Electromyographic signal data generation method based on deep learning

Abstract

The invention discloses an electromyographic signal data generation method based on deep learning, and belongs to the field of deep learning. The method comprises the following steps: firstly, building a confrontation model framework comprising a depth angle sequence encoder, a context encoder, an electromyographic signal generator and a discriminator, and placing the confrontation model framework on a hand joint of a user to generate electromyographic signal data; wherein the depth angle sequence encoder is used for encoding a hand joint angle sequence control signal, inputting the hand joint angle sequence control signal into the context encoder to generate potential information of an electromyographic signal, and generating a corresponding electromyographic signal by using the electromyographic signal generator; the electromyographic signal discriminator judges whether each time step accords with a real electromyographic signal mode or not; and then, training an electromyographic signal classification network based on a convolutional neural network by using the electromyographic signal data, and deploying the electromyographic signal classification network to an actual embedded device or a mobile terminal to perform real-time gesture recognition. According to the method, an existing electromyographic signal data set is effectively expanded, meanwhile, the sequence information is used as a control signal, and the data generation quality is effectively improved.

Description

Technical Field

[0001] The present invention belongs to the field of deep learning, and particularly relates to a method for generating electromyogram signal data based on deep learning. Background Art

[0002] Gestures play an important role in non-verbal communication. Traditional methods use cameras and optical motion capture devices to depict the movement trajectories of the human body, and then recognize gesture actions; surface electromyography (sEMG) signals provide another solution for the recognition of movement intentions including gestures.

[0003] sEMG is a bioelectric signal obtained by placing electromyography electrodes on the surface of the human skin, which is directly related to human behavior intentions. Even for patients with reduced limb muscle function and unable to control muscle contractions, the gesture action intentions can still be mapped by collecting the sEMG of specific muscles.

[0004] In recent years, sEMG has been widely applied in fields such as prosthetic control, medical devices, and human-computer interaction. Analyzing the collected sEMG using artificial intelligence algorithms can better help disabled people independently complete some daily interactions. Due to rich information, non-invasive to the human body, and easy access to acquisition devices, gesture action recognition based on sEMG has gradually become a research hotspot, and its core is to accurately distinguish sEMG signals of different gestures.

[0005] Adopting appropriate electromyography devices and preprocessing methods can effectively improve the gesture recognition rate, but more importantly, it is still the selection of the classifier. Traditional machine learning models have been maturely applied in this field, taking the extracted signal features as the input of the model, but overly complex feature sets will lead to the loss of sEMG information.

[0006] Common traditional machine learning models include linear discriminant analysis (LDA), artificial neural network (ANN), support vector machine (SVM), k-nearest neighbor (KNN), and naive Bayes (NB), etc.

[0007] Deep learning models are a special branch of machine learning. Different from traditional machine learning that relies on manual feature extraction, they can take the original sEMG signal as the model input and allow feature extraction and model construction to be carried out simultaneously. Commonly used deep learning models include convolutional neural network (CNN), long short-term memory (LSTM), temporal convolutional network (TCN), gated recurrent unit (GRU), and deep belief network (DBN), etc. Compared with traditional machine learning, deep learning has a high recognition accuracy, but requires a large amount of data and a long training time.

[0008] In the field of gesture recognition, although gesture recognition based on surface electromyogram (sEMG) signals can achieve good results under certain experimental conditions, it still faces some challenges in practical applications, especially in terms of the generalization ability and transfer ability of the model. The relative scarcity of surface electromyogram signal datasets, as well as factors such as electrode patch positions and individual user differences, make the performance of gesture recognition systems based on surface electromyogram signals often greatly affected under different users and new gesture postures; specifically as follows:

[0009] 1) Scarcity of the dataset and model generalization ability;

[0010] Currently, surface electromyogram signal datasets are still relatively few, especially limited in terms of diversity and scale. Most existing surface electromyogram signal datasets often only cover a limited number of gesture types and a small sample size of subjects, resulting in insufficient diversity of the dataset when training the model and making it difficult to handle diverse gestures and user characteristics in practical applications. Due to the existence of physiological differences among users (such as muscle size, skin characteristics, etc.), the generalization ability is poor when facing different users, and the recognition accuracy is also prone to decline.

