Upper limb action classification and recognition method, device and equipment based on surface electromyogram signals and storage medium

Abstract

The invention relates to the field of voiceprint recognition, and discloses an upper limb action classification recognition method and device based on surface electromyogram signals, equipment and a storage medium. The method comprises the steps that signal decomposition is conducted, specifically, surface electromyogram signals are subjected to signal decomposition, different sub-signals are obtained, and the signal decomposition is deployed on edge equipment; model construction: constructing an independent 1D-CNN (Convolutional Neural Network) sub-model for each sub-signal, and calculating a fusion feature according to the weight of each sub-signal to obtain a 1D-CNN model; model training: dividing a training data set into a training set and a verification set, and training the 1D-CNN model to obtain a trained 1D-CNN model; the same signal decomposition is carried out on newly input original surface electromyogram signals, the trained 1D-CNN model is used for upper limb action classification and recognition, and classification and recognition are completed in a cloud server. According to the method, signal decomposition is adopted and deployed on edge equipment, so that the real-time problem of upper limb action classification is solved.

Description

Technical Field

[0001] The present invention relates to the field of voiceprint recognition, and in particular to a method, device, equipment and storage medium for upper limb movement classification and recognition based on surface electromyography signals. Background Art

[0002] With the advancement of science and technology and the development of biomedical engineering, surface electromyography (sEMG), as a non-invasive biological signal, has been widely used in fields such as rehabilitation robots, intelligent prostheses, and muscle strength monitoring.

[0003] By analyzing and processing surface electromyographic signals, real-time monitoring and recognition of upper limb movements can be achieved, thereby providing accurate control signals for rehabilitation training and auxiliary equipment. However, surface electromyographic signals themselves are highly complex and diverse, and actual application scenarios have high requirements for computing resources and response speed.

[0004] Therefore, it is necessary to design efficient and real-time algorithms and models for surface electromyography signal processing and upper limb movement recognition. Summary of the invention

[0005] The main purpose of the present invention is to solve the problem in the prior art that large changes may occur under different channel conditions, thereby reducing the recognition performance.

[0006] A first aspect of the present invention provides an upper limb movement classification and recognition method based on surface electromyography signals, the upper limb movement classification and recognition method based on surface electromyography signals comprising:

[0007] Signal decomposition: decompose the surface electromyography signal to obtain different sub-signals, and deploy the signal decomposition on the edge device; model construction: for each sub-signal, build an independent 1D-CNN sub-model, and calculate the fusion feature according to the weight of each sub-signal to obtain

[0008] 1D-CNN model; Model training: Divide the training data set into a training set and a validation set, train the 1D-CNN model, and obtain a trained 1D-CNN model. Classification and recognition: Perform the same signal decomposition on the newly input raw surface electromyography signal, and use the trained 1D-CNN model to classify and recognize upper limb movements, and the classification and recognition is completed on the cloud server.

[0009] As a preferred technical solution, the model construction includes:

[0010] For each sub-signal, multiple 1D-CNN sub-models are constructed, and each 1D-CNN sub-model independently learns the feature representation of the sub-signal; the weight of the sub-signal is multiplied by the corresponding sub-model feature, and the weighted features are added to obtain a fused feature; after the features of all the sub-signals are fused, one or more fully connected layers are added, and the output of the last fully connected layer is connected to an output node with the number of upper limb movement categories; the Softmax activation function is used to obtain the probability distribution of each upper limb movement category.

[0011] As a preferred technical solution, the weight of the sub-signal is obtained by the following steps:

[0012] Sub-signal feature extraction: extract features from each sub-signal; feature combination: combine the features of each sub-signal into a feature vector as the input of the weight calculation neural network; weight calculation neural network construction: build a small neural network to calculate the weight of each sub-signal; neural network training: use a labeled training data set to train the weight calculation neural network, and the optimization goal is to minimize the loss between the predicted weight and the true weight to obtain a trained weight calculation neural network; weight calculation deployment: deploy the trained weight calculation neural network on the edge device. When a new sub-signal is input, the sub-signal features are extracted and input into the weight calculation neural network to obtain the corresponding weight value.

