A hand action recognition method based on surface electromyogram signal feature collaborative optimization

Abstract

The application discloses a hand action recognition method based on feature cooperative optimization of surface electromyogram signals, and comprises the following steps: acquiring surface electromyogram signals; pre-processing the surface electromyogram signals to obtain a pure surface electromyogram signal sample set; extracting multi-domain statistical features from the pure surface electromyogram signal sample, and normalizing the multi-domain statistical features to obtain a multi-dimensional feature set; cooperatively optimizing the multi-dimensional feature set through a two-stage feature mutual feedback method to obtain an optimal feature subset; wherein the two-stage feature mutual feedback method comprises a feature weight screening method and a feature structure feedback correction method; and inputting the optimal feature subset into a machine learning classification algorithm to recognize hand actions and obtain a recognition result.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence technology, and in particular relates to a method for hand movement recognition based on feature-coordinated optimization of surface electromyography signals. Background Technology

[0002] With the rapid development of artificial intelligence, human-computer interaction, and rehabilitation engineering, hand movement recognition technology based on surface electromyography (SEMG) signals has been widely applied in intelligent prosthetic control, virtual reality operation, rehabilitation training, and wearable human-computer interaction systems. SEMG signals are non-invasive, real-time physiological electrical signals that reflect the neuromuscular activity state of muscle groups under different movements, making them an important source of information for hand movement recognition. For different movement recognition tasks, extracting stable and discriminative features from SEMG signals has always been a key technical challenge for improving recognition performance.

[0003] Existing research typically characterizes the time-varying properties of surface electromyography (EMG) signals by extracting multidimensional features from the time domain, frequency domain, time-frequency domain, and nonlinear domain, thereby constructing feature sets for hand movement recognition. These methods can reflect the multi-level attributes of the signal to some extent; however, as the feature dimension increases, strong correlations often exist between different features, and directly using high-dimensional feature sets can easily lead to feature space redundancy, increased model complexity, and decreased recognition performance. To reduce redundant information and improve feature utilization, researchers have proposed various feature optimization methods, including feature ranking based on statistical criteria, feature dimensionality reduction based on linear or nonlinear mappings, and feature reconstruction based on multi-domain fusion.

[0004] While existing methods have achieved some success in improving recognition accuracy, they still have several limitations. Firstly, most methods employ a sequential processing flow of "feature selection and feature mapping," lacking information exchange and synergy between the different stages of feature optimization. This results in insufficient representativeness and a loose structure in the mapped feature set. Secondly, existing optimizations largely focus on signal preprocessing or classifier parameters, with limited systematic research on optimizing the internal structural relationships within the feature layer. This makes it difficult to improve the generalization ability and stability of the recognition model from the perspective of feature synergy.

[0005] Therefore, how to achieve synergistic optimization between feature selection and feature mapping while maintaining the integrity of feature information, and how to establish a feature processing framework with dynamic feedback mechanism to improve the discriminativeness and structural compactness of feature sets, has become a key problem that urgently needs to be solved in current hand action recognition technology based on surface electromyography signals. Summary of the Invention

[0006] This invention proposes a feature-coordinated optimization method for hand movement recognition based on surface electromyography signals to solve the problems existing in the background technology.

[0007] To achieve the above objectives, the present invention provides a method for hand movement recognition based on feature-coordinated optimization of surface electromyography signals, comprising:

[0008] Acquire surface electromyography signals;

[0009] Preprocessing of surface electromyography (EMG) signals yields a clean sample set of surface EMG signals;

[0010] Multi-domain statistical features are extracted from pure surface electromyography signal samples, and the multi-domain statistical features are normalized to obtain a multi-dimensional feature set.

[0011] The optimal feature subset is obtained by synergistic optimization of the multidimensional feature set through a two-stage feature feedback method; wherein the two-stage feature feedback method includes a feature weight screening method and a feature structure feedback correction method.

[0012] The optimal feature subset is input into a machine learning classification algorithm to recognize hand movements and obtain the recognition results.

[0013] Optionally, the process of acquiring surface electromyography signals includes:

[0014] Surface electromyography sensors were placed at the flexor carpi ulnaris and extensor carpi radialis muscles of the subject's forearm.

[0015] The surface electromyography (EMG) signals of the muscles were collected synchronously during the execution of different hand movements, with a sampling frequency of 500 Hz.