[0011] In addition, surface electromyogram signals have strong individual differences. Even for the same person, the myoelectric signals generated during gesture operations at different times and in different environments may vary greatly.

[0012] The lack of sufficient and diverse training data makes it difficult for the model to accurately adapt to these changes, thus affecting its performance under different users or new postures.

[0013] 2) Poor transfer ability under new gesture postures

[0014] Another challenge of surface electromyography (sEMG) signals is the impact of posture changes on recognition performance. Gesture recognition not only requires accurate recognition of specific gestures but also the ability to adapt to posture changes. Traditional sEMG-based recognition methods usually have weak transfer ability when facing new gesture postures (such as different gesture angles, tension of hand muscles, etc.). Because new postures may cause significant changes in EMG signals, or even be completely different from the signals in the training dataset.

[0015] Due to the limited variety of postures in the dataset, the model fails to be fully trained under diverse postures. Therefore, when facing unseen postures, the transfer ability of the model is significantly insufficient, leading to a substantial drop in recognition accuracy. To improve the performance of the model in new postures, a large amount of additional training data is usually required, which is often infeasible in practical applications.

[0016] 3) Adaptation problems caused by user differences

[0017] Differences among users (such as muscle structure, skin conductivity, movement habits, etc.) directly affect the characteristics of sEMG signals. Therefore, the model often performs well within the user group of the training dataset, but when facing new users, due to individual differences, the performance will significantly decline. Traditional methods often rely on a large amount of labeled individual data for training, and the collection of labeled data is very expensive and time-consuming, making it difficult for the model to handle user differences.

[0018] Existing transfer learning-based methods attempt to transfer the model trained on one user or dataset to another user. However, they still face problems such as scarce training samples and excessive individual differences, and the transfer effect is often not ideal, especially in the case of lack of sufficient diverse data support.

[0019] To improve the performance of sEMG signals in practical applications, improvements need to be made in multiple directions:

[0020] First, increase the diversity and scale of the dataset. By collecting more sEMG signals from different users and under different postures, the generalization ability and transfer ability of the model can be effectively enhanced. In addition, enhancing the adaptability of the model, especially developing training strategies that can adapt to individual user differences, such as personalized transfer learning methods, can also significantly improve the performance in practical scenarios.

[0021] Although existing methods attempt to solve the above problems, the scarcity, individual differences, and posture change problems of sEMG signals are still the main challenges faced by current technologies. Therefore, in applications of gesture recognition based on sEMG signals, more research and technological innovation are needed to improve its stability and accuracy under various complex environments and user differences. Summary of the Invention

[0022] In view of the problems faced by existing surface electromyogram signal (sEMG)-based gesture recognition methods in practical applications, such as scarce data and poor model generalization ability, the present invention provides a deep learning-based electromyogram signal data generation method, which can effectively expand the existing electromyogram signal dataset and contribute to solving the problem of scarce data and improving the model generalization ability. At the same time, the present invention uses sequence information as a control signal, which can effectively improve the quality of data generation and reduce the optimization difficulty.