[0013] As a preferred technical solution, the signal decomposition step includes:

[0014] Signal conversion: convert the surface electromyography signal into a graph structure, each surface electromyography signal is regarded as a node, the connection between the nodes is constructed according to the similarity or distance between the surface electromyography signals, and the connection between the nodes is determined using a graph construction method; build a graph convolutional neural network (GCNN): the GCNN captures the local structure and long-distance dependencies between signals. The GCNN consists of multiple graph convolution layers and activation functions. Each graph convolution layer can update the node features, and the GCNN model is deployed on the edge device; train the GCNN model: use the labeled surface electromyography signal data to train the GCNN model, learn an effective signal decomposition representation, and obtain a trained GCNN model; signal decomposition: use the trained GCNN model to decompose the input surface electromyography signal to obtain different sub-signals.

[0015] As a preferred technical solution, the signal conversion step includes:

[0016] Signal division: Divide the surface electromyographic signal into windows of equal length, each window contains a certain number of signal samples, and the windows will be used as nodes represented by the graph structure; Feature extraction: For each window, calculate the features of each window, and use the features as the feature vectors of the nodes; Calculate the distance between nodes: Calculate the distance between nodes between the feature vectors of each pair of nodes, and the distance between nodes is used to measure the similarity between nodes; Construct a graph structure: According to the distance between nodes, use the k-nearest neighbor method to determine the connection between nodes, and each node is only connected to its k nearest neighbor nodes; Graph weight setting: For connected nodes, assign weights according to the distance between the connected nodes, and the weight is the inverse of the distance. The weight reflects the similarity between nodes, thereby helping the graph convolutional neural network capture local information.

[0017] As a preferred technical solution, the construction of a graph convolutional neural network includes:

[0018] Build an input layer: receive the facial electromyography signal converted into a graph structure representation, the input data includes a node feature matrix and an adjacency matrix, each row of the node feature matrix corresponds to a feature vector of a node, and the adjacency matrix represents the connection and weight between nodes; build a graph convolution layer: the graph convolution layer is used to update node features; build a global pooling layer: add a global pooling layer to integrate the information of all nodes; build a fully connected layer: after the global pooling layer, add one or more fully connected layers, the fully connected layer is used to integrate features and output the final classification result; build an output layer: the output of the last fully connected layer is connected to the output node with the number of categories, and the Softmax activation function is used to obtain the probability distribution of each category.

[0019] As a preferred technical solution, the construction of the graph convolution layer includes:

[0020] Node feature propagation: multiply the adjacency matrix by the node feature matrix to obtain the propagated node feature matrix. Feature transformation: multiply the propagated node feature matrix by the weight matrix to obtain the updated node features. Activation function: apply the activation function to the updated node features to complete the graph convolution layer construction.

[0021] A second aspect of the present invention provides an upper limb movement classification and recognition device based on surface electromyography signals, comprising:

[0022] Signal decomposition module: used to decompose the surface electromyography signal to obtain different sub-signals, and the signal decomposition module is deployed on the edge device; model construction module: used to construct an independent 1D-CNN sub-model for each sub-signal, and obtain the fusion feature according to the weight calculation of each sub-signal to obtain the 1D-CNN model; model training module: used to divide the training data set into a training set and a validation set, train the 1D-CNN model, and obtain a trained 1D-CNN model; classification and recognition module: used to perform the same signal decomposition on the newly input original surface electromyography signal, and use the trained 1D-CNN model to classify and recognize upper limb movements, and the classification and recognition is completed on the cloud server.

[0023] The third aspect of the present invention provides an electronic device, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via lines; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned upper limb movement classification and recognition method based on surface electromyography signals.

[0024] A fourth aspect of the present invention provides a computer-readable storage medium, which stores instructions. When the computer-readable storage medium is run on a computer, the computer executes the above-mentioned upper limb movement classification and recognition method based on surface electromyography signals.

[0025] The present invention proposes a surface electromyography signal processing method based on the combination of edge computing and cloud computing. First, signal decomposition is performed on the edge device to decompose the original surface electromyography signal into different sub-signals. For each sub-signal, an independent 1D-CNN sub-model is constructed. Through weight calculation, the features of each sub-signal are fused to obtain an integrated 1D-CNN model. Next, the training data set is divided into a training set and a validation set, and the 1D-CNN model is trained to obtain a trained model. Finally, the upper limb movement classification and recognition of the newly input surface electromyography signal is completed on the cloud server.

