Identity identification device and method and wearable device
By performing signal preprocessing and feature extraction on the electrocardiogram signal, extracting the Mel cepspectral coefficients, and using neural networks in the deep learning classification device for identity identification, the problem of excessive changes in the electrogram signal in the existing technology center is solved, and the accurate identification of user identity is achieved.
Patent Information
- Application Number
- CN202410883352.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-12-27
- Filing Date
- 2024-07-03
- Publication Date
- 2025-06-27
AI Technical Summary
It is difficult to accurately identify the user's identity through the electrocardiogram signal in the prior art, especially when the user's electrocardiogram signal changes too much, it is prone to identify errors.
The digital signal processing device is used to perform signal preprocessing and feature extraction on the ECG signal, extract the Mel cepspectral coefficients as the input feature vector, and identity identification is used using the trained neural network in the deep learning classification device.
By extracting the Mel cepspectral coefficients of the ECG signal and using a neural network for identification, the user's identity can be accurately identified without identification errors due to excessive changes in the user's ECG signal.
Smart Images
Figure CN120217338A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to an identity recognition technology, and particularly to an identity recognition device, method, and wearable device that use electrocardiogram signals to identify user identities. Background Art
[0002] Traditional identity recognition methods, such as passwords, fingerprints, iris, or face recognition, have some limitations and risks. For example, passwords may be stolen, fingerprints and irises may be replicated, and face recognition may be cracked through disguise. Therefore, some practices have started to use biometric signals that are not easily replicated for identity recognition, and electrocardiogram signals have been proposed to be used for identity recognition.
[0003] However, electrocardiogram signals are prone to change with an individual's psychological and physiological health conditions. In view of this, some researchers have tried to extract the features of electrocardiogram signals as the basis for user identity recognition. However, the existing techniques still have cases of misidentification due to excessive changes in the electrocardiogram signals of users. Therefore, how to extract features from electrocardiogram signals and use a model to accurately identify user identities without misidentifying due to excessive changes in the electrocardiogram signals of users has become an urgent need in the relevant industry. Summary of the Invention
[0004] It can be understood from the above description that the technical problem to be solved by the present invention is how to extract the features of electrocardiogram signals and establish a model to identify user identities without misidentifying due to excessive changes in the electrocardiogram signals of users.
[0005] To solve the above-mentioned problems in the prior art, an embodiment of the present invention provides an identity recognition device, and this identity recognition device includes a digital signal processing device and a deep learning classification device electrically connected to the digital signal processing device. The digital signal processing device is used to perform signal preprocessing on the first electrocardiogram signal to generate a second electrocardiogram signal, and then perform feature extraction on the second electrocardiogram signal to obtain a plurality of mel cepstral coefficients of the second electrocardiogram signal as an input feature vector. The deep learning classification device has a trained neural network, and is used to use the neural network to generate an identity recognition result according to the input feature vector, where the identity recognition result indicates whether the first electrocardiogram signal belongs to the user corresponding to the trained neural network, so as to identify the user's identity thereby.
[0006] To solve the above-mentioned problems in the prior art, an embodiment of the present invention further provides a wearable device, and this wearable device includes an identity recognition device having an electrocardiogram signal acquisition device and a communication device. The communication device is electrically connected to the identity recognition device and is used to transmit the identity recognition result and the first electrocardiogram signal or the second electrocardiogram signal to a computer device communicatively connected to the communication device.
[0007] To solve the above-mentioned prior problems, embodiments of the present invention further provide an identity recognition method, and this identity recognition method at least includes the following steps: performing signal preprocessing on a first electrocardiogram signal to generate a second electrocardiogram signal; performing feature extraction on the second electrocardiogram signal to obtain a plurality of mel cepstral coefficients of the second electrocardiogram signal as an input feature vector; and using a trained neural network to generate an identity recognition result according to the input feature vector, where the identity recognition result indicates whether the first electrocardiogram signal belongs to the user corresponding to the trained neural network, so as to thereby recognize the identity of the user.