[0023] The deep learning-based electromyogram signal data generation method specifically comprises the following steps:

[0024] Step 1: Build a generative adversarial model framework and place it on the user's hand joints to generate electromyogram signal data;

[0025] The generative adversarial model framework includes a deep angle sequence encoder, a context encoder, an electromyogram signal generator, and an electromyogram signal discriminator;

[0026] Step 101: The deep angle sequence encoder is used to encode the hand joint angle sequence control signal and input it into the context encoder;

[0027] Step 102: At each time step t, the GRU unit used in the lower layer of the context encoder receives the concatenation of the electromyogram signal s t and Gaussian noise ∈ t and outputs the feature vector i t and the gate signal g t :

[0028] i t ,g t = GRU(s t ,∈ t ,g t-1 ),

[0029] Step 103: The upper layer of the context encoder uses the Ang2Gist unit to combine the output vector i t and the story context h t to generate the output Gist vector o t , encoding and generating the latent information of the electromyogram signal;

[0030] The output Gist vector o t is:

[0031] o t ,h t = Ang2Gist(i t ,h t-1 ),

[0032] Step 104: The story context ht Updated by the Ang2Gist unit to reflect changes in potential context information;

[0033] The specific update formula is as follows:

[0034]

[0035] r t = σ r (W r i t + U r h t-1 + b r ),

[0036] h t = (1 - z t ) ⊙ h t-1 + z t ⊙ σ h (W h i t + U h h t-1 + b h ),

[0037] o t = Filter(i t ) * h t

[0038] U z , is the first layer of learnable network parameters in Ang2Gist, used to encode the feature vectors i t and h t-1 inputs and output the intermediate layer feature vector At the same time, Ang2Gist also contains another set of learnable network parameters σ r , W r , U r , b r to encode the feature vectors i t and h t-1 inputs, and then output the intermediate layer feature vector r t ; σ h , W h , U h , b h are used to encode the intermediate feature vector and r t , and output the h at the next moment t ; i t is convolved and multiplied by h t to obtain the Gist vector o t .

[0039] Step 105: Output the potential information to the electromyogram signal generator to generate corresponding electromyogram signals;

[0040] The electromyogram signal generator uses a deep convolutional network to gradually upsample according to the Gist vector o at each time step t to generate electromyogram signals t Step 106: The electromyogram signal discriminator measures the generated electromyogram signals

[0041] and the real electromyogram signal s to obtain the matching degree m at a given initial hand joint electromyogram signal information h t for discriminating whether each time step o conforms to the real electromyogram signal pattern: t

[0042]

[0043]

[0044] Step 107: The electromyogram encoder encodes the hand joint angle sequence into a feature vector. After concatenating the matching degrees corresponding to all time steps, the global consistency score is calculated through a fully connected layer and a sigmoid function;

[0045] P = FC(sigmoid([m 0 , m 1 , …, m t ))

[0046] FC is the fully connected layer, and P is the obtained classification probability;

[0046] The calculation formula of the MSE error is:

[0047] The calculation formula of the GAN adversarial loss is:

[0048] When the MSE error is lower than 0.01, the signal feature distribution matching degree is high, and the GAN generator and discriminator are in a certain state, it indicates that the signals are similar enough.

[0049] Step 2: Use the generated electromyogram signal data to train the electromyogram signal classification network;

[0050] First, a 10 Hz high-pass filter is used to remove the low-frequency noise components in the electromyogram signal data and extract the high-frequency noise components of the electromyogram signals;

[0051] Then, the square root algorithm is used to further optimize the electromyogram signal data of the high-frequency noise components.

[0052] Finally, an electromyogram (EMG) signal classification network is constructed based on a convolutional neural network (CNN). The optimized EMG signal data is used for training to extract spatio-temporal features in the signals and predict the corresponding gesture category sequences. Through continuous training, the EMG signal classification network gradually learns the mapping relationship between EMG signals and gestures, achieving high-accuracy gesture recognition.

[0053] Step 3: Deploy the trained EMG signal classification network based on a convolutional neural network (CNN) to an actual embedded device or a mobile terminal for real-time gesture recognition.

[0054] The advantages of the present invention are as follows:

[0055] 1) The method for generating EMG signal data based on deep learning in the present invention enhances the diversity of the data set: It can control the generation of EMG signal data for new gestures. This controllable generation method can not only effectively expand the existing EMG signal data set, but also further improve the generalization ability of the model by generating signal data with diversity. It provides an effective solution to solve the problem of data scarcity and improve the robustness of the model.

