Electromyographic signal identification method, device and equipment

Abstract

The invention provides an electromyographic signal recognition method, device and equipment, which can be applied to the technical field of electromyographic signal decoding. The method comprises the following steps: acquiring electromyographic signals of a plurality of channels, and preprocessing the electromyographic signals of the plurality of channels to obtain an electromyographic data set; feature extraction is carried out on the myoelectricity data set to obtain a myoelectricity feature set, and the myoelectricity feature set comprises a plurality of time domain features; performing feature extraction and spatial mapping on the plurality of time domain features based on a target electromyographic signal recognition model to obtain a multi-dimensional initial vector; and taking minimization of the distance between the multi-dimensional initial vector and the center of a preset fixed anchor point as an objective function, and based on a target electromyographic signal identification model, identifying the plurality of time domain features to obtain an identification result used for indicating whether an action corresponding to the target electromyographic signal is a known action or not.

Description

Technical Field

[0001] This disclosure relates to the field of electromyography (EMG) signal decoding technology, and more specifically to an EMG signal recognition method, apparatus, and device. Background Technology

[0002] Electromyography (EMG) signals are bioelectrical signals generated by changes in the electrical potential of muscle fiber cells during muscle contraction or relaxation. With the development of EMG signal acquisition and processing technologies, the application of EMG signals in control systems for end-effector control has received increasing attention. Current EMG signal recognition models primarily target known categories of EMG signals. However, they struggle to effectively capture key classification information from EMG signals in the complex and varied classification requirements of real-world scenarios. This results in low accuracy for unknown categories and large fluctuations in accuracy for known categories, failing to meet the practical needs of EMG signal decoding. Summary of the Invention

[0003] In view of the above problems, this disclosure provides a method, apparatus and device for electromyography signal recognition.

[0004] According to a first aspect of this disclosure, an electromyography (EMG) signal recognition method is provided, comprising: acquiring EMG signals from multiple channels and preprocessing the EMG signals from multiple channels to obtain an EMG dataset; extracting features from the EMG dataset to obtain an EMG feature set, wherein the EMG feature set includes multiple temporal features; performing feature extraction and spatial mapping on the multiple temporal features based on a target EMG signal recognition model to obtain a multidimensional initial vector; using the minimization of the distance between the multidimensional initial vector and the center of a preset fixed anchor point as the objective function, recognizing the multiple temporal features based on the target EMG signal recognition model to obtain a recognition result indicating whether the action corresponding to the target EMG signal is a known action.

[0005] The method involves acquiring electromyography (EMG) signals from multiple channels and preprocessing them to obtain an EMG dataset. This includes: performing noise removal processing on the EMG signals from multiple channels using a bandpass filter and a notch filter bank to obtain intermediate EMG signals from multiple channels; calculating target feature parameters for each channel based on the intermediate EMG signals, where the target feature parameters include time-domain feature parameters, frequency-domain feature parameters, and statistical distribution feature parameters; filtering out invalid channels and noisy channels from the multiple channels based on at least one of the time-domain feature parameters, frequency-domain feature parameters, and statistical distribution feature parameters of each channel, using a first preset condition and a second preset condition, respectively. Invalid channels represent channels whose signal quality does not meet the first preset condition, and noisy channels represent channels whose signal quality does not meet the second preset condition. The first preset condition and the second preset condition are different. Based on the invalid and noisy channels, determining the target channel and the target EMG signal transmitted through the target channel from the multiple channels; and sampling the target EMG signal based on a preset time window to obtain the EMG dataset.

[0006] Based on at least one of the time-domain characteristic parameters, frequency-domain characteristic parameters, and statistical distribution characteristic parameters of each of the multiple channels, invalid channels and noise channels are selected from the multiple channels using a first preset condition and a second preset condition, respectively. This includes: selecting invalid channels from the multiple channels using a first preset condition based on the time-domain characteristic parameters of each of the multiple channels; and selecting noise channels from the multiple channels other than invalid channels using a second preset condition based on at least one of the time-domain characteristic parameters, frequency-domain characteristic parameters, and statistical distribution characteristic parameters of each of the multiple channels.

[0007] The time-domain characteristic parameters include at least one of the following: root mean square value, standard deviation, variance, mean absolute value, peak-to-peak value, or amplitude; the frequency-domain characteristic parameters include spectral flatness; and the statistical distribution characteristic parameters include at least one of the following: kurtosis or spatial neighborhood correlation coefficient.

[0008] The preset fixed anchor center is a fixed point set based on the known action category corresponding to the sample electromyography (EMG) signal. The target EMG signal recognition model is obtained by training a neural network model based on a loss function, using the multidimensional initial vector of the samples and the preset fixed anchor center. The multidimensional initial vector of the samples is obtained by extracting features from multiple temporal features of the samples using the neural network model. The multiple temporal features of the samples are obtained by extracting features from the sample EMG dataset. The sample EMG dataset is obtained by preprocessing sample EMG signals from multiple channels. The loss function is used to ensure that the classification error of sample EMG signals of known categories is less than the classification error of sample EMG signals of unknown categories.

[0009] The sample electromyography (EMG) dataset is obtained by preprocessing sample EMG signals from multiple channels, including: performing noise removal processing on the sample EMG signals from multiple channels using bandpass filters and notch filters to obtain intermediate sample EMG signals from multiple channels; calculating the sample target feature parameters for each of the multiple channels based on the intermediate sample EMG signals from multiple channels, wherein the sample target feature parameters include sample time-domain feature parameters, sample frequency-domain feature parameters, and sample statistical distribution feature parameters; filtering out invalid channels and noisy channels from the multiple channels based on at least one of the sample time-domain feature parameters, sample frequency-domain feature parameters, and sample statistical distribution feature parameters from the multiple channels using a first preset condition and a second preset condition, wherein invalid channels represent channels whose signal quality does not meet the first preset condition, and noisy channels represent channels whose signal quality does not meet the second preset condition, and the first preset condition and the second preset condition are different; determining the target channel and the sample target EMG signal transmitted through the target channel from the multiple channels based on the invalid channel and the noisy channel; and sampling the sample target EMG signal based on a preset time window to obtain the sample EMG dataset.

[0010] The neural network model includes a residual fusion module and a convolutional fusion module. The residual fusion module includes a first residual fusion submodule, a second residual fusion submodule, and a first residual connection submodule. The multidimensional initial vector of the samples is obtained by extracting features from the temporal features of multiple samples based on the neural network model. This includes: using the first residual fusion submodule, the second residual fusion submodule, and the residual connection submodule to obtain multiple scale features of the temporal features of multiple samples to obtain intermediate sample feature vectors; and using the convolutional fusion module to perform spatial mapping on the intermediate sample feature vectors to obtain the multidimensional initial vector of the samples.