[0008] As described above, the identity recognition device, method, and wearable device of the present invention can accurately recognize the user's identity according to the electrocardiogram signal, and will not misrecognize due to excessive changes in the user's electrocardiogram signal. Brief Description of the Drawings
[0009] To make the above and other objects, features, advantages, and embodiments of the present invention more obvious and understandable, the description of the accompanying drawings is as follows:
[0010] Figure 1 It is a schematic block diagram of an identity recognition device according to an embodiment of the present invention;
[0011] Figure 2 It is a schematic flowchart of performing identity recognition by an identity recognition method according to an embodiment of the present invention;
[0012] Figure 3 It is a schematic flowchart of training a neural network by an identity recognition method according to an embodiment of the present invention; and
[0013] Figure 4 It is a schematic block diagram of an identity recognition system according to an embodiment of the present invention.
[0014] Symbol Explanation
[0015] 11: Electrocardiogram (ECG) signal acquisition device;
[0016] 12: Digital signal processing device;
[0017] 121: Signal preprocessing unit;
[0018] 122: Mel cepstral coefficient acquisition unit;
[0019] 13: Deep learning classification device;
[0020] S21~S24, S31~S34: Steps;
[0021] 41: Wearable device;
[0022] 42: Computer device;
[0023] 43: Cloud server. Detailed implementation
[0024] The main purpose of the present invention is to use electrocardiogram signals to identify users, and will not misidentify due to excessive changes in the electrocardiogram signals of users. An electrocardiogram signal is a record of an individual's electrocardiogram, which reflects the electrical activity of the individual's heart. The heart structure and electrocardiogram pattern of each person are unique, just like fingerprints and irises. Therefore, electrocardiogram signals as biometric features can be used as a unique and difficult-to-forge identity recognition marker.
[0025] Electrocardiogram signal unlocking has several advantages. First, electrocardiogram signals can be obtained non-contact, without the user performing specific actions or contacting special sensors. This not only facilitates the user, but also reduces the dependence on sensors and security risks. Second, electrocardiogram signal unlocking has higher stability and reliability. Compared with face unlocking and fingerprint unlocking, electrocardiogram signals are relatively less affected by factors such as physical state and mood, and thus are more consistent and stable. This means that electrocardiogram signal unlocking can still maintain a high degree of accuracy under different environmental conditions.
[0026] However, even though electrocardiogram signals are relatively less affected by factors such as physical state and mood, if the electrocardiogram signals of the user change too much in a short period of time, there will also be large linear and non-linear changes in the high-frequency part of the electrocardiogram signals, resulting in misidentification. Obviously, most of the high-frequency part of the electrocardiogram signals may be useless information for identity recognition. In addition, if the electrocardiogram signals of the user change too much in a short period of time and the low-frequency part of the electrocardiogram is extracted as a linear feature, this approach will also reduce the accuracy of identity recognition. In view of this, the feature of the electrocardiogram signal selected by the present invention is the mel cepstral coefficient of the electrocardiogram signal. The mel cepstral coefficient is evenly distributed at the frequencies of the mel scale, and the energy spectrum has been processed by a mel filter and logarithmized. Therefore, a large amount of useless information in the high frequency has been excluded and the features of the low-frequency part have been logarithmized, which enables the neural network to accurately identify the user's identity without misidentifying due to excessive changes in the user's electrocardiogram signals.
[0027] Please refer to Figure 1 , Figure 1It is a schematic block diagram of the identity recognition device according to an embodiment of the present invention. The identity recognition device includes an electrocardiogram signal acquisition device 11, a digital signal processing device 12, and a deep learning classification device 13. The electrocardiogram signal acquisition device 11 is electrically connected to the digital signal processing device 12, and the digital signal processing device 12 is electrically connected to the deep learning classification device 13. The electrocardiogram signal acquisition device 11 is used to sense a user and acquire the user's electrocardiogram signal. Further, for different users, the electrocardiogram signals acquired by the electrocardiogram signal acquisition device 11 will be different from each other. Therefore, the electrocardiogram signal can be used for user identity recognition. In addition, the electrocardiogram signal acquisition device 11 is built into the identity recognition device in this embodiment. In other embodiments, the electrocardiogram signal acquisition device 11 can be an external device independent of the identity recognition device.