[0056] 2) The method for generating EMG signal data based on deep learning in the present invention improves the data generation quality by using sequence information: Different from traditional EMG signal generation methods, the present invention can generate more accurately realistic EMG signals by using the hand joint angle sequence as a control signal. The introduction of sequence information can not only ensure the quality of data generation, but also help reduce the optimization difficulty, thereby improving the training efficiency.

[0057] 3) The method for generating EMG signal data based on deep learning in the present invention adopts a generative adversarial model (GAN) framework. The hand joint angle sequence is encoded by a deep angle sequence encoder, and the EMG signal is generated by a generator. The discriminator is used to improve the quality and consistency of the generated signal, ensuring that the generated EMG signal is closer to the real signal and enhancing the reliability of the model.

[0058] 4) The method for generating EMG signal data based on deep learning in the present invention can effectively solve the overfitting problem caused by insufficient samples in traditional methods by synthesizing a large amount of high-quality EMG signal data, and further improve the generalization ability of the gesture recognition model. This enables the present invention to obtain good recognition performance in various complex scenarios.

[0059] 5) The method for generating EMG signal data based on deep learning in the present invention is not only applicable to the field of EMG signal recognition, but also provides new ideas for other applications based on signal generation. Especially in machine learning tasks that require a large amount of data for training, high-quality samples can be synthesized through a generative adversarial network, expanding the extensiveness of its applications. Brief Description of the Drawings

[0060] Figure 1 is a flowchart of a method for generating electromyogram (EMG) signal data based on deep learning according to the present invention;

[0061] Figure 2 is a schematic diagram of generating EMG signal data by building a generative adversarial model framework according to the present invention. Detailed Embodiments

[0062] The EMG signal generation system based on a generative adversarial network provided by the present invention will be further described in detail below with reference to the accompanying drawings and examples.

[0063] A method for generating EMG signal data based on deep learning according to the present invention converts noise into synthetic EMG data through a generative adversarial network; meanwhile, a hand joint angle sequence is used as a generation control signal to control the generation of EMG data for a new gesture. The EMG data for the new gesture can enrich the dataset, so as to train the classifier more fully and improve the generalization ability.

[0064] The method for generating EMG signal data based on deep learning, as Figure 1 shown, the specific steps are as follows:

[0065] Step 1: Build a generative adversarial model framework and place it on the user's hand joints to generate EMG signal data;

[0066] As Figure 2 shown, the generative adversarial model framework is based on sequence conditions and is unique in that it includes a joint angle sequence encoder (Angle Encoder) for dynamically tracking the hand joint angle sequence signal to control data generation, and at the same time, an EMG signal generator and a discriminator are configured to enhance the quality and consistency of the generated sequence.

[0067] Different from previous EMG signal generation methods, the present invention uses the hand joint angle sequence as a control signal to generate the corresponding EMG signal data for controlling the generation of a new gesture;

[0068] First, a deep angle sequence encoder is used to encode the hand joint angle sequence control signal, and the context encoder is input to dynamically generate the latent information of the EMG signal, and output to the EMG signal generator, and the corresponding EMG signal data is generated according to the control signal output by the context encoder; finally, the EMG signal discriminator measures the quality and consistency of the generated EMG signal data to ensure the similarity between the generated signal and the real EMG signal.

[0069] The specific steps are as follows:

[0070] Step 101: The depth angle sequence encoder is used to encode the hand joint angle sequence control signal and input it into the context encoder to generate the latent information of the electromyogram signal;

[0071] Joint Angle Sequence Encoder (Angle Encoder): Encodes the entire hand joint sequence into a low-dimensional vector h 0 , as the initial hidden state of the context encoder; uses two neural networks to implement the mean μ(S) and variance Σ(S), and samples h from the normal distribution N(μ(S), Σ(S)) 0 .

