Real-time identity recognition method, device and electronic equipment based on electromyography signals

Abstract

This invention provides a real-time identity recognition method, device, and electronic device based on electromyography (EMG) signals. The real-time identity recognition method includes: acquiring a target EMG signal characterizing limb movement from collected raw EMG signals; determining the target time-frequency features of the target EMG signal; and performing recognition processing on the target time-frequency features based on a preset convolutional neural network to obtain target identity information. Using this method, a terminal device can receive raw EMG signals collected by an EMG wristband in real time and perform time-frequency feature extraction and identity recognition processing on the received raw EMG signals, thereby achieving the goal of real-time identity recognition based on EMG signals and improving the recognition rate of identity information while meeting real-time requirements.

Description

Technical Field

[0001] This invention relates to the field of biometric identification technology, and in particular to a real-time identity recognition method, device, and electronic device based on electromyography signals. Background Technology

[0002] With the continuous and rapid development of Internet technology, biometric identification technology has received increasing attention and is being widely used to ensure the security of personal information. For example, biometric features such as face, fingerprint, iris, and voice are used in various identity recognition systems such as credit card transactions, ATM security, bank security, and smartphones. However, considering that the above biometric features are easy to forge, recreate, and counterfeit, how to improve the security of personal identity remains a hot research topic.

[0003] In related technologies, identity recognition based on electromyography (EMG) signals first involves segmenting the event EMG signal from the entire EMG signal stream, and then using artificial neural networks or convolutional neural networks (CNNs) to process the event EMG signal to obtain identity recognition information.

[0004] However, existing identity recognition methods based on electromyography (EMG) signals process EMG signals offline, and researchers focus on improving the recognition rate of EMG-based identity recognition in offline states, while ignoring the needs of actual application scenarios. As a result, they cannot meet the real-time requirements while ensuring the recognition rate. Summary of the Invention

[0005] This invention provides a real-time identity recognition method, device, and electronic device based on electromyography (EMG) signals, which addresses the shortcomings of existing technologies that cannot meet real-time requirements while ensuring recognition rate when performing identity recognition based on EMG signals. This invention aims to improve the recognition rate of identity information while meeting real-time requirements when performing identity recognition based on EMG signals.

[0006] This invention provides a real-time identity recognition method based on electromyography signals, comprising:

[0007] From the collected raw electromyographic signals, target electromyographic signals characterizing limb movement are obtained;

[0008] Determine the target time-frequency characteristics of the target electromyographic signal;

[0009] The target's time-frequency features are processed by a pre-defined convolutional neural network to obtain the target's identity information.

[0010] According to the present invention, a real-time identity recognition method based on electromyography (EMG) signals is provided, wherein acquiring target EMG signals characterizing limb movement from the acquired raw EMG signals includes:

[0011] The acquired raw electromyography (EMG) signals were processed by sliding window segmentation to obtain the EMG signal of the sliding window corresponding to each sliding window;

[0012] Calculate the average absolute value of the electromyographic signals in the sliding window;

[0013] Based on the average absolute value and the preset dynamic threshold, the target electromyographic signal characterizing limb movement is obtained.

[0014] According to the present invention, a real-time identity recognition method based on electromyography (EMG) signals is provided, wherein obtaining a target EMG signal characterizing limb movement based on the mean absolute value and a preset dynamic threshold includes:

[0015] A reference sliding window electromyographic signal is selected from the plurality of said sliding window electromyographic signals;

[0016] A preset dynamic threshold is determined based on the reference average absolute value of the reference sliding window electromyography signal;

[0017] Determine the target electromyographic signal representing limb movement that is greater than the preset dynamic threshold among the multiple average absolute values.

[0018] According to a real-time identity recognition method based on electromyography (EMG) signals provided by the present invention, determining the target time-frequency features of the target EMG signal includes:

[0019] The target electromyographic signal is divided into multiple consecutive target sub-bands;

[0020] Preset feature extraction is performed on each target sub-frequency band to obtain the target time-frequency features of the target electromyographic signal; wherein, the preset features include entropy features, percentile features, median features, mean features, variance features, standard deviation features, root mean square value features, zero-crossing rate features, and average zero-crossing rate features.

[0021] According to the present invention, a real-time identity recognition method based on electromyography signals is provided. When the target time-frequency feature is one-dimensional, the step of performing recognition processing on the target time-frequency feature based on a preset convolutional neural network to obtain target identity information includes:

[0022] The target time-frequency features are input into a pre-defined one-dimensional convolutional neural network for convolution, pooling, and fully connected processing to obtain the target feature vector.