[0028] The digital signal processing device 12 is used to perform pre-signal processing on the electrocardiogram signal acquired by the electrocardiogram signal acquisition device 11 to generate a pre-signal processed electrocardiogram signal, and then perform feature extraction on the pre-signal processed electrocardiogram signal to obtain multiple Mel cepstral coefficients of the pre-signal processed electrocardiogram signal as input feature vectors. The deep learning classification device 13 has a trained neural network and is used to generate an identity recognition result using the neural network according to the input feature vectors, where the identity recognition result indicates whether the electrocardiogram signal acquired by the electrocardiogram signal acquisition device 11 belongs to the user corresponding to the trained neural network, so as to identify the user's identity. Further, the neural network is, for example but not limited to, a backpropagation neural network, and any supervised learning and classification neural network can be used in the present invention.
[0029] Further, the digital signal processing device 12 includes a pre-signal processing unit 121 and a Mel cepstral coefficient acquisition unit 122. The pre-signal processing unit 121 is electrically connected to the Mel cepstral coefficient acquisition unit 122, and the pre-signal processing unit 121 and the Mel cepstral coefficient acquisition unit 122 are respectively electrically connected to the electrocardiogram signal acquisition device 11 and the deep learning classification device 13. The pre-signal processing unit 121 is used to perform pre-signal processing on the electrocardiogram signal acquired by the electrocardiogram signal acquisition device 11 to generate a pre-signal processed electrocardiogram signal. The Mel cepstral coefficient acquisition unit 122 is used to perform feature extraction on the pre-signal processed electrocardiogram signal to obtain multiple Mel cepstral coefficients of the pre-signal processed electrocardiogram signal as input feature vectors.
[0030] Please refer to Figure 1 and Figure 2 , Figure 2It is a schematic flowchart for identity recognition by the identity recognition method of the embodiment of the present invention. After the deep learning classification device 13 has trained the neural network, the identity recognition device can perform user identity recognition based on the electrocardiogram signal obtained by the electrocardiogram signal acquisition device 11. In step S21, the electrocardiogram signal acquisition device 11 senses the user to obtain an electrocardiogram signal. Then, in step S22, the signal preprocessing unit 121 is used to perform signal preprocessing on the electrocardiogram signal obtained by the electrocardiogram signal acquisition device 11 to generate a preprocessed electrocardiogram signal.
[0031] The process of signal preprocessing is described as follows. First, the electrocardiogram signal obtained by the electrocardiogram signal acquisition device 11 is segmented, that is, the electrocardiogram signal obtained by the electrocardiogram signal acquisition device 11 is divided into multiple signal frames, and each signal frame is, for example but not limited to, 512 sampling points. Then, the segmented signals after segmentation are multiplied by a window function, for example, multiplied by a Hamming Window and the multiple segmented signals. Then, the multiple segmented signals after multiplication are subjected to baseline drift correction, frequency selective filtering, and signal enhancement to generate a preprocessed electrocardiogram signal. Here, it should be noted that frequency selective filtering uses a low-pass filter to filter out noise.
[0032] After that, in step S23, the Mel cepstral coefficient acquisition unit 122 is used to extract features from the preprocessed electrocardiogram signal to obtain multiple Mel cepstral coefficients of the preprocessed electrocardiogram signal as input feature vectors. Then, in step S24, the deep learning classification device 13 uses the neural network to generate an identity recognition result based on the input feature vectors.
