X m (N-m+1), where X m (i)={x(i),x(i+1),…,x(i+m-1)},1≤i≤N-m+1,X m (i) represents the m consecutive values ​​of x starting from the i-th point.

[0076] 2) Define vector X m (i) With X m (j) The distance d[X m (i),X m (j)] is the absolute value of the maximum difference between the corresponding elements of the two, that is:

[0077] d[X m (i),X m (j)]=max k＝0,L,m-1 (x(i+k)-x(j+k)) (10)

[0078] 3) For a given X m (i), statistics X m (i) With X m The number of j (1≤j≤Nm,j≠i) whose distance between (j) is less than or equal to ri , for 1≤i≤Nm, define:

[0079]

[0080] Definition B (m) (r) is:

[0081]

[0082] 4) Increase the dimension to m+1 and calculate X m+1 (i) With X m+1 (j)(1≤j≤Nm,j≠i) The number of distances A less than or equal to r i , Defined as:

[0083]

[0084] Definition A m (r) is:

[0085]

[0086] So, B m (r) is the probability that two sequences match m points under similarity tolerance r, and A m (r) is the probability that the two sequences match m+1 points.

[0087] 5) The sample entropy calculation formula is:

[0088]

[0089] When N is a finite value, it can be estimated using the following formula:

[0090]

[0091] The value of SampEn is related to the selection of parameters r and m, and usually m=2, r=0.1-0.25STD, where STD (standard deviation) is the standard deviation of the sEMG sequence {X(n)}.

[0092] (9) Fuzzy entropy FuzzyEn, the calculation process includes:

[0093] 1) A surface electromyography signal time series X(N)=x(1),x(2),…,x(N) consisting of N data.

[0094] 2) Define the phase space dimension m (m≤N-2) and the similarity tolerance r, and reconstruct the phase space as:

[0095] X(i)=[u(i),u(i+1),…,u(i+m-1)]-u0(i) (17)

[0096] Where i = 1, 2, ..., N-m + 1,

[0097] 3) Introduce fuzzy membership function:

[0098]

[0099] Where r is the similarity tolerance.

[0100] 4) For i=1,2,…,N-m+1, calculate:

[0101]

[0102] Among them, j=1,2,…,N-m+1 and i≠j.

[0103] 5) For each i, find the average value:

[0104]

[0105] 6) Define Φ m (r) is:

[0106]

[0107] 7) The fuzzy entropy calculation formula is:

[0108]

[0109] When N is a finite value, it can be estimated using the following formula:

[0110] FuzzyEn(m,r,N)=lnΦ m (r)-lnΦ m+1 (r) (23)

[0111] The value of FuzzyEn is related to the selection of parameters r and m. Usually, m=2, r=0.15STD, and STD (standard deviation) are taken as the standard deviation of the sEMG sequence {X(n)}.

[0112] (10) Wavelet packet energy E, the calculation process includes:

[0113] 1) Perform wavelet packet decomposition on the EMG sequence. Suppose the EMG signal s(t) is decomposed to the jth resolution layer by wavelet packet decomposition to obtain 2 j subspace, the coefficient of each subspace is: D n ={d n (k), k=1,2,…,k}, k is the length of the wavelet packet coefficient.

[0114] Obtain the coefficient matrix constructed by the wavelet coefficients of each subspace after decomposition, and assume that the wavelet coefficients of all subspaces in the entire j-th resolution layer constitute a matrix:

[0115]

[0116] 2) Normalize the coefficient matrix. Then, divide [-1,1] into M equal-length intervals: [-1,a1), [a1,a2), [a2,a3), L, [a m-1 ,a m ),L,[a M-1 ,1], where a1,a2,a3,a m-1 ,a m ,L,a M-1 It is the dividing point between intervals, satisfying the following relationship: -1<a1<a2<L<a m-1 <a m <L<a M-1 <1.

[0117] For the sEMG signal wavelet packet coefficient D of the nth subspace n Falling in the interval [a m-1 ,a m ] is N, since the wavelet packet coefficient D n The length of is K.

