A progressive heart-shock-signal-based identity database construction method, an identity recognition method, an apparatus, and a storage medium

Abstract

The application discloses a progressive heart impact signal-based identity database construction method and identity recognition method and device and a storage medium, and belongs to the technical field of biological recognition. The method comprises the following steps: collecting heart impact signals of multiple users, storing the heart impact signals in the form of time and amplitude binary groups, and obtaining original signals; preprocessing the original signals to remove invalid signal segments; using multiple time sequence prediction modules to perform signal decomposition on the preprocessed signals, and extracting heartbeat signal segments; obtaining an intermediate feature vector by performing random masking and signal reconstruction on the heartbeat signals, performing comparative learning and metric learning based on the intermediate feature vector, and obtaining an identity feature vector of a user; regarding a feature vector set of each user as a cluster, determining the center of each cluster, and constructing a target identity database. An efficient, accurate and stable vital sign state monitoring algorithm is realized, so that accurate and efficient identity recognition is realized.

Description

Technical Field

[0001] This invention belongs to the field of biometric technology, and more specifically, relates to a progressive method for constructing an identity database based on cardiac impulse signals, an identity recognition method, a device, and a storage medium. Background Technology

[0002] In recent years, biometric identification has faced multiple challenges, including data security, privacy, technical accuracy, legal ethics, and system scalability. Currently used biometric identification methods include fingerprint recognition, but it is susceptible to contamination, damage, and forgery. While facial recognition is a contactless method, it is significantly affected by lighting and facial expressions, and raises prominent privacy and ethical concerns. Other methods, such as iris recognition, voiceprint recognition, and behavioral recognition, often experience inconveniences or are easily interfered with during testing.

[0003] Repeated personal authentication is a tedious yet crucial daily task for people accessing sensitive information and private assets. Inconspicuous biometric identification methods are gaining increasing attention due to their ability to protect individual rights while reducing the burden of repetitive checks. Ballistocardiography (BCG) is a technique that indirectly reflects cardiovascular activity by measuring the minute movements caused by the heartbeat. Because BCG signals contain information reflecting a person's cardiovascular activity, they have become a non-invasive biometric identification method. However, the complex composition and diverse waveforms of BCG signals pose challenges to the effective extraction of identity information.

[0004] Current identity recognition methods based on BCG signals mainly extract features using simple deep learning networks and compare them to stable ECG (Electrocardiogram) signals. However, these methods are limited by their high requirements for the quality of the test signals. Traditional feature extraction methods typically rely on fixed feature sets, making them vulnerable to individual differences and unable to handle subtle variations, resulting in low specificity when dealing with large amounts of identity recognition data. Furthermore, identity recognition based on ECG signals relies on additional data overhead and cannot achieve truly seamless authentication, nor can it address the open-set problem. Therefore, existing technologies suffer from high costs in acquiring BCG signals and corresponding ECG signals, and inaccurate recognition results due to low specificity. Summary of the Invention

[0005] In view of the shortcomings of related technologies, the purpose of this invention is to provide a progressive method for constructing an identity database based on cardiac impact signals, an identity recognition method, an apparatus and a storage medium, which aims to solve the problems of high cost of acquiring BCG signals and corresponding ECG signals, low specificity in identity recognition leading to inaccurate identity databases and inaccurate recognition results in the prior art.

[0006] To achieve the above objectives, the present invention provides a progressive method for constructing an identity database based on cardiac impact signals, comprising:

[0007] S1. Collect cardiac impact signals from multiple users and store them as time and amplitude pairs to obtain the raw signals; preprocess the raw signals to remove invalid signal segments;

[0008] S2. Perform signal decomposition on the preprocessed raw signal to extract the heartbeat signal segment q. i,j q i,j This represents the j-th signal segment of the i-th user; i ranges from 0 to N1, where N1 is the number of users, and j ranges from 1 to N2, where N2 is the number of signal segments.

[0009] S3, by analyzing the heartbeat signal segment q i,j After random masking, signal reconstruction is performed to obtain the heartbeat signal segment q. i,j A consistent reconstructed signal is obtained by acquiring the intermediate feature vector f from the signal encoding and decoding processes during signal reconstruction. i,j Based on the intermediate feature vector f i,j Comparative learning and metric learning are performed to obtain the user's feature vector;

[0010] S4. Treat each user's feature vector set as a cluster, determine the center of each cluster, and construct the target identity database.

[0011] Optionally, the preprocessing of the original signal includes: performing first-order difference processing on the original signal, calculating the mean and standard deviation of the signal using a sliding window on the first-order difference result, and marking segments with window energy fluctuations exceeding 2 standard deviations as invalid signal segments.