[0016] The hand movements include spherical grip, fingertip grip, hook grip, palm grip, side pinch, and cylindrical grip.

[0017] Optionally, the preprocessing of the surface electromyography signal includes:

[0018] Bandpass filtering is performed using a bandpass filter with a bandwidth of 15-500Hz.

[0019] Power frequency interference is suppressed by notch filtering at a frequency of 50Hz;

[0020] A sample set is constructed from the filtered signal using a sliding time window with a window length of 200ms and an overlap ratio of 40%.

[0021] Optionally, the process of extracting multi-domain statistical features from pristine surface electromyography (EMG) signal samples includes:

[0022] Extracting time-domain features: root mean square value, mean absolute value, and variance;

[0023] Extracting frequency domain features: mean frequency, median frequency;

[0024] Extracting time-frequency domain features: wavelet transform coefficients;

[0025] Extracting nonlinear features: sample entropy and fuzzy entropy.

[0026] Optionally, the multi-domain statistical features can be normalized using a min-max normalization method.

[0027] Optionally, the process of collaboratively optimizing the multidimensional feature set using a two-stage feature feedback method includes:

[0028] The correlation index between features and hand movements is calculated by using the feature weight screening method to obtain feature weight values. Features with feature weight values ​​lower than the average minus the standard deviation are discarded to obtain a weighted feature subset.

[0029] The weighted feature subset is mapped using the feature structure feedback correction method. The correlation coefficient between the original features and the mapped components is calculated, the feature structure stability value is calculated, and the feature weight values ​​are corrected by feedback. Features whose corrected weight values ​​are greater than the average minus the standard deviation are retained to obtain the structural feature subset. The structural feature subset is returned to the first stage for mutual feedback. If the sum of the rate of change of the feature weight value and the rate of change of the feature structure stability value is greater than a preset threshold in two consecutive mutual feedbacks, the mutual feedback process continues. When the sum is less than or equal to the preset threshold, the mutual feedback process is determined to have converged, and the optimal feature subset is obtained.

[0030] Optionally, the process of obtaining feature weight values ​​includes:

[0031] Calculate the feature correlation index:

[0032]

[0033] in, As a feature correlation index, For feature set The first in k One characteristic, K For the total number of features, y Tags for hand gestures. A feature selection operator is used to measure the correlation between features and labels. The mutual information method or the maximum correlation minimum redundancy method can be selected.

[0034] The feature weight values ​​are calculated as follows:

[0035]

[0036] in, For the first k The normalized weight values ​​of each feature.

[0037] Optionally, the process of feedback correction of feature weight values ​​includes:

[0038] Mapping the weighted feature subsets:

[0039]

[0040] in, For the first t The feature set after mapping in the second mutual feedback. d The mapped feature dimensions, For the eigenmap operator, principal component analysis, linear discriminant analysis, diffusion mapping, or Laplace eigenmap can be selected. For the parameter set of the mapping operator;

[0041] The correlation coefficient between the original features and the mapped components is calculated:

[0042]

[0043] in, The correlation coefficient between the original features and the mapped components. Original features and mapping components covariance, and The standard deviation of the two;

[0044] Calculate the stability value of the characteristic structure:

[0045]

[0046] in, S The feature structure stability value measures the overall stability of the mapping space. The importance weights of the mapped components satisfy the following conditions: , For the first j The eigenvalues ​​corresponding to each mapping component reflect the information contribution of that component in the mapping space;

[0047] The feedback correction for the feature weight values ​​is calculated as follows:

[0048]

[0049] in, For the revised first k Each feature weight value, The feedback balance coefficient is used to adjust the strength of the influence of mapping feedback in weight updates. For the first k The correlation coefficient between each feature and its corresponding mapping component.

[0050] Optionally, the machine learning classification algorithm employs a random forest, support vector machine, or Naive Bayes machine learning classifier.

