A Muscle Activation Pattern Coding Method for EMG Biometrics

Abstract

The present invention discloses a muscle activation pattern encoding method for electromyographic signal biometric identification, including: collecting multi-channel electromyographic signals of user objects and constructing data sets; performing denoising on the original electromyographic signals, identifying and segmenting event signal segments and non-event signal segments; using a sliding window to intercept the event signal segments, calculating the energy of each channel in each time window to obtain an energy distribution vector; performing cluster analysis on all energy distribution vectors, determining the optimal number of pattern categories, each category representing a different muscle activation pattern, using different letters for encoding, and converting the event signal segments into muscle activation pattern sequences. The present invention can extract muscle activation pattern sequences, is suitable for a variety of action scenarios such as signing and performing gestures, can effectively reduce intra-class differences, improve the performance of identity recognition or authentication systems, and is suitable for commercial wearable bracelets with few channels, greatly improving the practicality and convenience of identity recognition or authentication.

Description

Technical Field

[0001] The invention belongs to the technical field of biometric identification, and in particular relates to a muscle activation pattern encoding method for electromyographic signal biometric identification. Background Art

[0002] In recent years, with the increasing demand for combating security intrusions and malicious behaviors, researchers have proposed various security solutions. Biometrics, as an advanced identification or authentication method, requires comparing an individual's physiological or behavioral characteristics with the template entered during the registration phase. Traditional physiological characteristics, such as fingerprints, faces, and irises, are easy to steal and forge through cameras and other channels, and once leaked, they cannot be cancelled, which poses a security risk. In order to solve this problem, many methods combine behavioral characteristics with physiological characteristics. Due to the characteristics of behavioral characteristics that can be cancelled and are not easy to forge, combining them with physiological characteristics effectively improves the security of biometrics. Among them, the electrical activity of the nerve fibers that control the muscles will generate electromyographic signals (EMG), which contain both the individual's physiological characteristics and the individual's behavioral characteristics, and are most suitable for combination with behavioral characteristics; and surface electromyographic signals (sEMG) can be recorded by electrodes connected to the surface of the skin, so identity identification or authentication based on sEMG has better security and convenience.

[0003] The related technologies for identity authentication or recognition based on electromyographic signals usually require the user to sign or make a specific sequence of gestures. During this process, surface electromyographic signals are collected from the surface of the user's arm, and then event fragments are identified and segmented from the signal. After feature extraction, the obtained fragments are input into the classifier for identity authentication or recognition. The related technologies of the feature extraction part mainly include: ① Identity recognition technology based on muscle synergy electromyographic signals; ② Electromyographic signal identity recognition technology based on traditional features.

[0004] However, method ①, such as the muscle synergy features used in the literature [A. Asemi, K. Maghooli, FN Rahatabad, H. Azadeh, Handwritten signatures verification based on arm and hand muscles synergy, Biomedical Signal Processing and Control 76 (2022) 103697], requires the use of multiple sets of electrodes to collect electromyographic signals from multiple muscle parts of the arm. It is not suitable for wearable devices with a small number of channels and a concentrated measurement range. The measurement scheme is relatively complex and difficult to promote. Method ②, such as the traditional electromyographic signal features extracted in the literature [X. Jiang, X.Liu, J. Fan, X. Ye, C. Dai, EA Clancy, D. Farina, W. Chen, Enhancing IoTsecurity via cancelable HD-sEMG-based biometric authentication password, encoded by gesture, IEEE Internet of Things Journal 8 (22) (2021) 16535–16547], also has high requirements on the number of signal channels. The performance of this method will decrease rapidly as the number of channels decreases, and it is difficult to generalize to more convenient and practical wearable devices. The intra-class variation of traditional electromyographic features is large, and the performance is poor when multiple identity recognition or authentication is performed over multiple days; in addition, traditional electromyographic features are only applicable to relatively simple gestures, and the performance is poor when performing complex signature behaviors. Summary of the invention

[0005] In view of the above, the present invention provides a muscle activation pattern encoding method for electromyographic signal biometric identification, which is suitable for both the original more complex laboratory measurement scheme and the more convenient commercial bracelet measurement scheme, greatly improving the practicality and accuracy of identity recognition or authentication technology based on electromyographic signals.