[0023] Based on the preset mapping relationship between feature vectors and identity recognition information, the target identity information corresponding to the target feature vector is obtained.

[0024] According to the present invention, a real-time identity recognition method based on electromyography (EMG) signals, prior to the step of acquiring a target EMG signal characterizing limb movement from the acquired raw EMG signals, the method further includes:

[0025] The raw electromyographic (EMG) signals collected during the user's limb movement tasks within a target time period are acquired. The raw EMG signals are EMG signal streams and include EMG signals triggered by limb movement and EMG signals in non-limb movement states.

[0026] The present invention also provides a real-time identity recognition device based on electromyography signals, comprising:

[0027] The acquisition module is used to acquire target electromyographic signals that characterize limb movement from the collected raw electromyographic signals.

[0028] The determination module is used to determine the target time-frequency characteristics of the target electromyographic signal;

[0029] The identification module is used to perform identification processing on the time-frequency features of the target based on a preset convolutional neural network to obtain the target identity information.

[0030] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the real-time identity recognition method based on electromyography signals as described above.

[0031] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the real-time identity recognition method based on electromyography signals as described above.

[0032] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above-described real-time identity recognition methods based on electromyography signals.

[0033] This invention provides a real-time identity recognition method, device, and electronic device based on electromyography (EMG) signals. The method involves acquiring target EMG signals representing limb movement from collected raw EMG signals, determining the target time-frequency features of the target EMG signals, and then processing these features using a pre-defined convolutional neural network to obtain target identity information. This enables a terminal device to receive raw EMG signals collected by an EMG wristband in real time and perform time-frequency feature extraction and identity recognition processing on the received raw EMG signals, thereby achieving real-time identity recognition based on EMG signals and improving the recognition rate of identity information while meeting real-time requirements. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0035] Figure 1 This is a flowchart illustrating the real-time identity recognition method based on electromyography signals provided by the present invention.

[0036] Figure 2 This is a schematic diagram illustrating the principle of using discrete wavelet transform to decompose target electromyographic signals provided by the present invention;

[0037] Figure 3 This is a schematic diagram of the structure of the real-time identity recognition device based on electromyography signals provided by the present invention;

[0038] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0040] With the rapid development of science and technology, biometric identification technology has gained increasing attention to ensure the security of personal information. Many biometric identification methods have been derived from this technology, using unique physical characteristics to identify an individual or verify their claimed identity. Currently, facial, fingerprint, iris, and voice biometrics are widely used in various identity verification systems, such as credit card transactions, ATM security, bank security, and smartphones. However, biometric identification methods utilizing unique physiological characteristics are vulnerable to attack because these characteristics can still be forged, recreated, and counterfeited. For example, voice authentication systems are susceptible to spoofing attacks, where imposters attempt to deceive the authentication system using pre-recorded or synthesized voice samples of the victim.

[0041] Utilizing liveness signatures for biometric identification is one method to overcome the shortcomings of traditional biometric identification technologies. Electromyography (EMG) signals are complex signals controlled by the nervous system, relying on the anatomical and physiological characteristics of muscles. Because the direct connection between a complete muscle, the complete central nervous system, and the brain is individual and directly related to each person's physiology, it possesses universality, stability, uniqueness, and measurability. Therefore, EMG signals can be used in identity recognition systems. In recent years, there has been considerable research on EMG-based identity recognition methods. For example, using artificial neural network algorithms to process EMG signals acquired through electrodes, an identity recognition accuracy of 81.6% has been achieved; another example is using CNN algorithms to process EMG signals acquired through electrodes, which can improve the recognition accuracy to 95.0%. However, because existing EMG-based identity recognition systems are based on offline processing, meaning the EMG signals to be processed have already been completely segmented from the whole signal, they are not suitable for real-life scenarios.

[0042] In summary, all existing electromyography (EMG) signal-based identity recognition methods are based on offline processing. Researchers focus only on the recognition rate of EMG-based identity recognition in offline conditions, while ignoring the needs of real-world application scenarios.

[0043] Based on this, the present invention provides a real-time identity recognition method, device, and electronic device based on electromyography signals, which satisfies both real-time performance and ensures a high recognition rate. The following is a combination of... Figures 1-4 This invention describes a real-time identity recognition method, apparatus, and electronic device based on electromyography (EMG) signals.

