Geometric feature-based sEMG signal muscle force analysis method under polar coordinates

Abstract

The invention discloses an sEMG signal muscle force analysis method based on geometric features under polar coordinates, which comprises the following steps: synchronously acquiring sEMG signals by adopting multiple channels based on the corresponding relationship between finger flexion and extension actions and forearm muscle groups, and marking sEMG channels corresponding to different fingers; performing data preprocessing on the marked signals; performing coordinate transformation on the preprocessed signal to obtain a polar coordinate graph of the sEMG signal; extracting geometric feature parameters according to the polar coordinate graph, extracting polar coordinate features of each channel, and constructing a polar coordinate space cooperation characterization mechanism for representing a muscle cooperation strategy, movement mode classification and an abnormal state; and establishing a corresponding model of muscle force-geometric polar coordinate characteristics, and realizing real-time estimation and motion process judgment of the multi-finger cooperative motion sEMG. According to the invention, the sensitivity to gripping force change is greatly improved, and the anti-interference performance is obviously enhanced. The problem that the motion section is difficult to analyze is solved, and the classification accuracy is remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the field of biomedical signal processing, and particularly relates to a method for analyzing the muscle strength of sEMG signals based on geometric features in polar coordinates. Background Art

[0002] The rehabilitation robot technology based on surface electromyogram (sEMG) signals is a cutting-edge direction in the fields of robotics and rehabilitation, and has great advantages in improving the rehabilitation training effect. Compared with the rehabilitation training that relies on rehabilitation physicians to manually assist patients, rehabilitation robots can perform high-intensity and precise rehabilitation training for a long time, enhance the rehabilitation training effect and reduce the work intensity of physicians.

[0003] The generation of sEMG signals will occur prior to limb movement, and the human motion intention can be obtained for gesture prediction and recognition. Currently, the common methods for detecting muscle strength and muscle fatigue are to evaluate the motion of sEMG signals through sample entropy (SampEn), median frequency (MF), etc. Although these indicators can provide information about the intensity and frequency of muscle activity, they may not be sensitive enough to capture the subtle changes in muscle fatigue and muscle strength. In subsequent research, methods using deep learning were proposed for identification and prediction, but these methods are all based on time-domain features or frequency-domain features, and there is a problem that the feature values change not significantly, resulting in a low recognition rate. Therefore, it is urgent to study an efficient and reliable feature extraction method to improve the analysis of sEMG signals and the effect of reflecting muscle strength. This method should be able to effectively overcome the limitations of traditional methods and play a greater role in electromyogram signal processing. Summary of the Invention

[0004] Object of the Invention: Aiming at the above problems, the present invention proposes a method for analyzing the muscle strength of sEMG signals based on geometric features in polar coordinates, and constructs a high-precision dynamic muscle strength estimation model through the polar coordinate space mapping and geometric topological feature quantization of multi-channel sEMG signals.

[0005] Technical Solution: To achieve the object of the present invention, the technical solution adopted by the present invention is: A method for analyzing the muscle strength of sEMG signals based on geometric features in polar coordinates, including the following steps:

[0006] Based on the anatomical correspondence between finger flexion and extension movements and forearm muscle groups, multi-channel synchronous acquisition of sEMG signals is performed, and the sEMG channels corresponding to different fingers are marked;

[0007] The marked sEMG signals are preprocessed using a filter; the preprocessed sEMG signals are subjected to coordinate transformation using a signal processing algorithm to obtain the polar coordinate diagram of the sEMG signals;

[0008] Extract geometric feature parameters according to the outermost closed contour of the polar plot and construct a multi-dimensional feature vector;

[0009] Extract the polar coordinate feature parameters of each channel, construct a polar coordinate space collaborative representation mechanism to represent muscle coordination strategies, movement pattern classification and abnormal states, and provide parameters for rehabilitation evaluation and movement analysis;

[0010] Establish a non-linear mapping model of muscle strength-geometric polar coordinate features to realize real-time estimation of multi-finger coordinated movement sEMG and determination of the movement process.