[0072] The hand joint angle sequence is encoded into a low-dimensional vector through the reparameterization trick and used as the initial hidden state of the control signal. The reparameterization trick can be expressed by the following formula, and the goal is to make the random variable differentiable:

[0073] Original sampling (non-differentiable):

[0074]

[0075] Sampling after reparameterization: Introduce a random noise independent of the parameters

[0076]

[0077] Among the above parameters: The randomness of ε is independent of the parameters μ and σ. The generation process of becomes a deterministic calculation (μ and σ are the parameters output by the network)

[0078] The encoded hand joint angle h 0 can be written as h 0 = μ(S) + σ 2 (S) 1 / 2 ⊙ ∈ s , where ∈ s ~N(0, I).

[0079] The context encoder consists of two layers of recurrent neural networks (RNNs). The lower layer uses standard GRU units, and the upper layer uses the newly proposed Ang2Gist units.

[0080] Step 102: At each time step t, the GRU units used in the lower layer of the context encoder receive the concatenation of the electromyogram signal s t and the Gaussian noise ∈ t and output the feature vector i t and the gating signal g t :

[0081] i t , g t = GRU(st , ∈ t , g t-1 ),

[0082] Step 103: The upper layer of the context encoder uses the Ang2Gist unit to combine the output vector i t and the story context h t to generate the output Gist vector o t , encoding the potential information of the myoelectric signal;

[0083] The output Gist vector o t is:

[0084] o t , h t = Ang2Gist(i t , h t-1 ),

[0085] Step 104: The story context h t is updated through the Ang2Gist unit to reflect the change in the potential context information;

[0086] The specific update formula is as follows:

[0087]

[0088] r t = σ r (W r i t + U r h t-1 + b r ),

[0089] h h = (1 - z t ) ⊙ h t-1 + z t ⊙ σ h (W h i t + U h h t-1 + b h ),

[0090] o t = Filter(i t ) * h t

[0091] U z , is the first layer of learnable network parameters in Ang2Gist, used to encode the feature vectors i t and h t-1Input and output the intermediate layer feature vector z t ; At the same time, Ang2Gist also contains another set of learnable network parameters σ r , W r , U r , b r to encode the feature vectors i t and h t-1 as input, and then output the intermediate layer feature vector r t ; σ h , W h , U h , b h are used to encode the intermediate feature vectors and r t , and output the h at the next moment t ; i t is multiplied by the convolution of h t to obtain the Gist vector o t .

[0092] The potential information includes:

[0093] 1. Motion dynamics features: acceleration of joint movement, jitter pattern.

[0094] 2. Muscle co-activation pattern: activation timing and intensity ratio of different muscle groups.

[0095] 3. Individual adaptation parameters: influence of muscle stiffness and subcutaneous fat layer thickness on signal attenuation.

[0096] 4. Task semantic encoding: implicit identification of action type (fist clenching / extension).

[0097] Step 105, output the potential information to the EMG signal generator to generate the corresponding EMG signal;

[0098] The EMG signal generator (EMGs Generator) uses a deep convolutional network to generate the EMG signal t according to the Gist vector o at each time step t implemented using a deep convolutional network, and gradually upsample to generate a high-density EMG signal.

[0099] o t = G(o t )

[0100] It contains the following physical features:

[0101] 1. Time domain features: including waveform amplitude (μV level), pulse width (50 - 200 ms), and superposition form of motor unit action potential (MUAP);

[0102] 2. Frequency domain features: including energy concentrated in 20 - 500 Hz, peak frequency depending on muscle type, and high-frequency component attenuation characteristics (reflecting the signal propagation path);

[0103] 3. Spatial features: spatial correlation of multi-channel signals (such as topological consistency of electrode arrays) and temporal coordination patterns across muscle groups.

