Lower limb motion recognition method based on S-transformation energy concentration surface myoelectricity decoding

Abstract

A lower limb movement recognition method based on S-transformation energy concentration surface myoelectricity decoding comprises the steps that firstly, original myoelectricity signals are preprocessed, useful time periods are intercepted from the preprocessed signals through endpoint detection, then the signals in the time periods are subjected to segmented S transformation and energy concentration calculation, and the S transformation energy concentration surface myoelectricity decoding is obtained; adopting segmentation operation to extract signal features of a specified dimension; carrying out motion mode classification through an SVM, carrying out fusion analysis on multichannel signal features, and carrying out lower limb motion identification; segmented S transformation is carried out on the signals to keep time phase information of the signals, and meanwhile, the calculation process of energy concentration is optimized to improve the motion classification accuracy; an SVM multi-classifier is built, and the recognition accuracy of the lower limb motion mode is further improved through S-transformation energy concentration and feature fusion.

Description

Technical Field

[0001] The present invention relates to the technical field of human-computer interaction pattern recognition, and in particular to a method for lower limb movement recognition based on S-transform energy concentration surface electromyography decoding. Background Art

[0002] With the development of modern society, the number of lower limb amputees caused by factors such as population aging, war, disease, and traffic accidents continues to increase. Wearing suitable prostheses to compensate for the amputees' lack of motor function and enable them to take care of themselves and return to society is a key link in promoting harmonious social development and improving the living conditions of people with disabilities.

[0003] With the continuous development of human-computer interaction technology, the needs of amputees have shifted from bulky passive prostheses to intelligent prostheses, which can recognize human movement intentions, actively adjust output torque or joint angles, achieve a gait similar to that of healthy people, and are more comfortable to use.

[0004] Existing lower limb prostheses use mechanical sensing signals (Huang Pingao. Research on key technologies for sensing and identifying motion intentions of intelligent lower limb prostheses [D]. University of Chinese Academy of Sciences (Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences), 2020.), but the inherent time lag of mechanical signals causes their control process to lag behind human gait. Some prostheses are controlled by electromyographic signals, but the current lower limb electromyographic decoding mostly extracts simple time-frequency domain features of signals (Sánchez-Velasco LE, Arias-Montiel M, Guzmán-Ramírez E, Lugo-González EA low-cost EMG-controlled anthropomorphicrobotic hand for power and precision grasp. Biocybern Biomed Eng 2020; 40: 221–37.), which loses the phase information that is indispensable in motion pattern analysis, resulting in poor recognition of lower limb motion patterns of the human body and the inability to achieve natural interaction between patient intentions and prosthetic control. Therefore, it is necessary to explore methods to recognize human movement intentions through decoding of lower limb surface electromyographic signals, further improve the accuracy of lower limb movement intention recognition, and enrich the movement control modes of lower limb prostheses. Summary of the invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide a lower limb movement recognition method based on S transform energy concentration surface electromyography decoding, perform segmented S transform on the signal to retain its phase information, and optimize the energy concentration calculation process to improve the accuracy of movement classification; build an SVM multi-classifier, and further improve the accuracy of lower limb movement pattern recognition through S transform energy concentration and its feature fusion.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is:

[0007] A method for lower limb movement recognition based on S-transform energy concentration surface electromyography decoding, first preprocessing the original electromyographic signal, the preprocessed signal through endpoint detection to intercept the useful time period, then the signal in this time period is subjected to segmented S-transform and energy concentration calculation, and the segmentation operation is used to extract the signal features of the specified dimension; the motion pattern is classified by SVM, and the multi-channel signal features are fused and analyzed for lower limb movement recognition.

[0008] A method for lower limb movement recognition based on S-transform energy concentration surface electromyography decoding comprises the following steps:

[0009] 1) Signal preprocessing and active segment detection;

[0010] 2) Signal segmentation S transform and energy concentration calculation;

[0011] 3) Construction of support vector machine multi-classifier and correlation analysis of multi-channel electromyographic signals;

[0012] 4) Support vector machine motion classification based on multi-channel signal feature fusion analysis.

[0013] The signal preprocessing in step 1) includes: 1.1) 50 Hz notch filtering; 1.2) 30 Hz zero phase shift high pass filtering; a zero phase shift filter is used in the preprocessing process to retain the phase information of the original data sequence.

[0014] In step 1), the active segment detection adopts a double threshold detection method based on short-time energy and short-time variance. First, the initial short-time energy minimum threshold and the variance and low threshold that characterize the fluctuation are set. Both parameters are dynamically adjusted according to the actual situation. Then, the window length and window shift are defined to frame the signal.