[0011] Further, based on the anatomical correspondence between finger flexion and extension movements and forearm muscle groups, collect sEMG signals and perform labeling, including the following steps:

[0012] Build a finger-muscle model between the finger and the corresponding forearm muscle; based on the built model, place electromyography sensors at the corresponding forearm muscles;

[0013] Collect grip strength through a digital display hand dynamometer, collect sEMG signals using electromyography sensors under different grip strength conditions; based on the finger movement process, label the collected sEMG signals.

[0014] Further, the corresponding models built between the finger and the forearm muscle controlling the finger movement include: sEMG1, sEMG2, sEMG3, sEMG4, sEMG5;

[0015] Among them, sEMG1 represents the corresponding relationship built between the thumb and the abductor pollicis brevis and / or extensor pollicis longus, sEMG2 represents the corresponding relationship built between the index finger and the extensor digitorum, sEMG3 represents the corresponding relationship built between the middle finger and the extensor carpi ulnaris, sEMG4 represents the corresponding relationship built between the ring finger and the extensor carpi radialis brevis, and sEMG5 represents the corresponding relationship built between the little finger and the extensor digiti minimi;

[0016] Place the five electromyography sensors at the abductor pollicis brevis and / or extensor pollicis longus, extensor digitorum, extensor carpi ulnaris, extensor carpi radialis brevis and extensor digiti minimi of the forearm respectively, and collect sEMG signals at the five locations, denoted as sEMG1, sEMG2, sEMG3, sEMG4, sEMG5, which are used to control the thumb, index finger, middle finger, ring finger and little finger respectively.

[0017] Further, during the process of labeling the collected sEMG signals, define different finger states, including: initial state, bending state and recovery state;

[0018] The initial state represents the natural relaxation state of the finger maintained within the first 5 seconds during a round of sEMG signal acquisition; the bending state represents the bending state of the finger maintained for 10 seconds after 5 seconds during a round of sEMG signal acquisition; the recovery state represents the state where the finger returns to the relaxed state after ending the bending 15 seconds after the start of a round of sEMG signal acquisition.

[0019] Further, a notch filter is used to filter out the power frequency interference from the marked sEMG signal;

[0020] A band - pass filter is used to optimize the band - pass filtering of the processed sEMG signal to obtain the time - frequency relationship diagram of the sEMG signal.

[0021] Further, a signal processing algorithm is used to perform coordinate transformation on the pre - processed sEMG signal, and the steps are as follows:

[0022] The sEMG signal is subjected to FFT transformation to obtain the spectrum. The transformation formula of FFT is as follows:

[0023]

[0024] Among them, X[k] is the frequency component of the signal in the frequency domain, x[n] is the nth sampling point of the time - domain signal, N is the number of sampling points of the signal, k is the frequency index, k = 0, 1, …, N - 1, j is the imaginary unit, and e -jθ represents the exponential form of a complex number;

[0025] Calculate the real part a[k] and the imaginary part b[k], and the formulas are as follows:

[0026]

[0027] Calculate the amplitude |x(k)| and the argument θ(k) according to the real part and the imaginary part, and the formulas are as follows:

[0028]

[0029] Plot the amplitude and the argument in the polar coordinate graph to obtain the polar coordinate graph of the sEMG signal.

[0030] Further, according to the polar coordinate graph of the sEMG signal, extract the geometric features of the polar coordinates of the sEMG signal, including the perimeter, the area, and the maximum polar radius value. The steps are as follows:

[0031] Calculate the total length of the path described by the sEMG in the polar coordinate space, that is, the perimeter, which characterizes the spatial extensibility of the activation range. The calculation formula of the perimeter P is as follows:

[0032]

[0033] Among them, r i and r i+1are the amplitudes at the i-th and (i + 1)-th moments respectively, and θ i and θ i+1 are the angles at the i-th and (i + 1)-th moments respectively, and N is the number of sampling points of the signal;

[0034] By dividing the sEMG into several small sectors, the area of the sEMG in the polar coordinate space is calculated to characterize the overall activation intensity. The formula is as follows:

[0035] The area A of each small sector i :

[0036]

[0037] The total area A is the sum of the areas of all small sectors:

[0038]

[0039] Calculate the maximum polar radius value of the sEMG polar diagram to characterize the instantaneous maximum activation level in a specific action. The formula is as follows:

[0040] r max = max{|x(k)|},

[0041] where X[k] is the frequency component of the signal in the frequency domain, and |x(k)| is the signal amplitude.