[0012] Optionally, S2 specifically includes:

[0013] S21. Perform sequence decomposition on the preprocessed signal X. The decomposition formula is as follows:

[0014] X br =AvgPool(Padding(X))

[0015] X s =XX br

[0016] Where AvgPool represents the average pooling layer, Padding represents padding, and X br X represents the separated respiratory component. s Includes heart rate component and noise component;

[0017] S22. Use a sliding window of size L1 from X s Extract the first sample point Use a sliding window with a window size of L2 from X s Second sampling point Where L1 is the window size corresponding to 0.5s to 0.7s, L2 is the window size corresponding to 0.04s to 0.08s, and g is the median coordinate of the signal within the sliding window;

[0018] S23. Calculate the energy of the first sample point respectively. Energy of the second sample point The first and second sample points are input into a multilayer perceptron (MLP) for prediction, and the heartbeat features Y = MLP(X) are obtained respectively. s [g-1],…,X s [g-L2]) and noise characteristics Z=MLP(X s [g-1],…,X s [g-L1]), by The standardized heartbeat signal segment is obtained; the loss function of the multilayer perceptron (MLP) is: When inputting the first sample point, V in the loss function adopts... When inputting the second sample point, V in the loss function adopts...

[0019] Optionally, S3 specifically includes:

[0020] S31, By analyzing the heartbeat signal segment q i,j Perform random masking, and then encode the processed signal to obtain the intermediate feature vector f. i,j Based on the intermediate feature vector f i,j Decoding yields a reconstructed signal, which is then compared with the heartbeat signal segment q. i,j Consistent;

[0021] S32, the intermediate feature vector f i,j The input is fed into a Siamese network model with self-attention as its core, and the contrastive features r are obtained. i,j The loss function Loss2 of the Siamese network model is:

[0022]

[0023] Where k represents the k-th user and l represents the l-th signal segment;

[0024] S33, the comparison feature r i,j Inputting the data into a multilayer perceptron (MLP) for dimensionality reduction enhances the contrastive features r corresponding to different users. i,j The difference between them yields the metric feature s. i,j ; metric features s corresponding to different users i,j The distance between them is greater than or equal to a preset value m, and the metric feature s i,j As the user's feature vector; the loss function Loss3 expression of the multilayer perceptron (MLP) is:

[0025] Loss3=∑ i,j,k,l I i＝k (‖s i,j -s k,l ‖)+I i≠k (ReLU(m-‖s i,j -s k,l ||)).

[0026] Optionally, determining the center of each cluster includes: taking the feature vector s of each user... i,j The average value is used as the center of the corresponding cluster.

[0027] Secondly, the present invention also provides a progressive identity recognition method based on cardiac impulse signals, applied to a target identity database constructed as described in any of the first aspects, comprising:

[0028] The cardiac impulse signal of the user to be identified is acquired and stored as a pair of time and amplitude to obtain the original signal to be identified; the original signal to be identified is preprocessed to remove invalid signal segments.

[0029] The preprocessed raw signal to be identified is decomposed to extract the heartbeat signal fragment q. j q j This represents the j-th signal segment of the user; the value of j ranges from 1 to N2, where N2 is the number of signal segments;

[0030] By analyzing the heartbeat signal segment q j After random masking, signal reconstruction is performed to obtain the heartbeat signal segment q. i,j A consistent reconstructed signal is obtained by acquiring the intermediate feature vector f from the signal encoding and decoding processes during signal reconstruction. j Based on the intermediate feature vector f j Comparative learning and metric learning are performed to obtain the metric features of the user to be identified;

[0031] Calculate the Euclidean distances from the metric features of the user to be identified to the centers of all clusters in the target identity database, and obtain the identity features F of the user to be identified. i The identity feature F i The input is fed into a multi-head self-attention module encoder, and the result is fed into a softmax layer to obtain the identity feature F. i The probability of being authenticated as the user corresponding to each cluster;

[0032] When the maximum probability exceeds the preset threshold, the identity feature F will be... i The identity information is identified as having the highest probability.

[0033] Thirdly, the present invention also provides a progressive identity recognition device based on cardiac impulse signals, comprising:

[0034] The data acquisition and preprocessing module is used to collect cardiac impact signals from multiple users, store them as time and amplitude tuples to obtain the raw signals, and preprocess the raw signals to remove invalid signal segments.

[0035] The signal decomposition module is used to decompose the preprocessed raw signal and extract the heartbeat signal fragment q. i,j q i,j This represents the j-th signal segment of the i-th user; i ranges from 0 to N1, where N1 is the number of users, and j ranges from 1 to N2, where N2 is the number of signal segments.

[0036] The training module is used to train the heartbeat signal segment q i,j After random masking, signal reconstruction is performed to obtain the heartbeat signal segment q. i,j A consistent reconstructed signal is obtained by acquiring the intermediate feature vector f from the signal encoding and decoding processes during signal reconstruction. i,j Based on the intermediate feature vector f i,j Comparative learning and metric learning are performed to obtain the user's feature vector;

[0037] The identity database construction module is used to treat each user's feature vector set as a cluster, determine the center of each cluster, and construct the target identity database.