[0006] A muscle activation pattern encoding method for electromyographic signal biometric identification comprises the following steps:

[0007] (1) Collect multi-channel sEMG of user subjects and construct sEMG dataset;

[0008] (2) De-noising and identifying the multi-channel sEMG of the user subject, and segmenting it into multiple event signal segments and multiple non-event signal segments;

[0009] (3) For any event signal segment, a sliding window is used to intercept it and the energy of each channel in each time window is calculated to obtain an energy distribution vector, thereby forming an energy distribution vector sequence of the event signal segment;

[0010] (4) Cluster analysis is performed on the energy distribution vectors to determine the optimal number of pattern categories. Each category represents a muscle activation pattern and is encoded. Finally, the encoding is used to map the energy distribution vector sequence of the user object's event signal fragments into a coding sequence of muscle activation patterns.

[0011] Furthermore, in step (1), for the user group, multiple groups of electrodes, flexible electrode arrays or wearable electromyography acquisition devices are used to record the multi-channel sEMG of the arm muscles of each user subject during the execution of the signature or gesture action, so as to construct a sEMG data set.

[0012] Furthermore, the specific implementation of step (2) is as follows:

[0013] 2.1 Use a Butterworth filter with a passband of 20-500 Hz to bandpass filter the multi-channel sEMG to remove low-frequency motion artifacts, ECG interference, and high-frequency noise;

[0014] 2.2 Use a 50Hz notch filter to perform band-stop filtering on the multi-channel sEMG after bandpass filtering to remove power frequency interference;

[0015] 2.3 The multi-channel sEMG after band-stop filtering is subjected to envelope processing. The signal segments whose envelope absolute values ​​are greater than the set threshold are recorded as event signal segments of the user performing signature or gesture actions, and the remaining signal segments are recorded as non-event signal segments.

[0016] Furthermore, the specific implementation of step (3) is as follows:

[0017] 3.1 Using a sliding window with a fixed window length and fixed step size to intercept the event signal fragment, a window sequence is obtained W :

[0018]

[0019] in: W i Indicates the first segment to which the event signal is intercepted. i time window, i is a natural number and 1≤ i ≤ N , N is the number of time windows captured;

[0020] 3.2 For window sequencesW Any time window W i , the time window is calculated by the following formula W i The RMS value of sEMG in each channel:

[0021]

[0022] in: For time window Neidi c The RMS value of the channel sEMG, For time window Neidi c Channel sEMG n Sample values, L is a fixed window length, c is a natural number and 1≤ c ≤ C , C is the number of channels;

[0023] 3.3 Generate an energy vector sequence based on the RMS value of sEMG of each channel in each time window R :

[0024] ;

[0025]

[0026] in: For time window W i The energy vector composed of the RMS values ​​of sEMG of all channels;

[0027] 3.4 For any user object, the muscle activation threshold of each channel of the user object is calculated by the following formula:

[0028]

[0029] in, For user object c The muscle activation threshold of the channel, For user object r Non-event signal segment c sEMG of the channel, M is the number of non-event signal fragments of the user object, RMS ( ) represents the RMS value calculation function;

[0030] 3.5 For the energy vector Any RMS value in , using muscle activation threshold Correct it to get the corrected energy vector , the specific calculation method is as follows:

[0031] ;

[0032]

[0033] in: Corresponding to Corrected value;

[0034] 3.6 Energy vector Normalize to get the energy distribution vector , which then constitutes the energy distribution vector sequence of the event signal fragment NR , the specific calculation expression is as follows:

[0035] ;

[0036]

[0037] Furthermore, the specific implementation method of determining the optimal number of pattern categories in step (4) is as follows: first, a part of user objects are selected, and the energy distribution vectors of all event signal segments of these user objects are placed in a clustering pool for clustering analysis, and the optimal number of clusters is determined according to the clustering evaluation index; then, user objects are continuously added, and the energy distribution vectors of all event signal segments of the newly added user objects are added to the clustering pool for clustering analysis. When the number of users increases but the optimal number of clusters no longer increases, the optimal number of clusters at this time is used as the optimal number of pattern categories.