[0051] Compared with the prior art, the present invention has the following advantages and technical effects:

[0052] This invention provides a feature-coordinated optimization method for hand movement recognition based on surface electromyography (SEMG) signals. By deploying multi-channel SEMG sensors in key forearm muscle regions and employing a synchronous acquisition method, it can obtain raw SEMG signals that comprehensively reflect muscle activity, providing a high-quality data foundation for subsequent signal analysis. Preprocessing techniques using bandpass filtering, notch filtering, and sliding time windows effectively remove low-frequency drift, high-frequency noise, and power frequency interference from the SEMG signals, ensuring signal smoothness and temporal consistency, resulting in pure SEMG signal samples. Multi-domain statistical features, including time-domain, frequency-domain, time-frequency-domain, and nonlinear features, are extracted from these pure SEMG signal samples to comprehensively characterize... The dynamic changes and movement differences of electromuscular activity were plotted, and normalization techniques were used to normalize the extracted multidimensional features, eliminating the influence of channel differences and inconsistencies in feature dimensions between individuals, thus improving feature comparability and the stability of subsequent optimization. A two-stage feature feedback method was used to establish a feedback relationship between feature selection and mapping, realizing the coordinated adjustment of feature weights and dynamic correction of the mapping space structure, achieving a balance between feature expressive power and structural stability, effectively enhancing the discriminativeness, structural compactness, and anti-interference performance of the feature set. The optimal feature subset after co-optimization was input into a machine learning classification model to recognize different hand movements, achieving high-precision and robust recognition results.

[0053] The technical solution proposed in this invention significantly reduces feature redundancy and improves feature expression quality while maintaining the integrity of feature information, achieving high precision and high stability in hand movement recognition. This technical solution can be widely applied in scenarios such as intelligent prosthetic limb control, human-computer interaction, and rehabilitation training, enabling natural and precise movement control and interaction, providing users with a sensitive and stable operating experience, and possessing significant engineering practical value and broad prospects for promotion. Attached Figure Description

[0054] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0055] Figure 1 This is a schematic diagram of the overall method flow according to an embodiment of the present invention;

[0056] Figure 2 This is a flowchart of the surface electromyography signal preprocessing process according to an embodiment of the present invention;

[0057] Figure 3 This is a flowchart of the multi-domain statistical feature extraction process for surface electromyography signals according to an embodiment of the present invention;

[0058] Figure 4 This is a flowchart of the feature collaborative optimization process according to an embodiment of the present invention;

[0059] Figure 5 This is a flowchart illustrating the hand motion recognition process according to an embodiment of the present invention. Detailed Implementation

[0060] It should be noted that, without conflict, the various embodiments and their technical features in this application can be combined with each other to form other implementation methods.

[0061] It should be noted that the steps in the flowchart shown in the attached figure can be implemented in a computer system as computer-executable instructions. Although the flowchart illustrates the logical execution order of each step, in actual implementation, the execution order of some steps can be adjusted or executed in parallel according to the needs of the actual application.

[0062] This invention provides a method for hand movement recognition based on feature-coordinated optimization of surface electromyography signals, comprising the following steps:

[0063] Several surface electromyography (EMG) sensors were placed in the key muscle areas of the subject's forearm. The EMG signals generated by the subject when performing different hand movements (including ball grip, fingertip grip, hook grip, palm grip, side pinch and cylindrical grip) were acquired in a synchronous acquisition method to ensure that the acquired signals can comprehensively reflect the muscle group activity patterns under different movements.

[0064] The acquired surface electromyography (EMG) signals were preprocessed. A bandpass filter was used to remove low-frequency drift and high-frequency noise to retain the effective frequency bands related to hand movements; a notch filter was used to suppress power frequency interference; and a sample set was constructed from the filtered EMG signals using a sliding time window segmentation mechanism. Continuous time segments were extracted based on the window length and overlap ratio to ensure signal smoothness and temporal consistency, resulting in a clean and continuous EMG signal sample set.

[0065] Multi-domain statistical features, including time-domain features, frequency-domain features, time-frequency domain features, and nonlinear statistical features, are extracted from pure surface electromyography (EMG) signal samples to comprehensively characterize the dynamic features of hand movements and muscle activity patterns. The extracted multi-domain features are then normalized to eliminate amplitude differences and dimensional effects between different channels and subjects, improve the comparability of features and algorithm stability, and provide a unified data foundation for subsequent feature co-optimization.

[0066] A two-stage feature feedback method is used for collaborative optimization of the constructed feature set. The first stage selects a subset of highly discriminative weighted features based on the correlation and weight distribution between features and hand action labels. The second stage calculates the stability of the feature structure and the correlation of the mapping components in the mapping space, and feeds back the mapping results to correct the feature weight values, achieving information feedback and dynamic collaboration between feature selection and feature mapping. Through multiple rounds of feedback optimization, the feature layer and the mapping layer form a convergent closed loop, obtaining the optimal feature subset that balances feature discriminability and structural compactness.

