Multi-degree-of-freedom control system based on surface micro electric signals

Abstract

This invention discloses a multi-degree-of-freedom control system based on surface micro-electrical signals, comprising: an electromyography (EMG) signal acquisition module for acquiring surface EMG signals; a signal preprocessing module for preprocessing the acquired EMG signals; a feature extraction module for calculating the optimal parameters of multi-scale permutation entropy using a particle swarm optimization algorithm for the preprocessed signals; and based on the optimal parameters, performing multi-scale permutation entropy feature extraction on the preprocessed EMG signals to obtain feature data of the EMG signals; a signal classification module for classifying the EMG signals according to the feature dataset; and a feature matching module for matching the classified signals with preset actions. This invention solves the problem of limited application scenarios and efficiency of existing surface EMG signal-controlled robotic arms, which is caused by the low degree of freedom of the robotic arm.

Description

Technical Field

[0001] This invention belongs to the field of robotic arm control technology, and relates to a technical solution for controlling a robotic arm by electromyographic signals, and more specifically to a multi-degree-of-freedom control system based on surface micro-electrical signals. Background Technology

[0002] Surface electromyography (sEMG) is a type of electromyography (EMG) technique. EMG techniques are broadly classified into two categories: inserting EMG (IEMG) needle electrodes and surface EMG. Needle electrode techniques require inserting electrodes into muscle tissue to obtain electrical signals from the muscle. In contrast, surface EMG is painless and non-invasive, simply placing electrodes on the surface of the skin. The signals acquired through surface electrodes are aliased EMG signals generated by subcutaneous muscle tissue. This method can be applied in many fields such as motion analysis, rehabilitation examination, and human-computer interaction. It has minimal impact on the subject, is convenient to use, and is painless. Currently, surface EMG is widely used, for example, to control prosthetic limb movement and as a control signal for functional electrical stimulation to achieve a certain feedback regulation effect.

[0003] However, in the existing technology of controlling robotic arms with surface electromyography signals, the degree of freedom of the robotic arm is not high, which greatly limits the application scenarios and efficiency of the robotic arm. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, the present invention provides a multi-degree-of-freedom control system based on surface micro-electrical signals.

[0005] The present invention adopts the following technical solution:

[0006] A multi-degree-of-freedom control system based on surface micro-electrical signals includes the following modules:

[0007] Electromyography (EMG) signal acquisition module: used for acquiring surface EMG signals;

[0008] Signal preprocessing module: preprocesses the acquired electromyographic signals;

[0009] Feature extraction module: For the preprocessed signal, the particle swarm optimization algorithm is used to calculate the optimal parameters of the multi-scale permutation entropy; based on the optimal parameters, the multi-scale permutation entropy features of the preprocessed electromyography signal are extracted to obtain the feature data of the electromyography signal;

[0010] Signal classification module: Classifies electromyographic signals based on their feature dataset;

[0011] Feature matching module: Matches the classified signals with preset actions.

[0012] Preferably, in the signal preprocessing module, the preprocessing of the raw electromyographic signal includes: firstly, constructing a passive low-pass filter using capacitors and resistors to filter out high-frequency interference signals. In software, an IIR filter is used for signal processing; an nth-order IIR digital filter can be represented as...

[0013]

[0014] Among them, a i With b i The desired parameters are obtained through calculation, and these parameters also determine the filter's effectiveness. Since the frequency of upper limb surface electromyography (EMG) signals is approximately between 30Hz and 250Hz, the designed filter is 50Hz, 3rd order, with a bandwidth of 0.6Hz and a quality factor of 3. The final parameters are as follows:

[0015] a0 = 1, b0 = 0.998119

[0016] a1 = -1.898538, b1 = -1.898538

[0017] a² = 0.998119, b² = 0.998118

[0018] A high-pass filter is then used for further filtering. This high-pass filter is designed to allow only signals above 7Hz to pass through, with an upper cutoff frequency of 250Hz. After these two layers of filtering, noise can be effectively removed, significantly reducing the impact of power frequency interference in the original signal.

[0019] Preferably, in the signal preprocessing module, before filtering preprocessing, the sliding window analysis method is used to detect active segments in the original electromyography (EMG) signal to remove non-motion signals from the original signal. The window length is 300ms and the increment interval is 50ms. Active segment detection of the EMG signal using the sliding window analysis method is completed before filtering.

