WO2018072170A1 - Ecg signal-based identity recognition method and device - Google Patents

Abstract

An ECG signal-based identity recognition method and device. The method comprises steps of: acquiring multiple electrocardiogram data items of a first biometric information set sent by a first terminal and corresponding to a multi-lead ECG signal of a user, and employing the multiple electrocardiogram data items as an electrocardiogram data set of the user (S10); comparing the electrocardiogram data set against a pre-stored electrocardiogram data set (S20); and if the electrocardiogram data set matches the pre-stored electrocardiogram data set, then determining that the identity of the user is successfully recognized (S30). The method of the present invention improves accuracy of identity recognition according to an ECG signal.

Description

 ECG signal based identification method and device

Technical field

The present invention relates to the field of biometrics, and in particular, to an ECG signal-based identification method and apparatus.

Background technique

The ECG (electrocardiogram) signal is related to the physiological characteristics of the user's body. The ECG signals of different users are different. Therefore, the ECG signal can be used for user identification. When the user is identified by the ECG signal, an ECG signal of the user is usually collected by the corresponding ECG signal acquisition device, and then the identity is determined according to the ECG signal. However, because the ECG signal of the collected user may be different at different times, for example, the ECG signal of the user when it is calm may be different from the ECG signal after the motion, so that the identification is performed according to an ECG signal of the user. For example, if the user is originally the user but the identification fails, the accuracy of the identification using the ECG signal is not high.

Summary of the invention

The main object of the present invention is to provide an ECG signal-based identification method and apparatus, which aims to solve the technical problem that the accuracy of identification using ECG signals is not high.

To achieve the above object, the present invention provides an ECG signal-based identity recognition method, and the ECG signal-based identity recognition method includes the following steps:

Acquiring a plurality of ECG data corresponding to the multi-lead ECG signal of the user, and using the plurality of ECG data as the ECG data group of the user;

Comparing the ECG data set with a pre-stored ECG data set;

Determining that the identity of the user is successful when the ECG data set matches the pre-stored ECG data set;

The multi-lead ECG signal is collected by an ECG signal acquisition device, and the ECG signal acquisition device includes a plurality of acquisition terminals, and contacts the user through the plurality of acquisition terminals to collect a multi-lead ECG signal of the user.

Preferably, the acquisition terminal of the ECG signal acquisition device is a wrist-type acquisition terminal.

Preferably, before the step of acquiring the plurality of ECG data corresponding to the multi-lead ECG signal of the user, the method further includes:

After collecting the multi-lead ECG signal of the user by the ECG signal collecting device, performing denoising processing on the multi-lead ECG signal;

The step of acquiring multiple ECG data corresponding to the multi-lead ECG signal of the user includes:

Feature extraction is performed on the multi-lead ECG signal after the denoising process, and multiple ECG data corresponding to the multi-lead ECG signal are acquired.

Preferably, the step of performing feature extraction on the multi-lead ECG signal after the denoising process, and acquiring the plurality of ECG data corresponding to the multi-lead ECG signal includes:

And performing overall appearance feature extraction on the multi-lead ECG signal after the denoising process, and acquiring multiple ECG data corresponding to the overall appearance feature of the multi-lead ECG signal;

Performing wavelet coefficient feature extraction on the multi-lead ECG signal after the denoising process, and acquiring a plurality of ECG data corresponding to the wavelet coefficient feature of the multi-lead ECG signal;

And performing shape feature extraction on the multi-lead ECG signal after the denoising process, and acquiring a plurality of ECG data corresponding to the shape feature of the multi-lead ECG signal.

Preferably, the shape features include area features, perimeter features, form factor features, center of gravity features, and elongation features.

Preferably, the step of comparing the ECG data set with the pre-stored ECG data set comprises:

And adopting a preset multi-layer identification algorithm to sequentially sequentially, the plurality of ECG data corresponding to the overall appearance feature of the multi-lead ECG signal, the plurality of ECG data corresponding to the wavelet coefficient feature of the multi-lead ECG signal, and the The plurality of ECG data corresponding to the shape feature of the multi-lead ECG signal is compared with the pre-stored ECG data set.

Preferably, the multi-lead ECG signal is a multi-lead ECG signal of the user in a motion state.

In addition, in order to achieve the above object, the present invention further provides an ECG signal-based identity recognition apparatus, where the ECG signal-based identity recognition apparatus includes:

An acquiring module, configured to acquire a plurality of ECG data corresponding to the multi-lead ECG signal of the user, and use the plurality of ECG data as the ECG data group of the user;

a comparison module, configured to compare the ECG data set with a pre-stored ECG data set;

a processing module, configured to determine that the identity of the user is successful when the ECG data group matches the pre-stored ECG data group;

The multi-lead ECG signal is collected by an ECG signal acquisition device, and the ECG signal acquisition device includes a plurality of acquisition terminals, and contacts the user through the plurality of acquisition terminals to collect a multi-lead ECG signal of the user.

Preferably, the acquisition terminal of the ECG signal acquisition device is a wrist-type acquisition terminal.

Preferably, the ECG signal-based identity recognition device further includes:

a denoising module, configured to perform denoising processing on the multi-lead ECG signal after the multi-lead ECG signal is collected by the ECG signal collecting device;

The acquiring module is configured to perform feature extraction on the multi-lead ECG signal after the denoising process, and acquire multiple ECG data corresponding to the multi-lead ECG signal.

Preferably, the obtaining module is configured to:

And performing overall appearance feature extraction on the multi-lead ECG signal after the denoising process, and acquiring multiple ECG data corresponding to the overall appearance feature of the multi-lead ECG signal;

Performing wavelet coefficient feature extraction on the multi-lead ECG signal after the denoising process, and acquiring a plurality of ECG data corresponding to the wavelet coefficient feature of the multi-lead ECG signal;

And performing shape feature extraction on the multi-lead ECG signal after the denoising process, and acquiring a plurality of ECG data corresponding to the shape feature of the multi-lead ECG signal.