[0026] Through this method, the present invention makes full use of the advantages of edge computing and cloud computing, reasonably distributes computing tasks on edge devices and cloud servers, and reduces the demand for computing resources and response time. At the same time, the proposed 1D-CNN model has a high upper limb action recognition accuracy, providing reliable technical support for practical application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A flow chart of a method for upper limb movement classification and recognition based on surface electromyography signals provided by an embodiment of the present invention;

[0028] Figure 2 A schematic diagram of the structure of an upper limb movement classification and recognition device based on surface electromyography signals provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0030] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 The first embodiment of the upper limb action classification and recognition method based on surface electromyography signals in the embodiment of the present invention includes:

[0031] 101. Signal decomposition: decomposing the surface electromyography signal to obtain different sub-signals, wherein the signal decomposition is deployed on the edge device;

[0032] First, the original surface electromyography signal is preprocessed to remove noise and drift. On the edge device, the preprocessed signal is decomposed into different sub-signals using methods such as short-time Fourier transform (STFT) or wavelet transform. The decomposed sub-signals are transmitted to the cloud server.

[0033] In the present invention, due to the particularity of surface electromyographic signals, the length of the input signal is limited. When performing convolution operations, if the input signal is short, the following problems may be encountered when extracting features in the convolution layer:

[0034] Limited contextual information: Due to the short input signal, the network may have difficulty capturing longer-range contextual information when processing. This may cause the network to be unable to fully understand the relationship between different actions involved in the EMG signal, thus affecting recognition accuracy.

[0035] Overfitting: When the input signal is short, the neural network may easily fall into overfitting, that is, it performs well on the training set but performs poorly on the test set (new data). The reason for overfitting may be that the network is too complex or the training data is not enough to support the network to learn a stable model.

[0036] Insufficient feature extraction: Short input signals may cause the convolutional layers to be limited in extracting features. The convolutional layers extract local features by sliding over the input signal using a local receptive field. However, if the input signal is short, the number and diversity of local features may be limited, which may cause the network to be unable to fully learn a sufficiently rich feature representation.

[0037] Therefore, signal decomposition can decompose the original surface electromyography signal into different sub-signals, which helps to capture the signal information from multiple angles. At the same time, building independent 1D-CNN sub-models for different sub-signals can improve the adaptability and recognition accuracy of the model. Signal decomposition helps to reveal the potential patterns and structures in the signal, thereby improving the model's understanding of the signal characteristics.

[0038] The solution of the present invention achieves efficient recognition of upper limb movements by decomposing surface electromyographic signals and constructing independent 1D-CNN sub-models for each sub-signal. This method effectively extracts the local features of surface electromyographic signals while reducing the computational complexity, and is suitable for application scenarios with high real-time requirements. Completing classification and recognition on a cloud server can make full use of the powerful computing resources in the cloud and further improve recognition performance. Compared with the prior art, this solution is innovative in terms of signal decomposition, model construction and real-time performance.

[0039] Furthermore, more preferably, the signal decomposition of the present invention comprises the following steps:

[0040] Signal division: Divide the surface electromyographic signal into windows of equal length, each window contains a certain number of signal samples, and the windows will be used as nodes represented by the graph structure; Feature extraction: For each window, calculate the features of each window, and use the features as the feature vectors of the nodes; Calculate the distance between nodes: Calculate the distance between nodes between the feature vectors of each pair of nodes, and the distance between nodes is used to measure the similarity between nodes; Construct a graph structure: According to the distance between nodes, use the k-nearest neighbor method to determine the connection between nodes, and each node is only connected to its k nearest neighbor nodes; Graph weight setting: For connected nodes, assign weights according to the distance between the connected nodes, and the weight is the inverse of the distance. The weight reflects the similarity between nodes, thereby helping the graph convolutional neural network capture local information. Building an input layer: receiving the facial electromyography signal converted into a graph structure representation, the input data includes a node feature matrix and an adjacency matrix, each row of the node feature matrix corresponds to a feature vector of a node, and the adjacency matrix represents the connection and weight between nodes; Building a graph convolution layer: the graph convolution layer is used to update node features; the construction of the graph convolution layer includes: node feature propagation: multiplying the adjacency matrix with the node feature matrix to obtain a propagated node feature matrix. Feature transformation: multiplying the propagated node feature matrix with the weight matrix to obtain updated node features. Activation function: applying the activation function to the updated node features to complete the construction of the graph convolution layer. Building a global pooling layer: adding a global pooling layer to integrate the information of all nodes; building a fully connected layer: after the global pooling layer, adding one or more fully connected layers, the fully connected layer is used to integrate features and output the final classification result; building an output layer: the output of the last fully connected layer is connected to the output node with the number of categories, and the Softmax activation function is used to obtain the probability distribution of each category. Training the GCNN model: using the labeled surface electromyography signal data to train the GCNN model, learning an effective signal decomposition representation, and obtaining a trained GCNN model; Signal decomposition: using the trained GCNN model to decompose the input surface electromyography signal to obtain different sub-signals.

