SEMG recognition method based on AWPSO-SVM model, electronic equipment and storage medium

Abstract

The invention discloses an sEMG recognition method based on an AWPSO-SVM model, electronic equipment and a storage medium, belongs to the technical field of surface electromyogram signal recognition, and solves the problem of how to improve the recognition precision of surface electromyogram signals. Using a Sigmoid function as an adaptive weight update function to improve a particle swarm algorithm to obtain an AWPSO algorithm, and outputting an optimal penalty parameter and a kernel function radius through the AWPSO algorithm to optimize an SVM to obtain an AWPSO-SVM model; according to the method, the acceleration coefficient is adaptively controlled within a reasonable range, so that the efficiency in the speed updating process is ensured; a Sigmoid function is selected as an adaptive weighted update function, monotonous but relatively smooth change of an acceleration coefficient can be reflected, the larger the distance is, the larger the acceleration coefficient value is, and particles can be excited to seek an optimal solution as soon as possible, so that the recognition precision of the surface electromyogram signals is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of surface electromyographic signal recognition, and relates to a sEMG recognition method, an electronic device and a storage medium based on an AWPSO-SVM model. Background Art

[0002] A surface electromyographic signal (sEMG) refers to a bioelectrical signal generated by muscle groups related to corresponding movements during human movement, and this electrical signal can be detected by a specific sensor on the human epidermis.

[0003] Currently, a common method for surface electromyographic signal recognition is the PSO-SVM classification model, which combines the particle swarm optimization algorithm and the support vector machine model to obtain a better classification effect. In the common PSO-SVM classification model, considering that the SVM classifier is greatly affected by the penalty parameter and the kernel function parameter, the particle swarm optimization algorithm (PSO) is introduced to optimize the penalty parameter and the kernel function parameter in the SVM, and a set of penalty parameters and kernel function parameters with the smallest error in the SVM result are used for pattern recognition.

[0004] Since the particle swarm optimization algorithm is an optimization algorithm based on swarm intelligence, when the PSO-SVM classifier model is used for surface electromyographic signal recognition, it is easy to fall into a local optimal solution rather than a global optimal solution; during the process of optimizing the SVM by the particle swarm optimization algorithm, the convergence speed is easily affected by the control parameters of the particles such as the inertia weight and the acceleration coefficient. When used for surface electromyographic signal recognition, the convergence speed is slow and the search rate is low. Also, due to the inability to dynamically adjust the control parameters of the particles in the classifier, the classification recognition accuracy is low and the recognition precision is not ideal. Summary of the Invention

[0005] The technical solution of the present invention is used to solve the problem of how to improve the recognition accuracy of surface electromyographic signals.

[0006] The present invention solves the above technical problem through the following technical solutions:

[0007] The present invention provides a sEMG recognition method based on an AWPSO-SVM model, including:

[0008] S1. Feature extraction of surface electromyographic signals and dataset collection;

[0009] S2. Using the Sigmoid function as an adaptive weight update function to improve the particle swarm optimization algorithm to obtain the AWPSO algorithm;

[0010] S3. Optimize the SVM by outputting the optimal penalty parameter and kernel function radius through the AWPSO algorithm to obtain the AWPSO-SVM model.

[0011] Furthermore, the method for feature extraction and dataset collection of the surface electromyogram signals in S1 is as follows: extract the time-domain features and frequency-domain features of the surface electromyogram signals respectively, and store the time-domain features and frequency-domain features in pairs in the dataset.

[0012] Furthermore, the time-domain features in S1 include: mean absolute value, waveform length, and number of zero crossings.

[0013] Furthermore, the update rule of using the Sigmoid function as the adaptive weight update function in S2 is as follows:

[0014]

[0015] where the function F(·) represents the adaptive weighted update function, and g pi (k) and g gi (k) represent the distances from particle i to the individual extreme value and the global extreme value at the k-th iteration respectively.

