Identity recognition method based on bioelectric signal fusion
Through bioelectric signal fusion technology, combined with convolution-long and short-term memory neural network, the gesture password and user identity are identified, and the replication, imitation instability and daytime differences of existing biometric recognition technologies are solved, and the problem of replication, imitation and inter-day differences is achieved. Higher identity recognition accuracy and account security.
Patent Information
- Application Number
- CN202211416967.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-11-10
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2042-11-10
AI Technical Summary
The existing biometric identification technology has problems such as reproducibility, imitation instability and day-to-day differences, resulting in low identification accuracy and difficulty in ensuring account security.
Using the bioelectric signal-based identity recognition method, through the fusion of high-density surface electromyography and electrocardiogram signals, combined with convolution-long and short-term memory neural network, the gesture password is identified and the user identity is confirmed, and online learning is added to the model training. Module, update user data in real time to reduce daytime differences.
It improves the accuracy and system security of identity identification, provides changeable biometric templates, reduces the risk of malicious attacks, and reduces the misjudgment rate through the online learning module, enhancing the privacy and security of the account.
Smart Images

Figure CN115670446B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of machine learning and identity recognition technology, and in particular relates to an identity recognition method based on bioelectric signal fusion. Background Art
[0002] With the development of the information age, the information security of accounts has attracted the attention of the whole society. Many fields require strict identity recognition systems, such as secure payment, access to information, and even permission to use military weapons.
[0003] Biometric identification is an identification technology based on an individual's physiological or behavioral characteristics. However, current biometric identification templates still have shortcomings. For example, physiological characteristics such as fingerprints, faces, and DNA are easily copied and imitated, and behavioral characteristics such as voice and gait can also be imitated and are unstable. Bioelectric signals such as surface electromyography and electrocardiogram signals are directly related to each person's physiological conditions and are not easily exposed. Using them as biometric identification templates has the advantages of uniqueness, privacy, liveness, and universality. Summary of the invention
[0004] In the solution of the present invention, bioelectric signals are combined with gesture passwords, making biometric identification technology modifiable. When creating an account, the user selects and sorts gestures in the system's gesture pool and sets a gesture password for the account. Once the account's identification template is exposed, a new gesture password can be set, making biometrics easy to change while improving system security.
[0005] In addition, considering the day-to-day differences in bioelectric signals, the accuracy of identity recognition will be greatly reduced. In order to improve this problem, the present invention adds an online learning module during the model training process to update user data in a timely manner and iterate the network model. By detecting abnormal changes in bioelectric signals, it is also possible to determine whether the user is in a dangerous state.
[0006] In order to solve the defects and deficiencies in the prior art, the present invention proposes an identity recognition method based on bioelectric signal fusion, which analyzes the energy of high-density surface electromyographic signals, segments the gesture signal segments according to the set energy threshold, and uses the convolution (CNN) module in the convolution-long short-term memory (CNN-LSTM) neural network to identify whether the gesture action is correct. When it is determined to be the set correct gesture password, the input electromyographic signal is fused with features such as the electrocardiogram signal, and the bio-fusion features input this time are compared with the features predicted by the long short-term memory (LSTM) module to confirm the user's identity. After the identity is confirmed, the data is updated and iterated through the online learning module, and the network is trained again to reduce the misjudgment rate caused by daytime differences; if it is a malicious attack by other users, a warning or alarm process is taken.
[0007] The technical solution adopted by the present invention to solve the technical problem is:
[0008] An identity recognition method based on bioelectric signal fusion is characterized by: preprocessing the user's high-density surface electromyography signal and electrocardiogram signal, extracting features and fusing them to form a bio-fusion feature vector; inputting the bio-fusion features into a convolution-long short-term memory neural network, and identifying whether the gesture password is correct through the convolution module; and then confirming whether the user identity is correct through the long short-term memory module.
[0009] Furthermore, after the identity is confirmed, the online learning module is enabled to update and iterate the account data, and the network is trained again to reduce the misjudgment rate caused by daytime differences; if it is a malicious attack from other users, warnings or alarms are taken.
[0010] Furthermore, the user's high-density surface electromyographic signal and electrocardiographic signal are preprocessed, and the features are fused to form a biological fusion feature vector; specifically, the high-density surface electromyographic signal and electrocardiographic signal are preprocessed by filtering and denoising, the electromyographic signal energy is analyzed, and the electromyographic signal gesture segment is intercepted according to the energy threshold, the electromyographic signal feature is extracted, and the electrocardiographic signal is intercepted and the feature is extracted; when the neural network determines that the correct gesture password is obtained through the input electromyographic feature, the electromyographic signal and the electrocardiographic signal are feature-fused to form a biological fusion feature vector;
[0011] Before recognition, you need to select multiple gestures from the gesture pool and sort them, and set them as the gesture password for this account.