[0020] Preferably, in the feature extraction module, the optimal parameters of the multi-scale permutation entropy (MPE) are calculated using the particle swarm optimization (PSO) algorithm, including:

[0021] The ranges of each parameter of the multi-scale permutation entropy are set; among them, the ranges of each parameter of the multi-scale permutation entropy include: embedding dimension m, with a value range of m = 2 to 7; delay time t, with a value range of t = 1 to 5; and scale factor s, with a value range of s = 1 to 20.

[0022] The square function of the entropy skewness of the multi-scale arrangement is selected as the objective function, and its minimum value is sought; the entropy skewness of the multi-scale arrangement is expressed as:

[0023]

[0024] Among them, H p (X) is a time series X = {x i The sequence of permutation entropy for all scales of {i = 1, 2, ..., N} is represented as:

[0025] H p (X)={H p (1),H p (2),…,H p (s)}

[0026] and For sequence H p The mean of (X); H p (X) is the sequence H p The standard deviation of (X); E[.] is the expected value of the sequence;

[0027] Set the objective function of PSO as follows:

[0028] F(X) = Sk 2

[0029] The optimal parameters for the multi-scale permutation entropy are obtained through calculation.

[0030] Furthermore, in the feature extraction module, based on the optimal parameters, multi-scale permutation entropy feature extraction is performed on the preprocessed electromyography (EMG) signal to obtain the feature data of the EMG signal, as follows:

[0031] Time series coarsening:

[0032] For a time series X = {x} of length N i The sequence {i = 1, 2, ..., N} is coarsened to obtain a coarse-grained sequence:

[0033]

[0034] Where s is the scaling factor; [N / s] represents rounding down N / s. When s = 1, the coarse-grained sequence is the original sequence.

[0035] Phase space reconstruction for each coarse-grained sequence:

[0036]

[0037] Where m is the embedding dimension and τ is the delay factor for the time-reconstructed sequence. Sort in ascending order:

[0038]

[0039] Simplified as a sequence of symbols:

[0040] S(j)={j1,j2,…,j m}

[0041] There are m! (where ! represents the factorial of m) possible permutations. Calculate the probability of each permutation type:

[0042]

[0043] Where the numerator represents the frequency of each permutation and the denominator represents the total number of all permutations, the permutation entropy of the time series at multiple scales is calculated as follows:

[0044]

[0045] in, hour, Reaching the maximum value ln(m!), the entropy of the multi-scale arrangement is... Normalization is performed:

[0046]

[0047] in The normalized permutation entropy value is the characteristic data of the electromyographic signal.

[0048] Preferably, the signal classification module uses a Harris Eagle optimized support vector machine to classify the features of the feature dataset.

[0049] Furthermore, in the signal classification module, the support vector machine (SVM) classification model is optimized based on the Harris Eagle algorithm (HHO). The Harris Eagle algorithm mainly consists of three parts: the search phase, the transformation phase, and the development phase. The specific steps are as follows:

[0050] Step 1: Population initialization. The population is initialized using optimal point set optimization. The upper bound (ub) and lower bound (lb) of the search space are determined. The problem dimension r = 2. The position of each individual is initialized:

[0051] X i =2cos(2π / p)*(ub-lb)+lb,1≤k≤r,(p-3) / 2≥r

[0052] Among them, X i Let represent the position of the i-th individual in the population, where i ranges from [1, N], N is the size of the Harris Eagle population, r is the optimal point, and p is the smallest prime number that satisfies the conditions. The basic principle of the optimal point method is as follows:

[0053] Let G s If G is an s-dimensional Euclidean geometric space, then r∈G s ,but

[0054] Pn (i)=(r1i1,r2i2...r n i n ), 1≤i≤n

[0055] n represents the sample size, P n (i) represents the set of good points, r represents a good point, r = 2cos(2π / p), 1 ≤ i ≤ n, and here p is taken as 7;

[0056] Step 2: Calculate the initial fitness and set the position of the individual with the best fitness as the current prey position; Fitness calculation method:

[0057]

[0058]

[0059]