Preferably, the shape features include area features, perimeter features, form factor features, center of gravity features, and elongation features.

Preferably, the comparison module is used to:

And adopting a preset multi-layer identification algorithm to sequentially sequentially, the plurality of ECG data corresponding to the overall appearance feature of the multi-lead ECG signal, the plurality of ECG data corresponding to the wavelet coefficient feature of the multi-lead ECG signal, and the The plurality of ECG data corresponding to the shape feature of the multi-lead ECG signal is compared with the pre-stored ECG data set.

Preferably, the multi-lead ECG signal is a multi-lead ECG signal of the user in a motion state.

The method and device for identifying an ECG signal based on the present invention obtains a plurality of ECG data as an ECG data set and pre-stored ECG data by acquiring a plurality of ECG data corresponding to the multi-lead ECG signal of the user. The group performs the comparison. When it matches the pre-stored ECG data set, it is determined that the user's identity is successful, and the identity of the user is identified by using multiple ECG data, thereby improving the accuracy of the identity recognition.

DRAWINGS

1 is a schematic flow chart of a first embodiment of an ECG signal-based identification method according to the present invention;

2 is a schematic structural diagram of an optional ECG signal collection device according to various embodiments of the present invention;

3 is a schematic flow chart of a second embodiment of an ECG signal-based identification method according to the present invention;

4 is a schematic diagram of functional modules of a first embodiment of an ECG signal-based identity recognition apparatus according to the present invention;

FIG. 5 is a schematic diagram of functional modules of a second embodiment of an ECG signal-based identity recognition apparatus according to the present invention.

detailed description

It is understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

The invention provides an identity recognition method based on ECG signals. Referring to FIG. 1, FIG. 1 is a schematic flowchart diagram of a first embodiment of an ECG signal-based identification method according to the present invention. In this embodiment, the ECG signal-based identification method includes the following steps:

Step S10: Acquire a plurality of ECG data corresponding to the multi-lead ECG signal of the user, and use the plurality of ECG data as the ECG data group of the user, wherein the multi-lead ECG signal is collected by the ECG signal collection device. Collecting, the ECG signal collecting device includes a plurality of collecting terminals, and contacting the user through the plurality of collecting terminals to collect the multi-lead ECG signal of the user;

The ECG (electrocardiogram) signal is related to the physiological characteristics of the user's body. The ECG signals of different users are different. Therefore, the ECG signal can be used for user identification. When the user is identified by using the ECG signal, an ECG signal of the user is usually collected by the corresponding ECG signal acquisition device, and then the identity is determined according to the ECG data corresponding to the ECG signal. However, since the user's ECG signal will be different in different states, for example, the ECG signal of the user in a calm state will be different from the ECG signal after the motion, so that the user is originally but the identity Identifying faulty error conditions, therefore, the accuracy of using ECG signals for identification is not high.

In order to improve the accuracy of the identification using the ECG signal, in the embodiment, the user identification system is composed of a terminal such as a smart phone, a PAD (tablet computer), a PC (personal computer), and a corresponding ECG signal collection device. For example, the ECG signal acquisition device establishes a wireless connection with the PC terminal through Bluetooth to form a user identity recognition system. The ECG signal collection device includes multiple acquisition terminals. When the identity of the user needs to be recognized, the user wears the ECG signal acquisition device, and contacts the user through multiple acquisition terminals of the ECG signal acquisition device to collect the user's multi-channel. Link ECG signal. For example, as shown in FIG. 2, the ECG signal acquisition device is a wrist-type acquisition device, and the ECG signal acquisition device includes a plurality of (five shown) wrist-type acquisition terminals. After the user wears the ECG signal acquisition device on the wrist of the user, the user's multi-lead ECG signal is collected through the plurality of wrist-type acquisition terminals of the ECG signal acquisition device. Then, the terminal acquires a plurality of ECG data corresponding to the multi-lead ECG signal according to the collected multi-lead ECG signal, and the ECG data group corresponding to the multi-lead ECG signal of the user is formed by the plurality of ECG data.

Step S20, comparing the ECG data set with a pre-stored ECG data set;

In this embodiment, the terminal further stores in advance an ECG data group corresponding to the multi-lead ECG signals of the plurality of users. After acquiring the ECG data group corresponding to the current user's multi-lead ECG signal, the terminal compares the ECG data group corresponding to the current user's multi-lead ECG signal with the pre-stored ECG data group.

Step S30, when the ECG data group matches the pre-stored ECG data group, it is determined that the user's identity recognition is successful.

After comparing the ECG data group corresponding to the current user's multi-lead ECG signal with the pre-stored ECG data set, if the current user's multi-lead ECG signal corresponds to the ECG data group and the pre-stored ECG data The group matches. At this time, it is determined that the user's identity is successful; otherwise, it is determined that the user's identity fails. For example, a preset matching degree threshold is preset, and if the matching degree between the ECG data group corresponding to the current user's multi-lead ECG signal and the pre-stored ECG data group is greater than the preset matching degree threshold, the identity of the user is determined. The recognition was successful.

The preset matching degree threshold may be set according to a corresponding threshold selection strategy, for example, using two threshold selection strategies, a proportional coefficient training method and a minimum similarity method, respectively, to set a preset matching degree threshold, and then comparing the experimental results. In this way, the applicable data and actual occasions of the two strategies are analyzed and summarized. When setting the preset matching degree threshold, the appropriate threshold selection strategy is selected according to the characteristics of the ECG data and the application.