[0011] The convolutional fusion module includes a convolutional fusion submodule, a second residual connection submodule, and an activation function. It uses the convolutional fusion module to spatially map the intermediate sample feature vectors to obtain a multidimensional initial vector for the samples. This includes: using the convolutional fusion submodule to extract local features from the intermediate sample feature vectors to obtain local sample feature vectors; using the activation function to perform a nonlinear transformation on the local sample feature vectors to obtain intermediate local sample feature vectors; and using the second residual connection submodule to fuse the local sample feature vectors and the intermediate local sample feature vectors to obtain the multidimensional initial vector for the samples.

[0012] A second aspect of this disclosure provides an electromyography (EMG) signal recognition device, comprising: an acquisition module for acquiring multi-channel EMG signals and preprocessing the multi-channel EMG signals to obtain an EMG dataset; a first extraction module for extracting features from the EMG dataset to obtain an EMG feature set, wherein the EMG feature set includes multiple temporal features; a second extraction module for extracting features and spatially mapping multiple temporal features based on a target EMG signal recognition model to obtain a multidimensional initial vector; and a recognition module for recognizing multiple temporal features based on the target EMG signal recognition model, using minimizing the distance between the multidimensional initial vector and the center of a preset fixed anchor point as the objective function, to obtain a recognition result indicating whether the action corresponding to the target EMG signal is a known action.

[0013] A third aspect of this disclosure provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0014] A fourth aspect of this disclosure also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0015] The fifth aspect of this disclosure also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0016] According to the electromyography (EMG) signal recognition method provided in this disclosure, the quality of EMG signals can be improved by preprocessing the acquired EMG signals from multiple channels. Then, feature extraction is performed on the preprocessed EMG dataset to obtain an EMG feature set. The target EMG signal recognition model is used to extract and spatially map multiple time-domain features to obtain a multidimensional initial vector. The objective function is to minimize the distance between the multidimensional initial vector and the preset fixed anchor center. The target EMG signal recognition model is then used for recognition to obtain a recognition result indicating whether the action corresponding to the target EMG signal is a known action. The preset fixed anchor center is set according to the known action category. Combined with the multidimensional initial vector, the target EMG signal recognition model can achieve high-precision recognition of unknown action categories while maintaining high accuracy in recognizing known action categories. This improves the robustness and generalization ability of the target EMG signal recognition model in complex environments and provides new ideas and technical support for the intelligent application of future human-computer interaction devices. Attached Figure Description

[0017] The foregoing contents, as well as other objects, features, and advantages of this disclosure, will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0018] Figure 1 This diagram schematically illustrates an application scenario of the electromyography signal recognition method according to an embodiment of the present disclosure.

[0019] Figure 2 A flowchart illustrating an electromyography signal recognition method according to an embodiment of the present disclosure is shown schematically.

[0020] Figure 3 A flowchart illustrating the preprocessing of electromyographic signals according to an embodiment of the present disclosure is shown schematically.

[0021] Figure 4 A schematic diagram illustrating a neural network model according to an embodiment of the present disclosure is shown.

[0022] Figure 5 A flowchart illustrating the identification of electromyographic signals according to an embodiment of the present disclosure is shown schematically.

[0023] Figure 6 A schematic diagram illustrating the structure of an electromyography signal recognition device according to an embodiment of the present disclosure; and

[0024] Figure 7 A block diagram schematically illustrates an electronic device suitable for implementing an electromyography signal recognition method according to an embodiment of the present disclosure. Detailed Implementation

[0025] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0026] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0027] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0028] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0029] In the technical solution disclosed herein, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation entry points are provided for users to choose to authorize or refuse.

[0030] In scenarios involving automated decision-making using personal information, the methods, devices, and systems provided in this disclosure all offer users corresponding entry points for choosing to agree to or reject the automated decision-making results. If the user chooses to reject, the process proceeds to the expert decision-making stage. Here, "automated decision-making" refers to the activity of automatically analyzing and evaluating an individual's behavioral habits, interests, or economic, health, and credit status through computer programs, and then making a decision. Here, "expert decision-making" refers to the activity of making decisions by personnel who specialize in a particular field, possess specialized experience, knowledge, and skills, and have reached a certain level of professional expertise.

[0031] With the development of electromyography (EMG) signal acquisition and processing technologies, the application of EMG signals in control systems to achieve end-effector control has attracted increasing attention. In related technologies, EMG control systems typically follow a basic process of "data acquisition and preprocessing, feature extraction and selection, model training and optimization, pattern classification and recognition, and model evaluation and improvement" to gradually build the control framework of the EMG control system. Therefore, the key factors affecting the performance of EMG control systems mainly include the quality of the EMG signal, the number of extracted features, and the model structure.

[0032] However, due to limitations in the hardware performance of acquisition devices, the sampling quality and the number of extractable features of electromyography (EMG) signals are often limited. Therefore, related technologies improve the pattern recognition accuracy of the system by refining the model architecture. Most of the work in these technologies still focuses on model training and performance optimization in closed-set environments, i.e., recognition models based on known categories of EMG signals. However, in practical applications, the recognition performance may degrade due to interference from unknown categories of data. In other words, once encountering EMG signals of unknown categories, the model in a closed-set environment struggles to process them effectively, affecting decoding reliability. Therefore, how to effectively identify and distinguish unknown categories of data in complex scenarios has become a major problem in improving the accuracy and reliability of EMG control systems.

[0033] In view of this, embodiments of the present disclosure provide an electromyography (EMG) signal recognition method, comprising: acquiring EMG signals from multiple channels and preprocessing the EMG signals from multiple channels to obtain an EMG dataset; extracting features from the EMG dataset to obtain an EMG feature set, wherein the EMG feature set includes multiple temporal features; performing feature extraction and spatial mapping on the multiple temporal features based on a target EMG signal recognition model to obtain a multidimensional initial vector; using the minimization of the distance between the multidimensional initial vector and the center of a preset fixed anchor point as the objective function, recognizing the multiple temporal features based on the target EMG signal recognition model to obtain a recognition result indicating whether the action corresponding to the target EMG signal is a known action.

[0034] Figure 1 The diagram illustrates an application scenario of the electromyography signal recognition method according to an embodiment of the present disclosure.