[0044] Reference Figure 1 This is a flowchart illustrating the real-time identity recognition method based on electromyography (EMG) signals provided by the present invention. The executing entity of the real-time identity recognition method based on EMG signals can be a real-time identity recognition device based on EMG signals. This device can be implemented as part or all of a terminal device through software, hardware, or a combination of both. The terminal device must at least possess identity recognition functionality. Optionally, the terminal device can be a personal computer (PC), portable device, laptop computer, smartphone, tablet computer, portable wearable device, or other electronic devices. The present invention does not limit the specific form of the terminal device.

[0045] It should be noted that the executing entity of the following method embodiments can be part or all of the aforementioned terminal devices. The following method embodiments are described using a terminal device as the executing entity.

[0046] like Figure 1 As shown, this real-time identity recognition method based on electromyography signals includes the following steps:

[0047] Step 110: Obtain the target electromyographic signal representing limb movement from the collected raw electromyographic signals.

[0048] Specifically, the terminal device can receive raw electromyographic (EMG) signals collected by an EMG wristband. The EMG wristband is preferably the MYO EMG wristband, developed by the Thalmic laboratory for collecting EMG signals. The MYO EMG wristband features low power consumption, low cost, small size, light weight, and ease of wear. Therefore, the MYO EMG wristband can be worn on the subject's arm. While the subject demonstrates different hand gestures (such as opening the palm, clenching the fist, bending to the left, etc.), the eight electrodes in the MYO EMG wristband record the subject's raw EMG signals in real time. The captured raw EMG signals are then transmitted to the terminal device via Bluetooth serial port, where they can be saved as files or processed directly. Since the raw EMG signals may contain EMG signals representing non-arm movements, or may lack start and end time markers for arm movements, the terminal device does not directly process the raw EMG signals online. Instead, it first extracts the target EMG signal representing limb movement (such as arm movement) from the raw EMG signals to improve the accuracy of identification.

[0049] Step 120: Determine the target time-frequency characteristics of the target electromyographic signal.

[0050] Specifically, for target electromyographic (EMG) signals characterizing limb movement, terminal devices can use the discrete wavelet transform (DWT) method to determine the target time-frequency features of the target EMG signal. This is because the DWT can decompose the target EMG signal into approximate (low-frequency) coefficients and detail (high-frequency) coefficients. For an N-level DWT, one array of approximate coefficients and N arrays of detail coefficients can be obtained, where N is a positive integer; for example... Figure 2 As shown, when N is 2, the target electromyography signal can be decomposed twice to obtain one approximate coefficient array and two detail coefficient arrays. Then, feature extraction is performed based on each coefficient array. Feature extraction is usually used to extract the identification information in the target electromyography signal, that is, the features that represent the identity of the experimenter, so as to minimize the complexity and reduce the cost of signal processing.

[0051] Step 130: Based on a preset convolutional neural network, the time-frequency features of the target are processed to obtain the target identity information.

[0052] Specifically, the terminal device inputs the target time-frequency features of the target electromyography (EMG) signal into a preset convolutional neural network (CNN) for identity recognition. Considering that the CNN incorporates an attention mechanism and is derived from artificial neural networks, it possesses very powerful expressive capabilities. Therefore, training the CNN yields the desired preset CNN, which also has the function of mapping features to identity information. Based on this, by inputting the target time-frequency features into the preset CNN, a high-level mapping of the target time-frequency features can be achieved, thereby obtaining the target's identity information.

[0053] It should be noted that since training convolutional neural networks has been covered in many published documents and patents, it will not be elaborated here.

[0054] The present invention provides a real-time identity recognition method based on electromyography (EMG) signals. This method acquires target EMG signals representing limb movement from the collected raw EMG signals, determines the target time-frequency features of the target EMG signals, and then processes these features using a pre-defined convolutional neural network to obtain the target identity information. This enables a terminal device to receive raw EMG signals collected by an EMG wristband in real time and perform time-frequency feature extraction and identity recognition processing on the received raw EMG signals, thereby achieving real-time identity recognition based on EMG signals and improving the recognition rate of identity information while meeting real-time requirements.

[0055] Optionally, prior to step 110, the method may further include:

[0056] The raw electromyographic (EMG) signals collected during the user's limb movement tasks within a target time period are acquired. The raw EMG signals are EMG signal streams and include EMG signals triggered by limb movement and EMG signals in non-limb movement states.

[0057] The target duration can be determined based on the number of times the user repeats the task and the data collection time for each task. For example, if a task is repeated twice and the data collection time for each task is 8 seconds, the corresponding target duration is 16 seconds. There can be multiple users, and each user can be an experimenter.