[0016] Furthermore, the update formulas for the velocity and position of the i-th particle in the AWPSO algorithm during the population evolution process in S2 are as follows:

[0017]

[0018] where represents the acceleration constant determined by g pi (k); represents the acceleration constant determined by g gi (k), v i (k + 1) represents the velocity of particle i at the (k + 1)-th iteration, ω represents the inertia weight, v i (k) represents the velocity of particle i at the k-th iteration, r 1 , r 2 represent two random numbers within the range of [0, 1], x i (k + 1) represents the position of particle i at the (k + 1)-th iteration, and x i (k) represents the position of particle i at the k-th iteration.

[0019] Furthermore, the method for optimizing the SVM by outputting the optimal penalty parameter and kernel function radius through the AWPSO algorithm to obtain the AWPSO-SVM model is specifically as follows:

[0020] 1) Initialize the parameters of the AWPSO algorithm;

[0021] 2) Read the original surface electromyogram signal dataset and labels;

[0022] 3) Update the individual best position and the global best position;

[0023] 4) Calculate the inertia according to w = w1 - (w1 - w2) * k / maxiter, where w1 is the initial inertia weight, w2 is the final inertia weight, k is the current iteration number, and maxiter is the maximum iteration number;

[0024] 5) Calculate the distances of each particle relative to the individual best position and the global best position according to gpi(k) = pi(k) - xi(k) and ggi(k) = pg(k) - xi(k), where gpi(k) is the distance relative to the individual best position, ggi(k) is the distance relative to the global best position, pi(k) is the current position of particle i at the k-th iteration, xi(k) is the individual best position of particle i at the k-th iteration, and pg(k) is the global best position;

[0025] 6) Update the velocity and position of each particle according to xi(k + 1) = xi(k) + vi(k + 1);

[0026] 7) Update the particle acceleration coefficient according to the Sigmoid function where e is the natural logarithm, a represents the steepness of the curve, b represents the peak value of the curve, c represents the abscissa value of the center point of the curve, d is a normal value, and D is the input of the function;

[0027] 8) Determine whether the maximum iteration number is reached. If the maximum iteration number is not reached, repeat steps 2) to 8) until the maximum iteration number is reached;

[0028] 9) If the maximum iteration number is reached, output the optimal penalty parameter and the kernel function radius parameter, and use the optimal penalty parameter and the kernel function radius parameter to train and test the SVM model.

[0029] Further, the parameters for initializing the AWPSO algorithm include: particle swarm parameter setting, initialization of particle positions and velocities, adaptive weight parameter setting, setting of learning factors, and setting of the maximum iteration number.

[0030] Further, the method for training and testing the SVM model using the optimal penalty parameter and the kernel function radius parameter is as follows:

[0031] Import the numpy and svm modules;

[0032] Use the train_test_split function to divide the surface electromyogram signal dataset into a surface electromyogram signal training set and a surface electromyogram signal test set;

[0033] Create an SVM model using svm.SVC and pass in the optimal penalty parameter and kernel function radius parameter;

[0034] Call the fit method to train the model;

[0035] Use the surface electromyogram signal test set for prediction and calculate the accuracy of the model.

[0036] The present invention also provides an electronic device, including a memory and a processor. The memory is used to store a program that supports the processor to execute the above sEMG recognition method based on the AWPSO - SVM model, and the processor is configured to execute the program stored in the memory.

[0037] The present invention also provides a storage medium, on which a computer program is stored. When the computer program is run by a processor, it executes the steps of the above sEMG recognition method based on the AWPSO - SVM model.

[0038] The advantages of the present invention are as follows:

[0039] Through feature extraction and dataset collection of surface electromyogram signals, the present invention uses the Sigmoid function as an adaptive weight update function to improve the particle swarm algorithm to obtain the AWPSO algorithm. The optimal penalty parameter and kernel function radius are output by the AWPSO algorithm to optimize SVM, thus obtaining the AWPSO - SVM model; the acceleration coefficient is adaptively controlled within a reasonable range, ensuring the efficiency in the speed update process and improving the convergence speed; selecting the Sigmoid function as the adaptive weighted update function can reflect the monotonic but relatively smooth change of the acceleration coefficient. The larger the distance, the larger the acceleration coefficient value; it can encourage the particles to seek the optimal solution as quickly as possible, thereby improving the accuracy of surface electromyogram signal recognition. Brief Description of the Drawings