For example, taking the self-acquisition ECG signal database and the standard ECG signal database as an example, it is found through experiments that the proportional coefficient training method has higher recognition accuracy than the minimum similarity method in the self-acquired ECG signal database, but in the standard ECG signal database. The minimum similarity method has higher recognition accuracy than the proportional coefficient training method. By comparison, we find that in the identification algorithms of the two databases, the error acceptance rate of the minimum similarity method is higher than that of the proportional coefficient training method, and the error rejection rate of the proportional coefficient method is higher than the minimum similarity. This shows that for the same feature extraction algorithm, the same data, the minimum matching degree selection preset matching degree threshold is smaller than the proportional coefficient training selected preset matching degree threshold. In this case, the data in the standard ECG signal database is small, the waveform is stable, and the abrupt waveform is also small, so the small preset matching threshold can distinguish itself from others. The ECG data in the self-collecting ECG signal database is relatively noisy, and the waveform has a certain distortion. Therefore, the difference between itself and itself, and the self and others are small, and it is not suitable to use a small preset matching threshold. Therefore, in actual application, for the characteristics and applications of the ECG data, an appropriate threshold selection strategy is selected to set the preset matching degree threshold.

Further, since the user cannot always be in a calm state when collecting the multi-lead ECG signal of the user, in this embodiment, the collected multi-lead ECG signal is a multi-lead ECG signal of the user in a motion state.

In the solution provided by the embodiment, by acquiring multiple ECG data corresponding to the multi-lead ECG signal of the user, the obtained plurality of ECG data are compared as the ECG data group with the pre-stored ECG data group. When matching with the pre-stored ECG data set, it is determined that the user's identity is successful, because the identity of the user is identified by multiple ECG data, thereby improving the accuracy of the identity recognition.

Further, as shown in FIG. 3, a second embodiment of the ECG signal-based identification method of the present invention is proposed based on the first embodiment. In the second embodiment, before the step S10, the method further includes:

Step S40, after collecting the multi-lead ECG signal of the user by the ECG signal collecting device, performing denoising processing on the multi-lead ECG signal;

The step S10 includes:

Step S11: Perform feature extraction on the multi-lead ECG signal after the denoising process, and acquire multiple ECG data corresponding to the multi-lead ECG signal.

In this embodiment, when the user's multi-lead ECG signal is collected by the ECG signal acquisition device, since the contact point between the ECG signal acquisition device and the user's human body occurs during the acquisition process, the human body itself and the circuit leads also enable the acquisition. The ECG signal contains a lot of noise. The common noise of ECG signals mainly includes baseline drift, power frequency interference and myoelectric interference, which will greatly affect the accuracy of identification. Therefore, after the user's multi-lead ECG signal is acquired, the multi-lead ECG signal is first denoised.

Power frequency interference is generated by the electromagnetic radiation generated by the mains power and the electromagnetic coupling generated by the sensor circuit. The frequency of myoelectric interference covers the spectrum of the entire ECG signal, and its spectral distribution is 5~2000. Hz, there are many complex physiological electrical signals in the human body. A certain bioelectricity is a signal when it is needed to study it. The ECG signal may be noise when it is collected, that is, the noise caused by the human bioelectricity other than the measured physiological variable becomes a muscle. Electrical interference. Therefore, myoelectric interference is random. During the process of collecting ECG signals by using the ECG signal acquisition device, the movement of the human body and the movement of the contact point with the device may cause a baseline drift of the ECG signal, and the baseline drift appears as a vertical offset of the reference line in the waveform time domain. The baseline offset is low frequency interference, and its spectrum is generally below 0.5 Hz.

Among various noises, power frequency interference has the greatest impact on ECG signals. In the prior art, there are many mature and effective methods for eliminating power frequency interference regardless of hardware settings or software filters. In this embodiment, a notch filter and a Butterworth band rejection filter are used to eliminate 50 Hz power frequency noise, respectively. Then, the myoelectric interference of the ECG signal is eliminated by the wavelet threshold denoising method. The specific steps of the wavelet threshold denoising method are as follows:

(1) Multi-layer wavelet decomposition of ECG signals to obtain wavelet coefficients at different scales;

(2) The wavelet coefficient is based on a threshold or is zeroed or contracted;

(3) Reconstruction of wavelet coefficients after zero or contraction

Eliminating baseline drift is a critical step in eliminating noise because baseline drift can cause serious errors in the detection of critical waveform amplitudes and slope areas of ECG signals. The baseline drift frequency is lower, less than 0.5 Hz. After multi-layer wavelet decomposition of the original ECG signal, the baseline drift is filtered in the low-frequency component, and the low-frequency component where the baseline drift is set to zero can eliminate the baseline drift in the wavelet domain. As a comparison, the moving window median filtering method is used to eliminate the baseline drift. The final signal-to-noise ratio and root mean square error comparison are used to select which denoising method is optimal. The experimental results are shown in Table 1:

Table 1

Denoising algorithm combination SNR (Signal to Noise Ratio) RMSE (root mean square error) Wavelet Threshold Denoising Method + Wavelet Reconstruction Denoising Method + Notch Filter 82.2530 9.6580e-04 Wavelet Threshold Denoising + Moving Window Median Filter + Notch Filter 89.9935 0.0076 Wavelet Threshold Denoising + Wavelet Reconstruction Denoising + Butterworth Band Rejection Filter 78.7360 2.3676e-04 Wavelet Threshold Denoising + Moving Window Median Filtering + Butterworth Band Rejection Filter 79.4153 2.3178e-04

It can be seen from the experimental results that the combination of wavelet threshold denoising method, moving window median filtering and Butterworth band rejection filter is better. Therefore, this combination scheme is used to denoise the user's multi-lead ECG signal.