[0015] Assume that the EMG signal of the nth frame is represented by x n (m), the frame length is N, calculate its short-time energy E n As shown in formula (1), calculate its variance and S n As shown in formula (2),

[0016]

[0017]

[0018] Where m is the sample index of the signal, indicating the sample point in each frame, and m is an integer from 0 to N-1, indicating the position of each sampling point in this frame of signal. Represents the mean of the signal.

[0019] The signal segmented S transform and energy concentration calculation in step 2) are specifically as follows: intercept the data length of an action cycle as x, set a sliding window with a window length and a step length, perform S transform on each segment of data, set the frequency interval, and obtain a time-frequency matrix; as shown in formula (3), the 1 / q root value of the absolute value of the signal after S transform is superimposed and summed in the time-frequency domain, and then the summation result is squared q-wise; by optimizing the width of the window function in the S transform, a higher energy concentration is obtained for the signal with rapidly changing frequency components, and its energy concentration calculation is shown in formula (4); the energy concentration measurement method is used as shown in formula (5), the entire time-frequency concentration is decomposed, and the window width at the discrete point is used to improve its time-frequency concentration;

[0020]

[0021]

[0022]

[0023] Where: CM - signal energy concentration; t - time; f - frequency; S x (t,f)——Signal S transformation matrix, F——discrete frequency range; CM1——time-frequency concentration, q——nonlinear parameter, controlling the degree of focus, the larger the q, the more emphasis is placed on the high-value area; CM2——normalized time-frequency concentration, ——S x2 The normalized form of (t,f), CM3 - generalized time-frequency concentration; α - low-order power parameter, usually emphasizing global distribution, β - high-order power parameter, usually emphasizing local concentration, ——It’s S. x3 (t,f) is in normalized form to ensure the scale consistency of time-frequency representation.

[0024] In step 3), the construction of the support vector machine multi-classifier is as follows: design a multi-classifier. For a given m classes, train a binary classifier for every two classes in the m classes, and the total number of binary classifiers is m(m-1) / 2; for a data that needs to be classified, it needs to be predicted by all classifiers, and the final class attribute is determined by voting.

[0025] The multi-channel EMG signal correlation analysis in step 3) is specifically as follows: by analyzing the correlation of EMG signals, exploring the difference in classification effects between single-channel signals and multi-channel signal fusion, and finding the best lower limb muscle combination to improve the recognition accuracy of human lower limb movements.

[0026] The support vector machine motion classification based on multi-channel signal feature fusion analysis in step 4) is specifically as follows: the similarities between the lower limb muscles of different subjects are different, and it is necessary to perform similarity calculation on the multi-channel signal features of the subjects, and select the channels with better classification effects in the muscle signals of the rectus femoris, vastus medialis, biceps femoris, semitendinosus, tibialis anterior, medial gastrocnemius, etc. for fusion; by comparing the classification results of single-channel signals, the multi-channel signal features of the rectus femoris, biceps femoris and medial gastrocnemius are fused and analyzed, the fusion coefficient is allocated by the control variable method, and then the features with different weights are input into the classifier for motion classification.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] The present invention analyzes the shortcomings of existing input features for motion recognition, and proposes a lower limb motion recognition method based on S-transform energy concentration surface electromyography decoding. The S-transform is a combination of wavelet transformation and short-time Fourier transform. It has the characteristics of Fourier transform and can provide different resolutions at different frequencies by relying on window functions. The present invention extracts the simple time-frequency domain features and S-transform energy concentration features of lower limb electromyography signals, and uses support vector machines for classification. On this basis, the similarity of the six-channel muscles of the lower limbs is further analyzed, and the accuracy of motion recognition is improved by multi-channel signal feature fusion. Compared with simple time-frequency domain statistical feature analysis, the present invention can effectively improve the classification accuracy, and the classification accuracy can be further improved by fusion of channel features such as rectus femoris, biceps femoris and medial gastrocnemius. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a flow chart of lower limb movement recognition according to an embodiment of the present invention.

[0030] Figure 2 It is a flow chart of signal activity segment detection according to an embodiment of the present invention.

[0031] Figure 3 It is the result of the signal segmentation S transformation in the embodiment of the present invention.

[0032] Figure 4 These are the energy concentration calculation results of the four movements in the embodiment of the present invention.

[0033] Figure 5 It is a flow chart of the SVM multi-classification process according to an embodiment of the present invention.