[0041] This scheme converts surface electromyography signals into graph structure representation, so that the model can capture the local structure and pattern in the signal; using graph convolutional neural network (GCNN) for feature learning instead of traditional convolutional neural network (CNN), GCNN can better capture local information in graph structure data. By building an end-to-end GCNN model, the input surface electromyography signal is directly decomposed, simplifying the signal processing process; using the global pooling layer to integrate the information of all nodes, the model can better process signals of different scales. By adopting this scheme, the surface electromyography signal can be effectively decomposed into different sub-signals, which can then be used to identify and analyze upper limb movements. In addition, the use of the GCNN model can better capture the local structure and pattern in the surface electromyography signal and improve the recognition accuracy.

[0042] 102. Model construction: for each of the sub-signals, construct an independent 1D-CNN sub-model, obtain fusion features according to the weight calculation of each of the sub-signals, and obtain the 1D-CNN model;

[0043] For each sub-signal, an independent 1D-CNN sub-model is constructed. 1D-CNN has a strong time domain feature extraction capability and can effectively capture the local features of surface electromyography signals. According to the weight calculation of each sub-signal, the features of each sub-model are fused to obtain the integrated 1D-CNN model. This helps to improve the adaptability of the model to different sub-signal features, thereby improving recognition accuracy.

[0044] As a more preferred technical solution, model construction includes the following steps:

[0045] For each sub-signal, multiple 1D-CNN sub-models are constructed, and each 1D-CNN sub-model independently learns the feature representation of the sub-signal;

[0046] The purpose of this step is to better capture the features in the sub-signals. Since different 1D-CNN sub-models may have different structures, hyperparameters, or initializations, they are able to learn the feature representations of sub-signals from multiple perspectives. This helps improve the generalization ability and recognition accuracy of the model.

[0047] Multiplying the weight of the sub-signal with the corresponding sub-model feature, adding the weighted features to obtain a fusion feature;

[0048] The purpose of assigning weights to sub-signals is to distinguish the importance of different sub-signals for upper limb movement recognition. This allows the model to focus more on sub-signal features with higher weights (i.e., more important for recognition) when fusing features. Multiply the weights of the sub-signals with the corresponding sub-model features, add the weighted features, and obtain the fused features: The purpose of this step is to integrate the features of all sub-signals so that the model can make full use of the information in each sub-signal. By multiplying the weights with the corresponding sub-model features and then adding them together, the features of different sub-signals can be fused in a targeted manner, thereby improving the overall recognition performance.

[0049] After the features of all the sub-signals are fused, one or more fully connected layers are added, and the output of the last fully connected layer is connected to the output nodes having the number of upper limb action categories;

[0050] Using the Softmax activation function, the probability distribution of each upper limb action category is obtained.

[0051] After the fusion features, add one or more fully connected layers, and the output of the last fully connected layer is connected to the output node with the number of upper limb action categories: The role of the fully connected layer is to integrate the feature representations of each sub-model, further extract global features and classify them. The number of nodes output by the fully connected layer should be equal to the number of upper limb action categories, so as to output the probability distribution of each action category. Use the Softmax activation function to obtain the probability distribution of each upper limb action category: The role of the Softmax activation function is to convert the node value output by the fully connected layer into a probability distribution, so that the value of each node is between 0 and 1 and the sum of all node values ​​is 1. This can intuitively represent the predicted probability of each upper limb action category, which is convenient for action classification.