[0042] Furthermore, according to the polar diagram of the sEMG signal, extract the polar coordinate features of the sEMG signal, including the mean value of the polar radius, the standard deviation of the angle, and the area entropy. The steps are as follows:

[0043] Calculate the mean value of the polar radius to characterize the comprehensive energy output during muscle contraction. The formula is as follows:

[0044]

[0045] where M is the number of channels, x m is the amplitude of the sEMG signal, and k is the k-th channel;

[0046] Calculate the standard deviation of the angle σ θ , to characterize the muscle coordination stability. The formula is as follows:

[0047]

[0048] where M is the number of channels, θ i is the phase angle corresponding to the i-th moment, is the mean value;

[0049] Calculate the area entropy H A to reflect the complexity of the motion pattern. The formula is as follows:

[0050]

[0051] Among them, P(A) is the probability density of the polar coordinate area, A is the polar coordinate area, and r i is the amplitude at the i-th moment, and θ i and θ i+1 are the angles at the i-th and (i + 1)-th moments respectively, and N is the number of sampling points of the signal.

[0052] Furthermore, a corresponding model of muscle strength - polar coordinate geometric features is established, and the model is used to identify and determine motion signals, including the following steps:

[0053] Extract the polar coordinate geometric features and polar coordinate features extracted from sEMG signals under different muscle strength states as the input of the SVR model; label the corresponding muscle strength for each sEMG signal sample as the output label of the SVR model;

[0054] Divide the labeled data set into a training set and a test set; train the model using support vector regression; use the mean square error MSE and the coefficient of determination R 2 to evaluate the fitting effect of the model.

[0055] Beneficial effects: Compared with the prior art, the technical solution of the present invention has the following beneficial technical effects:

[0056] 1. Polar coordinate space collaborative representation mechanism: Breaking through the limitations of traditional time - frequency domain analysis, through the mapping of sEMG information and polar coordinate features, the sEMG signal is mapped on the polar coordinate distribution map, realizing the visualization of the spatial collaborative activation pattern of muscle groups.

[0057] 2. Geometric topology feature system: Innovatively propose a geometric feature set centered on perimeter, area, and maximum polar diameter value, and analyze the evolution law of muscle activation regions from the perspective of topology. Experiments show that its sensitivity to grip force changes is greatly improved compared with traditional root mean square (RMS) features, and the anti - interference ability is significantly enhanced.

[0058] 3. Dynamic transfer learning model: Design an incremental SVR algorithm, combine polar coordinate features and geometric features, realize the analysis of the start, end of the motion segment and the degree of force application within the motion segment, and solve the difficult problem of analyzing the motion segment of an action. SVR can efficiently handle nonlinear problems, adapt to high - dimensional features, and can process small - sample data.

[0059] 4. Multimodal Fusion Architecture: By integrating the polar coordinate geometric features of surface electromyography (sEMG) and applying Kalman filtering technology to fuse spatio-temporal information, when dealing with complex grasping tasks, the present invention achieves a judgment accuracy of up to 98.89% and 99.7% for the start and end of movement respectively, significantly exceeding existing methods. In addition, in terms of evaluating the force exerted during movement, such as distinguishing low intensity, medium intensity, and high intensity, the classification accuracies of the present invention reach 91.2%, 95.3%, and 96.7% respectively, showing a significant improvement compared to the accuracy of a single signal modality. Description of the Drawings

[0060] Figure 1 This is a flowchart of the method of the present invention.

[0061] Figure 2 These are the muscle blocks corresponding to the placement of five electromyography sensors.

[0062] Figure 3 For a force of 5 kg

[0063] Figure 4 For a force of 10 kg

[0064] Figure 5 For the filtered 5 kg force

[0065] Figure 6 For the filtered 10 kg force

[0066] Figure 7 For a force of 5 kg Polar coordinate diagram of the signal every 0.5 seconds.

[0067] Figure 8 These are the outermost circles and the r values of each outer circle of each polar coordinate diagram for 5 kg.

[0068] Figure 9 For a force of 10 kg Polar coordinate diagram of the signal every 0.5 seconds.

[0069] Figure 10 These are the outermost circles and the r values of each outer circle of each polar coordinate diagram for 10 kg.

[0070] Figure 11 These are the area values for a force of 5 kg.