[0040] Figure 1 It is a flowchart of the sEMG recognition method based on the AWPSO - SVM model of the present invention;

[0041] Figure 2 It is a flowchart of obtaining the AWPSO - SVM model by optimizing the SVM model through the AWPSO algorithm of the present invention;

[0042] Figure 3 It is a diagram of the actual classification and predicted classification of the training set of the method of the present invention compared with the SVM optimized by the PSO algorithm;

[0043] Figure 4 It is a diagram of the actual classification and predicted classification of the test set of the method of the present invention compared with the SVM optimized by the PSO algorithm;

[0044] Figure 5Fitness curve diagram of the method of this invention for optimizing SVM by comparing with PSO algorithm

[0045] Figure 6 Actual classification and predicted classification diagram of the training set of SVM optimized by the method of this invention

[0046] Figure 7 Actual classification and predicted classification diagram of the test set of SVM optimized by the method of this invention

[0047] Figure 8 Fitness curve diagram of SVM optimized by the method of this invention Detailed implementation manners

[0048] To make the objectives, technical solutions and advantages of the embodiments of this invention clearer, the technical solutions in the embodiments of this invention will be clearly and completely described below in conjunction with the embodiments of this invention. Apparently, the described embodiments are some but not all of the embodiments of this invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of this invention without making creative efforts shall fall within the protection scope of this invention.

[0049] The technical solutions of this invention will be further described below in conjunction with the accompanying drawings of the specification and specific embodiments:

[0050] Embodiment 1

[0051] As Figure 1 shown, the sEMG recognition method based on the AWPSO - SVM model in the embodiment of this invention includes:

[0052] Step 1, Feature extraction of surface electromyography signals and dataset collection

[0053] Extract the time - domain features and frequency - domain features of surface electromyography signals respectively, and store the time - domain features and frequency - domain features in pairs in the dataset; where the time - domain features include the average absolute value, waveform length, and number of zero - crossings. When applying the frequency - domain features in sEMG signal feature extraction, the frequency - domain features are used to analyze the fatigue degree of relevant muscles during continuous limb movements.

[0054] Step 2, Propose the Adaptive Weighted Particle Swarm Optimization (AWPSO) algorithm

[0055] Improve the particle swarm optimization algorithm by using the Sigmoid function as an adaptive weight update function. Adaptively update the control parameters, the inertia weight and the acceleration coefficient, according to the output of the Sigmoid function. Use the Sigmoid function to describe the relationship between the acceleration coefficient and the distance, where the distance is from the particle to its personal best pbest and the global best gbest. This can effectively improve the search ability of the particle swarm optimization algorithm, being able to find the global optimal solution and obtain a satisfactory convergence speed.

[0056] The update rule of using the Sigmoid function as the adaptive weight update function proposed by the present invention is as follows:

[0057]

[0058] Among them, the function F(·) represents the adaptive weighted update function, and g pi (k) and g gi (k) respectively represent the distances from the particle i to the personal best pbest and the global best gbest at the k-th iteration.

[0059] During the population evolution process of the AWPSO algorithm, the velocity and position of the i-th particle are updated according to the following formulas:

[0060]

[0061] Among them, represents the acceleration constant determined by g pi (k); represents the acceleration constant determined by g gi (k), v i (k + 1) represents the velocity of the particle i at the (k + 1)-th iteration, ω represents the inertia weight, v i (k) represents the velocity of the particle i at the k-th iteration, r 1 , r 2 represent two random numbers within the range of [0, 1], x i (k + 1) represents the iterative position of the particle i at the (k + 1)-th time, x i (k) represents the iterative position of the particle i at the k-th time.

[0062] Step 3, output the optimal penalty parameter and kernel function radius through the AWPSO algorithm to optimize the SVM model, thereby obtaining the AWPSO-SVM model. Divide the surface electromyogram signal dataset into a surface electromyogram signal training set and a surface electromyogram signal test set, and train and test the AWPSO-SVM model.