[0067] The obtained optimal feature subset is input into a machine learning classification algorithm to identify different hand movements of the subjects (including ball grip, fingertip grip, hook grip, palm grip, side pinch and cylindrical grip), thereby achieving accurate identification of hand movement categories.

[0068] This invention achieves a coordinated balance between feature representation capability and structural stability through dynamic mutual feedback of feature layers. While maintaining the integrity of feature information, it effectively suppresses redundancy and distortion, significantly improving the accuracy, stability, and generalization performance of hand movement recognition. It can be widely applied to systems such as intelligent prostheses, rehabilitation training, and wearable monitoring, and has good practicality and promotional value.

[0069] This invention proposes a feature-co-optimized hand movement recognition method based on surface electromyography (SEMG) signals. Utilizing synchronous multi-channel SEMG signal acquisition technology, raw SEMG signals containing rich muscle activity information are acquired when subjects perform standardized hand movements. Signal preprocessing techniques are used to preprocess the raw SEMG signals, suppressing noise and motion artifacts to obtain clean and stable SEMG signals. A sliding time window segmentation mechanism is employed for sample construction, ensuring signal smoothness while improving data utilization. Multi-domain statistical analysis methods are used to quantify the sample data, extracting time-domain, frequency-domain, and nonlinear statistical features. These features are then normalized to eliminate amplitude differences between different channels and subjects, constructing a multi-dimensional feature set that comprehensively describes the dynamic patterns of hand movements and muscle activity patterns. A two-stage feature feedback method is used to achieve bidirectional information interaction between feature selection and feature mapping, obtaining an optimal feature subset that balances feature discriminativity and structural compactness. Finally, a machine learning classification algorithm is used to model the optimal feature subset, achieving high-precision, high-robustness, and strong generalization ability in hand movement recognition.

[0070] The technical solutions of the present invention will be described in detail with reference to the accompanying drawings, such as... Figure 1 As shown, the present invention provides a hand movement recognition method based on feature co-optimization of surface electromyography signals, which specifically includes the following:

[0071] Step 1: Acquisition of surface electromyography signals:

[0072] Multichannel surface electromyography (EMG) sensors were deployed in key muscle areas of the subject's forearm, including the flexor carpi ulnaris and extensor carpi radialis, to cover the major muscle groups involved in hand flexion and extension movements. Simultaneous acquisition was used to record EMG signals generated when subjects performed different hand movements (ball grip, fingertip grip, hook grip, palm grip, lateral pinch, and cylindrical grip) at a sampling frequency of 500 Hz to ensure temporal resolution and motion response sensitivity.

[0073] like Figure 2 As shown, step two: surface electromyography signal preprocessing:

[0074] The surface electromyography (EMG) signals were processed using bandpass filtering, notch filtering, and a sliding time window mechanism to obtain a clean set of surface EMG signal samples. Specifically, this included:

[0075] The acquired surface signals were preprocessed to improve signal purity and temporal stability. A bandpass filter with a bandwidth of 15-500Hz was used to filter out low-frequency drift and high-frequency noise, retaining the effective components of the surface electromyography (EMG) signal. A 50Hz notch filter was used to suppress power frequency interference, obtaining the filtered signal. A sliding time window with a window length of 200ms and an overlap ratio of 40% was used to construct samples of the filtered continuous signal, ensuring signal smoothness and temporal consistency, resulting in a pure surface EMG signal sample set.

[0076] like Figure 3 As shown, step three: extraction of multi-domain statistical features of surface electromyography signals:

[0077] Multi-domain statistical features were extracted from pure surface electromyography (EMG) signal samples, and these features were normalized to obtain a multi-dimensional feature set. Specifically, this included:

[0078] Multi-domain statistical features are extracted from pure surface electromyography (EMG) signal samples to characterize the muscle activity patterns and dynamic modes of hand movements at multiple levels. The extracted multi-domain statistical features are normalized, mapping the feature values ​​to the interval [0,1] to eliminate amplitude inconsistencies caused by electrode differences or individual differences, resulting in a multi-dimensional feature set.