[0012] Furthermore, the feature is input into the convolution-long short-term memory neural network, and the convolution module is used to identify whether the gesture password is correct; and then the user identity is confirmed through the long short-term memory module; specifically: first, the electromyographic feature is constructed into an electromyographic feature image and input into the convolution module to identify whether it is a correct gesture password; if the gesture password is correct, the long short-term memory module is used to predict the existing biological fusion features of the account, and the predicted features are compared with the fit of the current fusion features to verify the user identity.
[0013] Furthermore, the high-density surface electromyographic signals of m×n channels are preprocessed by filtering and denoising, the energy value of the electromyographic signals is calculated, the electromyographic signal segments of k gestures are intercepted according to the set energy threshold, the features are selected, and the electromyographic feature images of size m×n are constructed; the electromyographic feature images of k gestures are sequentially input into the convolution-long short-term memory neural network, the gestures are recognized and the gesture sequence is judged through the convolution module, and it is confirmed whether it is the correct gesture password: if it is judged to be an incorrect gesture password, the system is interrupted to deny the access application; if it is a correct gesture, the next step of judgment is carried out;
[0014] When the gesture password is entered correctly, the ECG signal is preprocessed by filtering, denoising, and benchmark segmentation, and the features of the processed data are extracted to obtain And combined with the high-density surface electromyography signal features extracted by the convolution module i=1,...,k are fused to form a biological feature vector xs; the existing h biological fusion features j=1,...,h Input the long short-term memory module to predict the current biometric feature x s ', compare the actual feature xs with the predicted feature x s 'Compare and determine whether it is the correct identity;
[0015] Calculate the absolute percentage error r(%) for the input features:
[0016]
[0017] Set a threshold a. When r(%)≤a, it is a correct identity and the data is saved in the online learning module. When r(%)>a, it is an incorrect identity and the access application is rejected with a warning or alarm.
[0018] Furthermore, the specific working method of the online learning module is: when the convolution-long short-term memory neural network confirms the correct identity, the online learning module is started to perform real-time training of the network model; the biological fusion features of this time are saved, the features with a longer timeline are deleted, and the convolution-long short-term memory neural network is retrained using the updated data set to prepare for the next identity recognition.
[0019] Compared with the prior art, the present invention and its preferred solution can fuse the high-density surface electromyographic signal of the forearm with the user's electrocardiogram signal when the user makes a gesture password set by himself, form a bio-fusion feature, and input it into the CNN-LSTM neural network to identify the user. The network combines the temporal and spatial characteristics of bioelectric signals, and can effectively solve the similarity problem and daytime difference problem existing in identity recognition. When it is determined that the input signal is not the biometric feature of the user, a warning or alarm processing is performed. At the same time, an online learning module is added to update the user's own data and train the network model in real time, so that the problem of large daytime differences in bioelectric signals can be well solved.
[0020] It fuses high-density surface electromyography signals with electrocardiogram signals to form a biometric feature vector, proposes a CNN-LSTM neural network for user gesture and identity recognition, and adds an online learning module to issue a warning or alarm when a signal is not recognized as that of the user. This ensures the security and privacy of user accounts and improves the accuracy of the neural network. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:
[0022] Figure 1 This is a flowchart of identity recognition based on bioelectric signal fusion according to an embodiment of the present invention.
[0023] Figure 2 This is a flowchart of identity recognition based on a CNN-LSTM neural network according to an embodiment of the present invention.
[0024] Figure 3 It is a structural diagram of a CNN-LSTM neural network according to an embodiment of the present invention.
[0025] Figure 4 It is a flow chart of the online learning module of an embodiment of the present invention. DETAILED DESCRIPTION
[0026] In order to make the features and advantages of this patent more obvious and easy to understand, the following embodiments are specifically described in detail as follows:
[0027] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.
[0028] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.
[0029] The present invention discloses an identity recognition method based on bioelectric signal fusion. By preprocessing high-density surface electromyographic signals and electrocardiographic signals, extracting features and fusing them to form a bio-fusion feature vector; inputting the bio-fusion features into a convolution-long short-term memory (CNN-LSTM) neural network, and using a convolution (CNN) module to identify whether the gesture password is correct. Then, using a long short-term memory (LSTM) module to confirm whether the user's identity is correct. After the identity is confirmed, the online learning module is enabled, the account data is updated and iterated, and the network is trained again to reduce the misjudgment rate caused by daytime differences; if it is a malicious attack by other users, a warning or alarm process is taken.