[0058] Specifically, when the limb movement is a gesture and the electromyography (EMG) signal is determined to be acquired using an EMG bracelet (such as the MYO EMG bracelet), the EMG bracelet can be worn at the radial humeral joint height of at least one participant to record in real time the EMG signals representing muscle activity triggered when the participant performs gesture actions such as clenching a fist or opening a palm, as well as the EMG signals in the non-gesture state when the participant does not make any gestures. Furthermore, each participant repeats the same task M times, where M is a positive integer. The acquisition time for each task can be 8 seconds, and each participant needs to complete one or more hand gestures within 8 seconds, with each gesture lasting 1.5 seconds. For example, when using the MYO electromyography (EMG) bracelet to collect EMG signals, the sampling rate of the MYO EMG bracelet is 200Hz / s, that is, each sample data is 200×8×8, and each experimenter has M sample data. At this time, the M sample data collected by each experimenter are the raw EMG signals collected by the corresponding MYO EMG bracelet within the target time period. Each raw EMG signal is an EMG signal stream, and each EMG data stream can be transmitted to the terminal device in real time through Bluetooth serial port. Thus, the terminal device receives the raw EMG signals collected when the user performs limb movement tasks within the target time period.

[0059] It should be noted that the target duration can be preset in the terminal device, or it can be set when the myoelectric bracelet is worn on each participant's arm, so that the electrodes on the surface of the myoelectric bracelet can collect signals within the target duration. Furthermore, task execution instructions can also be set in the terminal device. The task execution instructions carry limb movement task information, including the task to be performed, the number of times the task is repeated, and the duration of each task. The task execution instructions can be automatically generated when the participant's arm is wearing the myoelectric bracelet. When the task execution instructions are generated, the limb movement task information can be displayed in text and / or audio form, so that the participant can perform the limb movement task based on the limb movement task information seen and / or heard.

[0060] The real-time identity recognition method based on electromyography (EMG) signals provided by this invention acquires raw EMG signals by obtaining information about the user's limb movement tasks within a target time period. Since the raw EMG signals received by the terminal device are an EMG signal stream that includes EMG signals triggered by limb movement and EMG signals in non-limb movement states, the terminal device can process the raw EMG signals online, thereby meeting the real-time requirements of practical application scenarios.

[0061] Optionally, the specific implementation process of step 110 may include:

[0062] First, the acquired raw electromyography (EMG) signals are processed by sliding window segmentation to obtain the EMG signal of the sliding window corresponding to each sliding window; second, the average absolute value of the EMG signal of the sliding window is calculated; finally, based on the average absolute value and a preset dynamic threshold, the target EMG signal representing limb movement is obtained.

[0063] Specifically, the raw electromyographic (EMG) signals acquired by the terminal device are an EMG signal stream, including EMG signals triggered by limb movement and EMG signals in non-limb movement states. Since the raw EMG signals neither indicate when the gesture action begins nor ends, and in practical applications, user gesture recognition is only performed once, the system with identity recognition functionality within the terminal device should, in accordance with the design standards of minimizing memory, power consumption budget, and processing time, use limb detection methods to extract target EMG signals representing limb movement from the raw EMG signals as the switch for identity recognition. Furthermore, when the limb detection method is specifically a gesture detection method, a peak detection algorithm can be further used to detect the target EMG signal corresponding to the gesture time in real time. That is, a sliding window is used to segment the raw EMG signal, and the size of the sliding window and the step size of adjacent sliding windows are determined based on the recognition rate and the reaction time of the entire recognition system. Then, the average absolute value of the EMG signal corresponding to the sliding window is calculated. Since the average absolute value is used to measure the distance of a set of numbers from its average value within the sliding window, and the larger the average absolute value, the higher the probability of the event occurring. Therefore, the method of searching for local maxima by the average absolute value of electromyographic signals from adjacent sliding windows and a preset dynamic threshold can be used to detect limb movement signals (such as gesture detection), thereby determining the target electromyographic signal representing limb movement from the original electromyographic signal.

[0064] It should be noted that when the electromyographic signal of the sliding window is represented as X = {x1, x2, ..., x...} m When the sliding window electromyography signal is in the range of}, the formula for calculating the average absolute value can be: x i This represents the i-th data point in the electromyography (EMG) signal from the sliding window. X represents the mean value of the electromyographic signal X within the sliding window, and MAD represents the mean absolute value of the electromyographic signal X within the sliding window.

[0065] The real-time identity recognition method based on electromyography (EMG) signals provided by this invention performs sliding window segmentation on the acquired raw EMG signals and calculates the average absolute value of the EMG signals in each sliding window. This achieves the purpose of determining the target EMG signal representing limb movement based on the change of adjacent average absolute values ​​and its relationship with a preset dynamic threshold. This method can not only detect limb movement events in real time, but also lay the foundation for subsequent real-time identity recognition.