[0079] The extracted multi-domain features include:

[0080] Time-domain characteristics: root mean square value, mean absolute value, variance;

[0081] Frequency domain characteristics: mean frequency, median frequency;

[0082] Time-frequency domain characteristics: wavelet transform coefficients;

[0083] Nonlinear characteristics: sample entropy, fuzzy entropy.

[0084] like Figure 4 As shown, step four: Co-optimization processing of surface electromyography signal features:

[0085] The extracted multidimensional feature set is used to obtain the optimal feature subset through a two-stage feature feedback method. Specifically, this includes:

[0086] The obtained feature set is processed by a two-stage feature feedback method to achieve collaborative optimization of feature selection and feature mapping, thereby obtaining the optimal feature subset with high discriminativeness and structural stability.

[0087] The first stage is the feature weight screening method: the correlation index between features and hand movements is calculated, and the feature weight values ​​are statistically analyzed. Features with feature weight values ​​lower than the average minus the standard deviation are discarded to obtain a weighted feature subset.

[0088] The formula for calculating the feature correlation index is:

[0089] (1)

[0090] in, As a feature correlation index, For feature set The first in k One characteristic, K For the total number of features, y Tags for hand gestures. Feature selection operators are used to measure the correlation between features and labels. Mutual information method or maximum correlation minimum redundancy method can be selected.

[0091] The formula for calculating feature weights is:

[0092] (2)

[0093] in, For the first k The normalized weight values ​​of each feature.

[0094] The second stage is the feature structure feedback correction method: the weighted feature subset is mapped, the correlation coefficient between the original feature and the mapped component is calculated, the feature structure stability value is calculated, and the feature weight value is corrected by feedback. Features with corrected weight values ​​greater than the average minus the standard deviation are retained to obtain the structural feature subset. The structural feature subset is returned to the first stage for mutual feedback. If the sum of the rate of change of the feature weight value and the rate of change of the feature structure stability value is greater than a preset threshold in two consecutive mutual feedbacks, the mutual feedback process continues. When the sum is less than or equal to the preset threshold, the mutual feedback process is determined to have converged, and the optimal feature subset is obtained.

[0095] The formula for calculating the weighted feature subset mapping is:

[0096] (3)

[0097] in, For the first t The feature set after mapping in the second mutual feedback. d The mapped feature dimensions, For the eigenmap operator, principal component analysis, linear discriminant analysis, diffusion mapping, or Laplace eigenmap can be selected. For the parameter set of the mapping operator.

[0098] The formula for calculating the correlation coefficient between the original features and the mapped components is:

[0099] (4)

[0100] in, The correlation coefficient between the original features and the mapped components. Original features and mapping components covariance, and Let be the standard deviation of the two.

[0101] The formula for calculating the stability value of the characteristic structure is:

[0102] (5)

[0103] in, S The feature structure stability value measures the overall stability of the mapping space. The importance weights of the mapped components satisfy the following conditions: , For the first j The eigenvalues ​​corresponding to each mapping component reflect the information contribution of that component in the mapping space.

[0104] The formula for calculating the feature weight value feedback correction is as follows:

[0105] (6)

[0106] in, For the revised first k Each feature weight value, The feedback balance coefficient is used to adjust the strength of the influence of mapping feedback in weight updates. For the first k The correlation coefficient between each feature and its corresponding mapping component.

[0107] like Figure 5 As shown, step five: hand motion recognition:

[0108] The optimal feature subset is input into a machine learning classification algorithm to achieve accurate recognition of hand movements. Specifically, this includes:

[0109] The optimal subset of features is input into a machine learning classification algorithm for hand gesture recognition. The machine learning classification algorithm can include random forest, support vector machine, or Naive Bayes. This process achieves high-precision recognition of six typical hand gestures (spherical grip, fingertip grip, hook grip, palm grip, side pinch, and cylindrical grip).

[0110] The above description is merely a preferred embodiment of this application and does not constitute a limitation on the scope of protection of this application. Any equivalent substitutions or improvements made by those skilled in the art without departing from the core concept of this application should be considered to fall within the scope of protection of this application. The scope of protection of this application is defined by the appended claims.