[0104] Step 106, the EMG signal discriminator measures the generated EMG signal against the real EMG signal s t , and at a given initial hand joint EMG signal information h o , the matching degree m t is used to determine whether each time step conforms to the real EMG signal pattern:

[0105] The EMG signal discriminator ensures the similar features between the generated signal and the real EMG signal by evaluating the matching degree between the generated EMG signal and the control signal and the global consistency of the signal.

[0106] The discriminator evaluates both global consistency (global signal) and local consistency (frame-level signal) simultaneously:

[0107]

[0108] D(x) = D glocal (f(x)) + D local (x t )

[0109] In the EMG task:

[0110] The global discriminator D global : determines whether the entire generated EMG signal sequence matches the motion intention.

[0111] The local discriminator D local : discriminates whether each time step conforms to the real EMG signal pattern.

[0112] Step 107, the EMG encoder encodes the hand joint angle sequence into a feature vector. After concatenating the matching degrees corresponding to all time steps, the global consistency score is calculated through a fully connected layer and a sigmoid function;

[0113] P = FC(sigmoid([m 0 , m 1 , …, m t ))

[0114] FC is the fully connected layer, and P is the obtained classification probability;

[0115] The formula for calculating the MSE error is as follows:

[0116] The formula for calculating the GAN adversarial loss is as follows:

[0117] When the MSE error is lower than 0.01, the signal feature distribution has a high degree of matching, and the GAN generator discriminator indicates that the signals are sufficiently similar.

[0118] Step 2: Use the generated electromyogram signal data to train the electromyogram signal classification network;

[0119] To reduce noise data and enhance effective electromyogram signals, preprocess the generated electromyogram signal data, including: 1) High-pass filtering: Use a 10 Hz high-pass filter to remove the low-frequency noise components in the generated signals, thereby extracting the high-frequency components of the electromyogram signals and improving the effectiveness of the signals; 2) Square root algorithm optimization: Use the square root algorithm to further optimize the electromyogram signals of the preprocessed high-frequency noise components, improving the quality and classification effect of the signals;

[0120] Then, build an electromyogram signal classification network based on a convolutional neural network (CNN), and use the optimized electromyogram signal data for training. This network can effectively extract the spatio-temporal features in the signals and predict the corresponding gesture category sequences. Through continuous training, the electromyogram signal classification network gradually learns the mapping relationship between electromyogram signals and gestures, achieving high-accuracy gesture recognition.

[0121] Training process: Use the Adam optimizer to update the parameters of the depth angle sequence encoder, context encoder, electromyogram signal generator, and electromyogram signal discriminator. The parameters of the electromyogram signal discriminator are updated in two for loops, while the parameters of the electromyogram generator are updated in both loops. Use different mini-batch sizes and different time steps to update the generator and discriminator to accelerate the training convergence.

[0122] Step 3: Deploy the trained electromyogram signal classification network based on a convolutional neural network (CNN) to an actual embedded device or mobile terminal for real-time gesture recognition.

[0123] Deploy the trained electromyogram signal classification network to an actual application environment, such as an embedded device or a mobile terminal, for real-time gesture recognition. This system can be widely applied in fields such as intelligent prosthetic control, brain-computer interfaces, and rehabilitation medicine to help users achieve more natural and accurate gesture control.

[0124] The myoelectric signal data generation system described in the present invention expands the existing myoelectric signal dataset by synthesizing high-quality myoelectric signal data, thereby improving the generalization ability and robustness of the gesture recognition model. It is applicable to the gesture recognition task, and by generating myoelectric signals consistent with real data, it improves the recognition accuracy of the model under different users and new gesture postures. The generative adversarial network framework can generate diverse myoelectric signal data, support the adaptation to user individual differences and gesture changes, thereby improving the model's transfer ability and recognition accuracy.