[0071] Figure 12 These are the area values for a force of 10 kg.

[0072] Figure 13 These are the results of the support vector regression model test set. Detailed implementation mode

[0073] The technical solution of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0074] Aiming at the technical bottleneck that traditional sEMG muscle strength analysis relies on time-frequency domain features and is difficult to characterize the spatial distribution of muscle co-activation and is easily interfered by individual differences, the present invention constructs a high-precision muscle strength dynamic estimation model through polar coordinate space mapping and geometric topological feature quantization of multi-channel sEMG signals. Its core process is as follows: First, based on the anatomical correspondence between finger flexion and extension movements and forearm muscle groups (forearm abductor pollicis brevis or extensor pollicis longus, extensor digitorum, flexor carpi radialis, flexor carpi ulnaris, flexor digitorum superficialis), multi-channel synchronous acquisition of sEMG signals is used, and the sEMG channels corresponding to different fingers are marked to achieve precise alignment of electromyographic signals and actions; secondly, power frequency noise and motion artifacts are eliminated through band-pass filtering (20 - 150 Hz) and 50 Hz notch filtering, the independent component analysis (ICA) is combined to separate the electromyographic signal sources, and the individual amplitude differences are eliminated by using sliding window normalization (window length 0.5 s). The signal points of each window of the filtered sEMG are mapped into amplitude-phase parameters in the polar coordinate system (amplitude: Phase: ). Then, geometric feature parameters are extracted according to the outermost closed contour of the polar coordinate diagram, including perimeter (reflecting the spatial extensibility of the activation range), area (characterizing the overall activation intensity), and maximum polar diameter value (instantaneous maximum activation level in a specific action) to construct a multi-dimensional feature vector. Subsequently, the polar coordinate feature parameters of each channel are extracted, the mean value of the polar radius ( reflecting the overall activation intensity), the standard deviation of the angle ( characterizing the muscle co-activation stability), and the area entropy (H A =-∑P(A)logP(A), reflecting the complexity of the motion pattern) are used to construct a polar coordinate space co-activation representation mechanism to represent muscle co-activation strategies, motion pattern classification, and abnormal states, providing key parameters for rehabilitation evaluation and motion analysis. Finally, a non-linear mapping model of muscle strength - geometric polar coordinate features is established based on the support vector regression (SVR) algorithm, the individual physiological differences are dynamically calibrated through a transfer learning framework, and the temporal consistency is optimized by using the Kalman filter to realize the real-time estimation of multi-finger co-activation sEMG and the determination of the motion process.

[0075] An embodiment of the present invention provides a recognition method for sEMG signal muscle strength analysis based on geometric features in polar coordinates, as Figure 1 shown, including the following steps:

[0076] S1. Based on the correspondence between fingers and forearm muscle blocks, collect sEMG signals and mark them;

[0077] S2. Use a filter and a signal processing algorithm to preprocess the marked sEMG signal;

[0078] S3. Perform coordinate transformation on the preprocessed sEMG signal to obtain the polar coordinate graph of the sEMG signal;

[0079] S4. Analyze the geometric features (perimeter, area, maximum polar radius value) of the polar coordinates of the sEMG signal;

[0080] S5. Analyze the polar coordinate features (average polar radius, standard deviation of angle, area entropy) of the sEMG signal;

[0081] S6. Establish a corresponding model between muscle strength and geometric features, and use the model to perform signal recognition and determination on the motion signal.

[0082] In one embodiment, based on the corresponding relationship between the fingers and the forearm muscles, collect the sEMG signal and perform marking, including the following steps:

[0083] S11. Build a finger-muscle model between the finger and the corresponding forearm muscle;

[0084] S12. Based on the built model, place the electromyography sensor at the corresponding forearm muscle;

[0085] S13. Based on the digital display dynamometer, use the electromyography sensor to collect the sEMG signal under different grip forces;

[0086] S14. Based on the finger movement process, mark the collected sEMG signal.

