Gesture recognition method and system based on electromyographic signals

Abstract

The invention provides a gesture recognition method and system based on electromyographic signals. The method comprises the steps that the electromyographic signals generated when a wrist or an arm executes gesture actions are converted into signal segments based on a window cutting method; clustering the signal segments based on an unsupervised learning method, and dividing the signal segments into a plurality of clusters; extracting an optimal feature from each cluster based on an optimal path search algorithm; and obtaining a gesture category according to the optimal feature of each cluster based on an SVM classifier. According to the method, the accuracy of gesture recognition is effectively improved through multi-stage optimization of signal processing and classification steps, and more efficient and more accurate gesture recognition can be realized by using a simple computing environment in a complex environment.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular to a gesture recognition method and system based on electromyographic signals. Background Art

[0002] Gesture recognition technology has been applied in many important fields. In terms of the operation of smart devices, it can effectively replace traditional input devices and provide a more natural way of interaction. In the field of medicine and rehabilitation, gesture recognition is used in muscle training, hand function recovery, surgical assistance, and medical education and training. At the same time, in language communication, it helps the communication between hearing or speech impaired groups and non-sign language users, greatly improving their quality of life. Therefore, the application potential of electromyographic signals in gesture recognition is huge and shows significant value.

[0003] At present, common gesture recognition technologies are mainly divided into three categories: computer vision, data gloves and electromyographic signals. In gesture recognition based on computer vision, cameras and depth sensors are used to record hand movements. However, gesture recognition is generally performed in a more complex environment, and the use of image recording is easily affected by external influences such as external lighting or partial occlusion of the image. Although the accuracy of gesture recognition based on data gloves is outstanding, data gloves and their related cables will bring great inconvenience to users. In contrast, electromyographic signals come from the human body itself and are less affected by the outside world. In addition, electromyographic sensors are small in size and can be easily attached to the skin surface. They are comfortable to wear and suitable for long-term use.

[0004] In recent years, with the development of computer vision, machine learning, deep learning and other technologies, gesture recognition based on electromyographic signals has gradually attracted attention. Deep learning has been widely used in various fields. Compared with traditional machine learning methods, it does not require manual feature extraction. They can classify objects into corresponding types through an end-to-end method, but deep learning models are usually large, have high requirements for equipment, and require a long time to train. Compared with deep learning, the outstanding advantage of traditional machine learning lies in its dependence on feature extraction, which can effectively mine patterns and regularities in data. Feature extraction, as a crucial part of the traditional machine learning process, directly affects the performance and generalization ability of the model. Through carefully designed features, the model can focus on the most representative information, thereby ensuring classification accuracy while avoiding overfitting and reducing equipment requirements.

[0005] Currently, there are few patents on the market for gesture recognition based on electromyographic signals. Some related patents, such as the use of multi-stream multi-feature neural networks and selection-based sparse Transformer models for gesture recognition, or cross-user electromyographic pattern recognition based on motor unit activity, have high requirements for equipment, are not conducive to environmental deployment, and have long training times, which will reduce the scope of application. Summary of the invention

[0006] The present invention provides a gesture recognition method and system based on electromyographic signals, which are used to solve the defects of gesture recognition in the prior art, such as large amount of computation, high requirements on equipment and long training time, and realize the extraction of optimal feature signals from electromyographic signals and classification using machine learning, thereby reducing computing resources and training time while maintaining accuracy.

[0007] The present invention provides a gesture recognition method based on electromyographic signals, comprising:

[0008] Based on the window cutting method, the electromyographic signal of the wrist or arm when performing gestures is converted into signal segments;

[0009] Clustering the signal segments based on an unsupervised learning method to divide the signal segments into a plurality of clusters;

[0010] Extract the best features from each cluster based on the optimal path search algorithm;

[0011] The gesture category is obtained based on the optimal features of each cluster based on the SVM classifier.

[0012] According to a gesture recognition method based on electromyographic signals provided by the present invention, the electromyographic signals of the wrist or arm when performing gesture actions are converted into signal segments based on a window cutting method, including:

[0013] The electromyographic signals are sampled and classified based on a data segmentation method, and invalid signals in the sampled signals are eliminated;

[0014] Merge the sampled signals with the same label;

[0015] The merged signal is divided into signal segments based on the window segmentation method.