[0060] Among them, R i Recall evaluates how many positive samples are ultimately predicted successfully out of all positive samples, denoted by P. i The Precision Evaluation (PEE) measures the number of true positive samples among those predicted as positive. TP indicates that a positive sample was successfully predicted as positive, TN indicates that a negative sample was successfully predicted as negative, FP indicates that a negative sample was incorrectly predicted as positive, and FN indicates that a positive sample was incorrectly predicted as negative. r represents the number of dimensions in the classification target, and f(X) represents the number of true positive samples among those predicted as positive. i The value X represents the fitness score. The fitness scores are compared, and the individual with the lower fitness score is selected. i Position X as prey rabbit , that is, X rabbit It is all individuals X i The one with the lowest fitness;

[0061] Step 3: Position update. First, update the prey's escape energy, and then execute the corresponding position update strategy in the search or exploit behavior based on the escape energy and the generated random number.

[0062] The strategy for updating the position is determined based on the magnitude of the random value q:

[0063]

[0064] Where X(t) and X(t+1) are the positions of the individual in the current and next iterations, respectively, t is the iteration number, and X... rand (t) represents the randomly selected individual position, X rabbit(t) represents the prey position, i.e., the position of the individual with the optimal fitness, and is a random number between [0,1]. Specifically, information is obtained from the shared region when q is greater than or equal to 0.5; otherwise, information is obtained from the collaborators. The population average position is represented as:

[0065]

[0066] Based on the transformation of escape energy between search and different exploitation behaviors, the optimized perturbation escape energy is defined as:

[0067]

[0068] v = 1 + tan(π(rand() - 0.5))

[0069] Where E0 is the initial energy of the prey, a random number between [-1, 1], which is automatically updated in each iteration; t is the number of iterations; T is the maximum number of iterations; v represents the Cauchy mutation perturbation factor; and rand() represents a random number between [0, 1]. When |E|≥1, the search phase begins; when |E|<1, the development phase begins.

[0070] The development phase includes:

[0071] Determine the value of the random number r. If 0.5 ≤ |E| < 1 and r ≥ 0.5, adopt a soft encirclement strategy for position update.

[0072] X(t+1)=ΔX(t)-E|JX rabbit (t)-X(t)|

[0073] Where, ΔX(t)=X rabbit X(t) represents the difference between the prey's location and the individual's current location, and J is a random number between [0,2].

[0074] When |E| < 0.5 and e ≥ 0.5, a hard siege strategy is adopted for position update:

[0075] X(t+1)=X rabbit (t)-E|ΔX(t)|

[0076] When 0.5 ≤ |E| < 1 and r < 0.5, a soft encirclement strategy of asymptotic rapid dive is adopted for position update:

[0077]

[0078] Y = X rabbit (t)-E|JX rabbit (t)-X(t)|

[0079] Z = Y + S*LF(n)

[0080] Where f() is the fitness function, S is a 2-dimensional random vector with elements being random numbers between [0,1], n represents the dimension, and LF() is the mathematical expression for Lévy flight:

[0081]

[0082] Where u and μ represent random numbers between [0,1], β is a constant with a value of 1.5, and Υ(1+β) represents the gamma function;

[0083] When |E| < 0.5 and r < 0.5, a gradual, rapid dive hard encirclement strategy is adopted for position update:

[0084]

[0085] Y = X rabbit (t)-E|JX rabbit (t)-X m (t)|

[0086] Z = Y + S*LF(n)

[0087] Step 4: Calculate fitness. Calculate the individual fitness after the location update and compare it with the prey fitness value. If the individual fitness value after the location update is better than the prey fitness value, then the location of the individual with the better fitness value is taken as the new prey location.

[0088] Step 5: Repeat steps 3 and 4. When the algorithm reaches the maximum number of iterations, output the global optimal value of HHO and establish a classification model using the optimal parameters, penalty factor C and kernel radius g.

[0089] The method for remotely controlling a six-degree-of-freedom robotic arm using electromyography (EMG) signals includes: sampling EMG signals from the arm surface using an electrode sleeve worn by the user; performing preliminary processing on the EMG signals, including filtering and noise reduction, and extracting features from the preprocessed EMG signals; the user performing arm movements according to corresponding action commands, and a classification model forming a classification model based on the user's EMG signal features; feeding the extracted features into the trained classification model for signal classification; and sending action commands to the robotic arm based on the signal classification results to complete one movement control operation.