[0035] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0036] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0037] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0038] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0039] It should be noted that the electromyography (EMG) signal recognition method provided in this embodiment can generally be executed by server 105. Correspondingly, the EMG signal recognition device provided in this embodiment can generally be located in server 105. The EMG signal recognition method provided in this embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the EMG signal recognition device provided in this embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0040] It should be understood that Figure 1 The number of first terminal devices, second terminal devices, third terminal devices, networks, and servers shown in the diagram is merely illustrative. Depending on implementation needs, any number of first terminal devices, second terminal devices, third terminal devices, networks, and servers can be included.

[0041] Figure 2 A flowchart illustrating an electromyography signal recognition method according to an embodiment of the present disclosure is shown schematically.

[0042] like Figure 2 As shown, the electromyography signal recognition method 200 of this embodiment includes operations S210 to S240.

[0043] In operation S210, electromyography (EMG) signals from multiple channels are acquired, and the EMG signals from multiple channels are preprocessed to obtain an EMG dataset.

[0044] In operation S220, feature extraction is performed on the electromyography dataset to obtain the electromyography feature set;

[0045] In operation S230, feature extraction and spatial mapping are performed on multiple time-domain features based on the target electromyography signal recognition model to obtain a multidimensional initial vector;

[0046] In operation S240, with the objective function being the minimization of the distance between the multidimensional initial vector and the center of the preset fixed anchor point, multiple time-domain features are identified based on the target electromyography signal recognition model to obtain a recognition result indicating whether the action corresponding to the target electromyography signal is a known action.

[0047] Electromyography (EMG) signals characterize the voltage waveforms recorded on or inside the body, representing the bioelectrical activity generated when muscle cells are excited. Essentially, they are the temporal and spatial superposition of action potentials from thousands of motor units. EMG signals can also be used to indicate muscle contraction state, joint torque, or movement intention. In one implementation, EMG signals can be acquired using surface electrodes. In another, they can be acquired using needle electrodes.

[0048] In one implementation, multiple channels of electromyography (EMG) signals can be acquired synchronously based on multiple electrode pairs. By preprocessing the EMG signals from multiple channels, an EMG dataset can be obtained. Since EMG signals inherently have a relatively low signal-to-noise ratio, preprocessing can include filtering the EMG signals. Furthermore, because EMG signals can be continuous, and the corresponding movements are variable, preprocessing can also include setting time windows for the EMG signals to reduce interference caused by signal abrupt changes during movement transitions.

[0049] Electromyographic (EMG) signals can be grouped based on a preset step size. In one implementation, the preset step size can be 1 / 8 second. By grouping the EMG signals, multiple groups of EMG signals can be obtained. Feature extraction is performed on each group of EMG signals to obtain multiple EMG features corresponding to each group. The features obtained from feature extraction of multiple groups of EMG signals constitute an EMG feature set. The EMG feature set can include multiple time-domain features. In some implementations, the time-domain features can include at least one of the following: mean absolute value, number of zero-crossing points, waveform length, rate of change of sign, first-order autoregressive coefficient, second-order autoregressive coefficient, third-order autoregressive coefficient, fourth-order autoregressive coefficient, fifth-order autoregressive coefficient, or sixth-order autoregressive coefficient. In one implementation, the time-domain features can include at least one of the following: first-order autoregressive coefficient, second-order autoregressive coefficient, third-order autoregressive coefficient, fourth-order autoregressive coefficient, fifth-order autoregressive coefficient, and sixth-order autoregressive coefficient, as well as mean absolute value, number of zero-crossing points, waveform length, and rate of change of sign.

[0050] By extracting features from multiple temporal features, the temporal features can be transformed into more discriminative, low-dimensional features. Then, by spatial mapping the more discriminative, low-dimensional features, a multidimensional initial vector can be obtained. Each dimension of the multidimensional initial vector can indicate different action categories.

[0051] The preset fixed anchor center is set based on known actions. Each known action category can correspond to one preset fixed anchor center; therefore, multiple preset fixed anchor centers can be included. With the goal of minimizing the distance between the multidimensional initial vector and the preset fixed anchor center, a target electromyography (EMG) signal recognition model is used to identify multiple time-domain feature signals. This yields a recognition result indicating whether the action corresponding to the target EMG signal is a known action.

[0052] In one implementation, a target electromyography (EMG) signal recognition model can be used to identify the target EMG signal. When the action corresponding to the target EMG signal is a known action, the target EMG signal recognition model can output the specific action category corresponding to the target EMG signal. When the action corresponding to the target EMG signal is an unknown action, the target EMG signal recognition model marks the target EMG signal as an unknown category, thereby avoiding blurring the EMG signal of the unknown category into the known action category and affecting the recognition accuracy of the model.

[0053] Preprocessing the acquired electromyographic (EMG) signals from multiple channels improves their quality. Feature extraction is then performed on the preprocessed EMG dataset to obtain an EMG feature set. A target EMG signal recognition model is then used to extract and spatially map these features across multiple temporal domains, resulting in a multidimensional initial vector. The objective function is to minimize the distance between this multidimensional initial vector and a preset fixed anchor point. This minimized distance allows the target EMG signal recognition model to identify whether the action corresponding to the target EMG signal is a known action. The preset fixed anchor point is set based on known action categories. Combined with the multidimensional initial vector, the target EMG signal recognition model can achieve high-precision recognition of unknown action categories while maintaining high accuracy in recognizing known action categories. This improves the robustness and generalization ability of the target EMG signal recognition model in complex environments and provides new ideas and technical support for the intelligent application of future human-computer interaction devices.

[0054] Figure 3 A flowchart illustrating the preprocessing of electromyographic signals according to an embodiment of the present disclosure is shown.

[0055] like Figure 3 As shown, the electromyography signal preprocessing procedure of this embodiment includes operations S310 to S350.

[0056] The S310 is operated to acquire electromyographic signals from multiple channels.

[0057] In operation S320, a bandpass filter is used to process the electromyography (EMG) signals of multiple channels, resulting in EMG signals of multiple channels filtered by the bandpass filter.

[0058] In operation of S330, notch filter banks are used to process the electromyographic signals of multiple channels filtered by bandpass filters to obtain the intermediate electromyographic signals of multiple channels.

[0059] In operation S340, channel filtering is performed based on intermediate electromyographic signals from multiple channels to obtain the target channel and the target electromyographic signal transmitted through the target channel.

[0060] Operate S350 to sample the target electromyographic signal based on a preset time window to obtain an electromyographic dataset.

[0061] The following provides a more detailed explanation of the electromyography signal preprocessing procedure.