Claims

1. A method for generating electromyographic signal data based on deep learning, characterized in that: The specific steps are as follows: Step 1: Build a generative adversarial model framework and place it on the user's hand joints to generate electromyographic signal data; The generative adversarial model framework includes a deep angle sequence encoder, a context encoder, an electromyographic signal generator and an electromyographic signal discriminator; Step 101, a depth angle sequence encoder is used to encode a hand joint angle sequence control signal and input the signal into a context encoder; Step 102, at each time step t, the GRU unit used in the lower layer of the context encoder receives the electromyographic signal s t and Gaussian noise ∈ t The concatenation of the output feature vector i t and the gating signal g t : i t ,g t =GRU(s t ,∈ t ,g t-1 ), Step 103, the upper layer of the context encoder uses the Ang2Gist unit, combined with the output vector i t and story context t Generate output Gist vector o t , encoding the potential information of generating EMG signals; Output Gist vector o t for: o t ,h t =Ang2Gist(i t ,h t-1 ), Step 104, story context t Updated by Ang2Gist unit to reflect changes in underlying context information; The specific update formula is as follows: z t =σ z (W z i t +U z h t-1 +b z ), r t =σ r (W r i t +U r h t-1 +b r ), h t =(1-z t )⊙h t-1 +z t ⊙σ h (W h i t +U h h t-1 +b h ), o t =Filter(i t )*h t σ z ,W z ,U z ,b z is the first layer of learnable network parameters in Ang2Gist, used to encode the feature vector i t and h t-1 Input and output the intermediate layer feature vector z t ; At the same time, Ang2Gist also contains another set of learnable network parameters σ r ,W r ,U r ,b r Decoding is used to encode the feature vector i t and h t-1 Input, then output the intermediate layer feature vector r t ; σ h ,W h ,U h ,b h Used to encode the intermediate feature vector z t and r t , output the h of the next moment t ;i t After convolution and h t Multiply to get Gist vector o t ; Step 105: Output the potential information to the electromyographic signal generator to generate the corresponding electromyographic signal Step 106, the electromyographic signal discriminator measures the generated electromyographic signal Compared with the real electromyographic signal t , given the initial hand joint electromyographic signal information h o The matching degree m t , used to determine each time step Whether it conforms to the real electromyographic signal mode: Step 107, the myoelectric encoder encodes the hand joint angle sequence into a feature vector, concatenates the matching degrees corresponding to all time steps, and then calculates the global consistency score through a fully connected layer and a sigmoid function; P=FC(sigmoid([m0,m1,…,m t ])) FC is the fully connected layer, and P is the obtained classification probability; Step 2: Use the generated electromyographic signal data to train the electromyographic signal classification network; Step 3: Deploy the trained convolutional neural network (CNN)-based electromyographic signal classification network to actual embedded devices or mobile terminals for real-time gesture recognition.

2. A method for generating electromyographic signal data based on deep learning as claimed in claim 1, characterized in that: In step 105, the electromyographic signal generator uses a deep convolutional network to generate the Gist vector o at each time step t. t Generate electromyographic signals by stepwise upsampling 3. A method for generating electromyographic signal data based on deep learning as claimed in claim 1, characterized in that: In step 107, the MSE error calculation formula is: The GAN adversarial loss calculation formula is: When the MSE error is less than 0.01 and the GAN generator discriminator When Compared with the real electromyographic signal t Match.

4. A method for generating electromyographic signal data based on deep learning as claimed in claim 1, characterized in that: The step 2 is specifically as follows: First, a 10 Hz high-pass filter was used to remove the low-frequency noise components in the EMG signal data and extract the high-frequency noise components of the EMG signal; Then, the square root algorithm is used to further optimize the EMG signal data of high-frequency noise components; Finally, an EMG signal classification network is constructed based on a convolutional neural network. The optimized EMG signal data is used for training to extract the spatiotemporal features in the signal and predict the corresponding gesture category sequence. Through continuous training, the EMG signal classification network gradually learns the mapping relationship between EMG signals and gestures, achieving high-accuracy gesture recognition.

Images