[0016] According to a gesture recognition method based on electromyographic signals provided by the present invention, the signal segments are clustered based on an unsupervised learning method to divide the signal segments into multiple clusters, including:

[0017] Extract features from each signal segment based on the signal feature extraction method, and determine the optimal value of the number of clusters K based on the elbow criterion;

[0018] Determine K cluster centers, and assign each signal segment to the cluster center closest to it according to its characteristics, so as to form K clusters;

[0019] According to the characteristics of the signal segments in each cluster, calculate the new cluster center of each cluster;

[0020] According to the characteristics of each signal segment, each signal segment is assigned to a new cluster center that is closest to it, and K clusters are formed again until a preset termination condition is reached.

[0021] According to a gesture recognition method based on electromyographic signals provided by the present invention, an optimal value of the number of clusters K is determined based on the elbow criterion, including:

[0022] Calculating the sum of square errors of the clustering results of the signal fragments under different cluster numbers K;

[0023] Draw a curve based on the sum of squared errors corresponding to different cluster numbers K;

[0024] The elbow position of the curve is determined, and the preset cluster corresponding to the elbow position is used as the optimal value of the number of clusters K.

[0025] According to a gesture recognition method based on electromyographic signals provided by the present invention, the clustering results of the signal segments under different cluster numbers K are used to calculate the sum of square errors by the following formula:

[0026]

[0027] Among them, SSE K is the sum of squared errors corresponding to the number of clusters K, ||x i -c k || 2 Represents the feature x of the i-th signal segment i The squared distance to the center of the kth cluster.

[0028] According to a gesture recognition method based on electromyographic signals provided by the present invention, the signal segments are clustered based on an unsupervised learning method to divide the signal segments into multiple clusters, including:

[0029] Extract features from each signal segment based on the signal feature extraction method, and reorganize the features of each signal segment on all channels;

[0030] The features of the recombined signal fragments are optimized based on dimensionality reduction technology;

[0031] A clustering algorithm is used to perform unsupervised classification on the signal segments according to the optimized features of each signal segment to obtain multiple clusters.

[0032] According to a gesture recognition method based on electromyographic signals provided by the present invention, the optimal feature is extracted from each cluster based on an optimal path search algorithm, including:

[0033] Extract multiple features of each cluster based on feature extraction methods;

[0034] The optimal path search algorithm based on particle swarm optimization selects the best feature from multiple features of each cluster.

[0035] According to a gesture recognition method based on electromyographic signals provided by the present invention, the characteristics of each cluster include one or more of mean absolute value, root mean square, zero crossing rate, waveform amplitude, sequence length, variance, integrated electromyographic signal, mean absolute change, standard deviation, peak value, valley value, amplitude variability, power spectral density, mean square frequency and spectral entropy.

[0036] According to a gesture recognition method based on electromyographic signals provided by the present invention, an optimal path search algorithm based on particle swarm optimization selects the optimal feature from multiple features of each cluster, including:

[0037] Take each cluster as a particle and update the position of each particle in each generation using the following formula:

[0038]

[0039] in, and are the velocity and position of the ith particle in the tth generation, is the velocity of the ith particle in the t+1th generation, and g best are the historical optimal position and global optimal position of the ith particle, c1 and c2 are acceleration constants, r1 and r2 are random variables, and ω is the inertia weight;

[0040] The position update formula of each particle is:

[0041]

[0042] in, is the position of the ith particle in the t+1th generation;

[0043] By performing global search and local optimization in multiple features of each cluster, and taking weighted accuracy as the objective function to measure the quality of the particle's features, the optimal feature is selected for each particle. The formula of the objective function is:

[0044]

[0045] Among them, WA is the objective function value, C is the number of clusters, ω k is the number of signal segments in the kth cluster, Acck is the accuracy of the kth cluster.

[0046] The present invention also provides a gesture recognition system based on electromyographic signals, comprising:

[0047] A segmentation module, used for converting the myoelectric signals of the wrist or arm when performing gesture actions into signal segments based on a window cutting method;

[0048] A clustering module, used for clustering the signal segments based on an unsupervised learning method, and dividing the signal segments into a plurality of clusters;

[0049] An extraction module, used to extract the best features from each cluster based on an optimal path search algorithm;

[0050] The recognition module is used to obtain the gesture category according to the optimal features of each cluster based on the SVM classifier.