[0062] The method involves acquiring electromyography (EMG) signals from multiple channels and preprocessing them to obtain an EMG dataset. This includes: performing noise removal processing on the EMG signals from multiple channels using a bandpass filter and a notch filter bank to obtain intermediate EMG signals from multiple channels; calculating target feature parameters for each of the multiple channels based on the intermediate EMG signals; filtering out invalid and noisy channels from the multiple channels based on at least one of the time-domain feature parameters, frequency-domain feature parameters, and statistical distribution feature parameters of each channel, using a first preset condition and a second preset condition, respectively; determining the target channel and the target EMG signal transmitted through the target channel based on the invalid and noisy channels; and sampling the target EMG signal based on a preset time window to obtain the EMG dataset.

[0063] Noise cancellation can be performed on multi-channel electromyography (EMG) signals using bandpass filters and notch filter banks. In some implementations, a fourth-order Butterworth bandpass filter can be used to eliminate DC and high-frequency noise. The fourth-order Butterworth bandpass filter can be a 10-500 Hz filter. After noise cancellation by the bandpass filter, the EMG signal can be further processed using a notch filter bank to eliminate power frequency noise interference. In one implementation, the EMG signal after noise cancellation by the bandpass filter can be processed using a notch filter bank including two notch filters to eliminate power frequency noise interference. The notch filters can be 50 Hz or their harmonics.

[0064] By performing noise cancellation processing on the electromyography (EMG) signals from multiple channels using a bandpass filter and a notch filter bank, the intermediate EMG signals from multiple channels can be obtained.

[0065] By calculating the intermediate electromyographic (EMG) signals from multiple channels, target characteristic parameters for each channel can be obtained. These target characteristic parameters include time-domain characteristic parameters, frequency-domain characteristic parameters, and statistical distribution characteristic parameters. The time-domain characteristic parameters characterize the intensity or temporal fluctuation stability of the muscle electrical activity associated with the intermediate EMG signals. The frequency-domain characteristic parameters characterize the frequency-domain distribution uniformity of the muscle electrical activity associated with the intermediate EMG signals. The statistical distribution characteristic parameters characterize the waveform steepness, peak characteristics, and waveform similarity of adjacent intermediate EMG signals associated with the intermediate EMG signals.

[0066] Invalid channels represent channels whose signal quality does not meet a first preset condition, and noisy channels represent channels whose signal quality does not meet a second preset condition. The first and second preset conditions can be different. By eliminating invalid and noisy channels, the quality of the target electromyographic signal can be improved. Based on at least one of the time-domain characteristic parameters, frequency-domain characteristic parameters, and statistical distribution characteristic parameters of each of the multiple channels, invalid and noisy channels can be screened from the multiple channels using the first and second preset conditions, respectively.

[0067] In one implementation, invalid channels and noise channels are selected from multiple channels based on at least one of the time-domain characteristic parameters, frequency-domain characteristic parameters, and statistical distribution characteristic parameters of each channel, respectively, using a first preset condition and a second preset condition. This may include: selecting invalid channels from multiple channels based on the time-domain characteristic parameters of each channel, using a first preset condition; and selecting noise channels from the other channels besides invalid channels from multiple channels based on at least one of the time-domain characteristic parameters, frequency-domain characteristic parameters, and statistical distribution characteristic parameters of each channel, using a second preset condition.

[0068] In some implementations, the time-domain characteristic parameters include at least one of the following: root mean square value, standard deviation, variance, mean absolute value, peak-to-peak value, or amplitude; the frequency-domain characteristic parameters include spectral flatness; and the statistical distribution characteristic parameters include at least one of the following: kurtosis or spatial neighborhood correlation coefficient.

[0069] The root mean square (RMS) value reflects the intensity of muscle electrical activity associated with intermediate electromyography (EMG) signals, closely approximating the actual physiological level of muscle contraction and showing a positive correlation with muscle tension. The standard deviation reflects the degree of fluctuation in the intermediate EMG signal; a larger standard deviation indicates more severe amplitude fluctuations. Variance also reflects the degree of fluctuation. The mean absolute value reflects the average activity level of the intermediate EMG signal. The peak-to-peak value reflects the maximum fluctuation range of the intermediate EMG signal, encompassing the overall span of both positive and negative peaks. Amplitude reflects the instantaneous maximum force exertion capacity of the muscle. Spectral flatness reflects the uniformity of frequency domain energy distribution; its value ranges from 0 to 1. A flatness close to 1 indicates a flat spectrum, while a flatness close to 0 indicates a concentrated spectrum. Kurtosis reflects the peak characteristics of the waveform. The spatial neighborhood correlation coefficient reflects the spatial continuity of muscle activation; a higher correlation coefficient indicates synchronous activation of adjacent muscle regions.

[0070] Invalid channels can be filtered out from multiple channels based on their respective time-domain characteristic parameters, using a first preset condition. In some implementations, invalid channels can be determined based on their respective root mean square (RMS), standard deviation, amplitude, variance, mean absolute value (MAV), or peak-to-peak value (PVP). Each time-domain characteristic parameter can correspond to a preset threshold. The MMS threshold corresponding to the RMS value is determined by multiplying 0.1 by the RMS values ​​of all channels. The standard deviation threshold corresponding to the standard deviation is determined by multiplying 0.1 by the standard deviation of all channels. The amplitude threshold corresponding to the amplitude can be 0.05 mV. The amplitude threshold corresponding to the variance can be 1e-6. The mean absolute value threshold corresponding to the mean absolute value is determined by multiplying 0.1 by the mean absolute value of all channels. The PVP threshold corresponding to the peak-to-peak value is determined by multiplying 0.05 by the mean absolute value of all channels. The spectral flatness threshold corresponding to the spectral flatness can be 0.6. The kurtosis threshold corresponding to the kurtosis can be 10. The preset threshold for the spatial neighborhood correlation coefficient, which corresponds to the spatial neighborhood correlation coefficient, can be 0.1.

[0071] The first preset condition is met if any two of the following conditions are met: the root mean square value is less than the root mean square value preset threshold, the standard deviation is less than the standard deviation preset threshold, the amplitude is less than the amplitude preset threshold, the variance is less than the variance preset threshold, the mean absolute value is less than the mean absolute value preset threshold, or the peak-to-peak value is less than the peak-to-peak value preset threshold.

[0072] For example, if the standard deviation of the intermediate electromyography (EMG) signal in a certain channel is less than a preset standard deviation threshold and the variance is less than 1e-6, then the first preset condition is met, and the channel can be considered an invalid channel. As another example, if the standard deviation of the intermediate EMG signal in a certain channel is less than a preset standard deviation threshold and the mean absolute value is less than a preset mean absolute value threshold, then the first preset condition is met, and the channel can be considered a noise channel.