Claims

1. A hand movement recognition method based on feature co-optimization of surface electromyography signals, characterized in that, include: Acquire surface electromyography signals; Preprocessing of surface electromyography (EMG) signals yields a clean sample set of surface EMG signals; Multi-domain statistical features were extracted from a pure set of surface electromyography signal samples, and the multi-domain statistical features were normalized to obtain a multi-dimensional feature set. The optimal feature subset is obtained by synergistic optimization of the multidimensional feature set through a two-stage feature feedback method; wherein the two-stage feature feedback method includes a feature weight screening method and a feature structure feedback correction method. The process of collaboratively optimizing a multidimensional feature set using a two-stage feature feedback method includes: The correlation index between features and hand movements is calculated by using the feature weight screening method to obtain feature weight values. Features with feature weight values ​​lower than the average minus the standard deviation are discarded to obtain a weighted feature subset. The weighted feature subset is mapped using the feature structure feedback correction method. The correlation coefficient between the original feature and the mapped component is calculated, the feature structure stability value is calculated, and the feature weight value is corrected by feedback. Features whose corrected feature weight value is greater than the average minus the standard deviation are retained to obtain the structural feature subset. The structural feature subset is returned to the first stage for mutual feedback. If the sum of the rate of change of the feature weight value and the rate of change of the feature structure stability value is greater than a preset threshold in two consecutive mutual feedbacks, the mutual feedback process continues. When the sum is less than or equal to the preset threshold, the mutual feedback process is determined to have converged, and the optimal feature subset is obtained. The process of obtaining feature weight values ​​includes: Calculate the feature correlation index: in, As a feature correlation index, For multidimensional feature sets The first in k One characteristic, K For the total number of features, y Tags for hand gestures. Feature selection operators are used to measure the correlation between features and labels; The feature weight values ​​are calculated as follows: in, For the first k Normalized weight values ​​for each feature; The process of feedback correction of feature weight values ​​includes: Mapping the weighted feature subset: in, For the first t The multidimensional feature set after mapping in the secondary feedback loop d The mapped feature dimensions, For feature mapping operators, For the parameter set of the mapping operator; The correlation coefficient between the original features and the mapped components is calculated: in, The correlation coefficient between the original features and the mapped components. Original features and mapping components covariance, Original features standard deviation For mapping components Standard deviation; Calculate the stability value of the characteristic structure: in, S The feature structure stability value measures the overall stability of the mapping space. The importance weights of the mapped components satisfy the following conditions: , For the first j The eigenvalues ​​corresponding to each mapping component reflect the information contribution of that component in the mapping space; The feedback correction for the feature weight values ​​is calculated as follows: in, For the revised first k Each feature weight value, The feedback balance coefficient is used to adjust the strength of the influence of mapping feedback in weight updates. For the first k The correlation coefficient between each feature and its corresponding mapping component; The optimal feature subset is input into a machine learning classification algorithm to recognize hand movements and obtain the recognition results.

2. The hand movement recognition method based on feature co-optimization of surface electromyography signals according to claim 1, characterized in that, The process of acquiring surface electromyography signals includes: Surface electromyography sensors were placed at the flexor carpi ulnaris and extensor carpi radialis muscles of the subject's forearm. Simultaneously acquire surface electromyographic signals of muscles during different hand movements, with a sampling frequency of 500Hz; The hand movements include spherical grip, fingertip grip, hook grip, palm grip, side pinch, and cylindrical grip.

3. The hand movement recognition method based on feature collaboration optimization of surface electromyography signals according to claim 1, characterized in that, The preprocessing of surface electromyography signals includes: Bandpass filtering is performed using a bandpass filter with a bandwidth of 15-500Hz. Power frequency interference is suppressed by notch filtering at a frequency of 50Hz; A sample set is constructed from the filtered signal using a sliding time window with a window length of 200ms and an overlap ratio of 40%.

4. The hand movement recognition method based on feature co-optimization of surface electromyography signals according to claim 1, characterized in that, The process of extracting multi-domain statistical features from a pure set of surface electromyography (EMG) signal samples includes: Extracting time-domain features: root mean square value, mean absolute value, and variance; Extracting frequency domain features: mean frequency, median frequency; Extracting time-frequency domain features: wavelet transform coefficients; Extracting nonlinear features: sample entropy and fuzzy entropy.

5. The hand movement recognition method based on feature collaborative optimization of surface electromyography signals according to claim 1, characterized in that, Multi-domain statistical features are normalized using the min-maximum normalization method.

6. The method according to claim 1, characterized in that, The machine learning classification algorithm used is a random forest, support vector machine, or Naive Bayes machine learning classifier.

Images