[0051] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the gesture recognition method based on electromyographic signals as described above is implemented.

[0052] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the gesture recognition method based on electromyographic signals as described above is implemented.

[0053] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned methods for gesture recognition based on electromyographic signals.

[0054] The gesture recognition method and system based on electromyographic signals provided by the present invention convert long sequence signals into short sequence signals by adopting window cutting, thereby effectively improving data processing efficiency and being able to more accurately capture local features and dynamic changes; then, the extracted signals are clustered by using an unsupervised learning method to automatically discover potential patterns in the signals, reduce manual intervention and enhance the adaptability and robustness of the model; finally, the most suitable feature signal is assigned to each cluster by optimal path selection, the feature selection process is optimized, the classification accuracy is improved and the generalization ability of the model is enhanced, so that it can adapt to a variety of application scenarios; the optimal feature signal is extracted from the electromyographic signal, and traditional machine learning is used for classification, which improves the recognition accuracy and reaction speed while reducing computing resources and training time, and can run efficiently on devices with low computing resource requirements; the accuracy of gesture recognition is effectively improved by optimizing the signal processing and classification steps in multiple stages, and more efficient and accurate gesture recognition can be achieved in a complex environment using a simple computing environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 This is one of the flow charts of the method for hand gesture recognition based on electromyographic signals provided by the present invention;

[0057] Figure 2 This is the second flow chart of the method for hand gesture recognition based on electromyographic signals provided by the present invention;

[0058] Figure 3 Schematic diagram of the overall network structure of the gesture recognition method based on electromyographic signals provided by the present invention;

[0059] Figure 4 It is a structural schematic diagram of a gesture recognition system based on electromyographic signals provided by the present invention;

[0060] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] Combine the following Figure 1 A method for hand gesture recognition based on electromyographic signals of the present invention is described, comprising:

[0063] Step 101, converting the electromyographic signal of the wrist or arm when performing a gesture into signal segments based on a window cutting method;

[0064] like Figure 2 As shown, an electromyographic signal collection device is used to collect electromyographic signals when the wrist or arm is performing gesture actions, and the collected electromyographic signals are saved as a .mat file.

[0065] The user can wear the armband, make gestures, connect the armband Bluetooth and computer Bluetooth, directly transmit the electromyographic signals to the computer, and process the electromyographic signals on the computer.

[0066] Firstly, the EMG signal is preprocessed and the long sequence signal of the EMG signal is converted into a short sequence signal, namely, a signal segment, by using the window cutting method.

[0067] Step 102, clustering the signal segments based on an unsupervised learning method to divide the signal segments into multiple clusters;

[0068] Unsupervised learning methods are used to reorganize and cluster the extracted signals to complete unsupervised preliminary classification.

[0069] Step 103, extracting the best feature from each cluster based on the optimal path search algorithm;

[0070] The most suitable feature signal is assigned to each cluster obtained after clustering through optimal path selection.

[0071] Step 104: Obtain gesture categories based on the optimal features of each cluster using a SVM (Support Vector Machine) classifier.

[0072] The machine learning classifier SVM classifier is used to perform the final classification according to the optimal features of each cluster to obtain the gesture category.

[0073] Before using the SVM classifier for recognition, the data set is divided into a training set, a validation set, and a test set. The training set data is used to train the SVM model, and the radial basis function (RBF) kernel is selected as the kernel function for gesture classification. The trained SVM model is used to classify and recognize gesture data. The gestures are divided into 52 gestures, and the final gesture action is determined based on the label output by the gesture classification.