[0034] Figure 6 This is the simple time-frequency domain feature recognition result of the embodiment of the present invention.

[0035] Figure 7 This is the result of S transformation energy concentration feature recognition in an embodiment of the present invention.

[0036] Figure 81 is a comparison of the rectus femoris signal classification results of an embodiment of the present invention.

[0037] Fig. 9 This is a comparison of the multi-channel fusion and single-channel classification results of electromyographic signals in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The technical scheme of the present invention is further described and illustrated in detail below in conjunction with the embodiments and drawings. The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0039] Reference Figure 1 , a method for lower limb movement recognition based on S-transform energy concentration surface electromyography decoding, comprising the following steps:

[0040] 1) Signal preprocessing and active segment detection;

[0041] 2) Signal segmentation S transform and energy concentration calculation;

[0042] 3) Construction of support vector machine multi-classifier and correlation analysis of multi-channel electromyographic signals;

[0043] 4) Support vector machine motion classification based on multi-channel signal feature fusion analysis.

[0044] The signal preprocessing in step 1) includes: 1.1) 50Hz notch filtering to remove power frequency interference; 1.2) 30Hz zero-phase shift high-pass filtering to remove motion artifacts; In the preprocessing process, a zero-phase shift filter is used to retain the phase information of the original data sequence, compensating for the disadvantage of the original filter phase distortion.

[0045] Reference Figure 2 In step 1), the active segment detection adopts a double threshold detection method based on short-time energy and short-time variance. First, the initial short-time energy minimum threshold and the variance and low threshold representing the fluctuation are set. Both parameters are dynamically adjusted according to the actual situation. Then, the window length and window shift are defined to frame the signal.

[0046] Assume that the EMG signal of the nth frame is represented by x n (m), the frame length is N, calculate its short-time energy E n As shown in formula (1), calculate its variance and S n As shown in formula (2),

[0047]

[0048]

[0049] Where m is the sample index of the signal, indicating the sample point in each frame, and m is an integer from 0 to N-1, indicating the position of each sampling point in this frame of signal. represents the mean value of the signal;

[0050] Reference Figure 3 , the signal segmentation S transform and energy concentration calculation in step 2) are specifically as follows: intercept the data length of an action cycle as x, set a sliding window with a window length (x / 25) and a step length (x / 25), perform S transform on each segment of data, set the frequency interval to 2Hz, and obtain a time-frequency matrix of size 16*x / 25; the signal segmentation S transform result, the darker the blue part, the lower its time-frequency energy, when the color gradually changes to yellow, it indicates that its time-frequency energy is higher; it can be seen that its energy is mainly concentrated between 100-300Hz; after S transform, the frequency, phase and energy amplitude of the surface electromyographic signal of the corresponding action are displayed in the three-dimensional graph;

[0051] As shown in formula (3), the 1 / q root values ​​of the absolute value of the signal after S transformation are superimposed and summed in the time-frequency domain, and then the summation result is squared and q-operated; by optimizing the width of the window function in the S transformation, a higher energy concentration can be obtained for the signal with rapidly changing frequency components, and its energy concentration calculation is shown in formula (4); the energy concentration measurement method is used as shown in formula (5), and the entire time-frequency concentration is decomposed, and the window width at the discrete points is fully utilized to improve its time-frequency concentration;

[0052]

[0053]

[0054]

[0055] Where: CM - signal energy concentration; t - time; f - frequency; S x (t,f)——Signal S transformation matrix; F——Discrete frequency range; CM1——Time-frequency concentration, q——Nonlinear parameter, controls the degree of focus, the larger the q, the more emphasis is placed on the high-value area; CM2——Normalized time-frequency concentration, ——S x2 The normalized form of (t,f), CM3 - generalized time-frequency concentration; α - low-order power parameter, usually emphasizing global distribution, β - high-order power parameter, usually emphasizing local concentration, ——It’s S. x3 (t,f) Normalized form to ensure the scale consistency of time-frequency representation;