[0052] In summary, this scheme achieves efficient recognition of upper limb movements by constructing multiple 1D-CNN sub-models and assigning weights to each sub-signal. This method effectively extracts local features of surface EMG signals while considering the importance of different sub-signals for recognition. This helps improve the generalization ability of the model and recognition accuracy.

[0053] As a more preferred technical solution, the weight of the sub-signal is obtained by the following steps:

[0054] Sub-signal feature extraction: extract features from each sub-signal; feature combination: combine the features of each sub-signal into a feature vector as the input of the weight calculation neural network; weight calculation neural network construction: build a small neural network to calculate the weight of each sub-signal; neural network training: use a labeled training data set to train the weight calculation neural network, and the optimization goal is to minimize the loss between the predicted weight and the true weight to obtain a trained weight calculation neural network; weight calculation deployment: deploy the trained weight calculation neural network on the edge device. When a new sub-signal is input, the sub-signal features are extracted and input into the weight calculation neural network to obtain the corresponding weight value.

[0055] Sub-signal feature extraction: Use 1D-CNN sub-model or other feature extraction methods to extract features from each sub-signal to obtain the feature vector of each sub-signal.

[0056] Feature combination: The feature vectors of each sub-signal are connected in sequence or combined in other ways into a total feature vector as the input of the weight calculation neural network.

[0057] Weight calculation neural network construction: Build a small neural network, such as a multi-layer perceptron (MLP), to calculate the weights of each sub-signal. The input layer of the neural network receives the total feature vector, and the number of nodes in the output layer is equal to the number of sub-signals. The activation function of the output layer can be selected as Softmax to ensure that the weight value is between 0 and 1 and the sum is 1.

[0058] Neural network training: Use labeled training data sets to train weight calculation neural networks. The optimization goal is to minimize the loss between the predicted weights and the true weights, such as the mean square error (MSE) loss. During the training process, the neural network parameters are adjusted through optimization algorithms such as gradient descent until the training converges to obtain a trained weight calculation neural network.

[0059] Weight calculation deployment: The trained weight calculation neural network is deployed on the edge device. When a new sub-signal is input, the sub-signal features are extracted and input into the weight calculation neural network to obtain the corresponding weight value.

[0060] The purpose of constructing a weighted neural network is to automatically learn the importance of each sub-signal in the upper limb movement recognition process and assign weights to each sub-signal in real time. Through the weighted neural network, the characteristics of different sub-signals can be fused in a targeted manner to improve the overall recognition performance. At the same time, the weighted neural network can adaptively adjust the weight allocation, so that the model can maintain a high recognition accuracy in different situations.

[0061] 103. Model training: Divide the training data set into a training set and a validation set, train the 1D-CNN model, and obtain a trained 1D-CNN model.

[0062] The expanded training data set is divided into a training set and a validation set. Each sub-model is trained using the training set, and performance is evaluated on the validation set, with hyperparameters adjusted to optimize model performance. After the training is completed, a trained 1D-CNN model is obtained.

[0063] 104. Classification and recognition: Perform the same signal decomposition on the newly input original surface electromyography signal, and use the trained 1D-CNN model to classify and recognize the upper limb movements. The classification and recognition is completed on the cloud server.

[0064] The newly input raw surface electromyography signal is preprocessed and decomposed, the same as the training stage. The trained 1D-CNN model is used to classify and recognize the upper limb movements of the decomposed sub-signals. The classification and recognition are completed on the cloud server, and the recognition results are returned to the edge device.

[0065] 1D-CNN has a strong time domain feature extraction capability and can effectively capture local features in surface electromyography signals. At the same time, compared with other models (such as fully connected neural networks, recurrent neural networks, etc.), 1D-CNN has lower computational complexity and is more suitable for application scenarios with higher real-time requirements.

[0066] The solution of the present invention has the following technical advantages:

[0067] Edge computing: By performing signal decomposition on the surface electromyography signal on the edge device, the original signal is decomposed into different sub-signals. This design not only reduces the computing burden of the cloud server, but also improves the real-time performance and response speed of the system.