[0073] The second preset condition is met if any two of the following conditions are met: the root mean square value is less than the root mean square value preset threshold, the standard deviation is less than the standard deviation preset threshold, the spectral flatness is greater than the spectral flatness preset threshold, the kurtosis is greater than the kurtosis preset threshold, or the spatial neighborhood correlation coefficient is less than the spatial neighborhood correlation coefficient preset threshold.

[0074] For example, if the spectral flatness of the intermediate electromyography (EMG) signal in a certain channel is greater than 0.6 and the kurtosis is greater than 10, then the second preset condition is met, and the channel can be a noise channel. As another example, if the spatial neighborhood correlation coefficient of the intermediate EMG signal in a certain channel is less than 0.1 and the kurtosis is greater than 10, then the second preset condition is met, and the channel can be a noise channel.

[0075] Given that invalid and noisy channels are identified, invalid and noisy channels can be filtered out from multiple channels to obtain the target channel. The intermediate electromyographic signal transmitted through the target channel is then identified as the target electromyographic signal.

[0076] The size of the preset time window can be determined based on requirements or the type of action. In one implementation, if the action duration is 6 seconds, the preset time window can be set to 5 seconds. The target electromyographic signal is sampled based on the preset time length to obtain an electromyographic dataset. Setting a preset time window can reduce interference caused by signal abrupt changes during action transitions.

[0077] By processing electromyographic (EMG) signals using bandpass and notch filter banks, noise interference can be eliminated, improving the quality of the EMG signal and obtaining an intermediate EMG signal. Based on this intermediate EMG signal, target feature parameters for multiple channels are calculated. Then, channel filtering is performed using first and second preset conditions to obtain the target channel and the target EMG signal transmitted through it. Acquiring the target EMG signal within a preset time window yields an EMG dataset. Channel filtering removes noisy and invalid signal channels, improving signal quality. The existence of the first and second preset conditions ensures the reliability of the filtering. Furthermore, the preset time window reduces interference caused by signal abrupt changes during action switching, further improving the quality of the target EMG signal.

[0078] The sample electromyography (EMG) signals can be EMG signals of known action categories. The preset fixed anchor point center is a fixed point set based on the known action category corresponding to the sample EMG signal. The target EMG signal recognition model can be obtained by training a neural network model based on a loss function, using the sample multidimensional initial vector and the preset fixed anchor point center. The sample multidimensional initial vector can be obtained by feature extraction from multiple sample temporal features using a neural network model. These multiple sample temporal features can be obtained by feature extraction from the sample EMG dataset, which can be obtained by preprocessing sample EMG signals from multiple channels. The loss function is designed to ensure that the classification error of sample EMG signals of known categories is less than the classification error of sample EMG signals of unknown categories.

[0079] The temporal features of the samples can be obtained by feature extraction from the sample electromyography dataset. In some implementations, the sample temporal features may include at least one of the following: mean absolute value, number of zero crossings, waveform length, rate of change of sign, first-order autoregressive coefficient, second-order autoregressive coefficient, third-order autoregressive coefficient, fourth-order autoregressive coefficient, fifth-order autoregressive coefficient, or sixth-order autoregressive coefficient. In one implementation, the sample temporal features may include at least one of the following: first-order autoregressive coefficient, second-order autoregressive coefficient, third-order autoregressive coefficient, fourth-order autoregressive coefficient, fifth-order autoregressive coefficient, and sixth-order autoregressive coefficient, as well as mean absolute value, number of zero crossings, waveform length, and rate of change of sign.

[0080] The sample electromyography (EMG) dataset is obtained by preprocessing sample EMG signals from multiple channels. This process may include: performing noise removal processing on the sample EMG signals from multiple channels using bandpass and notch filter banks to obtain intermediate sample EMG signals from multiple channels; calculating the target feature parameters for each channel based on the intermediate sample EMG signals from multiple channels, where the target feature parameters include sample time-domain feature parameters, sample frequency-domain feature parameters, and sample statistical distribution feature parameters; filtering out invalid channels and noisy channels from the multiple channels based on at least one of the sample time-domain feature parameters, sample frequency-domain feature parameters, and sample statistical distribution feature parameters from the multiple channels using a first preset condition and a second preset condition, respectively. Invalid channels represent channels whose signal quality does not meet the first preset condition, and noisy channels represent channels whose signal quality does not meet the second preset condition. The first preset condition and the second preset condition are different; determining the target channel and the target sample EMG signal transmitted through the target channel from the multiple channels based on the invalid and noisy channels; and sampling the target sample EMG signal based on a preset time window to obtain the sample EMG dataset.

[0081] Noise cancellation can be performed on sample electromyography (EMG) signals from multiple channels using bandpass filters and notch filter banks. In some implementations, a fourth-order Butterworth bandpass filter can be used to eliminate DC and high-frequency noise. The fourth-order Butterworth bandpass filter can be a 10-500 Hz filter. After noise cancellation by the bandpass filter, the sample EMG signals can be further processed using a notch filter bank to eliminate power frequency noise interference. In one implementation, the sample EMG signals after noise cancellation by the bandpass filter can be processed using a notch filter bank including two notch filters to eliminate power frequency noise interference. The notch filters can be 50 Hz or their harmonics.

[0082] By performing noise removal processing on the sample electromyography (EMG) signals from multiple channels using a bandpass filter and a notch filter bank, the intermediate EMG signals from multiple channels can be obtained.

[0083] By calculating the intermediate electromyographic (EMG) signals from multiple channels, the target feature parameters for each channel can be obtained. These target feature parameters include temporal feature parameters, frequency domain feature parameters, and statistical distribution feature parameters. The temporal feature parameters characterize the intensity or temporal fluctuation stability of the muscle electrical activity associated with the intermediate EMG signals. The frequency domain feature parameters characterize the frequency distribution uniformity of the muscle electrical activity associated with the intermediate EMG signals. The statistical distribution feature parameters characterize the waveform steepness, peak characteristics, and waveform similarity between adjacent intermediate EMG signals associated with the intermediate EMG signals.

[0084] Invalid channels represent channels whose signal quality does not meet a first preset condition, and noisy channels represent channels whose signal quality does not meet a second preset condition. The first and second preset conditions can be different. By eliminating invalid and noisy channels, the quality of the target electromyographic signal can be improved. Based on at least one of the sample time-domain characteristic parameters, sample frequency-domain characteristic parameters, and sample statistical distribution characteristic parameters of each channel, invalid and noisy channels can be screened from multiple channels using the first and second preset conditions, respectively.