After denoising the multi-lead ECG signal, feature extraction is performed on the de-cored multi-lead ECG signal to obtain a plurality of ECG data corresponding to the multi-lead ECG signal. Optionally, since the feature fusion is such that the features of each of the identified individuals are more unique, the recognition rate is higher. Feature fusion can enhance the recognition ability of the whole system, and the characteristics of the fusion will have the advantages of strong complementarity and low redundancy. Therefore, the multi-lead ECG signal after the denoising process is first subjected to feature fusion, and then the feature extraction of the fused multi-lead ECG signal is performed. For example, taking the dual-lead ECG signal as an example, the steps of merging the dual-lead ECG signal are as follows:

(1) First, a two-dimensional space matrix of N×N order is established, where N is the maximum value in the time domain of the electrocardiogram data. For dual-lead ECG data, two-dimensional ECG data is mapped into a two-dimensional space.

(2) For the N×N-order two-dimensional space matrix, dimension reduction processing is needed to facilitate feature extraction and distance calculation. Set a matrix window of r × r. If the value of the spatial matrix in the window is more than two small cells, the value is 1 after the dimension reduction.

(3) Since the dimensionality of the two-dimensional space matrix is still mostly zero, the coordinates can be used to store the sparse matrix in the form of val=(row, column, value). This sparse matrix is used as a feature vector after multi-dimensional ECG signal data fusion to carry out subsequent identification algorithms.

Next, different features are extracted for the sparse matrix obtained after data fusion, and then the obtained feature vectors are multi-layered. Feature extraction is mainly to extract the unique properties of ECG signals that are different from other people's ECG signals, such as overall appearance features, wavelet coefficient features, shape features, density distribution features, and so on. Specifically, the step S11 includes:

Step a, performing overall appearance feature extraction on the multi-lead ECG signal after the denoising process, and acquiring a plurality of ECG data corresponding to the overall appearance feature of the multi-lead ECG signal;

Step b, performing wavelet coefficient feature extraction on the multi-lead ECG signal after the denoising process, and acquiring a plurality of ECG data corresponding to the wavelet coefficient feature of the multi-lead ECG signal;

Step c: performing shape feature extraction on the multi-lead ECG signal after the denoising process, and acquiring a plurality of ECG data corresponding to the shape feature of the multi-lead ECG signal.

After performing denoising processing on the multi-lead ECG signal, each feature extraction is performed on the de-noised multi-lead ECG signal, specifically, the overall appearance characteristics of the de-noised multi-lead ECG signal are performed. Extracting and acquiring a plurality of ECG data corresponding to the overall appearance feature of the multi-lead ECG signal; performing wavelet coefficient feature extraction on the de-noised multi-lead ECG signal, and acquiring the wavelet coefficient feature of the multi-lead ECG signal correspondingly The electrocardiographic data is obtained by performing shape feature extraction on the de-noised multi-lead ECG signal, and acquiring a plurality of electrocardiogram data corresponding to the shape feature of the multi-lead ECG signal.

For example, for the overall appearance feature, the multi-lead ECG signal is mapped to a sparse matrix according to the feature fusion algorithm, and the binary image reflected by the sparse matrix is used as an overall appearance feature vector of the multi-lead ECG signal.

For the wavelet coefficient feature, the multi-lead ECG signal is decomposed into three layers, and the detail coefficients CD1-CD3 and approximation coefficient CA3 of each level after decomposition are mapped into two-dimensional space as two-dimensional data, and then sparsely stored after dimension reduction. The resulting matrix is used as a wavelet coefficient feature vector.

For shape features, it mainly includes area features, perimeter features, shape factor features, elongation features, center of gravity features, and the like. Among them, the shape factor feature describes the degree of roundness of the shape of the binary image, and the elongation feature reflects the slenderness of the image. For the sparse matrix obtained by the fusion of the multi-lead ECG signals, the sparse matrix of the same person's ECG signal has similarity, and different people have greater differences. So we can think of it as a binary image to extract shape features.

Further, the step S20 includes:

Step d: sequentially adopting a preset multi-layer identification algorithm to sequentially select a plurality of ECG data corresponding to the overall appearance feature of the multi-lead ECG signal, and multiple ECG data corresponding to the wavelet coefficient feature of the multi-lead ECG signal And comparing the plurality of ECG data corresponding to the shape feature of the multi-lead ECG signal with the pre-stored ECG data set.

In this embodiment, a multi-layer identification algorithm is used for identity identification. For example, in the case of three layers, multiple ECG data and wavelet coefficient features corresponding to the overall appearance feature of the user's multi-lead ECG signal are sequentially used. The ECG data and the plurality of ECG data corresponding to the shape feature are compared with the pre-stored ECG data set to identify the identity of the user. The specific recognition process of the multi-layer recognition algorithm for identity recognition is as follows:

First, the first layer identification is performed, and the overall appearance feature of the binary image represented by the sparse matrix is used as the feature vector of the first layer identification. Since most of the data of the mapped matrix is zero, if the appearance of the binary image corresponding to the sparse matrix corresponding to two different users is difficult to distinguish by the preset matching degree threshold, the second layer identification is needed, and the other one is compared. Features for identification.

In the second layer identification, the error sample identified by the first layer is the input of the second layer identification, specifically, the sample which is indistinguishable by the overall appearance feature identified by the first layer is in the second layer, each lead core The electrical data is separately subjected to three-layer wavelet decomposition, and the approximation coefficient and the detail coefficient are respectively mapped to the two-dimensional space and then the same dimensionality reduction processing as the first layer is performed. The sparse matrix at this time reflects the wavelet domain characteristics of the ECG signal, and these features are identified by the matching threshold corresponding to the wavelet coefficient feature. Samples that fail to identify during this layer recognition process are also input to the third layer for third layer identification.

In the third layer identification, the input samples respectively extract the shape features and density distribution features of the sparse matrix corresponding to the binary image. It is identified by the corresponding matching degree threshold. The recognition result of the third layer is output as the final recognition result.

The solution proposed in this embodiment, after collecting the multi-lead ECG signal of the user, performs denoising processing on the multi-lead ECG signal, thereby avoiding interference of the noise signal on the ECG signal, thereby further improving the identity of the ECG signal. The accuracy of the identification.