[0030] The present invention performs preprocessing such as filtering and denoising on high-density surface electromyographic signals and electrocardiographic signals, calculates the energy of high-density surface electromyographic signals, and segments the electromyographic signals according to the energy threshold, intercepts the effective signal segments of the electromyographic signals and electrocardiographic signals, extracts features and fuses them to form a biological feature vector, and inputs it into the CNN-LSTM neural network. First, the electromyographic features are constructed as electromyographic feature images and input into the CNN module to identify whether it is a correct gesture password. If the gesture password is correct, the existing biological fusion features of the account are predicted through the LSTM module, and the fit between the predicted features and the current fusion features is compared to verify the identity of the user.
[0031] The present invention eliminates the day-to-day differences in bioelectric signals by adding an online learning module during the model training process. When the identity is confirmed, the input data is saved to the account, the data with the furthest timeline in the account is deleted, and the network model is retrained after the data is updated.
[0032] The following is further described in detail with reference to the accompanying drawings:
[0033] (1) Identity recognition based on bioelectric signal fusion
[0034] The bioelectric signal recognition method of the present invention mainly includes five parts: data processing, feature extraction and fusion, gesture password recognition, identity discrimination and online learning module.
[0035] Previously, it was necessary to select and sort several gestures from the gesture pool of the system and set them as the gesture password of this account. Pre-process the high-density surface electromyographic signals and electrocardiographic signals by filtering and denoising, analyze the energy of the electromyographic signals, and intercept the electromyographic signal gesture segments according to the energy threshold, select appropriate features and form an electromyographic feature image; then evaluate the quality of the electrocardiographic signal and segment the reference points, fuse the electromyographic and electrocardiographic signals, and send them to the CNN-LSTM neural network model for training; first identify the gesture password based on the electromyographic feature image, and when the gesture password is correct, identify the identity through the biometric fusion features. After confirming the correct identity, start the online learning module, update and iterate the data and retrain the network; if it is an incorrect identity, issue a warning or alarm.
[0036] The flow chart of the identity recognition method based on bioelectric signal fusion is as follows Figure 1 shown.
[0037] (2) Identity recognition based on CNN-LSTM neural network
[0038] The identity recognition flow chart based on CNN-LSTM neural network is as follows Figure 2 shown.
[0039] In this implementation, the array electrodes of high-density surface electromyography signals provide high-resolution sensing points and capture images of the activation areas corresponding to the target muscles. The CNN neural network has great advantages in image processing. Combined with the muscle activation images provided by high-density surface electromyography signals, better gesture classification results can be achieved. The memory unit of the LSTM network can well filter past information and combine it with current information, modeling time domain information and improving the model's ability to express time domain information. The combination of the two can effectively collect spatial and temporal features and improve the accuracy of identity recognition.
[0040] The high-density surface electromyographic signals of m×n channels are preprocessed by filtering and denoising, the energy value of the electromyographic signals is calculated, the electromyographic signal segments of k gestures are intercepted according to the set energy threshold, the appropriate features are selected, and the electromyographic feature images of size m×n are constructed; the electromyographic feature images of k gestures are sequentially input into the CNN-LSTM neural network, and the CNN module is used to recognize the gestures and the sequence of gestures to determine whether it is the correct gesture password. If it is judged to be an incorrect gesture password, the system will be interrupted to deny the access application; if it is a correct gesture, the next step of judgment will be carried out.
[0041] When the gesture password is entered correctly, the ECG signal is preprocessed by filtering, denoising, and benchmark segmentation, and features are extracted from the processed data. And compared with the high-density surface electromyography signal features extracted by the CNN module (i=1,...,k) are fused to form the biological feature vector x s . The existing h biological fusion features (j=1,...,h) is sent to the LSTM module to predict the biometric feature x s ', the actual feature x s With the predicted feature x s 'Compare and determine whether it is the correct identity.
[0042] Calculate the absolute percentage error r(%) for the input features:
[0043]
[0044] A threshold a is set. When r(%)≥a, it is a correct identity and the data is saved in the online learning module. When r(%)<a, it is an incorrect identity and the access application is rejected for warning or alarm processing.
[0045] The CNN-LSTM neural network structure of the present invention is as follows Figure 3 shown.
[0046] (3) Online learning module
[0047] The present invention adopts bioelectric signal fusion technology and adds a network online learning module to reduce the day-to-day variability of bioelectric signals through data iteration and network update, thereby improving the accuracy of the method in identity recognition.