[0056] Reference Figure 4 In this embodiment, the energy concentration calculation results of the segmented S transform of the four actions of walking on flat ground, climbing stairs, descending stairs, and crossing obstacles retain 25 feature data of the original electromyographic signal, better retain the phase information of the signal, and the energy amplitude difference is more obvious. Figure 4 The figure shows the energy concentration of the rectus femoris in four actions (walking on flat ground, climbing stairs, descending stairs, and crossing obstacles). Each sub-graph corresponds to an action. The horizontal axis of each sub-graph is the number of features (from 1 to 25), and the vertical axis is the energy value. The distribution characteristics of the four actions are as follows: Walking on flat ground: the energy value is low and the fluctuation amplitude is relatively stable, indicating that the signal energy distribution is relatively uniform; Climbing stairs: an obvious energy peak appears near feature 15, reflecting that this action has a higher energy concentration on specific features; Descending stairs: similar to climbing stairs, the overall energy fluctuation is large, but the energy peak at feature 15 is not as high as that of climbing stairs; Crossing obstacles: the energy value fluctuates greatly, and the overall energy distribution is more dispersed than other actions. The energy values ​​of some features are higher, especially in the range of features 20-25; Features and phase information: the data retains 25 features, indicating that the original signal is fully decomposed through the feature extraction of segmented S transform, which not only retains the phase information, but also enhances the differentiation of energy amplitude in specific actions, which is conducive to improving the accuracy of motion recognition.

[0057] Reference Figure 5 In step 3), the construction of the support vector machine multi-classifier is as follows: design a multi-classifier. For a given m classes, train a binary classifier for every two classes in the m classes, and the total number of binary classifiers is m(m-1) / 2; to achieve four classifications, six classifiers are required, namely: class 1 and class 2, class 1 and class 3, class 1 and class 4, class 2 and class 3, class 2 and class 4, class 3 and class 4; for a data that needs to be classified, it needs to be predicted by all classifiers, and the final class attribute is determined by voting.

[0058] The multi-channel EMG signal correlation analysis in step 3) is specifically as follows: by analyzing the correlation of EMG signals, exploring the difference in classification effects between single-channel signals and multi-channel signal fusion, and finding the best lower limb muscle combination to improve the recognition accuracy of human lower limb movements.

[0059] Reference Figure 6 , Figure 7 , Figure 8 In this embodiment, multi-channel signal fusion is to perform a series of processing on the original electromyographic data, and obtain the simple time-frequency domain feature matrix and S-transform energy concentration feature matrix of the lower limb surface electromyographic signal respectively. The two types of features are classified by support vector machine, and the accuracy mean, minimum value, and accuracy concentration of the motion classification results are comprehensively analyzed. It can be found that the performance of the rectus femoris and medial gastrocnemius is relatively excellent; considering the experimental analysis results of the two methods, in practical applications, single-channel muscle signals are selected, and it is most appropriate to select rectus femoris signals for action classification and recognition. According to the judgment of the above pattern recognition effect, preparation is made for selecting appropriate channels for fusion.

[0060] Figure 6 The average classification accuracy of the six-channel electromyographic signal based on simple time-frequency domain features of the 10 subjects was 80.28% (rectus femoris), 82.87% (vastus medialis), 71.71% (biceps femoris), 83.06% (semitendinosus), 81.41% (tibialis anterior), and 82.59% (medial gastrocnemius); the maximum classification accuracy of the six-channel muscle signals was 91.89% (rectus femoris), 100. % (vastus medialis), 97.30% (biceps femoris), 94.60% (semitendinosus), 91.67% (tibialis anterior), 90.91% (medial gastrocnemius); the minimum classification accuracy of the six-channel muscle signals are: 67.86% (rectus femoris), 52.94% (vastus medialis), 47.06% (biceps femoris), 64% (semitendinosus), 50% (tibialis anterior), 58.82% (medial gastrocnemius).

[0061] Among them, the maximum classification accuracy of 5 subjects appeared in the classification of semitendinosus signals, the maximum classification accuracy of 3 subjects appeared in the classification of rectus femoris signals, and the maximum classification accuracy of 2 subjects appeared in the classification of vastus medialis signals; the maximum classification accuracy was 100%, which was the classification result of vastus medialis for subject 9, but by observing the overall classification results of vastus medialis, it was found that the mean and minimum values ​​of its classification accuracy were lower than the classification results of semitendinosus, which showed that the classification of muscle signals between different subjects was quite different, and the maximum mean accuracy was 83.06% (semitendinosus). In comprehensive comparison, the muscle channels with better classification performance based on time-frequency domain features were semitendinosus, rectus femoris and vastus medialis. Among them, the classification accuracy and stability of the first two were higher than those of signals from other channels. The classification effect of vastus medialis varied from person to person, and its stability was slightly insufficient compared with the classification results of semitendinosus and rectus femoris.