The invention further provides an identity recognition device based on an ECG signal. Referring to FIG. 4, FIG. 4 is a schematic diagram of functional modules of a first embodiment of an ECG signal-based identity recognition apparatus according to the present invention.

In this embodiment, the ECG signal-based identity recognition apparatus includes:

The obtaining module 10 is configured to acquire a plurality of ECG data corresponding to the multi-lead ECG signal of the user, and use the plurality of ECG data as the ECG data group of the user, wherein the multi-lead ECG signal is used by the ECG Collecting by the signal collecting device, the ECG signal collecting device includes a plurality of collecting terminals, and contacting the user through the plurality of collecting terminals to collect the multi-lead ECG signal of the user;

The ECG (electrocardiogram) signal is related to the physiological characteristics of the user's body. The ECG signals of different users are different. Therefore, the ECG signal can be used for user identification. When the user is identified by using the ECG signal, an ECG signal of the user is usually collected by the corresponding ECG signal acquisition device, and then the identity is determined according to the ECG data corresponding to the ECG signal. However, since the user's ECG signal will be different in different states, for example, the ECG signal of the user in a calm state will be different from the ECG signal after the motion, so that the user is originally but the identity Identifying faulty error conditions, therefore, the accuracy of using ECG signals for identification is not high.

In order to improve the accuracy of the identity recognition by using the ECG signal, in this embodiment, a user identity recognition system is formed by a terminal such as a smart phone, a PAD (tablet computer), a PC (personal computer), and a corresponding ECG signal collection device, and the terminal side is An identity recognition device based on an ECG signal is provided. For example, the ECG signal acquisition device establishes a wireless connection with the PC terminal through Bluetooth to form a user identity recognition system. The ECG signal collection device includes multiple acquisition terminals. When the identity of the user needs to be recognized, the user wears the ECG signal acquisition device, and contacts the user through multiple acquisition terminals of the ECG signal acquisition device to collect the user's multi-channel. Link ECG signal. For example, as shown in FIG. 2, the ECG signal acquisition device is a wrist-type acquisition device, and the ECG signal acquisition device includes a plurality of (five shown) wrist-type acquisition terminals. After the user wears the ECG signal acquisition device on the wrist of the user, the user's multi-lead ECG signal is collected through the plurality of wrist-type acquisition terminals of the ECG signal acquisition device. Then, the acquiring module 10 acquires multiple ECG data corresponding to the multi-lead ECG signal according to the collected multi-lead ECG signal, and forms the ECG data corresponding to the multi-lead ECG signal of the user from the plurality of ECG data. group.

The comparison module 20 is configured to compare the ECG data set with a pre-stored ECG data set;

In this embodiment, the terminal further stores in advance an ECG data group corresponding to the multi-lead ECG signals of the plurality of users. After the obtaining module 10 acquires the ECG data group corresponding to the current user's multi-lead ECG signal, the comparison module 20 compares the ECG data group corresponding to the current user's multi-lead ECG signal with the pre-stored ECG data group. Correct.

The processing module 30 is configured to determine that the identity of the user is successful when the ECG data group matches the pre-stored ECG data group.

After comparing the ECG data group corresponding to the current user's multi-lead ECG signal with the pre-stored ECG data set, if the current user's multi-lead ECG signal corresponds to the ECG data group and the pre-stored ECG data The group matches, at this time, the processing module 30 determines that the user's identity is successful; otherwise, the processing module 30 determines that the user's identity recognition failed. For example, a preset matching degree threshold is set in advance. If the matching degree between the ECG data group corresponding to the multi-lead ECG signal of the current user and the pre-stored ECG data group is greater than the preset matching degree threshold, the processing module 30 determines The user's identity was successfully identified.

The preset matching degree threshold may be set according to a corresponding threshold selection strategy, for example, using two threshold selection strategies, a proportional coefficient training method and a minimum similarity method, respectively, to set a preset matching degree threshold, and then comparing the experimental results. In this way, the applicable data and actual occasions of the two strategies are analyzed and summarized. When setting the preset matching degree threshold, the appropriate threshold selection strategy is selected according to the characteristics of the ECG data and the application.

For example, taking the self-acquisition ECG signal database and the standard ECG signal database as an example, it is found through experiments that the proportional coefficient training method has higher recognition accuracy than the minimum similarity method in the self-acquired ECG signal database, but in the standard ECG signal database. The minimum similarity method has higher recognition accuracy than the proportional coefficient training method. By comparison, we find that in the identification algorithms of the two databases, the error acceptance rate of the minimum similarity method is higher than that of the proportional coefficient training method, and the error rejection rate of the proportional coefficient method is higher than the minimum similarity. This shows that for the same feature extraction algorithm, the same data, the minimum matching degree selection preset matching degree threshold is smaller than the proportional coefficient training selected preset matching degree threshold. In this case, the data in the standard ECG signal database is small, the waveform is stable, and the abrupt waveform is also small, so the small preset matching threshold can distinguish itself from others. The ECG data in the self-collecting ECG signal database is relatively noisy, and the waveform has a certain distortion. Therefore, the difference between itself and itself, and the self and others are small, and it is not suitable to use a small preset matching threshold. Therefore, in actual application, for the characteristics and applications of the ECG data, an appropriate threshold selection strategy is selected to set the preset matching degree threshold.

Further, since the user cannot always be in a calm state when collecting the multi-lead ECG signal of the user, in this embodiment, the collected multi-lead ECG signal is a multi-lead ECG signal of the user in a motion state.

In the solution provided by the embodiment, the plurality of ECG data corresponding to the multi-lead ECG signal of the user is obtained by the obtaining module 10, and the plurality of ECG data obtained by the comparison module 20 are used as the ECG data group and the pre-stored ECG data. The group performs the comparison. When it matches the pre-stored ECG data set, the processing module 30 determines that the user's identity is successful, because the identity of the user is identified by multiple ECG data, thereby improving the accuracy of the identity recognition. Sex.