[0074] This embodiment uses window cutting to convert long sequence signals into short sequence signals, which effectively improves data processing efficiency and can more accurately capture local features and dynamic changes; then uses unsupervised learning methods to cluster the extracted signals, automatically discovers potential patterns in the signals, reduces manual intervention and enhances the adaptability and robustness of the model; finally, the most suitable feature signal is assigned to each cluster through optimal path selection, the feature selection process is optimized, the classification accuracy is improved, and the generalization ability of the model is enhanced, so that it can adapt to a variety of application scenarios; the optimal feature signal is extracted from the electromyographic signal, and traditional machine learning is used for classification, which improves the recognition accuracy and response speed while reducing computing resources and training time, and can run efficiently on devices with low computing resource requirements; through multi-stage optimization of signal processing and classification steps, the accuracy of gesture recognition is effectively improved, and more efficient and accurate gesture recognition can be achieved in a complex environment using a simple computing environment.

[0075] Based on the above embodiment, in this embodiment, the myoelectric signal of the wrist or arm when performing gestures is converted into signal segments based on the window cutting method, including:

[0076] Classifying the electromyographic signals after sampling based on a data segmentation method, and removing invalid signals during rest periods in the sampled signals;

[0077] Merge the sampled signals with the same label;

[0078] The merged signal is divided into signal segments based on the window segmentation method.

[0079] This embodiment uses a window cutting method to convert long sequence signals into short sequence signals, thereby generating new signal samples for subsequent analysis.

[0080] The gesture classification dataset based on electromyographic signals can be derived from the public dataset NinaPro DB5, which uses two armbands to collect arm data and converts it into a .mat file. The required data signals and corresponding labels are extracted from the .mat file. In the preprocessing part, all sampled signals are first extracted with a dimension of (6353315,16), and invalid signals during the rest period are excluded. After excluding invalid signals, the dimension is (2384532,16). Then, signal samples with the same label are merged. After the merging is completed, the window segmentation method is applied to segment different gesture signals, and the window distance is set to 500 and the moving distance is 125. Each signal becomes the same shape (18897,16,500).

[0081] On the basis of the above embodiment, in this embodiment, the signal segments are clustered based on an unsupervised learning method, and the signal segments are divided into multiple clusters, including:

[0082] Extract features from each signal segment based on the signal feature extraction method, and determine the optimal value of the number of clusters K based on the elbow criterion;

[0083] Determine K cluster centers, and assign each signal segment to the cluster center closest to it according to its characteristics, so as to form K clusters;

[0084] According to the characteristics of the signal segments in each cluster, calculate the new cluster center of each cluster;

[0085] According to the characteristics of each signal segment, each signal segment is assigned to a new cluster center that is closest to it, and K clusters are formed again until a preset termination condition is reached.

[0086] A clustering algorithm is used to perform an unsupervised preliminary classification of the signal fragments as follows:

[0087] First, K cluster centers are selected. In the allocation phase, each signal segment is assigned to the cluster center closest to it to form K clusters. The i-th signal segment x i Assigned to the cluster center c that is closest to it k , the specific formula is:

[0088] Assign(x i ) = argmin||x i -c k ||

[0089] Among them, ||x i -c k || represents the i-th signal segment x i and the kth cluster center c k The Euclidean distance between .

[0090] In the update phase, all signal segments are assigned to their respective clusters, and the new cluster center of each cluster is calculated. The cluster center c k Updated to the mean of all signal fragments in the cluster, the specific formula is:

[0091]

[0092] Among them, C k represents all signal fragments in the kth cluster, |C k | is the number of signal segments in the kth cluster. The allocation phase and the update phase are repeated until the cluster center no longer changes significantly or the maximum number of iterations is reached.

[0093] Based on the above embodiment, in this embodiment, the optimal value of the number of clusters K is determined based on the elbow criterion, including:

[0094] Calculating the sum of square errors of the clustering results of the signal fragments under different cluster numbers K;

[0095] Draw a curve based on the sum of squared errors corresponding to different cluster numbers K;

[0096] The elbow position of the curve is determined, and the preset cluster corresponding to the elbow position is used as the optimal value of the number of clusters K.

[0097] The optimal number of clusters K is determined according to the elbow criterion, and the number of categories of unsupervised classification is determined. In order to determine the optimal number of clusters K, the clustering results of the signal segments under different numbers of clusters K are firstly calculated by the following formula to calculate the sum of squared errors (SSE):

[0098]

[0099] Among them, SSE Kis the sum of squared errors corresponding to the number of clusters k, ||x i -c k || 2 Represents the feature x of the i-th signal segment i The squared distance to the center of the kth cluster.