[0038] A muscle activation pattern coding system for myoelectric signal biometric identification, used to execute the above muscle activation pattern coding method, the muscle activation pattern coding system comprising:

[0039] A signal processing module, used for denoising and identifying the multi-channel sEMG of the user object, and segmenting it into a plurality of event signal segments and a plurality of non-event signal segments;

[0040] The energy distribution calculation module uses a sliding window to intercept any event signal segment and calculates the energy of each channel in each time window to obtain an energy distribution vector, thereby forming an energy distribution vector sequence of the event signal segment;

[0041] The cluster coding module is used to perform cluster analysis on the energy distribution vector to determine the optimal number of pattern categories. Each category represents a muscle activation pattern and encodes it. Finally, the encoding is used to map the energy distribution vector sequence of the event signal fragment of the user object into a coding sequence about the muscle activation pattern.

[0042] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor is used to execute the computer program to implement the above-mentioned muscle activation pattern encoding method for electromyographic signal biometric identification.

[0043] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned muscle activation pattern encoding method for myoelectric signal biometric identification.

[0044] Based on the above technical solution, the present invention has the following beneficial technical effects:

[0045] 1. The present invention can extract muscle activation pattern sequences and is suitable for various scenarios such as signing and performing gestures.

[0046] 2. The muscle activation pattern encoding of the present invention can effectively reduce intra-class differences and improve the performance of identity recognition or authentication systems.

[0047] 3. Compared with wearing multiple sets of electrodes or electrode arrays on the arm, the present invention is suitable for commercial wearable bracelets with few channels, which greatly improves the practicality and convenience of identity recognition or authentication. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 The present invention is a flowchart of a muscle activation pattern encoding method for electromyographic signal biometric identification according to an embodiment of the present invention.

[0049] Figure 2 The figure is a schematic diagram of a usage scenario of an embodiment of the present invention.

[0050] Figure 3 It is a schematic diagram of energy distribution vector calculation, cluster analysis, and muscle activation pattern encoding in an embodiment of the present invention.

[0051] Figure 4 This is a mapping diagram of muscle activation pattern encoding corresponding to different energy distributions when a four-channel acquisition device is used to record a user's electromyographic signal in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to describe the present invention more specifically, the technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0053] like Figure 1 As shown, this embodiment provides a muscle activation pattern encoding method for electromyographic signal biometric identification, comprising the following steps:

[0054] Step 1: Use an electromyography acquisition device to record multi-channel electromyography signals of the user's arm when the user performs actions such as signing or gestures.

[0055] like Figure 2 As shown, this embodiment uses a 4-channel (Ch.1, Ch.2, Ch.3, Ch.4) wearable electromyographic bracelet 202 to record the multi-channel surface electromyographic signals of the arm muscles of the user subject 201 during the signing process, and constructs a sEMG data set for cluster analysis on the host computer 203.

[0056] Step 2: De-noise the EMG signal, identify and segment the event signal segments and non-event signal segments.

[0057] 2-1 Use a Butterworth filter with a passband of 20~500Hz to bandpass filter the acquired signal to remove low-frequency motion artifacts, ECG interference and high-frequency noise;

[0058] 2-2 Use a 50Hz notch filter to perform band-stop filtering on the collected signal to remove power frequency interference;

[0059] 2-3 Envelope processing is performed on the processed signal. The segments whose envelope absolute values ​​are greater than 10% of the maximum value are the segments when the user performs events such as signing or gestures, which are recorded as event signal segments, and the remaining segments are non-event signal segments;

[0060] 2-4 The above segments are segmented to divide the sEMG data set into an event signal segment group and a non-event signal segment group.