[0087] In one embodiment, the corresponding model built between the finger and the forearm muscle that controls the finger movement includes: sEMG1, sEMG2, sEMG3, sEMG4, sEMG5;

[0088] Among them, sEMG1 represents the corresponding relationship built between the thumb and the abductor pollicis brevis and / or extensor pollicis longus, sEMG2 represents the corresponding relationship built between the index finger and the extensor digitorum, sEMG3 represents the corresponding relationship built between the middle finger and the extensor carpi ulnaris, sEMG4 represents the corresponding relationship built between the ring finger and the extensor carpi radialis brevis, and sEMG5 represents the corresponding relationship built between the little finger and the extensor digiti minimi.

[0089] Specifically, as Figure 2 shown, place the five electromyography sensors at the abductor pollicis brevis and / or extensor pollicis longus, extensor digitorum, extensor carpi ulnaris, extensor carpi radialis brevis, and extensor digiti minimi of the forearm respectively, collect the sEMG signals at the five places, denoted as sEMG1, sEMG2, sEMG3, sEMG4, sEMG5, which are used to control the thumb, index finger, middle finger, ring finger, and little finger respectively.

[0090] In one embodiment, different states of the finger include: an initial state, a bent state, and a restored state.

[0091] Specifically, the initial state represents the natural relaxation state of the finger maintained within the first 5 seconds during a round of sEMG signal acquisition; the bent state represents the bent state of the finger maintained for 10 seconds after 5 seconds during a round of sEMG signal acquisition; the restored state represents the state where the finger ends bending and returns to the relaxed state after 15 seconds during a round of sEMG signal acquisition.

[0092] Specifically, taking the fist clenching action as an example, the average value of the sEMG signals of five channels is taken as the signal of this action, that is, the average value of sEMG1, sEMG2, sEMG3, sEMG4, and sEMG5, denoted as of 5 kg force and 10 kg force respectively as Figure 3 and Figure 4 shown.

[0093] In one embodiment, using a filter and a signal processing algorithm to perform data preprocessing on the labeled sEMG signals to obtain a time-frequency relationship diagram of the sEMG signals, including the following steps;

[0094] S21. Use a notch filter to filter out the power frequency interference from the labeled sEMG signals;

[0095] S22. Use a band-pass filter to optimize the band-pass filtering of the processed sEMG signals;

[0096] Specifically, the filter includes a 50 Hz notch filter and a 20 Hz - 150 Hz Butterworth band-pass filter. Since the effective frequency range of the surface electromyogram signal is 20 Hz - 150 Hz, first use the 50 Hz notch filter to filter out the power frequency interference, and then use the Butterworth band-pass filter for band-pass filtering. The signals of 5 kg force and 10 kg force after filtering are as shown in Figure 5 and Figure 6 shown.

[0097] In one embodiment, based on the signal processing algorithm, perform coordinate transformation on the optimized sEMG signals to obtain a polar coordinate diagram of the sEMG signals, including the following steps:

[0098] S31. Perform FFT transformation on the preprocessed signals to obtain the spectrum. The transformation formula of FFT is as follows:

[0099]

[0100] Where: X[k] is the frequency component of the signal in the frequency domain, x[n] is the nth sampling point of the time-domain signal, N is the number of sampling points of the signal, k is the frequency index, k = 0, 1, …, N - 1, j is the imaginary unit, and e -jθ represents the exponential form of a complex number;

[0101] S32. Calculate the real part a[k] and the imaginary part b[k]. The formulas are as follows:

[0102] Real part:

[0103] Imaginary part:

[0104] S33. Calculate the amplitude and the argument based on the real part and the imaginary part. The calculation formulas are as follows:

[0105] Amplitude:

[0106] Argument:

[0107] S34. Plot the amplitude and the argument in a polar coordinate graph.

[0108] Specifically, the polar coordinates of the signal with a 5 kg force every 0.5 seconds are as Figure 7 shown. The outermost circle and the r values of each outer circle of each polar coordinate graph for 5 kg are as Figure 8 shown. The polar coordinates of the signal with a 10 kg force every 0.5 seconds are as Figure 9 shown. The outermost circle and the r values of each outer circle of each polar coordinate graph for 10 kg are as Figure 10 shown.