[0118] 3) Calculate the energy of each wavelet packet coefficient using the formula:

[0119]

[0120] The wavelet packet energy represents the energy of the frequency band by calculating the square sum of coefficients, j represents the j-th level wavelet packet decomposition, k represents the k-th wavelet packet coefficient, n represents the j-th level wavelet packet coefficient index, 1≤n≤N, N represents the number of coefficients of the j-th level wavelet packet decomposition.

[0121] 4) According to the characteristics of sEMG signals, the value of wavelet packet energy is obtained from the frequency band where the identifiable signal is located, which is used to describe the energy distribution of the signal in different frequency components.

[0122] S4, using binary particle swarm optimization algorithm to reduce the dimension of multi-muscle group multi-modal feature parameters and determine the optimal electromyographic feature matrix, such as Figure 2 As shown, specifically including:

[0123] S401, initialize the particle population, including taking the number of multi-muscle group multi-modal characteristic parameters as the particle space dimension d, setting the number of particles n, setting the maximum number of iterations kmax, randomly generating the particle initial position information vector X and the flight speed vector V, recording the initial state position and speed as the current local optimal solution, and importing the multi-muscle group multi-modal characteristic parameter data set.

[0124] Among them, the candidate particle X is expressed as follows:

[0125]

[0126] Then set the initial position of the candidate solution of the particle in the search space:

[0127]

[0128] S402. Use the Sigmoid function to perform binary conversion on the particles, form a new feature parameter subset from the conversion result, substitute the new feature parameter subset into the MobileNet V2 neural network for classification and recognition, and use the negative value of the indicator "F1 score" obtained by the MobileNet V2 neural network as the fitness function value.

[0129] Use the Sigmoid function to binarize the particles. The expression is as follows:

[0130]

[0131] S403. Update the individual best position according to the fitness function value and current position of each particle. The position of each particle represents a potential solution. The individual best position of the i-th particle in the k-th iteration is recorded as

[0132] S404, update the global best position according to the individual historical best solution and the current individual best position, and record the individual best position of the i-th particle in the k-th iteration

[0133] S405: Update the speed and position of all particles and record the global optimal result in the particle swarm.

[0134] Among them, the speed of the i-th particle in the k+1th iteration is and location They are:

[0135]

[0136] Where: and They represent the current position and velocity of the i-th particle at the k-th iteration respectively; is the global optimal position of the particle swarm at the kth iteration, the subscript g is the index of the optimal particle position; c1 and c2 represent the acceleration factors of the particles; r1 and r2 are random numbers in the interval [0,1]; ω represents the inertia weight that is linearly related to the number of iterations.

[0137] S406, determine whether the number of iterations reaches the maximum number of iterations kmax, if so, go to S407, otherwise go to S402.

[0138] S407, saving the global optimal result and obtaining the electromechanical optimal characteristic matrix.

[0139] Step S5: input the optimal electromyography feature matrix into the pre-trained MobileNet V2 neural network model and output the motion intention prediction result.

[0140] like Figure 3 As shown in the figure, the Bayesian optimization algorithm is used to optimize the MobileNet V2 neural network model. The specific process includes:

[0141] S501, input the optimal electromyographic feature matrix into the initialized MobileNet V2 neural network, use all hyperparameters of the MobileNetV2 neural network as the input of the objective function, use the optimization model based on the Bayesian optimization algorithm to optimize and obtain the average AUC of cross-validation 5 times, use the fully connected neural network as a proxy function to fit the average AUC of cross-validation 5 times, and create a hyperparameter search space;

[0142] S502: Randomly select a hyperparameter combination from the hyperparameter search space through the acquisition function, and calculate the objective function value y corresponding to the current hyperparameter combination i And record it in the data set. According to the result of the proxy function fitting, the acquisition function selects the next hyperparameter combination;

[0143] S503. Use the new hyperparameter combination to build a MobileNet V2 neural network, train the model using the training data, and return it. Record the parameter combination and its objective function value, and update the data set. The current hyperparameter combination and objective function value are used to refit the proxy function.

[0144] S504: Determine whether the iteration is finished. If the iteration is finished, proceed to S505. If it is not finished, return to S502, and the acquisition function will reselect the next hyperparameter combination.