[0062] Figure 7 The average accuracy of the six-channel electromyographic signal feature motion recognition of the 10 subjects based on the S-transform energy concentration feature was 94.20% (rectus femoris), 93.20% (vastus medialis), 95.58% (biceps femoris), 92.26% (semitendinosus), 91.93% (tibialis anterior), and 93.73% (medial gastrocnemius); the maximum classification accuracy of the six-channel muscle signals was 100% (rectus femoris), 97. 30% (vastus medialis), 100% (biceps femoris), 100% (semitendinosus), 100% (tibialis anterior), 100% (medial gastrocnemius); the minimum classification accuracy of the six-channel muscle signals are: 85.71% (rectus femoris), 70.59% (vastus medialis), 78.57% (biceps femoris), 77.42% (semitendinosus), 64.29% (tibialis anterior), 83.33% (medial gastrocnemius).

[0063] Except for the vastus medialis, the maximum classification accuracy of the other five channel muscle signals reached 100%. Comprehensive analysis of their accuracy mean, minimum, and accuracy concentration shows that the rectus femoris and medial gastrocnemius performed better. Considering the analysis results of the two methods, it is most appropriate to use single-channel muscle signals for action classification and recognition in practical applications.

[0064] Reference Figure 8 Through the above analysis, it is found that the classification effect of the simple time-frequency domain features is better for the semitendinosus and rectus femoris, and the S-transform energy concentration feature is better for the medial gastrocnemius and rectus femoris. Taking the rectus femoris signal as an example, the classification results of ten subjects under the two methods are statistically analyzed. It can be seen that for each subject, the classification effect of the S-transform energy concentration method is better than the classification effect of the simple time-frequency domain features. The statistics show that the accuracy of the former is 93.70%, and the accuracy of the latter is 80.71%. Further statistics on the classification effect of the six-channel signals of the ten subjects show that 91.67% of the results are better for the S-transform energy concentration method. It can be seen that the S-transform energy concentration method does have a better pattern recognition effect than extracting simple time-frequency domain features.

[0065] Reference Fig. 9 In step 4), the support vector machine motion classification based on multi-channel signal feature fusion analysis is specifically as follows: the similarities between the lower limb muscles of different subjects are different, and it is necessary to perform similarity calculation on the multi-channel signal features of the subjects, and select the channels with better classification effects among the muscle signals such as the rectus femoris, vastus medialis, biceps femoris, semitendinosus, tibialis anterior, and medial gastrocnemius for fusion; by comparing the classification results of single-channel signals, in order to further improve the accuracy, the multi-channel signal features of the rectus femoris, biceps femoris, and medial gastrocnemius are fused and analyzed, and the fusion coefficient is allocated by the control variable method, and then the features with different weights are input into the classifier for motion classification. Through the fusion of multi-channel muscle signal features, the accuracy of subject motion recognition can be effectively improved.

[0066] In the above analysis, it can be seen that the classification result of the rectus femoris is better. In order to further improve the accuracy, the multi-channel signal features of the rectus femoris, biceps femoris and medial gastrocnemius are fused and analyzed. The multi-channel fusion classification results of the four subjects are compared with the single-channel classification results. Fig. 9 In the middle electromyographic signal channel, channel 1 represents the rectus femoris, channel 2 represents the vastus medialis, channel 3 represents the biceps femoris, channel 4 represents the semitendinosus, channel 5 represents the tibialis anterior, channel 6 represents the medial gastrocnemius, and channel 7 represents the fusion analysis results of the rectus femoris, biceps femoris, and medial gastrocnemius.

[0067] Assuming that the fusion coefficients of the three are a, b, and c respectively, in the experiment of subject one, it was found that when a=1, b=0.2, and c=0.5, the classification accuracy reached 100%, and the classification accuracy of the subject using the single-channel signal of the rectus femoris was 85.7140%. Similarly, in the experiments of subjects two and four, the multi-channel fusion method was used for classification, which can obtain higher classification accuracy. In the experiment of subject three, it was found that when a=1, b=0.2~0.7, and c=0.3, the classification accuracy can reach 100%, and the classification accuracy of the subject using the single-channel signal of the rectus femoris was 96.9697%. It can be seen that the fusion of multi-channel muscle signal features can effectively improve the accuracy of subject action recognition.

[0068] The above implementation modes are only used to illustrate the present invention rather than to limit the present invention. Ordinary technicians in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, all equivalent technical solutions also belong to the scope of the present invention. The patent protection scope of the present invention should be defined by the claims.