[0109] In one embodiment, analyze according to the polar coordinate graph of the sEMG signal, and extract the geometric features (perimeter, area, maximum polar radius value) of the polar coordinates of the sEMG signal, including the following steps;

[0110] S41. According to the polar coordinate graph of the sEMG signal, calculate the total length of the path depicted by the sEMG in the polar coordinate space, that is, the perimeter. The calculation formula for the perimeter P is as follows:

[0111]

[0112] Where: r i and r i+1 are the amplitudes at the i-th and (i + 1)-th moments respectively, θ i and θ i+1 are the angles at the i-th and (i + 1)-th moments respectively, and N is the number of sampling points of the signal;

[0113] S42. Approximately calculate the area of sEMG in the polar coordinate space by dividing sEMG into several small sectors. The calculation formula is as follows:

[0114] The area A of each small sector i :

[0115]

[0116] The total area A is the sum of the areas of all small sectors:

[0117]

[0118] S43. Calculate the maximum polar radius value of the sEMG polar diagram to reflect the instantaneous maximum activation level. The calculation formula is as follows:

[0119] r max = max{|x(k)|},

[0120] where X[k] is the frequency component of the signal in the frequency domain, and |x(k)| is the signal amplitude. Specifically, the geometric characteristic values (taking the area value as an example) of 5 kg force and 10 kg force are respectively as Figure 11 and Figure 12 shown.

[0121] In one embodiment, analyze according to the polar diagram of the sEMG signal, and extract the polar coordinate features (mean polar radius, angular standard deviation, area entropy) of the sEMG signal, including the following steps;

[0122] S51. Calculate the mean polar radius according to the polar diagram of the sEMG signal, which characterizes the comprehensive energy output during muscle contraction. The formula is as follows:

[0123]

[0124] where: M is the number of channels, x m is the sEMG signal amplitude, and k is the k-th channel;

[0125] S52. Calculate the angular standard deviation σ θ , which characterizes the muscle coordination stability. The formula is as follows:

[0126]

[0127] where: M is the number of channels, θ i is the phase angle corresponding to the i-th moment, is the mean value;

[0128] S53. Calculate the area entropy H AUsed to reflect the complexity of the motion pattern, the formula is as follows:

[0129]

[0130] Among them, P(A) is the probability density of the polar coordinate area, A is the polar coordinate area, r i is the amplitude at the i-th moment, θ i and θ i+1 are the angles at the i-th and (i + 1)-th moments respectively, and N is the number of sampling points of the signal.

[0131] Specifically, a high mean polar radius indicates a large muscle activation intensity, and a low mean polar radius reflects a muscle resting or low-intensity contraction state, which can be used to evaluate the dominant muscle group and action intensity classification during the force generation phase. A high angle standard deviation indicates poor muscle coordination stability, which is common in the late stage of fatigue or compensatory movements. A low angle standard deviation indicates a stable muscle coordination pattern, which can be used to evaluate the rehabilitation training effect. A high entropy value indicates a complex action pattern or the muscle is in the initial stage of fatigue. A low entropy value indicates strong action repeatability, a simple action pattern or the muscle in the late stage of fatigue, which can be used for motion pattern classification and fatigue warning. Specifically, the results of the five-channel sEMG polar coordinate feature analysis are shown in Table 1. High σ θ + high H A (such as channel 1): A complex and unstable activation pattern, which may be for multi-task switching or the initial stage of fatigue (the muscle tries various strategies to maintain the action). Low σ θ + low H A (such as channel 5): A stable and single activation, which may be for the efficient energy utilization stage or the late stage of fatigue (the muscle locks into a single contraction pattern). High σ θ + low H A : Sudden strong contraction (such as weightlifting), quickly relaxing after a short high-intensity activity.

[0132]

[0133] Table 1

[0134] In one embodiment, a corresponding model of muscle strength - polar coordinate geometric features is established, and the model is used to perform signal recognition and determination on the motion signal, including the following steps:

[0135] S61. By analyzing the sEMG signals under different muscle strength states, extract the corresponding polar coordinate geometric features (perimeter, area, maximum polar radius value, mean polar radius, angle standard deviation, area entropy), denoted as x1, x2,..., x n , as the input of the SVR model;

[0136] S62. Perform data annotation, and label the corresponding muscle strength intensity for each sEMG signal sample, which is divided into low intensity, medium intensity, and high intensity, as the output label of the SVR model;

[0137] S63. Dataset division, divide the labeled dataset into a training set and a test set, with a ratio of 70% training set and 30% test set;

[0138] S64. Regression model selection and training, use the support vector regression (SVR) training model, and its regression function is:

[0139]

[0140] x = {x1, x2,..., x n},

[0141] where: W is the weight vector, x is the feature vector, is the kernel function that maps the input features to a high-dimensional space, and b is the bias term;

[0142] S65. Model evaluation, use the mean squared error (MSE) and the coefficient of determination (R 2 ) to indicate the model fitting effect, and its calculation formula is as follows:

[0143]

[0144] where, y i is the true value of the i-th sample, is the predicted value of the i-th sample, n is the number of samples, is the mean of the true values.