[0063] As Figure 2 shown, the specific method for obtaining the AWPSO-SVM model is as follows:

[0064] 1) Initialize the parameters of the adaptive weighted particle swarm, including setting the particle swarm parameters, initializing the particle positions and velocities, setting the adaptive weight parameter, setting the learning factors, and setting the maximum number of iterations. Among them, setting the particle swarm parameters includes setting the number of particles N and the dimension D. It is preset that N is 60, and the value range of N is 20 - 100. The dimension D is set according to the complexity of the problem to determine the dimension of the particles, which is the same as the number of variables of the optimization problem. The initial position of the particles is randomly generated using a normal distribution, and the initial velocity of the particles is set to 0, with a value range of [-Vmax, Vmax]. The adaptive weight parameter includes weight initialization and weight update strategy. The weight initialization parameter is initialized to a constant value, preset to 1, and the weight update strategy dynamically adjusts the weight according to the historical best position and the global best position of the particles. The learning factors (c1, c2) are set to constants, initially set to 2, where c1 is the attraction of the particle to its own best position, and c2 is the attraction of the particle to the global best position. The number of iterations T is the maximum number of iterations, with a value range of 50 - 500, preset to 100.

[0065] Initializing the parameters of the adaptive weighted particle swarm is the basis for ensuring the effectiveness of the algorithm. By reasonably setting the number of particles, positions, velocities, weights, and learning factors, the convergence speed and optimization effect of the algorithm can be improved.

[0066] 2) Read the original surface electromyogram signal dataset and labels. Read the surface electromyogram signal feature data stored in the dataset and remove the outliers in the dataset. The format of the dataset can be selected as CSV, Excel, and text formats, or it can also be in the form of a NumPy array.

[0067] 3) Update the individual best position Pbest and the global best position Gbest. By continuously updating these values, the particles can better explore the search space and find the global optimal solution. Among them, the condition for updating Pbest is that the fitness value (objective function value) of the current particle is better than the Pbest fitness value recorded previously. When updating Pbest, first calculate the fitness value of the current particle. If the current fitness value is better than the fitness value of Pbest, then update Pbest and its corresponding fitness value. The condition for updating the global best position Gbest is that the fitness value of the current particle is better than the global best fitness value. When updating Gbest, traverse all the particles, find the particle with the optimal fitness value, and then update Gbest to the current position of this particle.

[0068] 4) Calculate the inertia according to w = w1 - (w1 - w2) * k / maxiter, where w1 is the initial inertia weight, w2 is the final inertia weight, k is the current iteration number, and maxiter is the maximum number of iterations. After calculating the current inertia weight (w) using the above formula for calculating inertia each time, output the current inertia weight.

[0069] 5) Calculate the distances of each particle relative to Pbest and Gbest according to gpi(k) = pi(k) - xi(k) and ggi(k) = pg(k) - xi(k), where gpi(k) is the distance relative to Pbest, ggi(k) is the distance relative to Gbest, pi(k) is the current position of particle i at the k-th iteration, xi(k) is the Pbest position of particle i at the k-th iteration, and pg(k) is the global best position Gbest. The above calculations can help us understand the relative positions of the particles and guide the particles towards better solutions during the update process.

[0070] 6) Update the velocity and position of each particle according to xi(k + 1) = xi(k) + vi(k + 1).

[0071] 7) Update the particle acceleration coefficient according to the Sigmoid function where e is the natural logarithm, a represents the steepness of the curve, b represents the peak value of the curve, c represents the abscissa value of the center point of the curve, d is a normal value, and D is the input of the function.

[0072] 8) Determine whether the maximum number of iterations has been reached. If the maximum number of iterations has not been reached, repeat steps 2) to 8) until the maximum number of iterations is reached.

[0073] 9) If the maximum number of iterations is reached, output the optimal penalty parameter and kernel function radius parameter, and use the optimal penalty parameter and kernel function radius parameter to train and test the SVM model.