[0085] In some implementations, the time-domain characteristic parameters of the samples include at least one of the following: root mean square value, standard deviation, variance, mean absolute value, peak-to-peak value, or amplitude; the frequency-domain characteristic parameters of the samples include spectral flatness; and the statistical distribution characteristic parameters of the samples include at least one of the following: kurtosis or spatial neighborhood correlation coefficient.

[0086] Given that invalid and noisy channels are identified, invalid and noisy channels can be screened out from multiple channels to obtain the target channel. The intermediate electromyographic signal of the sample transmitted through the target channel is determined as the target electromyographic signal of the sample.

[0087] The electromyography (EMG) signals of the target sample are sampled based on a preset time window to obtain the sample EMG dataset.

[0088] By processing the sample electromyography (EMG) signal using a bandpass filter and a notch filter bank, noise interference in the signal can be eliminated, improving the quality of the EMG signal and obtaining the intermediate EMG signal of the sample. Based on the intermediate EMG signal, the target feature parameters of each of the multiple channels are calculated. Then, channel filtering is performed using the first and second preset conditions to obtain the target channel and the sample target EMG signal transmitted through the target channel. The sample target EMG signal is then acquired based on a preset time window to obtain the sample EMG dataset. Channel filtering can remove noisy and invalid signal channels, improving signal quality. Furthermore, due to the existence of the preset time window, interference caused by signal abrupt changes during action switching can be reduced, further improving the quality of the sample target EMG signal.

[0089] The multidimensional initial vector of samples is obtained by extracting features from the temporal features of multiple samples based on a neural network model. It can include: using the first residual fusion submodule, the second residual fusion submodule and the residual connection submodule to obtain multiple scale features of the temporal features of multiple samples to obtain intermediate sample feature vectors; and using the convolution fusion module to perform spatial mapping on the intermediate sample feature vectors to obtain the multidimensional initial vector of samples.

[0090] The neural network model may include a residual fusion module and a convolutional fusion module. The residual fusion module may include a first residual fusion submodule, a second residual fusion submodule, and a first residual connection submodule. The convolutional fusion module includes a convolutional fusion submodule, a second residual connection submodule, and an activation function.

[0091] In one implementation, the first residual fusion submodule may include a convolutional layer with a 3x3 kernel and a convolutional layer with a 1x1 kernel, and the second residual fusion submodule may also include a convolutional layer with a 3x3 kernel and a convolutional layer with a 1x1 kernel. A convolutional fusion submodule may include two convolutional layers with 3x3 kernels.

[0092] The first residual fusion submodule, the second residual fusion submodule, and the residual connection submodule can be used to obtain multiple scale features of the temporal features of multiple samples, resulting in intermediate sample feature vectors. The convolutional fusion module performs spatial mapping on the intermediate sample feature vectors to obtain multidimensional initial vectors for the samples. This can include: using the convolutional fusion submodule to extract local features from the intermediate sample feature vectors, obtaining local sample feature vectors; using an activation function to perform a nonlinear transformation on the local sample feature vectors, obtaining intermediate local sample feature vectors; and using the second residual connection submodule to fuse the local sample feature vectors and the intermediate local sample feature vectors to obtain the multidimensional initial vectors for the samples.

[0093] By using a convolutional fusion submodule to extract local features from the intermediate sample feature vectors, local sample feature vectors can be obtained. Then, an activation function is used to perform a nonlinear transformation on the local sample feature vectors to obtain intermediate local sample feature vectors. Finally, a second residual connection submodule is used to fuse the local sample feature vectors and the intermediate local sample feature vectors to obtain the multidimensional initial vector of the samples. In one implementation, the activation function can be a Gaussian Error Linear Unit (GELU) function.

[0094] By utilizing the first residual fusion submodule, the second residual fusion submodule, and the residual connection submodule to obtain multiple scale features of the temporal features of multiple samples, the intermediate sample feature vector can be obtained. The input features and output features of the residual fusion module are directly added through the residual connection submodule to avoid feature loss during transformation and ensure effective transfer and fusion of cross-layer features. The convolutional fusion module overcomes the problems of incomplete feature capture and feature alignment by a single convolutional kernel through the synergy of multi-scale feature capture by multiple convolutional kernels and nonlinear processing of activation functions.

[0095] Figure 4 A schematic diagram of a neural network model according to an embodiment of the present disclosure is shown.

[0096] like Figure 4 As shown, the input data is fed into the first residual fusion submodule, the second residual fusion submodule, and the first residual connection submodule of the residual fusion module, respectively. The input data may include sample temporal features. The first residual fusion submodule may include a convolutional layer with a 3x3 kernel and a convolutional layer with a 1x1 kernel. The second residual fusion submodule may also include a convolutional layer with a 3x3 kernel and a convolutional layer with a 1x1 kernel. The outputs of the first and second residual fusion submodules can be directly added to the sample temporal features to obtain the intermediate sample feature vector. The intermediate sample feature vector is processed by a convolutional layer with a 3x3 kernel and then input into the convolutional fusion submodule of the convolutional fusion module. The convolutional fusion submodule can include two convolutional layers with 3x3 kernels. After processing by the convolutional fusion submodule, a local sample feature vector can be obtained. The local sample feature vector is then processed by a convolutional layer with a 1x1 kernel and input into an activation function. It is then input into another convolutional layer with a 1x1 kernel to output an intermediate local sample feature vector. The intermediate local sample feature vector is directly added to the local sample feature vector output by the convolutional fusion submodule to obtain a multidimensional initial vector of samples. The multidimensional initial vector of samples is processed by the last convolutional layer included in the convolutional fusion module and then processed by a pooling layer and a fully connected layer before being output.

[0097] Figure 5 A flowchart illustrating the identification of electromyographic signals according to an embodiment of the present disclosure is shown.

[0098] like Figure 5 As shown, multiple channels of electromyography (EMG) signals are acquired. These signals are preprocessed to obtain an EMG dataset. Feature extraction is then performed on the EMG dataset to obtain an EMG feature set. This feature set is divided into a training feature set and a validation feature set. The features in the training feature set are those of known categories, while the validation feature set includes features of both known and unknown categories. The training feature set is input into a neural network model, and the model is trained based on a multidimensional initial vector of samples and a preset fixed anchor center to obtain the target EMG signal recognition model. The multidimensional initial vector of samples is obtained by extracting features from the temporal domain of multiple samples during the neural network model training process. Combining this with the preset fixed anchor center allows the introduction of the open set method into the neural network model, resulting in the target EMG signal recognition model after training. The verification electromyographic feature set is input into the target electromyographic signal recognition model. The target electromyographic signal recognition model can determine the rejection score by the distance between the multidimensional initial vector and the center of the preset fixed anchor point. If the rejection score output by the target electromyographic signal recognition model is greater than the preset score threshold, the electromyographic signal is identified as an unknown action of an unknown category. If the rejection score is not greater than the preset score threshold, the electromyographic signal is identified as a known action of a known category.