[0145] Specifically, the prediction results of the support vector regression model test set are as Figure 13 shown. According to the confusion matrix, it can be seen that the classification accuracies of the start of movement, end of movement, low-intensity force, medium-intensity force, and high-intensity force are as high as 98.89%, 99.7%, 91.2%, 95.3%, and 96.7%.

Claims

1. A method for analyzing the muscle strength of sEMG signals based on geometric features in polar coordinates, characterized in that, Including the following steps: Based on the anatomical correspondence between finger flexion and extension movements and forearm muscle groups, multi-channel synchronous acquisition of sEMG signals is performed, and the sEMG channels corresponding to different fingers are marked; The marked sEMG signals are preprocessed using a filter; the signal processing algorithm is used to perform coordinate transformation on the preprocessed sEMG signals to obtain the polar coordinate diagram of the sEMG signals; Geometric feature parameters are extracted according to the outermost closed contour of the polar coordinate diagram, and a multi-dimensional feature vector is constructed; The polar coordinate feature parameters of each channel are extracted, and a polar coordinate space collaborative representation mechanism is constructed to represent muscle coordination strategies, movement pattern classification, and abnormal states, providing parameters for rehabilitation evaluation and motion analysis; A non-linear mapping model of muscle strength-geometric polar coordinates is established to realize real-time estimation of sEMG of multi-finger coordinated movement and determination of the movement process.

2. The method for analyzing muscle strength of sEMG signals based on geometric features in polar coordinates according to claim 1, wherein Based on the anatomical correspondence between finger flexion and extension movements and forearm muscle groups, sEMG signals are collected and marked, including the following steps: A finger-muscle model is built between the finger and the corresponding forearm muscle; based on the built model, electromyography sensors are placed at the corresponding forearm muscles; Grip strength is collected through a digital display dynamometer, and sEMG signals are collected using the electromyography sensors under different grip strength conditions; based on the finger movement process, the collected sEMG signals are marked.

3. The method for analyzing muscle strength of sEMG signals based on geometric features in polar coordinates according to claim 2, characterized in that, The corresponding model built between the finger and the forearm muscle that controls finger movement includes: sEMG1, sEMG2, sEMG3, sEMG4, sEMG5; Among them, sEMG1 represents the corresponding relationship built between the thumb and the abductor pollicis brevis and / or extensor pollicis longus, sEMG2 represents the corresponding relationship built between the index finger and the extensor digitorum, sEMG3 represents the corresponding relationship built between the middle finger and the extensor carpi ulnaris, sEMG4 represents the corresponding relationship built between the ring finger and the extensor carpi radialis brevis, and sEMG5 represents the corresponding relationship built between the little finger and the extensor digiti minimi; Five electromyography sensors are respectively placed at the abductor pollicis brevis and / or extensor pollicis longus, extensor digitorum, extensor carpi ulnaris, extensor carpi radialis brevis, and extensor digiti minimi of the forearm to collect sEMG signals at the five locations, denoted as sEMG1, sEMG2, sEMG3, sEMG4, sEMG5, which are respectively used to control the thumb, index finger, middle finger, ring finger, and little finger.

4. The method for analyzing muscle strength of sEMG signals based on geometric features in polar coordinates according to claim 3, characterized in that During the process of marking the collected sEMG signals, different finger states are defined, including: initial state, bending state, and recovery state; The initial state represents the natural relaxation state maintained by the finger within the first 5 seconds during a round of sEMG signal acquisition; the bending state represents the bending state maintained by the finger for 10 seconds after 5 seconds during a round of sEMG signal acquisition; the recovery state represents the state where the finger ends bending and returns to the relaxed state after 15 seconds during a round of sEMG signal acquisition.