[0066] Optionally, obtaining the target electromyographic signal characterizing limb movement based on the mean absolute value and a preset dynamic threshold includes:

[0067] First, a reference sliding window electromyographic signal is selected from multiple sliding window electromyographic signals; then, a preset dynamic threshold is determined based on the reference average absolute value of the reference sliding window electromyographic signal; finally, a target electromyographic signal representing limb movement is determined from among the multiple average absolute values ​​that are greater than the preset dynamic threshold.

[0068] Specifically, to avoid interference from noise signals, this invention uses a preset dynamic threshold as the judgment standard. That is, for each sliding window's corresponding electromyographic (EMG) signal and its average absolute value, the terminal device can first select a reference sliding window EMG signal from multiple sliding window EMG signals. For example, the EMG signal corresponding to the first sliding window can be used as the reference sliding window EMG signal. Then, the average absolute value of the reference sliding window EMG signal can be used as the reference average absolute value. The reference average absolute value is further determined as a preset dynamic threshold, such as three times the reference average absolute value. Then, according to the order in which the sliding window EMG signals are obtained (excluding the reference sliding window EMG signal), the average absolute value of the corresponding sliding window EMG signal is compared with the dynamic preset threshold. When the average absolute value is greater than the dynamic preset threshold for the first time, the corresponding sliding window EMG signal can be determined as the target EMG signal representing limb movement.

[0069] It should be noted that when the preset dynamic threshold is determined based on the average absolute value of the electromyographic (EMG) signal in the sliding window corresponding to the first sliding window, the sliding window can be moved twice over the original EMG signal, and the average absolute value of the EMG signal in the sliding window corresponding to the two sliding windows can be calculated. Based on this, it can be determined whether the average absolute value of the EMG signal in the sliding window corresponding to the second sliding window is greater than the preset dynamic threshold. If it is not greater, it can be moved once more, and the average absolute value of the EMG signal in the sliding window corresponding to the third sliding window can be calculated. It can then be determined whether the average absolute value of the EMG signal in the sliding window corresponding to the third sliding window is greater than the preset dynamic threshold. This process is repeated until a target EMG signal representing limb movement that is greater than the preset dynamic threshold is determined.

[0070] The real-time identity recognition method based on electromyography (EMG) signals provided by this invention selects a reference sliding window EMG signal and a reference average absolute value of the reference sliding window EMG signal from multiple sliding window EMG signals to determine a preset dynamic threshold. Furthermore, it determines the target EMG signal representing limb movement corresponding to a multiple average absolute value that is greater than the preset dynamic threshold. This avoids interference from noise signals, ensures the accuracy of subsequent identity recognition, and also meets the real-time requirements.

[0071] Optionally, the specific implementation process of step 120 may include:

[0072] First, the target electromyographic signal is divided into multiple consecutive target sub-bands; then, preset features are extracted for each target sub-band to obtain the target time-frequency features of the target electromyographic signal.

[0073] The preset features include entropy features, percentile features, median features, mean features, variance features, standard deviation features, root mean square value features, zero-crossing rate features, and average zero-crossing rate features.

[0074] Specifically, considering that the Discrete Wavelet Transform (DWT) can be implemented as a filter bank and decompose a signal into multiple continuous sub-bands, and that compared to the continuous wavelet transform method for characterizing the time-frequency characteristics of a signal, the DWT avoids the problem of redundant parameters severely impacting computational performance during processing; furthermore, the DWT can simultaneously improve processing speed. Therefore, for the target electromyography (EMG) signal obtained by the terminal device, the DWT can be used to divide it into multiple continuous target sub-bands. For an N-level DWT, the target EMG signal can be divided into N+1 target sub-bands, where N sub-bands are specifically represented by N detail coefficient arrays, and each sub-band is specifically represented by one approximation coefficient array.

[0075] Then, preset features are extracted for each target sub-frequency band. In order to minimize implementation complexity and reduce signal processing costs, this invention adopts statistical methods to extract 12 different types of preset features for each target sub-frequency band, including: entropy feature, 5th percentile feature, 25th percentile feature, 75th percentile feature, 95th percentile feature, median feature, mean feature, variance feature, standard deviation feature, root mean square value feature, zero-crossing rate feature, and average zero-crossing rate feature. The 12 different types of preset features extracted for each target sub-frequency band are determined as the target time-frequency features of the target electromyographic signal.