[0061] Step 3: Use a sliding window to intercept the event signal fragments of multiple channels, calculate the total energy of each channel in the window, and form an energy distribution vector, such as Figure 3 As shown:

[0062] 3-1 This embodiment uses a sliding window with a window length of 100ms and a step length of 50ms to intercept the multi-channel signal of the event signal segment, and each segment obtains a window sequence :

[0063]

[0064] 3-2 Calculate the root mean square value RMS of the multi-channel electromyographic signal in each window of the event signal segment window sequence:

[0065]

[0066] in, For Window No. c The RMS value of the channel, For Window The EMG signal sequence of the c-th channel, The length of the sliding window.

[0067] The RMS values ​​of all channels in the sliding window sequence together constitute an energy vector sequence :

[0068]

[0069] in, For Window The energy vector composed of the RMS values ​​of all channels is:

[0070]

[0071] 3-3 Calculate the RMS value of the non-event signal segment, and use 1.2 times of this value as the detection threshold for muscle activation :

[0072]

[0073] in: For user object c The muscle activation threshold of the channel, For non-event signal segment c The EMG signal sequence of the channel, M The number of non-event signal fragments for the user object.

[0074] 3-4 Perform threshold detection on the energy vector of the event signal segment. If the RMS value of a channel in a certain time window exceeds the threshold of the channel, it is considered that the muscle of the channel is activated in the time window and the value is retained. Otherwise, the RMS value is set to zero to obtain the corrected energy vector :

[0075]

[0076] in, For Window The corrected RMS value of the cth channel is calculated as follows:

[0077]

[0078] 3-5 Normalize the corrected energy vector of each window to obtain the energy distribution vector , which then constitutes the energy distribution vector sequence of the event signal fragment NR :

[0079] ;

[0080]

[0081] Step 4: Perform cluster analysis on all energy distribution vectors to obtain several energy distribution classes, which are coded with different letters and correspond to different muscle activation patterns. According to this mapping, all event signal fragments are converted into muscle activation pattern sequences, such as Figure 3 As shown:

[0082] 4-1 Perform Gaussian mixture cluster analysis on the energy distribution vectors of all event signal fragments to determine the optimal number of categories. The specific operation is: perform multiple groups of cluster analysis, increase the number of users participating in the cluster analysis in each group, and select the optimal number of clusters for each group based on the Calinski-Harabasz score; when the number of users increases but the optimal number of clusters no longer increases, the most optimal number of clusters at this time is the optimal number of model categories.

[0083] 4-2 All the categories clustered under the optimal number of model categories are coded using letters ABCDEFG, such as Figure 4 As shown, each category represents a muscle activation pattern.

[0084] 4-3 According to Figure 4 The mapping rules convert the user's event signal fragments into a sequence of muscle activation patterns composed of letters.

[0085] The above description of the embodiments is to facilitate the understanding and application of the present invention by those skilled in the art. It is obvious that those skilled in the art can easily make various modifications to the above embodiments and apply the general principles described herein to other embodiments without creative work. Therefore, the present invention is not limited to the above embodiments. Improvements and modifications made by those skilled in the art to the present invention based on the disclosure of the present invention should be within the protection scope of the present invention.