[0033] Further, the process of feature extraction is described as follows. First, the electrocardiogram (ECG) signal after signal preprocessing is segmented, that is, the ECG signal after signal preprocessing is divided into multiple signal frames, and each signal frame is, for example but not limited to, 512 sampling points. Next, the multiple segmented signals can be selectively pre-emphasized. The pre-emphasis process is to perform high-pass filtering on the multiple segmented signals using a high-pass filter to highlight the resonance peaks in the high frequency. After that, a window function is used to multiply the multiple segmented signals (after segmentation or after pre-emphasis processing). The window function is a Hamming window, but this is not limiting. Then, the multiple segmented signals after multiplication are converted from the time domain to the frequency domain to generate multiple segmented frequency domain signals. The time domain to frequency domain conversion is, for example but not limited to, the fast Fourier transform. Next, a Mel filter is used to perform Mel filtering on the multiple segmented frequency domain signals to obtain multiple Mel scales. The Mel filter is a triangular band-pass filter. After that, the logarithmic energy value of each Mel scale is calculated, and the multiple logarithmic energy values are subjected to discrete cosine transform to obtain a cepstrum diagram, and multiple Mel cepstral coefficients are obtained from the cepstrum diagram. For example, 13 Mel cepstral coefficients at different frequencies can be obtained from the cepstrum diagram, but the present invention is not limited thereto. In other embodiments, the difference of the 13 Mel cepstral coefficients at different frequencies in the cepstrum diagram can also be selectively calculated, and at this time, the input feature vector is the 13 Mel cepstral coefficients at different frequencies and the difference of the 13 Mel cepstral coefficients at different frequencies.
[0034] Please then refer to Figure 1 and Figure 3 , Figure 3 which is a schematic flowchart of training a neural network of the identity recognition method according to an embodiment of the present invention. The process of training the neural network of the identity recognition device is described as follows. First, in step S31, the identity recognition device first obtains multiple ECG signals for training and corresponding multiple identity recognition results. In step S32, the signal preprocessing unit 121 performs signal preprocessing on the multiple ECG signals for training to generate multiple ECG signals for training after signal preprocessing. Then, in step S33, the Mel cepstral coefficient obtaining unit 122 performs feature extraction on the multiple ECG signals for training after signal preprocessing to obtain multiple Mel cepstral coefficients of the multiple ECG signals for training after signal preprocessing as input feature vectors. Next, in step S34, the deep learning classification device 13 uses the input feature vectors of the multiple ECG signals for training and the corresponding multiple identity recognition results to train the neural network to establish a neural network that can be used for identity recognition.
[0035] Please refer to Figure 1 and Figure 4 , Figure 4It is a schematic block diagram of the identity recognition system according to an embodiment of the present invention. The identity recognition system includes a wearable device 41, a computer device 42, and a cloud server 43. The wearable device 41 includes the aforementioned identity recognition device and a communication device (not shown in the figure). The communication device is electrically connected to the identity recognition device and is used to transmit the identity recognition result and the electrocardiogram signal after signal preprocessing (or the electrocardiogram signal without signal preprocessing) to the computer device 42 that the communication device is communicatively connected to. The computer device 42 is, for example but not limited to, a tablet, a computer, or a smart phone, and the computer device 42 is communicatively connected to the cloud server 43 and is used to transmit the identity recognition result and the electrocardiogram signal after signal preprocessing (or the electrocardiogram signal without signal preprocessing) to the cloud server 43. In this way, the cloud server 43 can record the user's identity and record and analyze multiple electrocardiogram signals to feedback health reminder information to the user.
[0036] In summary, the present invention obtains multiple Mel cepstral coefficients of the electrocardiogram signal as input feature vectors and inputs the input feature vectors into a trained neural network to generate an identity recognition result. Since the Mel cepstral coefficients are evenly distributed in the frequency of the Mel scale, and the energy spectrum is processed by a Mel filter and logarithmically transformed, a large amount of useless information in the high frequency has been excluded and the features in the low frequency part have been logarithmically transformed. This enables the neural network to accurately recognize the user's identity without misidentifying due to excessive changes in the user's electrocardiogram signal.
[0037] The present invention is only disclosed in the preferred embodiments herein. However, any person skilled in the art should understand that the above embodiments are only used to describe the present invention and are not intended to limit the scope of the patent rights claimed by the present invention. Any equivalent or equivalent changes or substitutions to the above embodiments should be construed as being covered by the spirit or scope of the present invention. Therefore, the protection scope of the present invention should be based on what is defined by the scope of the claims.