Further, as shown in FIG. 5, a second embodiment of the ECG signal-based identity recognition apparatus of the present invention is proposed based on the first embodiment. In the second embodiment, the ECG signal-based identity recognition apparatus further includes:

The de-noising module 40 is configured to perform denoising processing on the multi-lead ECG signal after the multi-lead ECG signal is collected by the ECG signal collecting device;

The acquiring module 10 is configured to perform feature extraction on the multi-lead ECG signal after the denoising process, and acquire multiple ECG data corresponding to the multi-lead ECG signal.

In this embodiment, when the user's multi-lead ECG signal is collected by the ECG signal acquisition device, since the contact point between the ECG signal acquisition device and the user's human body occurs during the acquisition process, the human body itself and the circuit leads also enable the acquisition. The ECG signal contains a lot of noise. The common noise of ECG signals mainly includes baseline drift, power frequency interference and myoelectric interference, which will greatly affect the accuracy of identification. Therefore, after the user's multi-lead ECG signal is acquired, the de-noising module 40 first performs denoising processing on the multi-lead ECG signal.

Power frequency interference is generated by the electromagnetic radiation generated by the mains power and the electromagnetic coupling generated by the sensor circuit. The frequency of myoelectric interference covers the spectrum of the entire ECG signal, and its spectral distribution is 5~2000. Hz, there are many complex physiological electrical signals in the human body. A certain bioelectricity is a signal when it is needed to study it. The ECG signal may be noise when it is collected, that is, the noise caused by the human bioelectricity other than the measured physiological variable becomes a muscle. Electrical interference. Therefore, myoelectric interference is random. During the process of collecting ECG signals by using the ECG signal acquisition device, the movement of the human body and the movement of the contact point with the device may cause a baseline drift of the ECG signal, and the baseline drift appears as a vertical offset of the reference line in the waveform time domain. The baseline offset is low frequency interference, and its spectrum is generally below 0.5 Hz.

Among various noises, power frequency interference has the greatest impact on ECG signals. In the prior art, there are many mature and effective methods for eliminating power frequency interference regardless of hardware settings or software filters. In this embodiment, the denoising module 40 uses a notch filter and a Butterworth band rejection filter to eliminate 50 Hz power frequency noise, respectively. Then, the myoelectric interference of the ECG signal is eliminated by the wavelet threshold denoising method. The specific process of the wavelet threshold denoising method is as follows:

(1) Multi-layer wavelet decomposition of ECG signals to obtain wavelet coefficients at different scales;

(2) The wavelet coefficient is based on a threshold or is zeroed or contracted;

(3) Reconstruction of wavelet coefficients after zero or contraction

Eliminating baseline drift is a critical step in eliminating noise because baseline drift can cause serious errors in the detection of critical waveform amplitudes and slope areas of ECG signals. The baseline drift frequency is lower, less than 0.5 Hz. After multi-layer wavelet decomposition of the original ECG signal, the baseline drift is filtered in the low-frequency component, and the low-frequency component where the baseline drift is set to zero can eliminate the baseline drift in the wavelet domain. As a comparison, the moving window median filtering method is used to eliminate the baseline drift. The final signal-to-noise ratio and root mean square error comparison are used to select which denoising method is optimal. The experimental results are shown in Table 1.

It can be seen from the experimental results that the combination of wavelet threshold denoising method, moving window median filtering and Butterworth band rejection filter is better. Therefore, the denoising module 40 uses this combination scheme to perform the user's multi-lead ECG signal. noise.

After the denoising module 40 performs denoising processing on the multi-lead ECG signal, the acquiring module 10 performs feature extraction on the de-noised multi-lead ECG signal, and acquires multiple ECGs corresponding to the multi-lead ECG signal. data. Optionally, since the feature fusion is such that the features of each of the identified individuals are more unique, the recognition rate is higher. Feature fusion can enhance the recognition ability of the whole system, and the characteristics of the fusion will have the advantages of strong complementarity and low redundancy. Therefore, the multi-lead ECG signal after the denoising process is first subjected to feature fusion, and then the feature extraction of the fused multi-lead ECG signal is performed. For example, taking the dual-lead ECG signal as an example, the process of merging the dual-lead ECG signals is as follows:

(1) First, a two-dimensional space matrix of N×N order is established, where N is the maximum value in the time domain of the electrocardiogram data. For dual-lead ECG data, two-dimensional ECG data is mapped into a two-dimensional space.

(2) For the N×N-order two-dimensional space matrix, dimension reduction processing is needed to facilitate feature extraction and distance calculation. Set a matrix window of r × r. If the value of the spatial matrix in the window is more than two small cells, the value is 1 after the dimension reduction.

(3) Since the dimensionality of the two-dimensional space matrix is still mostly zero, the coordinates can be used to store the sparse matrix in the form of val=(row, column, value). This sparse matrix is used as a feature vector after multi-dimensional ECG signal data fusion to carry out subsequent identification algorithms.

Next, different features are extracted for the sparse matrix obtained after data fusion, and then the obtained feature vectors are multi-layered. Feature extraction is mainly to extract the unique properties of ECG signals that are different from other people's ECG signals, such as overall appearance features, wavelet coefficient features, shape features, density distribution features, and so on. Specifically, the obtaining module 10 is configured to:

And performing overall appearance feature extraction on the multi-lead ECG signal after the denoising process, and acquiring multiple ECG data corresponding to the overall appearance feature of the multi-lead ECG signal;

Performing wavelet coefficient feature extraction on the multi-lead ECG signal after the denoising process, and acquiring a plurality of ECG data corresponding to the wavelet coefficient feature of the multi-lead ECG signal;

And performing shape feature extraction on the multi-lead ECG signal after the denoising process, and acquiring a plurality of ECG data corresponding to the shape feature of the multi-lead ECG signal.