[0099] Based on the above-described electromyography (EMG) signal recognition method, this disclosure also provides an EMG signal recognition device. The following will be combined with... Figure 6 The device is described in detail.

[0100] Figure 6 A schematic block diagram of an electromyography signal recognition device according to an embodiment of the present disclosure is shown.

[0101] like Figure 6 As shown, the electromyography signal recognition device 600 of this embodiment includes an acquisition module 610, a first extraction module 620, a second extraction module 630, and a recognition module 640.

[0102] The acquisition module 610 is used to acquire multi-channel electromyography (EMG) signals and preprocess the multi-channel EMG signals to obtain an EMG dataset. In one embodiment, the acquisition module 610 can be used to perform the operation S210 described above, which will not be repeated here.

[0103] The first extraction module 620 is used to extract features from the electromyography (EMG) dataset to obtain an EMG feature set, wherein the EMG feature set includes multiple time-domain features. In one embodiment, the first extraction module 620 can be used to perform the operation S220 described above, which will not be repeated here.

[0104] The second extraction module 630 is used to extract features and spatially map multiple time-domain features based on the target electromyography signal recognition model to obtain a multi-dimensional initial vector. In one embodiment, the second extraction module 630 can be used to perform the operation S230 described above, which will not be repeated here.

[0105] The recognition module 640 is used to identify multiple time-domain features based on the target electromyography (EMG) signal recognition model, with the objective function being minimizing the distance between the multidimensional initial vector and the center of the preset fixed anchor point, to obtain a recognition result indicating whether the action corresponding to the target EMG signal is a known action. In one embodiment, the recognition module 640 can be used to perform the operation S240 described above, which will not be repeated here.

[0106] According to an embodiment of this disclosure, the acquisition module 610 includes: a first acquisition submodule, configured to perform noise cancellation processing on electromyographic signals from multiple channels using a bandpass filter and a notch filter bank to obtain intermediate electromyographic signals from multiple channels; a second acquisition submodule, configured to calculate target feature parameters for each of the multiple channels based on the intermediate electromyographic signals from the multiple channels, wherein the target feature parameters include time-domain feature parameters, frequency-domain feature parameters, and statistical distribution feature parameters; a third acquisition submodule, configured to filter out invalid channels and noise channels from the multiple channels based on at least one of the time-domain feature parameters, frequency-domain feature parameters, and statistical distribution feature parameters of each of the multiple channels, respectively using a first preset condition and a second preset condition, wherein invalid channels represent channels whose signal quality does not meet the first preset condition, and noise channels represent channels whose signal quality does not meet the second preset condition, and the first preset condition is different from the second preset condition; a fourth acquisition submodule, configured to determine a target channel and a target electromyographic signal transmitted through the target channel from the multiple channels based on the invalid channel and the noise channel; and a fifth acquisition submodule, configured to sample the target electromyographic signal based on a preset time window to obtain an electromyographic dataset.

[0107] According to embodiments of this disclosure, the third acquisition submodule includes: a first acquisition unit, configured to filter out invalid channels from the multiple channels based on the time-domain characteristic parameters of each of the multiple channels using a first preset condition; and a second acquisition unit, configured to filter out noise channels from the other channels besides invalid channels from the multiple channels based on at least one of the time-domain characteristic parameters, frequency-domain characteristic parameters, and statistical distribution characteristic parameters of each of the multiple channels using a second preset condition. The time-domain characteristic parameters include at least one of the following: root mean square value, standard deviation, variance, mean absolute value, peak-to-peak value, or amplitude; the frequency-domain characteristic parameters include spectral flatness; and the statistical distribution characteristic parameters include at least one of the following: kurtosis or spatial neighborhood correlation coefficient.

[0108] According to embodiments of this disclosure, any plurality of modules among the acquisition module 610, the first extraction module 620, the second extraction module 630, and the identification module 640 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in one module. According to embodiments of this disclosure, at least one of the acquisition module 610, the first extraction module 620, the second extraction module 630, and the identification module 640 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the acquisition module 610, the first extraction module 620, the second extraction module 630, and the identification module 640 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0109] Figure 7 A block diagram schematically illustrates an electronic device suitable for implementing an electromyography signal recognition method according to an embodiment of the present disclosure.

[0110] like Figure 7 As shown, an electronic device 700 according to an embodiment of the present disclosure includes a processor 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage portion 708 into RAM (Random Access Memory). The processor 701 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 701 may also include onboard memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0111] RAM 703 stores various programs and data required for the operation of electronic device 700. Processor 701, ROM 702, and RAM 703 are interconnected via bus 704. Processor 701 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 702 and / or RAM 703. It should be noted that the programs may also be stored in one or more memories other than ROM 702 and RAM 703. Processor 701 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.

[0112] According to embodiments of this disclosure, the electronic device 700 may further include an I / O interface 705, which is also connected to a bus 704. The electronic device 700 may also include one or more of the following components connected to the I / O interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 710 as needed so that computer programs read from it can be installed into the storage section 708 as needed.

[0113] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0114] According to embodiments of this disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, RAM (Random Access Memory), ROM (Read Only Memory), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this disclosure, the computer-readable storage medium may include ROM 702 and / or RAM 703 and / or one or more memories other than ROM 702 and RAM 703 described above.

[0115] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the electromyography signal recognition method provided in the embodiments of this disclosure.

[0116] When the computer program is executed by the processor 701, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0117] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 709, and / or installed from a removable medium 711. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0118] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 709, and / or installed from the removable medium 711. When the computer program is executed by the processor 701, it performs the functions defined in the system of this disclosure embodiment. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0119] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can execute entirely on a user's computing device, partially on a user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0120] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0121] Those skilled in the art will understand that the features described in the various embodiments of this disclosure can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. In particular, the features described in the various embodiments of this disclosure can be combined and / or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0122] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A method for recognizing electromyographic signals, characterized in that, The method includes: Acquire electromyographic signals from multiple channels and preprocess the electromyographic signals from the multiple channels to obtain an electromyographic dataset; Feature extraction is performed on the electromyography dataset to obtain an electromyography feature set, wherein the electromyography feature set includes multiple time-domain features; Based on the target electromyography signal recognition model, feature extraction and spatial mapping are performed on the multiple time-domain features to obtain a multi-dimensional initial vector; Using the minimization of the distance between the multidimensional initial vector and the center of the preset fixed anchor point as the objective function, the multiple time-domain features are identified based on the target electromyography signal recognition model to obtain a recognition result indicating whether the action corresponding to the target electromyography signal is a known action.