[0145] S505. Extract the best parameter combination in the data set according to the evaluation score, build the optimal MobileNet V2 neural network model, and train the model using the training set data.

[0146] S506. Import the test set to test the constructed model, and use the test results to judge the feasibility of the model to obtain the final MobileNet V2 neural network model.

[0147] This embodiment also provides a system for predicting movement intention based on surface electromyography signals of multiple muscle groups, including:

[0148] Myoelectric signal acquisition module, used to collect surface myoelectric signals of multiple muscle groups;

[0149] The electromyographic sequence segmentation module is used to detect and segment the action sequence of surface electromyographic signals of multiple muscle groups and obtain the segmented electromyographic sequence;

[0150] A multi-muscle group multi-modal feature parameter acquisition module is used to extract multi-dimensional electromyographic feature values ​​from the segmented electromyographic sequences to obtain multi-muscle group multi-modal feature parameters;

[0151] The module for obtaining the optimal electromyographic feature matrix is ​​used to reduce the dimension of multi-modal feature parameters of multiple muscle groups using a binary particle swarm optimization algorithm to determine the optimal electromyographic feature matrix. The MobileNet V2 neural network is used to classify and identify the binary converted particle data set, and the negative value of the F1 score obtained by the MobileNet V2 neural network is used as the fitness function value.

[0152] The recognition prediction module is used to input the optimal electromyography feature matrix into the pre-trained MobileNet V2 neural network model and output the motion intention prediction result.

[0153] This embodiment provides a wearable electronic device, including: one or more processors and a memory, wherein one or more programs are stored in the memory, and the one or more programs include instructions for executing the above-mentioned method of deep learning prediction of movement intention using surface electromyography signals of multiple muscle groups.

[0154] The electronic device of the present invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0155] Multiple components in the device are connected to the I / O interface, including: input units, such as keyboards, mice, etc.; output units, such as various types of displays, speakers, etc.; storage units, such as disks, optical disks, etc.; and communication units, such as network cards, modems, wireless communication transceivers, etc. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunication networks.

[0156] The processing unit performs the various methods and processes described above, such as methods S1 to S5. For example, in some embodiments, methods S1 to S5 may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via a ROM and / or a communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of methods S1 to S5 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute methods S1 to S5 in any other appropriate manner (e.g., by means of firmware).

[0157] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip systems (SOCs), load programmable logic devices (CPLDs), and the like.

[0158] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0159] In the context of the present invention, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0160] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for predicting movement intention based on surface electromyographic signals of multiple muscle groups, characterized in that: include: S1, collect surface electromyographic signals of multiple muscle groups; S2, performing motion sequence detection and segmentation on surface electromyographic signals of multiple muscle groups to obtain segmented electromyographic sequences; S3, extracting multi-dimensional electromyographic feature values ​​from the segmented electromyographic sequences to obtain multi-muscle group multi-modal feature parameters; S4. Use binary particle swarm optimization algorithm to reduce the dimension of multi-muscle group multi-modal feature parameters and determine the optimal electromyographic feature matrix; use MobileNet V2 neural network to classify and identify the binary converted particle data set, and use the negative value of the F1 score obtained by MobileNet V2 neural network as the fitness function value; S5. Input the optimal electromyography feature matrix into the pre-trained MobileNet V2 neural network model and output the motion intention prediction result.

2. The method for predicting movement intention based on surface electromyography signals of multiple muscle groups according to claim 1 is characterized in that: The multi-muscle surface electromyographic signals include the surface electromyographic signals of the rectus femoris, the surface electromyographic signals of the vastus medialis, the surface electromyographic signals of the vastus lateralis, the surface electromyographic signals of the semitendinosus and the surface electromyographic signals of the biceps femoris.

3. The method for predicting movement intention based on surface electromyography signals of multiple muscle groups according to claim 1 is characterized in that: After collecting the surface electromyographic signals of multiple muscle groups, the method also includes denoising the surface electromyographic signals of multiple muscle groups, specifically: using a 50Hz industrial frequency notch filter to remove industrial frequency interference in the surface electromyographic signals of multiple muscle groups, and using a 20Hz-500Hz Butterworth bandpass filter to filter out high-frequency noise in the surface electromyographic signals of multiple muscle groups.