[0100] Then, we plotted the SSE curves for different numbers of clusters and found the elbow position in the curve. The number of clusters corresponding to the elbow position is the optimal number of clusters. The error reduction rate at this position slowed down significantly, indicating that the effect of further increasing the number of clusters on improving model performance gradually decreased.

[0101] On the basis of the above embodiment, in this embodiment, the signal segments are clustered based on an unsupervised learning method, and the signal segments are divided into multiple clusters, including:

[0102] Extract features from each signal segment based on the signal feature extraction method, and reorganize the features of each signal segment on all channels;

[0103] The features of the recombined signal fragments are optimized based on dimensionality reduction technology;

[0104] A clustering algorithm is used to perform unsupervised classification on the signal segments according to the optimized features of each signal segment to obtain multiple clusters.

[0105] like Figure 3 As shown in the figure, before using the unsupervised learning method to cluster the signal fragments, the signal fragments are reorganized first. By disrupting the original features, the signal feature extraction method is used to comprehensively analyze the three-dimensional signal fragments (samples, channels, time steps) and extract discriminative features. Subsequently, the signal is reorganized to convert the three-dimensional signal into a two-dimensional signal (sample, feature dimension). Next, the dimensionality reduction technology is applied to optimize the signal to reduce redundant data and improve processing efficiency. Finally, the clustering algorithm is used to perform unsupervised preliminary classification of the signal fragments based on the extracted features, and similar signal fragments are classified into the same category.

[0106] Based on the above embodiment, in this embodiment, the optimal features are extracted from each cluster based on the optimal path search algorithm, including:

[0107] Extract multiple features of each cluster based on feature extraction methods;

[0108] The optimal path search algorithm based on particle swarm optimization selects the best feature from multiple features of each cluster.

[0109] A variety of advanced feature extraction methods are used to extract features from the signal fragments in each cluster. The optimal path search algorithm of particle swarm optimization (PSO) is used for different clusters to select the optimal features of each cluster. The optimal path search algorithm of particle swarm optimization can deeply mine and filter out the most discriminative features in high-dimensional signal data.

[0110] On the basis of the above embodiments, the feature extraction method is used in this embodiment to extract from three aspects: time domain, frequency domain and time-frequency domain respectively. The features of each cluster include mean absolute value (MAV), root mean square (RMS), zero crossing rate (ZC), waveform amplitude (WAMP), sequence length (SL), variance (VAR), integrated electromyographic signal (IEMG), mean absolute change (MAS), standard deviation (SD), peak value (Peak), trough value (Trough), amplitude variability (Amplitude_Variability), power spectral density (PSD), mean square frequency (MNF) and spectral entropy (Spectral_Entropy) One or more.

[0111] In this embodiment, the original signal in each cluster is subjected to signal feature extraction along the time step, and the signal finally obtained is a two-dimensional signal (sample, number of channels×number of feature extraction methods).

[0112] On the basis of the above embodiment, the optimal path search algorithm based on particle swarm optimization in this embodiment selects the optimal feature from multiple features of each cluster, including:

[0113] PSO simulates the foraging behavior of bird flocks and uses particles to move in the search space to find the optimal solution. Each cluster is regarded as a particle, and the position of each particle in each generation is updated by the following formula:

[0114]

[0115] in, and are the velocity and position of the ith particle in the tth generation, is the velocity of the ith particle in the t+1th generation, and g best are the historical optimal position and global optimal position of the ith particle, c1 and c2 are acceleration constants, r1 and r2 are random variables, and ω is the inertia weight;

[0116] During the particle update process, the particle is guided to approach a better solution by combining its own experience and group experience. The position update formula of each particle is:

[0117]

[0118] in, is the position of the ith particle in the t+1th generation;

[0119] By performing global search and local optimization in multiple features of each cluster, and taking weighted accuracy as the objective function to measure the quality of the particle's features, the optimal feature is selected for each particle. The formula of the objective function is:

[0120]

[0121] Among them, WA is the objective function value, C is the number of clusters, ω k is the number of signal segments in the kth cluster, Acc k is the accuracy of the kth cluster.