[0048] The flow chart of the online learning module in this embodiment is as follows Figure 4 shown.
[0049] When the CNN-LSTM neural network confirms the correct identity, the online learning module is started to train the network model in real time. The biological fusion features of this time are saved, the features with a longer timeline are deleted, and the CNN-LSTM neural network is retrained with the updated data set to prepare for the next identity recognition.
[0050] The present invention adopts an online learning module in conjunction with the LSTM module in CNN-LSTM. By screening the information at the past moment and combining it with the information at the current moment, it can effectively reduce the daytime variability of bioelectric signals and improve the accuracy of the identity recognition method based on bioelectric signal fusion.
[0051] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.
[0052] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0053] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0054] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0055] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.
[0056] This patent is not limited to the above-mentioned optimal implementation mode. Anyone can derive other forms of identity recognition methods based on bioelectric signal fusion under the inspiration of this patent. All equal changes and modifications made according to the scope of the patent application of the present invention should be covered by this patent.
Claims
1. An identity recognition method based on bioelectric signal fusion, characterized in that: By preprocessing high-density surface electromyographic signals and electrocardiographic signals, extracting features and fusing them to form a biological fusion feature vector; inputting the biological fusion features into the convolution-long short-term memory neural network, using the convolution module to identify whether the gesture is correct; and then using the long short-term memory module to confirm whether the identity is correct; Preprocess the high-density surface electromyography signal and electrocardiogram signal, extract the features and fuse them to form a biological fusion feature vector; Specifically, the high-density surface electromyographic signals and electrocardiographic signals are pre-processed by filtering and denoising, the electromyographic signal energy is analyzed and the electromyographic signal gesture segment is intercepted according to the energy threshold, the electromyographic signal features are extracted, and the electrocardiographic signal is intercepted and features are extracted; When the neural network determines that the inputted electromyographic features are the correct gesture password, the electromyographic signal and the electrocardiographic signal are fused to form a biological fusion feature vector; Before recognition, you need to select multiple gestures from the gesture pool and sort them, and set them as the gesture password for this account; The biometric fusion feature is input into the convolution-long short-term memory neural network, and the convolution module is used to identify whether the gesture password is correct; and the long short-term memory module is used to confirm the user identity; specifically, the electromyographic feature is first constructed into an electromyographic feature image and input into the convolution module to identify the gesture to determine whether it is a correct gesture password; if the gesture password is correct, the long short-term memory module is used to predict the existing biometric fusion feature of the account corresponding to the identity, and the predicted feature is compared with the fit of the current fusion feature to verify the user identity; The high-density surface electromyographic signals of m×n channels are filtered and denoised, and the energy value of the electromyographic signals is calculated. The electromyographic signal segments of k gestures are intercepted according to the set energy threshold, and the features are selected to construct an electromyographic feature image of size m×n. The electromyographic feature images of k gestures are sequentially input into the convolution-long short-term memory neural network, and the convolution module is used to recognize the gestures and determine the gesture sequence to confirm whether it is the correct gesture password: if it is judged to be an incorrect gesture password, the system is interrupted to deny the access request; If it is a correct gesture, proceed to the next step of judgment; When the gesture password is entered correctly, the ECG signal is preprocessed by filtering, denoising, and benchmark segmentation, and the features of the processed data are extracted to obtain And combined with the high-density surface electromyography signal features extracted by the convolution module i = 1 ,..., k Fusion to form the biological feature vector x s; The existing h Biofusion features j = 1 ,..., h Input the long short-term memory module to predict the biological feature x of this time s ' , The actual feature x s With the predicted feature x s 'Compare and determine whether it is the correct identity; Calculate the absolute percentage error r(%) for the input features: Set a threshold a. When r(%)≤a, it is a correct identity and the data is saved in the online learning module. When r(%)>a, it is an incorrect identity and the access application is rejected with a warning or alarm.
2. The identity recognition method based on bioelectric signal fusion according to claim 1 is characterized in that: Once the identity is confirmed, the online learning module is enabled to iterate the data of the account corresponding to the identity and train the network again to reduce the misjudgment rate caused by daytime differences. If it is a malicious attack from other users, warnings or alarms will be taken.
3. The identity recognition method based on bioelectric signal fusion according to claim 2 is characterized in that: The specific working method of the online learning module is as follows: when the convolution-long short-term memory neural network confirms the correct identity, the online learning module is started to perform real-time training of the network model; the biological fusion features of this time are saved, the features with a long time line are deleted, and the convolution-long short-term memory neural network is retrained with an updated data set to prepare for the next identity recognition.