[0074] The method of using the optimal penalty parameter and kernel function radius parameter to train and test the SVM model is as follows:

[0075] ① Import the numpy and svm modules;

[0076] ② Use the train_test_split function to divide the surface electromyogram signal dataset into a surface electromyogram signal training set and a surface electromyogram signal test set;

[0077] ③ Create an SVM model using svm.SVC and pass in the optimal penalty parameter and kernel function radius parameter;

[0078] ④ Call the fit method to train the model;

[0079] ⑤ Use the surface electromyogram signal test set for prediction and calculate the accuracy rate of the model.

[0080] Experimental verification

[0081] Based on the method of the present invention, the surface electromyogram signals of 10 subjects were collected as a data set for experiments. A total of 50 groups of data, that is, 450 sample data, were obtained for each action, and a total of 200 groups of data, that is, 1,800 sample data, were obtained for the four actions. A total of 40 groups, that is, 360 sample data, for each action were used for the training of the classifier, and the remaining 10 groups, that is, 90 sample data, were used for the test experiment. Finally, a total of 1,440 data for the four actions were used as training samples, and the remaining 360 data were used as test samples. The experimental results are as Figures 3 to 8 shown:

[0082] From Figure 3 it can be seen that the classification accuracy rate of the training set in the PSO-SVM classifier is 94.86%;

[0083] From Figure 4 it can be seen that the classification accuracy rate of the test set in the PSO-SVM classifier is 88.61%;

[0084] From Figure 5 it can be seen that the PSO optimization of the SVM parameters converges to the optimal solution after 12 iterations;

[0085] From Figure 6 it can be seen that the classification accuracy rate of the training set in the AWPSO-SVM classifier can reach 97.63%;

[0086] From Figure 7 it can be seen that the classification accuracy rate of the test set in the AWPSO-SVM classifier can reach 92.22%;

[0087] From Figure 8 it can be seen that the AWPSO algorithm optimizes the SVM parameters and converges to the optimal solution after 7 iterations.

[0088] In summary, compared with the PSO-optimized SVM algorithm (i.e., the PSO-SVM algorithm) used as a control, the method of the present invention can improve the accuracy rates of the training set and the test set. At the same time, by observing the respective fitness curves, it can be found that the method of the present invention can achieve the final classification effect after fewer evolutionary times.

[0089] In order to better solve the problems of slow convergence speed and low recognition accuracy in the traditional PSO-SVM classification model, the present invention introduces an adaptive weighting strategy, fully utilizes the distances of each particle to its individual optimal position and the global optimal position in each iteration, takes the calculated distances as inputs, selects the Sigmoid function as the adaptive weighting update function, adaptively updates the control parameters according to the output of the Sigmoid function, and constructs an AWPSO-SVM classification and recognition model, which can not only improve the recognition accuracy of the algorithm but also obtain a satisfactory convergence speed when used for surface electromyogram signal recognition.

[0090] Embodiment 2

[0091] An electronic device includes a memory and a processor. The memory is used to store a program that supports the processor to execute the sEMG recognition method based on the AWPSO-SVM model in Embodiment 1, and the processor is configured to execute the program stored in the memory.

[0092] Embodiment 3

[0093] A storage medium stores a computer program, and when the computer program is run by a processor, it executes the steps of the sEMG recognition method based on the AWPSO-SVM model in Embodiment 1.

[0094] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. The sEMG recognition method based on the AWPSO-SVM model is characterized by: include: S1, feature extraction of surface electromyography signals and data set collection; S2, using Sigmoid function as adaptive weight update function to improve particle swarm algorithm and obtain AWPSO algorithm; S3. The AWPSO algorithm is used to output the optimal penalty parameter and kernel function radius to optimize the SVM and obtain the AWPSO-SVM model.

2. The sEMG recognition method based on the AWPSO-SVM model according to claim 1, characterized in that: The method for feature extraction and data set collection of the surface electromyography signal described in S1 is as follows: extract the time domain features and frequency domain features of the surface electromyography signal respectively, and store the time domain features and frequency domain features in pairs in the data set.