[0122] This implementation requires the user to wear an armband, make gestures, and connect the armband Bluetooth to the computer Bluetooth to directly transmit the electromyographic signal to the computer, which processes the signal. The signal fragments of any 500 time steps of the gesture action can be intercepted, the signal fragments are reorganized, and classified once through unsupervised clustering. The optimal signal features of the cluster are extracted, and the signal features are put into SVM for classification. According to the gesture situation, a total of 52 gestures are divided.

[0123] The following is a description of a gesture recognition system based on electromyographic signals provided by the present invention. The gesture recognition system based on electromyographic signals described below and the gesture recognition method based on electromyographic signals described above can be referred to in correspondence with each other.

[0124] like Figure 4 As shown, the system includes a segmentation module 401, a clustering module 402, an extraction module 403 and a recognition module 404, wherein:

[0125] The segmentation module 401 is used to convert the electromyographic signal of the wrist or arm when performing gesture actions into signal segments based on the window cutting method;

[0126] The clustering module 402 is used to cluster the signal segments based on an unsupervised learning method, and divide the signal segments into multiple clusters;

[0127] The extraction module 403 is used to extract the best feature from each cluster based on the optimal path search algorithm;

[0128] The recognition module 404 is used to obtain the gesture category according to the optimal features of each cluster based on the SVM classifier.

[0129] This embodiment uses window cutting to convert long sequence signals into short sequence signals, which effectively improves data processing efficiency and can more accurately capture local features and dynamic changes; then uses unsupervised learning methods to cluster the extracted signals, automatically discovers potential patterns in the signals, reduces manual intervention and enhances the adaptability and robustness of the model; finally, the most suitable feature signal is assigned to each cluster through optimal path selection, the feature selection process is optimized, the classification accuracy is improved, and the generalization ability of the model is enhanced, so that it can adapt to a variety of application scenarios; the optimal feature signal is extracted from the electromyographic signal, and traditional machine learning is used for classification, which improves the recognition accuracy and response speed while reducing computing resources and training time, and can run efficiently on devices with low computing resource requirements; through multi-stage optimization of signal processing and classification steps, the accuracy of gesture recognition is effectively improved, and more efficient and accurate gesture recognition can be achieved in a complex environment using a simple computing environment.

[0130] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute a gesture recognition method based on electromyographic signals, the method comprising: converting the electromyographic signals of the wrist or arm when performing gesture actions into signal segments based on a window cutting method; clustering the signal segments based on an unsupervised learning method to divide the signal segments into multiple clusters; extracting the optimal features from each cluster based on an optimal path search algorithm; and obtaining a gesture category based on the optimal features of each cluster based on a SVM classifier.

[0131] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0132] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the gesture recognition method based on electromyographic signals provided by the above-mentioned methods, and the method includes: converting the electromyographic signals of the wrist or arm when performing gesture actions into signal segments based on a window cutting method; clustering the signal segments based on an unsupervised learning method, and dividing the signal segments into multiple clusters; extracting the optimal features from each cluster based on an optimal path search algorithm; and obtaining the gesture category based on the optimal features of each cluster based on a SVM classifier.

[0133] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the gesture recognition method based on electromyographic signals provided by the above-mentioned methods, the method comprising: converting the electromyographic signals of the wrist or arm when performing gesture actions into signal segments based on a window cutting method; clustering the signal segments based on an unsupervised learning method, and dividing the signal segments into multiple clusters; extracting the optimal features from each cluster based on an optimal path search algorithm; and obtaining the gesture category based on the optimal features of each cluster based on a SVM classifier.

[0134] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0135] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A gesture recognition method based on electromyographic signals, characterized in that: include: Based on the window cutting method, the electromyographic signal of the wrist or arm when performing gestures is converted into signal segments; Clustering the signal segments based on an unsupervised learning method to divide the signal segments into a plurality of clusters; Extract the best features from each cluster based on the optimal path search algorithm; The gesture category is obtained based on the optimal features of each cluster based on the SVM classifier.

2. The method for gesture recognition based on electromyographic signals according to claim 1, characterized in that: Based on the window cutting method, the electromyographic signal of the wrist or arm when performing gestures is converted into signal segments, including: The electromyographic signals are sampled and classified based on a data segmentation method, and invalid signals in the sampled signals are eliminated; Merge the sampled signals with the same label; The merged signal is divided into signal segments based on the window segmentation method.