[0090] This invention addresses the problem of limited application scenarios and efficiency in existing surface electromyography (EMG) signal-controlled robotic arms, which suffers from low degrees of freedom. Furthermore, in EMG signal feature extraction, this invention employs a multi-scale permutation entropy feature extraction method. Compared to existing wavelet transform algorithms, this method offers advantages in handling feature information from multiple time series. In signal classification, a support vector machine classification model optimized using the Harris Hawk algorithm is preferred. The population is initialized using a set of optimal points, avoiding uneven spatial distribution during random population generation, improving the ergodicity of the initial population, and eliminating uncertainties introduced by random initialization. An information-sharing update method is added to the position update process, and a Cauchy mutation perturbation factor is introduced when calculating escape energy, effectively preventing the HHO algorithm from getting trapped in local optima. Attached Figure Description

[0091] Figure 1 This is a flowchart of the signal processing involved in the control system of this invention;

[0092] Figure 2 This is a flowchart for obtaining entropy across multiple scales;

[0093] Figure 3 This is the flowchart of the Harris Eagle algorithm;

[0094] Figure 4 This is a block diagram of the multi-degree-of-freedom control system based on minute surface electrical signals of the present invention;

[0095] Figure 5 This is a comparison between randomly generating 100 positions and generating them using a set of optimal points (the right figure shows the generation using a set of optimal points).

[0096] Figure 6 This is the schematic diagram of a filter circuit. Detailed Implementation

[0097] The present invention will now be further described with reference to preferred embodiments and accompanying drawings.

[0098] Reference Figure 4 This embodiment discloses a multi-degree-of-freedom control system based on surface micro-electrical signals, comprising the following modules:

[0099] Electromyography (EMG) signal acquisition module: used for acquiring surface EMG signals;

[0100] Signal preprocessing module: preprocesses the acquired electromyographic signals;

[0101] Feature extraction module: For the preprocessed signal, the particle swarm optimization algorithm is used to calculate the optimal parameters of the multi-scale permutation entropy; based on the optimal parameters, the multi-scale permutation entropy features of the preprocessed electromyography signal are extracted to obtain the feature data of the electromyography signal;

[0102] Signal classification module: Classifies electromyographic signals based on their feature dataset;

[0103] Feature matching module: Matches the classified signals with preset actions.

[0104] like Figure 1-4 As shown, the following is a more detailed explanation of the use of the above control system and the signal processing process. The specific steps are as follows:

[0105] Step 1, surface electromyography signal sampling;

[0106] The user correctly puts the sampling sleeve on their arm, and the FDC flexible electrode array contacts the user's skin. The sampling sleeve, equipped with an electromyography (EMG) signal acquisition module, extracts EMG signals from the skin surface and sends them to the processing center for further processing through the acquisition channel.

[0107] Step 2, signal data preprocessing;

[0108] The signal preprocessing module in the processing center first uses a sliding window analysis method to segment the raw surface electromyography (EMG) signal using an incremental function and obtains feature vectors through the acquisition channel. The window length is 300ms and the increment interval is 50ms. Afterwards, filtering and noise reduction are performed, mainly filtering out various noises including power frequency interference and motion artifacts. Filtering is performed simultaneously in hardware and software. The hardware filtering uses a passive low-pass filter constructed with capacitors and resistors, while the software uses an IIR filter. The high-pass filter is designed to allow signals above 7Hz to pass through, with a cutoff frequency upper limit of 250Hz. The specific circuit principle is as follows... Figure 6 As shown.

[0109] Step 3, Feature Extraction;

[0110] The feature extraction module uses the particle swarm optimization (PSO) algorithm to calculate the optimal parameters of the multi-scale permutation entropy (MPE). Based on the optimal parameters of the MPE, multi-scale permutation entropy features are extracted from the preprocessed electromyography (EMG) signals to obtain the feature dataset of the EMG signals.

[0111] Step 4: Signal classification and gesture pairing;

[0112] The signal classification module uses the Harris Hawk algorithm (HHO) to optimize the support vector machine (SVM) classification model, and optimizes the penalty factor and kernel function parameters in the SVM to construct the classification model.