2. The method according to claim 1, characterized in that, The process of acquiring electromyographic (EMG) signals from multiple channels and preprocessing the EMG signals from these channels to obtain an EMG dataset includes: The electromyographic signals of the multiple channels are subjected to noise cancellation processing using a bandpass filter and a notch filter bank to obtain the intermediate electromyographic signal of the multiple channels. Based on the intermediate electromyographic signals of the multiple channels, the target feature parameters of each of the multiple channels are calculated, wherein the target feature parameters include time-domain feature parameters, frequency-domain feature parameters, and statistical distribution feature parameters; Based on at least one of the time-domain characteristic parameters, frequency-domain characteristic parameters, and statistical distribution characteristic parameters of each of the multiple channels, invalid channels and noise channels are selected from the multiple channels using a first preset condition and a second preset condition, respectively. The invalid channel represents a channel whose signal quality does not meet the first preset condition, and the noise channel represents a channel whose signal quality does not meet the second preset condition. The first preset condition and the second preset condition are different. Based on the invalid channel and the noise channel, a target channel and the target electromyographic signal transmitted through the target channel are determined from the plurality of channels; The target electromyographic signal is sampled based on a preset time window to obtain an electromyographic dataset.

3. The method according to claim 2, characterized in that, The step of filtering out invalid channels and noise channels from the plurality of channels based on at least one of the time-domain characteristic parameters, frequency-domain characteristic parameters, and statistical distribution characteristic parameters of each of the plurality of channels, respectively, using a first preset condition and a second preset condition, includes: Based on the time-domain characteristic parameters of each of the multiple channels, invalid channels are filtered out from the multiple channels using a first preset condition; Based on at least one of the time-domain characteristic parameters, frequency-domain characteristic parameters, and statistical distribution characteristic parameters of each of the multiple channels, noise channels are selected from the other channels (excluding invalid channels) using a second preset condition.

4. The method according to claim 3, characterized in that, The time-domain feature parameters include at least one of the following: root mean square value, standard deviation, variance, mean absolute value, peak-to-peak value, or amplitude; the frequency-domain feature parameters include spectral flatness; and the statistical distribution feature parameters include at least one of the following: kurtosis or spatial neighborhood correlation coefficient.

5. The method according to claim 1, characterized in that, The preset fixed anchor center is a fixed point set based on the known action category corresponding to the sample electromyography (EMG) signal. The target EMG signal recognition model is obtained by training a neural network model based on a loss function, according to the multidimensional initial vector of the samples and the preset fixed anchor center. The multidimensional initial vector of the samples is obtained by feature extraction of multiple sample time-domain features by the neural network model. The multiple sample time-domain features are obtained by feature extraction of the sample EMG dataset. The sample EMG dataset is obtained by preprocessing sample EMG signals from multiple channels. The loss function is used to make the classification error of sample EMG signals of known categories smaller than the classification error of sample EMG signals of unknown categories.

6. The method according to claim 5, characterized in that, The sample electromyography (EMG) dataset is obtained by preprocessing sample EMG signals from multiple channels, including: The sample electromyography (EMG) signals from the multiple channels are subjected to noise cancellation processing using a bandpass filter and a notch filter bank to obtain the intermediate sample EMG signals from the multiple channels. Based on the intermediate electromyographic signals of the samples from the multiple channels, the target feature parameters of each of the multiple channels are calculated, wherein the target feature parameters of the samples include sample time domain feature parameters, sample frequency domain feature parameters, and sample statistical distribution feature parameters; Based on at least one of the sample time-domain feature parameters, sample frequency-domain feature parameters, and sample statistical distribution feature parameters of each of the multiple channels, invalid channels and noise channels are screened out from the multiple channels using a first preset condition and a second preset condition, respectively. The invalid channel represents a channel whose signal quality does not meet the first preset condition, and the noise channel represents a channel whose signal quality does not meet the second preset condition. The first preset condition and the second preset condition are different. Based on the invalid channel and the noisy channel, a target channel and the sample target electromyographic signal transmitted through the target channel are determined from the plurality of channels; The target electromyography (EMG) signal of the sample is sampled based on a preset time window to obtain the sample EMG dataset.

7. The method according to claim 5, characterized in that, The neural network model includes a residual fusion module and a convolutional fusion module. The residual fusion module includes a first residual fusion submodule, a second residual fusion submodule, and a first residual connection submodule. The multidimensional initial vector of the samples is obtained by feature extraction of multiple sample temporal features based on the neural network model, including: The first residual fusion submodule, the second residual fusion submodule, and the residual connection submodule are used to obtain multiple scale features of the temporal features of the multiple samples, and an intermediate sample feature vector is obtained. The convolutional fusion module is used to spatially map the feature vectors of the intermediate samples to obtain multidimensional initial vectors for the samples.

8. The method according to claim 7, characterized in that, The convolutional fusion module includes a convolutional fusion submodule, a second residual connection submodule, and an activation function. The convolutional fusion module is used to spatially map the intermediate sample feature vectors to obtain a multidimensional initial vector for the samples, including: The convolutional fusion submodule is used to extract local features from the intermediate sample feature vector to obtain the local sample feature vector. Using the activation function, a nonlinear transformation is performed on the local sample feature vector to obtain the intermediate local sample feature vector; Using the second residual connection submodule, the local sample feature vector and the intermediate local sample feature vector are fused to obtain the sample multidimensional initial vector.

9. An electromyography (EMG) signal recognition device, characterized in that, The device includes: The acquisition module is used to acquire multi-channel electromyography (EMG) signals and preprocess the multi-channel EMG signals to obtain an EMG dataset. The first extraction module is used to extract features from the electromyography dataset to obtain an electromyography feature set, wherein the electromyography feature set includes multiple time-domain features; The second extraction module is used to extract and spatially map the multiple time-domain features based on the target electromyography signal recognition model to obtain a multidimensional initial vector. The recognition module is used to identify the multiple time-domain features based on the target electromyography signal recognition model, with the objective function being minimizing the distance between the multidimensional initial vector and the center of the preset fixed anchor point, to obtain a recognition result indicating whether the action corresponding to the target electromyography signal is a known action.

10. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 8.

Images