[0068] Independent 1D-CNN sub-model: An independent 1D-CNN sub-model is constructed for each sub-signal. This design helps capture the characteristics of each sub-signal. The characteristics of each sub-signal are fused through weight calculation to obtain an integrated 1D-CNN model, thereby improving the accuracy of upper limb movement recognition.

[0069] Integration of edge computing and cloud computing: The upper limb movement classification and recognition of newly input surface electromyography signals are completed on the cloud server. This design enables the system to reduce the computing pressure of edge devices while ensuring real-time performance. In addition, the cloud server has more powerful computing power, which is conducive to improving recognition accuracy.

[0070] See also Figure 2 Another embodiment of the upper limb action classification and recognition device based on surface electromyography signals in the embodiment of the present invention includes:

[0071] 201. Signal decomposition module: used to decompose the surface electromyography signal to obtain different sub-signals, and the signal decomposition module is deployed on the edge device;

[0072] 202. Model construction module: used to construct an independent 1D-CNN sub-model for each sub-signal, obtain fusion features according to the weight calculation of each sub-signal, and obtain the 1D-CNN model;

[0073] 203. Model training module: used for dividing the training data set into a training set and a validation set, training the 1D-CNN model, and obtaining a trained 1D-CNN model;

[0074] 204. Classification and recognition module: used to perform the same signal decomposition on the newly input original surface electromyography signal, and use the trained 1D-CNN model to classify and recognize the upper limb movements. The classification and recognition is completed on the cloud server.

[0075] The electronic device of the present invention may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) (for example, one or more processors) and memories, and one or more storage media for storing applications or data (for example, one or more mass storage devices). Among them, the memory and the storage medium can be short-term storage or permanent storage. The program stored in the storage medium may include one or more modules, each of which may include a series of instruction operations in the electronic device. Furthermore, the processor can be configured to communicate with the storage medium and execute a series of instruction operations in the storage medium on the electronic device.

[0076] The electronic device may also include one or more power supplies, one or more wired or wireless network interfaces, one or more input and output interfaces, and / or one or more operating systems, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc.

[0077] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of a method for classifying and identifying upper limb movements based on surface electromyography signals.

[0078] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0079] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer,

[0080] The server, or network device, etc.) performs all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0081] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. 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 method for upper limb movement classification and recognition based on surface electromyography signals, characterized in that: The upper limb action classification and recognition method based on surface electromyography signals comprises: Signal decomposition: decomposing the surface electromyography signal to obtain different sub-signals, and the signal decomposition is deployed on the edge device; Model construction: for each sub-signal, an independent 1D-CNN sub-model is constructed, and the fusion features are obtained according to the weight calculation of each sub-signal to obtain the 1D-CNN model; Model training: dividing the training data set into a training set and a validation set, training the 1D-CNN model, and obtaining a trained 1D-CNN model; Classification and recognition: The same signal decomposition is performed on the newly input raw surface electromyography signal, and the trained 1D-CNN model is used to classify and recognize the upper limb movements. The classification and recognition is completed on the cloud server.

2. The upper limb movement classification and recognition method based on surface electromyography signals according to claim 1 is characterized in that: The model construction includes: For each sub-signal, multiple 1D-CNN sub-models are constructed, and each 1D-CNN sub-model independently learns the feature representation of the sub-signal; Multiplying the weight of the sub-signal with the corresponding sub-model feature, adding the weighted features to obtain a fusion feature; After the features of all the sub-signals are fused, one or more fully connected layers are added, and the output of the last fully connected layer is connected to the output nodes having the number of upper limb action categories; Using the Softmax activation function, the probability distribution of each upper limb action category is obtained.

3. The upper limb movement classification and recognition method based on surface electromyography signals according to claim 2 is characterized in that: The weights of the sub-signals are obtained by the following steps: Sub-signal feature extraction: extract features from each sub-signal; Feature combination: The features of each sub-signal are combined into a feature vector as the input of the weight calculation neural network; Weight calculation neural network construction: Build a small neural network to calculate the weight of each sub-signal; Neural network training: Use labeled training data sets to train the weight calculation neural network. The optimization goal is to minimize the loss between the predicted weight and the true weight to obtain a trained weight calculation neural network. Weight calculation deployment: The trained weight calculation neural network is deployed on the edge device. When a new sub-signal is input, the sub-signal features are extracted and input into the weight calculation neural network to obtain the corresponding weight value.