[0113] Step 5: After sorting, the signal controls the movement of the robotic arm;

[0114] The feature matching module matches the classified signals with the pre-set actions and sends the information to the main control chip of the robotic arm via the Bluetooth module. The robotic arm will then be driven by six digital servos to perform the corresponding actions.

[0115] Step 6: Connect to the host computer and observe the signal;

[0116] The Bluetooth module of the electromyography (EMG) acquisition sleeve can connect to the host computer via Bluetooth. After the user wears the sleeve, the power values ​​and other values ​​of the acquired EMG signals can be reflected in the host computer application, which is convenient for inspection and debugging.

[0117] like Figure 5 As shown, using a set of optimal points to initialize the population can avoid uneven distribution of the population across the entire space when it is randomly generated, improve the ergodicity of the initial population, and eliminate the uncertainties brought about by random initialization.

[0118] This invention discloses a technology for controlling a robotic arm using surface electromyography (EMG) signals. One usage process of this invention is as follows: the user wears an EMG acquisition sleeve, turns on the power of the EMG acquisition processor and the robotic arm, waits for the EMG processor and the robotic arm processor to connect via Bluetooth, the user makes an arm posture, the sleeve collects surface EMG signals through electrodes, the signals are preprocessed including filtering and signal selection, the preprocessed signals are subjected to feature extraction, signal classification and feature matching, and finally, commands are sent via Bluetooth to make the robotic arm perform actions.

[0119] The robotic arm structure involved in this invention is existing technology.

[0120] This invention is not limited to the specific embodiments described above. Those skilled in the art can make various adjustments or modifications within the scope of the claims, which do not affect the substantive content of this invention.

Claims

1. A multi-degree-of-freedom control system based on surface micro-electrical signals, characterized in that, Includes the following modules: Electromyography (EMG) signal acquisition module: used for acquiring surface EMG signals; Signal preprocessing module: preprocesses the acquired electromyographic signals; Feature extraction module: For the preprocessed signal, the particle swarm optimization algorithm is used to calculate the optimal parameters of the multi-scale permutation entropy; based on the optimal parameters, the multi-scale permutation entropy features of the preprocessed electromyography signal are extracted to obtain the feature data of the electromyography signal; Signal classification module: Classifies electromyographic signals based on their feature dataset; Feature matching module: Matches the classified signals with preset actions; In the feature extraction module, the particle swarm optimization algorithm calculates the optimal parameters for the multi-scale permutation entropy, as follows: The ranges of each parameter of the multi-scale permutation entropy are defined; the ranges of each parameter of the multi-scale permutation entropy include: embedding dimension m, with a value range of m=2~7; delay time t, with a value range of t=1~5; and scale factor s, with a value range of s=1~20. The objective function is the square function of the entropy skewness of the multi-scale arrangement, and its minimum value is sought. The entropy skewness of the multi-scale arrangement is expressed as: Among them, H p (X) is a time series The sequence of permutation entropy at all scales is represented as follows: and For sequence H p The mean of (X); For sequence H p The standard deviation of (X); The expectation of the sequence; Set the objective function of PSO as follows: The optimal parameters for the multi-scale permutation entropy are obtained through computation. In the feature extraction module, based on the optimal parameters, multi-scale permutation entropy feature extraction is performed on the preprocessed electromyography (EMG) signal to obtain the feature data of the EMG signal, as follows: Time series coarsening: For a time series of length N Coarse-graining was performed to obtain the coarse-grained sequence: Where s is the scaling factor, and s takes positive integer values; [N / s] represents rounding down N / s, and when s=1, the coarse-grained sequence is the original sequence; Phase space reconstruction for each coarse-grained sequence: Where m is the embedding dimension, As a delay factor, for the time-reconstructed sequence Sort in ascending order: Simplified as a sequence of symbols: There are m! possible permutations. Calculate the probability of each permutation type: Where, N l Let n represent the number of occurrences of each permutation, and n represent the dimension. Based on this, the permutation entropy of the time series at multiple scales can be calculated as follows: in, hour, Reaching the maximum value ln(m!), the multi-scale permutation entropy is... Normalization is performed: in, The normalized permutation entropy value is the characteristic data of the electromyographic signal.