After the denoising module 40 performs the denoising process on the multi-lead ECG signal, the acquiring module 10 performs each feature extraction on the de-cored-processed multi-lead ECG signal. Specifically, the acquiring module 10 performs the denoising process. The multi-lead ECG signal is used to extract the overall appearance feature, and obtain multiple ECG data corresponding to the overall appearance feature of the multi-lead ECG signal; perform wavelet coefficient feature extraction on the de-noised multi-lead ECG signal to obtain multi-channel Combining the plurality of ECG data corresponding to the wavelet coefficient feature of the ECG signal; performing shape feature extraction on the de-noised multi-lead ECG signal, and acquiring a plurality of ECG data corresponding to the shape feature of the multi-lead ECG signal.

For example, for the overall appearance feature, the obtaining module 10 maps the multi-lead ECG signal into a sparse matrix according to the feature fusion algorithm, and uses the binary image reflected by the sparse matrix as an overall appearance feature vector of the multi-lead ECG signal.

For the wavelet coefficient feature, the obtaining module 10 performs three-layer wavelet decomposition on the multi-lead ECG signal, and maps the detail coefficients CD1-CD3 and the approximate coefficient CA3 of each level after decomposition into two-dimensional space, and performs dimensional reduction. Sparse storage, the resulting matrix as a wavelet coefficient feature vector.

For shape features, it mainly includes area features, perimeter features, shape factor features, elongation features, center of gravity features, and the like. Among them, the shape factor feature describes the degree of roundness of the shape of the binary image, and the elongation feature reflects the slenderness of the image. For the sparse matrix obtained by the fusion of the multi-lead ECG signals, the sparse matrix of the same person's ECG signal has similarity, and different people have greater differences. So we can think of it as a binary image to extract shape features.

Further, the comparison module 20 is configured to:

And adopting a preset multi-layer identification algorithm to sequentially sequentially, the plurality of ECG data corresponding to the overall appearance feature of the multi-lead ECG signal, the plurality of ECG data corresponding to the wavelet coefficient feature of the multi-lead ECG signal, and the The plurality of ECG data corresponding to the shape feature of the multi-lead ECG signal is compared with the pre-stored ECG data set.

In this embodiment, the comparison module 20 uses a multi-layer identification algorithm for identity recognition. For example, taking three layers as an example, multiple ECG data and wavelet coefficient features corresponding to the overall appearance characteristics of the user's multi-lead ECG signal are sequentially performed. The corresponding plurality of ECG data and the plurality of ECG data corresponding to the shape feature are compared with the pre-stored ECG data group to identify the identity of the user. The specific recognition process of the comparison module 20 using the multi-layer recognition algorithm for identity recognition is as follows:

First, the first layer identification is performed, and the overall appearance feature of the binary image represented by the sparse matrix is used as the feature vector of the first layer identification. Since most of the data of the mapped matrix is zero, if the appearance of the binary image corresponding to the sparse matrix corresponding to two different users is difficult to distinguish by the preset matching degree threshold, the second layer identification is needed, and the other one is compared. Features for identification.

In the second layer identification, the error sample identified by the first layer is the input of the second layer identification, specifically, the sample which is indistinguishable by the overall appearance feature identified by the first layer is in the second layer, each lead core The electrical data is separately subjected to three-layer wavelet decomposition, and the approximation coefficient and the detail coefficient are respectively mapped to the two-dimensional space and then the same dimensionality reduction processing as the first layer is performed. The sparse matrix at this time reflects the wavelet domain characteristics of the ECG signal, and these features are identified by the matching threshold corresponding to the wavelet coefficient feature. Samples that fail to identify during this layer recognition process are also input to the third layer for third layer identification.

In the third layer identification, the input samples respectively extract the shape features and density distribution features of the sparse matrix corresponding to the binary image. It is identified by the corresponding matching degree threshold. The recognition result of the third layer is output as the final recognition result.

After the multi-lead ECG signal of the user is collected, the multi-lead ECG signal is denoised by the denoising module 40, thereby avoiding the interference of the noise signal on the ECG signal, thereby further improving the solution. The accuracy of identification using ECG signals.

It is to be understood that the term "comprises", "comprising", or any other variants thereof, is intended to encompass a non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes those elements. It also includes other elements that are not explicitly listed, or elements that are inherent to such a process, method, article, or device. An element that is defined by the phrase "comprising a ..." does not exclude the presence of additional equivalent elements in the process, method, item, or device that comprises the element.

The serial numbers of the embodiments of the present invention are merely for the description, and do not represent the advantages and disadvantages of the embodiments.

The above are only the preferred embodiments of the present invention, and are not intended to limit the scope of the invention, and the equivalent structure or equivalent process transformations made by the description of the present invention and the drawings are directly or indirectly applied to other related technical fields. The same is included in the scope of patent protection of the present invention.

Claims (20)