4. The upper limb movement classification and recognition method based on surface electromyography signals according to claim 1 is characterized in that: The signal decomposition step comprises: Signal conversion: converting the surface electromyographic signal into a graph structure, wherein each surface electromyographic signal is regarded as a node, and the connection between the nodes is constructed according to the similarity or distance between the surface electromyographic signals, and the connection between the nodes is determined using a graph construction method; Build a graph convolutional neural network (GCNN): The GCNN captures the local structure and long-distance dependencies between signals. The GCNN consists of multiple graph convolutional layers and activation functions. Each graph convolutional layer can update node features. The GCNN model is deployed on edge devices. Training the GCNN model: using the labeled surface electromyography signal data to train the GCNN model, learning an effective signal decomposition representation, and obtaining a trained GCNN model; Signal decomposition: Use the trained GCNN model to decompose the input surface electromyography signal to obtain different sub-signals.

5. The upper limb movement classification and recognition method based on surface electromyography signals according to claim 4 is characterized in that: The signal conversion step comprises: Signal partitioning: Divide the surface electromyographic signal into windows of equal length, each window containing a certain number of signal samples, and the windows will be used as nodes for graph structure representation; Feature extraction: For each window, calculate the features of each window, and use the features as the feature vector of the node; Calculating the inter-node distance: calculating the inter-node distance between the feature vectors of each pair of nodes, wherein the inter-node distance is used to measure the similarity between nodes; Constructing a graph structure: according to the distance between the nodes, the connection between the nodes is determined using the k-nearest neighbor method, and each node is only connected to its k nearest neighbor nodes; Graph weight setting: For connected nodes, weights are assigned according to the distance between the connected nodes. The weight is the inverse of the distance and reflects the similarity between nodes, thereby helping the graph convolutional neural network capture local information.

6. The upper limb movement classification and recognition method based on surface electromyography signals according to claim 5 is characterized in that: The construction of the graph convolutional neural network includes: Build the input layer: receive the facial electromyography signal converted into a graph structure representation, the input data includes a node feature matrix and an adjacency matrix, each row of the node feature matrix corresponds to a feature vector of a node, and the adjacency matrix represents the connection and weight between nodes; Building a graph convolution layer: The graph convolution layer is used to update node features; Build a global pooling layer: Add a global pooling layer to integrate the information of all nodes; Building a fully connected layer: After the global pooling layer, add one or more fully connected layers, which are used to integrate features and output the final classification results; Build the output layer: The output of the last fully connected layer is connected to the output node with the number of categories, and the Softmax activation function is used to obtain the probability distribution of each category.

7. The upper limb movement classification and recognition method based on surface electromyography signals according to claim 6 is characterized in that: The construction of the graph convolution layer includes: Node feature propagation: multiply the adjacency matrix by the node feature matrix to obtain a propagated node feature matrix. Feature transformation: multiply the propagated node feature matrix by the weight matrix to obtain updated node features. Activation function: Apply the activation function to the updated node features to complete the construction of the graph convolution layer.

8. An upper limb movement classification and recognition device based on surface electromyography signals, characterized in that: include: Signal decomposition module: used to decompose the surface electromyography signal to obtain different sub-signals. The signal decomposition module is deployed on the edge device; Model construction module: used for constructing an independent 1D-CNN sub-model for each sub-signal, obtaining fusion features according to the weight calculation of each sub-signal, and obtaining a 1D-CNN model; Model training module: used for dividing the training data set into a training set and a validation set, training the 1D-CNN model, and obtaining a trained 1D-CNN model; Classification and recognition module: used to perform the same signal decomposition on the newly input original surface electromyography signal, and use the trained 1D-CNN model to classify and recognize upper limb movements. The classification and recognition is completed on the cloud server.

9. An electronic device, comprising a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the electronic device executes the various steps of the upper limb movement classification and recognition method based on surface electromyography signals as described in any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the various steps of the upper limb movement classification and recognition method based on surface electromyography signals as described in any one of claims 1 to 7 are implemented.

Images