4. The method for predicting movement intention based on surface electromyography signals of multiple muscle groups according to claim 1 is characterized in that: The method of performing action sequence detection and segmentation on surface electromyographic signals of multiple muscle groups to obtain the segmented electromyographic sequence specifically comprises: using sliding window signal energy threshold detection and segmenting the starting position of the electromyographic action to obtain the segmented electromyographic sequence.

5. The method for predicting movement intention based on surface electromyography signals of multiple muscle groups according to claim 1 is characterized in that: The multi-dimensional electromyographic characteristic values ​​include electromyographic integral value, variance, root mean square value, wavelength, average power frequency, median frequency, wavelet packet energy, approximate entropy, sample entropy and fuzzy entropy.

6. The method for predicting movement intention based on surface electromyography signals of multiple muscle groups according to claim 1 is characterized in that: The binary particle swarm optimization algorithm is used to reduce the dimension of the characteristic parameters and determine the optimal electromyographic characteristic matrix, which specifically includes: S401, initializing the particle population, including taking the number of multi-muscle group multi-modal characteristic parameters as the particle space dimension d, setting the number of particles n, setting the maximum number of iterations kmax, randomly generating the particle initial position information vector X and the flight speed vector V, recording the initial state position and speed as the current local optimal solution, and importing the multi-muscle group multi-modal characteristic parameter data set; S402, using the Sigmoid function to perform binary conversion on the particles, using the conversion result to form a new feature parameter subset, substituting the new feature parameter subset into the MobileNet V2 neural network for classification and recognition, and using the negative value of the F1 score obtained by the MobileNet V2 neural network as the fitness function value; S403, based on the fitness function value and current position of each particle, update the individual best position, and record the individual best position of the i-th particle in the k-th iteration S404, update the global best position according to the individual historical best solution and the current individual best position, and record the individual best position of the i-th particle in the k-th iteration S405, update the speed and position of all particles, where the speed V of the i-th particle in the k+1th iteration is i k+1 and location They are: Where: and V i k They represent the current position and velocity of the i-th particle at the k-th iteration respectively; is the global optimal position of the particle swarm at the kth iteration, the subscript g is the index of the optimal particle position; c1 and c2 represent the acceleration factors of the particles; r1 and r2 are random numbers in the interval [0,1]; ω represents the inertia weight that is linearly related to the number of iterations; S406, determine whether the number of iterations reaches the maximum number of iterations kmax, if yes, go to S407, otherwise go to S402; S407, saving the global optimal result and obtaining the electromyography optimal feature matrix.

7. The method for predicting movement intention based on surface electromyographic signals of multiple muscle groups according to claim 1, characterized in that: In S4, the MobileNet V2 neural network model is pre-trained using a Bayesian optimization algorithm, wherein all hyperparameters of the MobileNet V2 convolutional neural network are used as inputs of the objective function, and the average AUC of multiple cross-validations is used as the optimization objective function.

8. A system based on the method for predicting movement intention of multiple muscle group surface electromyography signals according to claim 1, characterized in that: The system comprises: Myoelectric signal acquisition module, used to collect surface myoelectric signals of multiple muscle groups; The electromyographic sequence segmentation module is used to detect and segment the action sequence of surface electromyographic signals of multiple muscle groups and obtain the segmented electromyographic sequence; A multi-muscle group multi-modal feature parameter acquisition module is used to extract multi-dimensional electromyographic feature values ​​from the segmented electromyographic sequences to obtain multi-muscle group multi-modal feature parameters; The module for obtaining the optimal electromyographic feature matrix is ​​used to reduce the dimension of multi-modal feature parameters of multiple muscle groups using a binary particle swarm optimization algorithm to determine the optimal electromyographic feature matrix. The MobileNet V2 neural network is used to classify and identify the binary converted particle data set, and the negative value of the F1 score obtained by the MobileNet V2 neural network is used as the fitness function value. The recognition prediction module is used to input the optimal electromyography feature matrix into the pre-trained MobileNet V2 neural network model and output the motion intention prediction result.

9. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images