2. The multi-degree-of-freedom control system based on surface micro-electrical signals according to claim 1, characterized in that: In the signal preprocessing module, the preprocessing of electromyographic signals includes: using the sliding window analysis method to detect active segments of the electromyographic signals, with a window length of 300ms and an increment interval of 50ms.

3. The multi-degree-of-freedom control system based on surface micro-electrical signals according to claim 2, characterized in that: In the signal preprocessing module, the preprocessing of electromyographic signals also includes: firstly, filtering is performed by combining a passive low-pass filter and an infinite impulse response filter; then, filtering is performed by a high-pass filter, which is set to 7Hz~250Hz.

4. The multi-degree-of-freedom control system based on surface micro-electrical signals according to claim 1, characterized in that: The signal classification module uses a Harris Eagle optimized support vector machine to classify features in the feature dataset.

5. The multi-degree-of-freedom control system based on surface micro-electrical signals according to claim 4, characterized in that: In the signal classification module, an optimized Harris Eagle-optimized support vector machine is used to classify the features in the feature dataset, as detailed below: Step 1: Population initialization. The population is initialized using optimal point set optimization. The upper bound (ub) and lower bound (lb) of the search space are determined. The problem dimension (r) is 2. The position of each individual is initialized: Among them, X i Let represent the position of the i-th individual in the population, where i ranges from [1, N], N is the size of the Harris Eagle population, r is the optimal point, and p is the smallest prime number that satisfies the conditions. Step 2: Calculate the initial fitness and set the position of the individual with the best fitness as the current prey position; Fitness calculation method: Among them, R i Recall evaluates how many positive samples are ultimately predicted successfully out of all positive samples; P i The precision evaluation represents the number of true positive samples among those predicted as positive; TP indicates that a positive sample was successfully predicted as positive, TN indicates that a negative sample was successfully predicted as negative, FP indicates that a negative sample was incorrectly predicted as positive, FN indicates that a positive sample was incorrectly predicted as negative, r represents the number of dimensions of the classification target, and f(X) represents the number of dimensions of the classification target. i The value X represents the fitness score. The fitness scores are compared, and the individual with the lower fitness score is selected. i Position X as prey rabbit , that is, X rabbit It is all individuals X i The one with the lowest fitness; Step 3: Position update. First, update the prey's escape energy, and then execute the corresponding position update strategy in the search or exploit behavior based on the escape energy and the generated random number. The strategy for updating the position is determined based on the magnitude of the random value q: Where X(t) and X(t+1) are the positions of the individual in the current and next iterations, respectively, t is the iteration number, and X... rand (t) represents the randomly selected individual position, X rabbit (t) represents the prey position, i.e., the position of the individual with the optimal fitness, and is a random number between [0,1]. Specifically, information is obtained from the shared region when q is greater than or equal to 0.5; otherwise, information is obtained from the collaborators. The population average position is represented as: Based on the transformation of escape energy between search and different exploitation behaviors, the optimized perturbation escape energy is defined as: Where E0 is the initial energy of the prey, a random number between [-1, 1], automatically updated in each iteration, t is the iteration number, T is the maximum iteration number, v represents the Cauchy mutation perturbation factor, and rand() represents a random number between [0, 1]. Then the search phase begins. It entered the development stage at that time; The development phase includes: Determine the size of the random number r, when At that time, a soft encirclement strategy was adopted to update the position: in, J represents the difference between the prey's location and the individual's current location, where J is a random number between [0,2]. when At that time, a hard-line siege strategy was adopted to update the position: when At that time, a soft encirclement strategy of gradual, rapid dive is adopted for position updates: Where f() is the fitness function, S is a 2-dimensional random vector with elements being random numbers between [0,1], and LF() is the mathematical expression for Levi's flight: Where u and Represents a random number between [0, 1]. It is a constant, taking a value of 1.

5. Represented as the gamma function; when At that time, a gradual, rapid dive-and-surround hard encirclement strategy is adopted for position updates: Step 4: Calculate fitness. Calculate the individual fitness after the location update and compare it with the prey fitness value. If the individual fitness value after the location update is better than the prey fitness value, then the location of the individual with the better fitness value is taken as the new prey location. Step 5: Repeat steps 3 and 4. When the algorithm reaches the maximum number of iterations, output the global optimal value of HHO and establish a classification model using the optimal parameters, penalty factor C and kernel radius g.

Images