3. The method for gesture recognition based on electromyographic signals according to claim 1, characterized in that: Clustering the signal segments based on an unsupervised learning method to divide the signal segments into a plurality of clusters includes: Extract features from each signal segment based on the signal feature extraction method, and determine the optimal value of the number of clusters K based on the elbow criterion; Determine K cluster centers, and assign each signal segment to the cluster center closest to it according to its characteristics, so as to form K clusters; According to the characteristics of the signal segments in each cluster, calculate the new cluster center of each cluster; According to the characteristics of each signal segment, each signal segment is assigned to a new cluster center that is closest to it, and K clusters are formed again until a preset termination condition is reached.

4. The method for gesture recognition based on electromyographic signals according to claim 3, characterized in that: The optimal value of the number of clusters K is determined based on the elbow criterion, including: Calculating the sum of square errors of the clustering results of the signal fragments under different cluster numbers K; Draw a curve based on the sum of squared errors corresponding to different cluster numbers K; The elbow position of the curve is determined, and the preset cluster corresponding to the elbow position is used as the optimal value of the number of clusters K.

5. The method for gesture recognition based on electromyographic signals according to claim 4, characterized in that: The error sum of squares of the clustering results of the signal fragments under different cluster numbers K is calculated by the following formula: Among them, SSE K is the sum of squared errors corresponding to the number of clusters K, ||x i -c k || 2 Represents the feature x of the i-th signal segment i The squared distance to the center of the kth cluster.

6. The method for gesture recognition based on electromyographic signals according to claim 1, characterized in that: Clustering the signal segments based on an unsupervised learning method to divide the signal segments into a plurality of clusters includes: Extract features from each signal segment based on the signal feature extraction method, and reorganize the features of each signal segment on all channels; The features of the recombined signal fragments are optimized based on dimensionality reduction technology; A clustering algorithm is used to perform unsupervised classification on the signal segments according to the optimized features of each signal segment to obtain multiple clusters.

7. The method for gesture recognition based on electromyographic signals according to claim 1, characterized in that: Extract the best features from each cluster based on the optimal path search algorithm, including: Extract multiple features of each cluster based on feature extraction methods; The optimal path search algorithm based on particle swarm optimization selects the best feature from multiple features of each cluster.

8. The method for gesture recognition based on electromyographic signals according to claim 7, characterized in that: The characteristics of each cluster include one or more of mean absolute value, root mean square, zero crossing rate, waveform amplitude, sequence length, variance, integrated EMG signal, mean absolute change, standard deviation, peak, valley, amplitude variability, power spectral density, mean square frequency, and spectral entropy.

9. The method for gesture recognition based on electromyographic signals according to claim 7, characterized in that: The optimal path search algorithm based on particle swarm optimization selects the best features from multiple features of each cluster, including: Take each cluster as a particle and update the position of each particle in each generation using the following formula: in, and are the velocity and position of the ith particle in the tth generation, is the velocity of the ith particle in the t+1th generation, and g best are the historical optimal position and global optimal position of the ith particle, c1 and c2 are acceleration constants, r1 and r2 are random variables, and ω is the inertia weight; The position update formula of each particle is: in, is the position of the ith particle in the t+1th generation; By performing global search and local optimization in multiple features of each cluster, and taking weighted accuracy as the objective function to measure the quality of the particle's features, the optimal feature is selected for each particle. The formula of the objective function is: Among them, WA is the objective function value, C is the number of clusters, ω k is the number of signal segments in the kth cluster, Acc k is the accuracy of the kth cluster.

10. A gesture recognition system based on electromyographic signals, characterized in that: include: A segmentation module, used for converting the electromyographic signal of the wrist or arm when performing gesture actions into signal segments based on a window cutting method; A clustering module, used for clustering the signal segments based on an unsupervised learning method, and dividing the signal segments into a plurality of clusters; An extraction module, used to extract the best features from each cluster based on an optimal path search algorithm; The recognition module is used to obtain the gesture category according to the optimal features of each cluster based on the SVM classifier.

Images