3. The sEMG recognition method based on the AWPSO-SVM model according to claim 2, characterized in that: The time domain features in S1 include: average absolute value, waveform length and number of zero crossing points.

4. The sEMG recognition method based on the AWPSO-SVM model according to claim 1, characterized in that: The update rule using the Sigmoid function as the adaptive weight update function described in S2 is as follows: Among them, the function F(·) represents the adaptive weighted update function, g pi (k) and g gi (k) represent the distance from particle i to the individual extreme value and the group extreme value at the kth iteration, respectively.

5. The sEMG recognition method based on the AWPSO-SVM model according to claim 4, characterized in that: The update formula of the speed and position of the i-th particle in the population evolution process of the AWPSO algorithm described in S2 is as follows: in, Indicated by g pi (k) Determined acceleration constant; Indicated by g gi (k) Determine the acceleration constant, v i (k+1) represents the velocity of particle i at the k+1th iteration, ω represents the inertia weight, and v i (k) represents the velocity of particle i at the kth iteration, r1 and r2 represent two random numbers in the range [0,1], and x i (k+1) represents the position of particle i at the k+1th iteration, x i (k) represents the position of particle i at the kth iteration.

6. The sEMG recognition method based on the AWPSO-SVM model according to claim 1, characterized in that: The method of optimizing SVM by outputting the optimal penalty parameter and kernel function radius through the AWPSO algorithm to obtain the AWPSO-SVM model is as follows: 1) Initialize the parameters of the AWPSO algorithm; 2) Read the original surface electromyography signal data set and labels; 3) Update the individual best position and the global best position; 4) Calculate the inertia according to w = w1-(w1-w2)*k / maxiter, where w1 is the initial inertia weight, w2 is the final inertia weight, k is the current number of iterations, and maxiter is the maximum number of iterations; 5) Calculate the distance of each particle relative to the individual best position and the global best position according to gpi(k)=pi(k)-xi(k) and ggi(k)=pg(k)-xi(k), where gpi(k) is the distance relative to the individual best position, ggi(k) is the distance relative to the global best position, pi(k) is the current position of particle i at the kth iteration, xi(k) is the individual best position of particle i at the kth iteration, and pg(k) is the global best position; 6) Update the velocity and position of each particle according to xi(k+1)=xi(k)+vi(k+1); 7) According to the Sigmoid function Update the particle acceleration coefficient, where e is the natural logarithm, a represents the steepness of the curve, b represents the peak value of the curve, c represents the horizontal coordinate value of the center point of the curve, d is a normal value, and D is the input of the function; 8) Determine whether the maximum number of iterations has been reached. If not, repeat steps 2) to 8) until the maximum number of iterations has been reached. 9) If the maximum number of iterations is reached, the optimal penalty parameter and kernel function radius parameter are output, and the optimal penalty parameter and kernel function radius parameter are used to train and test the SVM model.

7. The sEMG recognition method based on the AWPSO-SVM model according to claim 6, characterized in that: The parameters for initializing the AWPSO algorithm include: particle swarm parameter settings, initialization particle positions and velocities, adaptive weight parameter settings, learning factor settings, and maximum number of iterations settings.

8. The sEMG recognition method based on the AWPSO-SVM model according to claim 6, characterized in that: The method for training and testing the SVM model using the optimal penalty parameter and kernel function radius parameter is: Import numpy and svm modules; Use the train_test_split function to divide the surface electromyography data set into a surface electromyography training set and a surface electromyography test set; Use svm.SVC to create an SVM model and pass in the optimal penalty parameter and kernel function radius parameter; Call the fit method to train the model; Use the surface electromyography signal test set to make predictions and calculate the accuracy of the model.

9. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the sEMG recognition method based on the AWPSO-SVM model as described in any one of claims 1 to 8, and the processor is configured to execute the program stored in the memory.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the sEMG recognition method based on the AWPSO-SVM model according to any one of claims 1 to 8 are performed.

Images