Claims

1. A muscle activation pattern encoding method for electromyographic signal biometric identification, comprising the following steps: (1) Collect multi-channel sEMG of user subjects and construct sEMG dataset; (2) De-noising and identifying the multi-channel sEMG of the user subject, and segmenting it into multiple event signal segments and multiple non-event signal segments; (3) For any event signal segment, a sliding window is used to intercept it and the energy of each channel in each time window is calculated to obtain an energy distribution vector, thereby forming an energy distribution vector sequence of the event signal segment. The specific implementation method is as follows: 3.1 Using a sliding window with a fixed window length and fixed step size to intercept the event signal fragment, a window sequence is obtained W : ; in: W i Indicates the first segment to which the event signal is intercepted. i time window, i is a natural number and 1≤ i ≤ N , N is the number of time windows captured; 3.2 For window sequences W Any time window W i , the time window is calculated by the following formula W i The RMS value of sEMG in each channel: ; in: For time window Neidi c The RMS value of the channel sEMG, For time window Neidi c Channel sEMG n Sample values, L is a fixed window length, c is a natural number and 1≤ c ≤ C , C is the number of channels; 3.3 Generate an energy vector sequence based on the RMS value of sEMG of each channel in each time window R : ; ; in: For time window W i The energy vector composed of the RMS values ​​of sEMG of all channels; 3.4 For any user object, the muscle activation threshold of each channel of the user object is calculated by the following formula: ; in, For user object c The muscle activation threshold of the channel, For user object r Non-event signal segment c sEMG of the channel, M is the number of non-event signal fragments of the user object, RMS ( ) represents the RMS value calculation function; 3.5 For the energy vector Any RMS value in , using muscle activation threshold Correct it to get the corrected energy vector , the specific calculation method is as follows: ; ; in: Corresponding to Corrected value; 3.6 Energy vector Normalize to get the energy distribution vector , which then constitutes the energy distribution vector sequence of the event signal fragment NR , the specific calculation expression is as follows: ; ; (4) Cluster analysis is performed on the energy distribution vectors to determine the optimal number of pattern categories. Each category represents a muscle activation pattern and is encoded. Finally, the encoding is used to map the energy distribution vector sequence of the user object's event signal fragments into a coding sequence of muscle activation patterns.

2. The muscle activation pattern encoding method for electromyographic signal biometric identification according to claim 1, characterized in that: In the step (1), for the user group, multiple groups of electrodes, flexible electrode arrays or wearable electromyography acquisition devices are used to record the multi-channel sEMG of the arm muscles of each user subject during the execution of the signature or gesture action, so as to construct an sEMG data set.

3. The muscle activation pattern encoding method for electromyographic signal biometric identification according to claim 1, characterized in that: The specific implementation of step (2) is as follows: 2.1 Use a Butterworth filter with a passband of 20-500 Hz to bandpass filter the multi-channel sEMG to remove low-frequency motion artifacts, ECG interference, and high-frequency noise; 2.2 Use a 50Hz notch filter to perform band-stop filtering on the multi-channel sEMG after bandpass filtering to remove power frequency interference; 2.3 The multi-channel sEMG after band-stop filtering is subjected to envelope processing. The signal segments whose envelope absolute values ​​are greater than the set threshold are recorded as event signal segments of the user performing signature or gesture actions, and the remaining signal segments are recorded as non-event signal segments.

4. The muscle activation pattern encoding method for electromyographic signal biometric identification according to claim 1, characterized in that: The specific implementation method of determining the optimal number of pattern categories in step (4) is as follows: first, a part of user objects are selected, and the energy distribution vectors of all event signal segments of these user objects are placed in a clustering pool for clustering analysis, and the optimal number of clusters is determined according to the clustering evaluation index; then, user objects are continuously added, and the energy distribution vectors of all event signal segments of the newly added user objects are added to the clustering pool for clustering analysis. When the number of users increases but the optimal number of clusters no longer increases, the optimal number of clusters at this time is used as the optimal number of pattern categories.

5. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: The processor is used to execute the computer program to implement a muscle activation pattern encoding method for electromyographic signal biometric identification as described in any one of claims 1 to 4.

6. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it implements a muscle activation pattern encoding method for electromyographic signal biometric identification as described in any one of claims 1 to 4.

Images