[0076] The real-time identity recognition method based on electromyography (EMG) signals provided by this invention first divides the target EMG signal into multiple continuous target sub-frequency bands, and then extracts preset features from each target sub-frequency band to obtain the target time-frequency features of the target EMG signal, thereby improving the computing performance and processing speed of the terminal device. Furthermore, since the preset features include entropy features, percentile features, median features, mean features, variance features, standard deviation features, root mean square value features, zero-crossing rate features, and average zero-crossing rate features, it can ensure the completeness and diversity of features required for identity recognition, thus providing a strong basis for the accuracy of subsequent identity recognition.

[0077] Optionally, when the target time-frequency feature is one-dimensional, the specific implementation process of step 130 may include:

[0078] First, the target time-frequency features are input into a preset one-dimensional convolutional neural network for convolution, pooling, and fully connected processing to obtain the target feature vector; then, based on the preset mapping relationship between the feature vector and identity recognition information, the target identity information corresponding to the target feature vector is obtained.

[0079] Specifically, because one-dimensional convolutional neural networks (CNNs) have achieved high performance in various applications such as biomedical data classification, structural health monitoring, and early diagnosis, traditional two-dimensional deep convolutional neural networks (DCNNs) only operate on two-dimensional data and cannot operate on one-dimensional data. Furthermore, compared to two-dimensional deep DCNNs, one-dimensional CNNs are relatively easier to train, have lower computational complexity, and are suitable for real-time, low-cost applications. Therefore, this invention uses a pre-defined one-dimensional CNN trained on a one-dimensional CNN to identify the identity information of the corresponding experimenter. That is, the pre-defined one-dimensional CNN can be used for identity recognition, and it can achieve a high degree of feature mapping between one-dimensional target time-frequency features and identity information, thereby identifying the experimenter's identity information, thanks to its powerful expressive capabilities.

[0080] Similar to traditional two-dimensional convolutional neural networks, the pre-defined one-dimensional convolutional neural network in this invention mainly consists of three layers: a convolutional layer, a pooling layer, and a fully connected layer. The input layer receives one-dimensional feature signals, and the output layer is a fully connected layer with the number of neurons equal to the number of recognition classes.

[0081] It should be noted that the main difference between one-dimensional and two-dimensional convolutional neural networks is that the two-dimensional matrix of the convolution kernel and feature mapping is replaced by a one-dimensional array. This means that a one-dimensional convolutional neural network not only possesses the very powerful expressive capabilities of a two-dimensional convolutional neural network, capable of representing the mapping from features to identity information, but is also easier to train, has lower computational complexity, and therefore better meets real-time requirements.

[0082] For example, when the target time-frequency feature of the target electromyographic signal is one-dimensional, it can be represented as a one-dimensional convolutional input vector [b, c, w] input into a preset one-dimensional convolutional neural network, which first undergoes one-dimensional convolution processing through its convolutional layer to obtain a convolutional output vector [b, c′, w]; where b is the batch size, c is the number of channels, w is the length of the input one-dimensional target time-frequency feature, c′ is the number of convolutional kernels in the convolutional layer, and the size of each convolutional kernel is [k, c], where k is the width of each convolutional kernel. Next, the convolutional output vector [b,c′,w] is pooled. If the pooling layer is downsampled by a factor of 2, the pooled output vector [b,c,w / 2] can be obtained. Finally, the pooled output vector [b,c,w / 2] is fully connected. Generally, the fully connected layer is the last layer of the pre-defined one-dimensional convolutional neural network. When the width of the pooled output vector [b,c,w / 2] input to the fully connected layer is 1, the target feature vector [b,n] can be obtained, where n is the number of target identity information. Finally, based on the pre-defined mapping relationship between the feature vector and the identity recognition information, the target identity information corresponding to the target feature vector is obtained, thus obtaining the target identity information for each experimenter, such as identity document (id), name, etc.

[0083] It should be noted that the mapping relationship between feature vectors and identity recognition information can be pre-set and stored in a pre-defined one-dimensional convolutional neural network. This mapping relationship can be the mapping relationship between multiple experimenters and their identity information in a batch of experiments, so that when the pre-defined one-dimensional convolutional neural network outputs the target feature vector [b,n], the corresponding target identity information can be identified.

[0084] The real-time identity recognition method based on electromyography signals provided by this invention first inputs the target time-frequency features into a preset one-dimensional convolutional neural network for convolution, pooling, and fully connected processing, and then outputs the target feature vector. Then, based on the preset mapping relationship between the feature vector and identity recognition information, the target identity information corresponding to the target feature vector is obtained. This method combines the identity recognition function, powerful attention mechanism, and mapping relationship of the preset one-dimensional convolutional neural network to achieve the goal of quickly obtaining the target identity information. At the same time, it can meet the requirements of minimizing memory, power consumption budget, and processing time for real-time identity recognition with relatively low computational complexity, thereby effectively reducing the cost of real-time identity recognition.