5. The method for sEMG signal muscle strength analysis based on geometric features in polar coordinates according to any one of claims 1-4, characterized in that, The marked sEMG signals are processed to filter out power frequency interference using a notch filter; The processed sEMG signals are optimized by band-pass filtering using a band-pass filter to obtain the time-frequency relationship diagram of the sEMG signals.

6. The method for analyzing the muscle strength of sEMG signals based on geometric features in polar coordinates according to claim 1, characterized in that The signal processing algorithm is used to perform coordinate transformation on the preprocessed sEMG signals, and the steps are as follows: The sEMG signal is subjected to FFT transformation to obtain the spectrum. The transformation formula of FFT is as follows: where X[k] is the frequency component of the signal in the frequency domain, x[n] is the nth sampling point of the time-domain signal, N is the number of sampling points of the signal, k is the frequency index, k = 0, 1, …, N - 1, j is the imaginary unit, and e -jθ represents the exponential form of a complex number; Calculate the real part a[k] and the imaginary part b[k]. The formula is as follows: Calculate the amplitude |x(k)| and the argument θ(k) according to the real part and the imaginary part. The formula is as follows: Plot the amplitude and the argument in the polar coordinate diagram to obtain the polar coordinate diagram of the sEMG signal.

7. The method for analyzing muscle strength of sEMG signals based on geometric features in polar coordinates according to claim 1, characterized in that According to the polar coordinate diagram of the sEMG signal, extract the geometric features of the polar coordinates of the sEMG signal, including the perimeter, area, and maximum polar radius value. The steps are as follows: Calculate the total length of the path described by the sEMG in the polar coordinate space, that is, the perimeter, which characterizes the spatial extensibility of the activation range. The calculation formula of the perimeter P is as follows: where r < and r <:1 are the amplitudes at the \(i\)-th and \((i + 1)\)-th instants respectively, and \(\theta < and \(\theta <:1 are the angles at the \(i\)-th and \((i + 1)\)-th instants respectively, and \(N\) is the number of sampling points of the signal; Calculate the area of the sEMG in the polar coordinate space by dividing the sEMG into several small sectors, which characterizes the overall activation intensity. The formula is as follows: The area A of each small sector < : The total area A is the sum of the areas of all small sectors: Calculate the maximum polar radius value of the sEMG polar coordinate diagram, which characterizes the instantaneous maximum activation level in a specific action. The formula is as follows: r max = max{|x(k)|}, Among them, X[k] is the frequency component of the signal in the frequency domain, and |x(k)| is the signal amplitude.

8. The method for analyzing muscle strength of sEMG signals based on geometric features in polar coordinates according to claim 1 or 6 or 7, characterized in that According to the polar coordinate diagram of the sEMG signal, extract the polar coordinate features of the sEMG signal, including the mean polar radius, the standard deviation of the angle, and the area entropy. The steps are as follows: Calculate the mean polar radius Characterize the comprehensive energy output during muscle contraction. The formula is as follows: where M is the number of channels, x A is the amplitude of the sEMG signal, and k is the k-th channel; Calculate the standard deviation of angles σ M , which represents the stability of muscle synergy. The formula is as follows: where M is the number of channels, and θ < is the phase angle corresponding to the i-th moment, is the mean value; Calculate the area entropy H P which is used to reflect the complexity of the motion pattern, and the formula is as follows: Among them, P(A) is the probability density of the polar coordinate area, A is the polar coordinate area, r < is the amplitude at the i-th moment, θ < and θ <:1 are the angles at the i-th and (i + 1)-th moments respectively, and N is the number of sampling points of the signal.

9. The method for analyzing the muscle strength of sEMG signals based on geometric features in polar coordinates according to claim 1, wherein Establish a corresponding model between muscle strength and polar coordinate geometric features, and use the model to perform signal recognition and determination on the motion signal. The steps include: Take the geometric features and polar coordinate features of the polar coordinates extracted from the sEMG signals under different muscle strength states as the input of the SVR model; label the corresponding muscle strength for each sEMG signal sample as the output label of the SVR model; Divide the labeled dataset into a training set and a test set; train a model using support vector regression; use the mean squared error MSE and the coefficient of determination R ( to evaluate the fitting effect of the model.

Images