Claims

1. A method for lower limb movement recognition based on S-transform energy concentration surface electromyography decoding, characterized in that: Firstly, the original electromyographic signal is preprocessed, and the useful time period is intercepted by endpoint detection. Then, the signal within this time period is subjected to segmented S transform and energy concentration calculation, and the signal features of the specified dimension are extracted by segmentation operation. The motion pattern is classified by SVM, and the multi-channel signal features are fused and analyzed for lower limb movement recognition.

2. A method for lower limb movement recognition based on S-transform energy concentration surface electromyography decoding, characterized in that: The following steps are involved: 1) Signal preprocessing and active segment detection; 2) Signal segmentation S transform and energy concentration calculation; 3) Construction of support vector machine multi-classifier and correlation analysis of multi-channel electromyographic signals; 4) Support vector machine motion classification based on multi-channel signal feature fusion analysis.

3. The method for identifying lower limb movements according to claim 2, characterized in that: The signal preprocessing in step 1) includes: 1.1) 50 Hz notch filtering; 1.2) 30 Hz zero phase shift high pass filtering; a zero phase shift filter is used in the preprocessing process to retain the phase information of the original data sequence.

4. The method for identifying lower limb movement according to claim 2, characterized in that: In step 1), the active segment detection adopts a double threshold detection method based on short-time energy and short-time variance. First, the initial short-time energy minimum threshold and the variance and low threshold that characterize the fluctuation are set. Both parameters are dynamically adjusted according to the actual situation. Then, the window length and window shift are defined to frame the signal. Assume that the EMG signal of the nth frame is represented by x n (m), the frame length is N, calculate its short-time energy E n As shown in formula (1), calculate its variance and S n As shown in formula (2), Where m is the sample index of the signal, indicating the sample point in each frame, and m is an integer from 0 to N-1, indicating the position of each sampling point in this frame of signal. Represents the mean of the signal.

5. The method for identifying lower limb movements according to claim 2, characterized in that: The signal segmented S transform and energy concentration calculation in step 2) are specifically as follows: intercept the data length of an action cycle as x, set a sliding window with a window length and a step length, perform S transform on each segment of data, set the frequency interval, and obtain a time-frequency matrix; as shown in formula (3), the 1 / q root value of the absolute value of the signal after S transform is superimposed and summed in the time-frequency domain, and then the summation result is squared q times; By optimizing the width of the window function in the S transform, a higher energy concentration can be obtained for signals with rapidly changing frequency components. The energy concentration calculation is shown in formula (4); The energy concentration measurement method is used as shown in formula (5), the entire time-frequency concentration is decomposed, and the window width at the discrete points is used to improve its time-frequency concentration; Where: CM - signal energy concentration; t - time; f - frequency; S x (t,f)——Signal S transformation matrix; F——Discrete frequency range; CM1——Time-frequency concentration, q——Nonlinear parameter, controls the degree of focus, the larger the q, the more emphasis is placed on the high-value area; CM2——Normalized time-frequency concentration, The normalized form of CM3 is the generalized time-frequency concentration; α is a low-order power parameter that emphasizes global distribution, and β is a high-order power parameter that emphasizes local concentration. ——It’s S. x3 (t,f) is in normalized form to ensure the scale consistency of time-frequency representation.

6. The method for identifying lower limb movements according to claim 2, characterized in that: In step 3), the construction of the support vector machine multi-classifier is as follows: design a multi-classifier. For a given m classes, train a binary classifier for every two classes in the m classes, and the total number of binary classifiers is m(m-1) / 2; for a data that needs to be classified, it needs to be predicted by all classifiers, and the final class attribute is determined by voting.

7. The method for identifying lower limb movements according to claim 2, characterized in that: The multi-channel EMG signal correlation analysis in step 3) is specifically as follows: by analyzing the correlation of EMG signals, exploring the difference in classification effects between single-channel signals and multi-channel signal fusion, and finding the best lower limb muscle combination to improve the recognition accuracy of human lower limb movements.

8. The method for identifying lower limb movements according to claim 2, characterized in that: In step 4), the support vector machine motion classification based on multi-channel signal feature fusion analysis is specifically as follows: since the similarities between the lower limb muscles of different subjects are different, it is necessary to perform similarity calculation on the multi-channel signal features of the subjects, and select the channels with good classification effects in the muscle signals of the rectus femoris, vastus medialis, biceps femoris, semitendinosus, tibialis anterior, and medial gastrocnemius for fusion; by comparing the classification results of single-channel signals, the multi-channel signal features of the rectus femoris, biceps femoris, and medial gastrocnemius are fused and analyzed, the fusion coefficient is allocated by the control variable method, and then the features with different weights are input into the classifier for motion classification.

Images