[0085] The real-time identity recognition device based on electromyography (EMG) signals provided by the present invention will be described below. The real-time identity recognition device based on EMG signals described below can be referred to in correspondence with the real-time identity recognition method based on EMG signals described above.

[0086] Figure 3 An example of a real-time identity recognition device based on electromyography signals is provided, such as Figure 3 As shown, the real-time identity recognition device 300 based on electromyography (EMG) signals includes: an acquisition module 310, used to acquire a target EMG signal representing limb movement from the acquired raw EMG signal; a determination module 320, used to determine the target time-frequency features of the target EMG signal; and an identification module 330, used to perform identification processing on the target time-frequency features based on a preset convolutional neural network to obtain target identity information.

[0087] Optionally, the acquisition module 310 can be used to perform sliding window segmentation processing on the acquired raw electromyography (EMG) signals to obtain the EMG signals of the sliding window corresponding to each sliding window; calculate the average absolute value of the EMG signals of the sliding window; and obtain the target EMG signal representing limb movement based on the average absolute value and a preset dynamic threshold.

[0088] Optionally, the determining module 320 can be specifically used to select a reference sliding window electromyographic signal from multiple sliding window electromyographic signals; determine a preset dynamic threshold based on the reference average absolute value of the reference sliding window electromyographic signal; and determine the target electromyographic signal representing limb movement that is greater than the preset dynamic threshold among the multiple average absolute values.

[0089] Optionally, the determining module 320 can also be used to divide the target electromyographic signal into multiple continuous target sub-frequency bands; perform preset feature extraction on each target sub-frequency band to obtain the target time-frequency features of the target electromyographic signal; wherein, the preset features include entropy features, percentile features, median features, mean features, variance features, standard deviation features, root mean square value features, zero-crossing rate features, and average zero-crossing rate features.

[0090] Optionally, the recognition module 330 can be used to input the target time-frequency features into a preset one-dimensional convolutional neural network for convolution, pooling and fully connected processing to obtain the target feature vector; based on the preset mapping relationship between the feature vector and the identity recognition information, the target identity information corresponding to the target feature vector is obtained.

[0091] Optionally, the acquisition module 310 can also be used to acquire the raw electromyographic signals collected when the user performs a limb movement task within a target duration. The raw electromyographic signals are electromyographic signal streams and include electromyographic signals triggered by limb movement and electromyographic signals in non-limb movement states.

[0092] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4As shown, the electronic device 400 may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communication interface 420, and the memory 430 communicate with each other through the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute a real-time identity recognition method based on electromyography signals, the method including:

[0093] From the collected raw electromyographic signals, target electromyographic signals characterizing limb movement are obtained;

[0094] Determine the target time-frequency characteristics of the target electromyographic signal;

[0095] The target's time-frequency features are processed by a pre-defined convolutional neural network to obtain the target's identity information.

[0096] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0097] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is able to execute the real-time identity recognition method based on electromyography signals provided by the above methods, the method comprising:

[0098] From the collected raw electromyographic signals, target electromyographic signals characterizing limb movement are obtained;

[0099] Determine the target time-frequency characteristics of the target electromyographic signal;

[0100] The target's time-frequency features are processed by a pre-defined convolutional neural network to obtain the target's identity information.

[0101] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the real-time identity recognition method based on electromyography signals provided by the methods described above, the method comprising:

[0102] From the collected raw electromyographic signals, target electromyographic signals characterizing limb movement are obtained;

[0103] Determine the target time-frequency characteristics of the target electromyographic signal;

[0104] The target's time-frequency features are processed by a pre-defined convolutional neural network to obtain the target's identity information.