An identification method based on an ECG ECG signal, characterized in that the ECG signal-based identification method comprises the following steps: Acquiring a plurality of ECG data corresponding to the multi-lead ECG signal of the user, and using the plurality of ECG data as the ECG data group of the user; Comparing the ECG data set with a pre-stored ECG data set; Determining that the identity of the user is successful when the ECG data set matches the pre-stored ECG data set; The multi-lead ECG signal is collected by an ECG signal acquisition device, and the ECG signal acquisition device includes a plurality of acquisition terminals, and contacts the user through the plurality of acquisition terminals to collect a multi-lead ECG signal of the user. The ECG signal-based identification method according to claim 1, wherein the acquisition terminal of the ECG signal acquisition device is a wrist-type acquisition terminal. The ECG signal-based identification method according to claim 1, wherein before the step of acquiring the plurality of ECG data corresponding to the multi-lead ECG signal of the user, the method further includes: After collecting the multi-lead ECG signal of the user by the ECG signal collecting device, performing denoising processing on the multi-lead ECG signal; The step of acquiring multiple ECG data corresponding to the multi-lead ECG signal of the user includes: Feature extraction is performed on the multi-lead ECG signal after the denoising process, and multiple ECG data corresponding to the multi-lead ECG signal are acquired. The ECG signal-based identification method according to claim 1, wherein the multi-lead ECG signal is a multi-lead ECG signal of a user in a motion state. The ECG signal-based identification method according to claim 1, wherein the multi-lead ECG signal after the denoising process is subjected to feature extraction, and multiple corresponding to the multi-lead ECG signal are acquired. The steps of ECG data include: And performing overall appearance feature extraction on the multi-lead ECG signal after the denoising process, and acquiring multiple ECG data corresponding to the overall appearance feature of the multi-lead ECG signal; Performing wavelet coefficient feature extraction on the multi-lead ECG signal after the denoising process, and acquiring a plurality of ECG data corresponding to the wavelet coefficient feature of the multi-lead ECG signal; And performing shape feature extraction on the multi-lead ECG signal after the denoising process, and acquiring a plurality of ECG data corresponding to the shape feature of the multi-lead ECG signal. The ECG signal-based identification method according to claim 5, wherein the step of comparing the ECG data set with the pre-stored ECG data set comprises: And adopting a preset multi-layer identification algorithm to sequentially sequentially, the plurality of ECG data corresponding to the overall appearance feature of the multi-lead ECG signal, the plurality of ECG data corresponding to the wavelet coefficient feature of the multi-lead ECG signal, and the The plurality of ECG data corresponding to the shape feature of the multi-lead ECG signal is compared with the pre-stored ECG data set. The ECG signal-based identification method according to claim 5, wherein the multi-lead ECG signal is a multi-lead ECG signal of a user in a motion state. The ECG signal-based identification method according to claim 5, wherein the shape feature comprises an area feature, a perimeter feature, a shape factor feature, a center of gravity feature, and an elongation feature. The ECG signal-based identification method according to claim 8, wherein the step of comparing the ECG data set with the pre-stored ECG data set comprises: And adopting a preset multi-layer identification algorithm to sequentially sequentially, the plurality of ECG data corresponding to the overall appearance feature of the multi-lead ECG signal, the plurality of ECG data corresponding to the wavelet coefficient feature of the multi-lead ECG signal, and the The plurality of ECG data corresponding to the shape feature of the multi-lead ECG signal is compared with the pre-stored ECG data set. The ECG signal-based identification method according to claim 9, wherein the multi-lead ECG signal is a multi-lead ECG signal of a user in a motion state. An identity recognition device based on an ECG signal, wherein the ECG signal-based identity recognition device comprises: An acquiring module, configured to acquire a plurality of ECG data corresponding to the multi-lead ECG signal of the user, and use the plurality of ECG data as the ECG data group of the user; a comparison module, configured to compare the ECG data set with a pre-stored ECG data set; a processing module, configured to determine that the identity of the user is successful when the ECG data group matches the pre-stored ECG data group; The multi-lead ECG signal is collected by an ECG signal acquisition device, and the ECG signal acquisition device includes a plurality of acquisition terminals, and contacts the user through the plurality of acquisition terminals to collect a multi-lead ECG signal of the user. The ECG signal-based identification device according to claim 11, wherein the acquisition terminal of the ECG signal acquisition device is a wrist-type acquisition terminal. The ECG signal-based identity recognition apparatus according to claim 11, wherein the ECG signal-based identity recognition apparatus further comprises: a denoising module, configured to perform denoising processing on the multi-lead ECG signal after the multi-lead ECG signal is collected by the ECG signal collecting device; The acquiring module is configured to perform feature extraction on the multi-lead ECG signal after the denoising process, and acquire multiple ECG data corresponding to the multi-lead ECG signal. The ECG signal-based identification device of claim 11, wherein the multi-lead ECG signal is a multi-lead ECG signal of a user in a motion state. The ECG signal-based identity recognition apparatus according to claim 11, wherein the acquisition module is configured to: And performing overall appearance feature extraction on the multi-lead ECG signal after the denoising process, and acquiring multiple ECG data corresponding to the overall appearance feature of the multi-lead ECG signal; Performing wavelet coefficient feature extraction on the multi-lead ECG signal after the denoising process, and acquiring a plurality of ECG data corresponding to the wavelet coefficient feature of the multi-lead ECG signal; And performing shape feature extraction on the multi-lead ECG signal after the denoising process, and acquiring a plurality of ECG data corresponding to the shape feature of the multi-lead ECG signal. The ECG signal-based identity recognition apparatus according to claim 15, wherein the comparison module is configured to: And adopting a preset multi-layer identification algorithm to sequentially sequentially, the plurality of ECG data corresponding to the overall appearance feature of the multi-lead ECG signal, the plurality of ECG data corresponding to the wavelet coefficient feature of the multi-lead ECG signal, and the The plurality of ECG data corresponding to the shape feature of the multi-lead ECG signal is compared with the pre-stored ECG data set. The ECG signal-based identification device of claim 15, wherein the multi-lead ECG signal is a multi-lead ECG signal of a user in a motion state. The ECG signal based identification device of claim 15 wherein the shape features comprise area features, perimeter features, form factor features, center of gravity features, and elongation features. The ECG signal-based identity recognition apparatus according to claim 18, wherein the comparison module is configured to: And adopting a preset multi-layer identification algorithm to sequentially sequentially, the plurality of ECG data corresponding to the overall appearance feature of the multi-lead ECG signal, the plurality of ECG data corresponding to the wavelet coefficient feature of the multi-lead ECG signal, and the The plurality of ECG data corresponding to the shape feature of the multi-lead ECG signal is compared with the pre-stored ECG data set. The ECG signal based identification device of claim 19, wherein the multi-lead ECG signal is a multi-lead ECG signal of a user in a motion state.