Claims
1. An identity recognition device, characterized in that: include: a digital signal processing device, configured to perform a signal pre-processing on a first electrocardiogram signal to generate a second electrocardiogram signal, and then perform a feature extraction on the second electrocardiogram signal to obtain a plurality of Mel-frequency cepstral coefficients of the second electrocardiogram signal as an input feature vector; as well as A deep learning classification device is electrically connected to the digital signal processing device and has a trained neural network. The deep learning classification device is used to use the neural network to generate an identity recognition result according to the input feature vector, wherein the identity recognition result indicates whether the first electrocardiogram signal belongs to a user corresponding to the trained neural network, so as to identify the identity of the user.
2. The identity recognition device according to claim 1, characterized in that: Further including: An electrocardiogram signal acquisition device is electrically connected to the digital signal processing device and is used to acquire the first electrocardiogram signal.
3. The identity recognition device as claimed in claim 2, characterized in that: The digital signal processing device comprises: a signal pre-processing unit, configured to perform the signal pre-processing on the first electrocardiogram signal to generate the second electrocardiogram signal; and A Mel-cepstral coefficient acquisition unit is electrically connected to the signal pre-processing unit and is used to perform the feature extraction on the second electrocardiogram signal to obtain a plurality of Mel-cepstral coefficients of the second electrocardiogram signal as the input feature vector.
4. The identity recognition device as claimed in claim 3, characterized in that: The feature extraction is to segment the second electrocardiogram signal, use a window function to multiply the multiple segmented signals after segmentation, and perform a time domain to frequency domain conversion on the multiple segmented signals after multiplication to generate multiple segmented frequency domain signals, use a Mel filter to perform a Mel filtering on the multiple segmented frequency domain signals to obtain multiple Mel scales, calculate a logarithmic energy value of each Mel scale, perform a discrete cosine transform on the multiple logarithmic energy values to obtain a cepstrum, and obtain the multiple Mel cepstrum coefficients from the cepstrum.
5. The identity recognition device as claimed in claim 3, characterized in that: The signal pre-processing is to segment the first electrocardiogram signal, and multiply the segmented multiple segmented signals by a window function, and then perform a baseline drift correction, a frequency selective filtering and a signal enhancement to generate the second electrocardiogram signal.
6. The identity recognition device as claimed in claim 2, characterized in that: The neural network is a reverse transfer neural network.
7. A wearable device, characterized in that: include: An identification device as claimed in any one of claims 2 to 5; as well as A communication device is electrically connected to the identity recognition device and is used to transmit the identity recognition result and the first electrocardiogram signal or the second electrocardiogram signal to a computer device to which the communication device is communicatively connected.
8. An identity recognition method, characterized in that: include: Performing a signal pre-processing on a first electrocardiogram signal to generate a second electrocardiogram signal; Performing a feature extraction on the second electrocardiogram signal to obtain a plurality of Mel-frequency cepstral coefficients of the second electrocardiogram signal as an input feature vector; as well as A trained neural network is used to generate an identity recognition result according to the input feature vector, wherein the identity recognition result indicates whether the first electrocardiogram signal belongs to a user corresponding to the trained neural network, so as to recognize an identity of the user.
9. The identity recognition method according to claim 8, characterized in that: The feature extraction is to segment the second electrocardiogram signal, use a window function to multiply the multiple segmented signals after segmentation, and perform a time domain to frequency domain conversion on the multiple segmented signals after multiplication to generate multiple segmented frequency domain signals, use a Mel filter to perform a Mel filtering on the multiple segmented frequency domain signals to obtain multiple Mel scales, calculate a logarithmic energy value of each Mel scale, perform a discrete cosine transform on the multiple logarithmic energy values to obtain a cepstrum, and obtain the multiple Mel cepstrum coefficients from the cepstrum.
10. The identity recognition method according to claim 8, characterized in that: The signal pre-processing is to segment the first electrocardiogram signal, and multiply the segmented multiple segmented signals by a window function, and then perform a baseline drift correction, a frequency selective filtering and a signal enhancement to generate the second electrocardiogram signal.