[0105] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0106] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0107] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A real-time identity recognition method based on myoelectric signals, characterized in that, The method comprises the steps of: The original myoelectric signal collected is used to obtain a target myoelectric signal representing limb movement; wherein the original myoelectric signal collected is used to obtain a target myoelectric signal representing limb movement, comprising: performing sliding window segmentation processing on the original myoelectric signal collected to obtain a sliding window myoelectric signal corresponding to each sliding window; calculating the average absolute value of the sliding window myoelectric signal; obtaining a target myoelectric signal representing limb movement based on the average absolute value and a preset dynamic threshold; the original myoelectric signal collected is subjected to sliding window segmentation processing to obtain a sliding window myoelectric signal corresponding to each sliding window, comprising: determining the size of the sliding window and the step length of adjacent sliding windows according to the recognition rate and the reflection time, and performing sliding window segmentation processing on the original myoelectric signal according to the size of the sliding window and the step length of adjacent sliding windows to obtain the sliding window myoelectric signal corresponding to each sliding window; the average absolute value MAD = 1 / x i is the i-th data in the sliding window myoelectric signal, is the mean value of the sliding window myoelectric signal X, and MAD is the average absolute value of the sliding window myoelectric signal; the preset dynamic threshold is used as a judgment standard; determining a target time-frequency feature of the target electromyographic signal; performing identification processing on the target time-frequency feature based on a preset convolutional neural network to obtain target identity information; when the target time-frequency feature is one-dimensional, the identification processing on the target time-frequency feature based on the preset convolutional neural network to obtain the target identity information comprises: inputting the target time-frequency feature into a preset one-dimensional convolutional neural network for convolution, pooling and full connection processing to obtain a target feature vector; and obtaining the target identity information corresponding to the target feature vector based on a preset mapping relationship between the target feature vector and identity recognition information. 2.The real-time identity recognition method based on myoelectric signals according to claim 1, characterized in that, The target electromyographic signal representing limb movement is obtained based on the average absolute value and a preset dynamic threshold, and the method comprises the steps of: selecting a reference sliding window electromyographic signal from the plurality of sliding window electromyographic signals; determining a preset dynamic threshold based on a reference average absolute value of the reference sliding window electromyographic signal; determining a target electromyographic signal representing limb movement from the plurality of average absolute values that is greater than the preset dynamic threshold. 3.The real-time identity recognition method based on myoelectric signals according to claim 1, characterized in that, The method comprises the steps of: dividing the target electromyographic signal into a plurality of continuous target sub-frequency bands; performing preset feature extraction on each target sub-frequency band to obtain a target time-frequency feature of the target electromyographic signal; wherein the preset features include entropy features, percentile features, median features, mean features, variance features, standard deviation features, root mean square value features, zero-crossing rate features and average zero-crossing rate features. 4.The real-time identity recognition method based on myoelectric signals according to claim 1, characterized in that, Before the step of obtaining a target electromyographic signal representing limb movement from the collected original electromyographic signal, the method further comprises the steps of: obtaining an original electromyographic signal collected when a user performs a limb movement task within a target time period, wherein the original electromyographic signal is an electromyographic signal stream and includes electromyographic signals triggered by limb movement and electromyographic signals in a non-limb movement state.

5. A real-time identity recognition device based on myoelectric signals, characterized by The method comprises the steps of: The acquisition module is configured to acquire a target electromyography signal representing limb movement from the collected raw electromyography signal; wherein the acquisition of the target electromyography signal representing limb movement from the collected raw electromyography signal comprises: performing a sliding window segmentation process on the collected raw electromyography signal to obtain a sliding window electromyography signal corresponding to each sliding window; calculating an average absolute value of the sliding window electromyography signal; and obtaining the target electromyography signal representing limb movement based on the average absolute value and a preset dynamic threshold; the sliding window segmentation process on the collected raw electromyography signal to obtain a sliding window electromyography signal corresponding to each sliding window comprises: determining the size of the sliding window and the step length of adjacent sliding windows according to the recognition rate and the reflection time, and performing the sliding window segmentation process on the raw electromyography signal according to the size of the sliding window and the step length of adjacent sliding windows to obtain the sliding window electromyography signal corresponding to each sliding window; and the average absolute value x i is the i-th data in the sliding window electromyography signal, is the mean value of the sliding window electromyography signal X, and MAD is the average absolute value of the sliding window electromyography signal; the preset dynamic threshold serves as a judgment criterion; a determination module configured to determine a target time-frequency feature of the target electromyographic signal; an identification module configured to perform identification processing on the target time-frequency feature based on a preset convolutional neural network to obtain target identity information; when the target time-frequency feature is one-dimensional, the identification processing on the target time-frequency feature based on the preset convolutional neural network to obtain the target identity information comprises: inputting the target time-frequency feature into a preset one-dimensional convolutional neural network for convolution, pooling and full connection processing to obtain a target feature vector; and obtaining the target identity information corresponding to the target feature vector based on a preset mapping relationship between the target feature vector and identity recognition information.

6. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the steps of the real-time identity recognition method based on electromyographic signals according to any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the real-time identity recognition method based on electromyographic signals according to any one of claims 1 to 4.

8. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the real-time identity recognition method based on electromyographic signals according to any one of claims 1 to 4.

Images