[0038] Fourthly, the present invention also provides a computer-readable storage medium storing computer instructions, the computer instructions being configured to cause a processor to execute and implement, as described in any one of the first aspects, the progressive identity database construction method based on cardiac impulse signals, or the progressive identity recognition method based on cardiac impulse signals as described in the second aspect.

[0039] Compared with the prior art, the above-described technical solutions conceived in this invention can achieve the following beneficial effects:

[0040] 1. This invention provides a progressive identity recognition method based on cardiac impulse signals. Using BCG signals as the test signal, and employing non-orthogonal projection to perform dimensionality reduction, feature extraction, and signal recovery, the heartbeat component is accurately extracted from the composite signal, generating a refined heartbeat waveform. Progressive metric learning (using contrastive learning and metric learning sequentially) is used to project the heartbeat segments into a latent space to obtain identity feature information and promote the separation of unique features for each individual, creating accurate individual matching identifiers for identity recognition. This solution provides an efficient, accurate, and stable vital sign monitoring algorithm, thereby achieving accurate and efficient identity recognition. It is applicable to multiple fields, including professional medical diagnosis, smart home systems, and personal health care, providing users with a more reliable vital sign monitoring solution.

[0041] 2. This invention provides a progressive identity recognition method based on cardiac impulse signals. It involves sampling signals of a specific length at both large and small scales and predicting subsequent signal energy. Then, a loss function is used to correct the original data, resulting in more accurate heartbeat signal components and minimizing noise interference in the BCG signal used for heartbeat segmentation. The method decomposes the subject's cardiac impulse signal, i.e., orthogonalizes the heartbeat feature vectors, improving the algorithm's usability and robustness. By clustering user feature vectors, the vectors of different subjects are projected into the accumulation space, and subject clusters are distinguished using a certain distance threshold. This solution addresses the high data acquisition cost of methods requiring ECG comparison for identity recognition and the inaccuracy of the identity database due to low specificity during recognition. It achieves low-cost, continuous, and highly efficient identity recognition. Attached Figure Description

[0042] Figure 1 This is a flowchart illustrating a progressive identity database construction method based on cardiac impact signals provided by the present invention.

[0043] Figure 2 This is a signal decomposition diagram provided by the present invention, wherein (a) is the original signal diagram, (b) is the respiratory trend diagram, (c) is the artifact noise diagram, and (d) is the filtered heartbeat diagram;

[0044] Figure 3 It is a combined representation of a heat map;

[0045] Figure 4 This is a diagram illustrating the segmentation of heartbeats at rest and after strenuous exercise;

[0046] Figure 5This is a visualization of the results of metric learning, where (a) is the original image of the identity information encoding, (b) is the feature after contrastive learning, and (c) is the final feature after metric learning. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0048] The following description, in conjunction with a preferred embodiment, illustrates the content involved in the above embodiments.

[0049] Example 1

[0050] like Figure 1 As shown, this embodiment of the invention provides a progressive method for constructing an identity database based on cardiac impact signals, including:

[0051] S1. Collect cardiac impact signals from multiple users and store them as time and amplitude pairs to obtain the raw signals; preprocess the raw signals to remove invalid signal segments;

[0052] S2. Perform signal decomposition on the preprocessed raw signal to extract the heartbeat signal segment q. i,j q i,j This represents the j-th signal segment of the i-th user; i ranges from 0 to N1, where N1 is the number of users, and j ranges from 1 to N2, where N2 is the number of signal segments.

[0053] S3, by analyzing the heartbeat signal segment q i,j After random masking, signal reconstruction is performed to obtain the heartbeat signal segment q. i,j A consistent reconstructed signal is obtained by acquiring the intermediate feature vector f from the signal encoding and decoding processes during signal reconstruction. i,j Based on the intermediate feature vector f i,j Comparative learning and metric learning are performed to obtain the user's feature vector;

[0054] S4. Treat each user's feature vector set as a cluster, determine the center of each cluster, and construct the target identity database.

[0055] This invention uses BCG signals as test signals and extracts feature vectors to construct an identity database through segmented heartbeats and continuous heartbeat segments. First, signal acquisition and preprocessing are performed, storing each user's heartbeat signal data separately and removing invalid signal segments. In this embodiment, a PVDF flexible piezoelectric film sensor is used for data collection, typically embedded in mattresses or cushions, suitable for long-term, non-intrusive signal detection. Signal separation is then performed on the preprocessed data. To accurately represent user identity using BCG signals, single or continuous heartbeats are extracted as identification criteria. A three-peak detection method is employed, identifying three consecutive energy value spikes, defined as peaks in the BCG signal. The middle peak of these three consecutive energy spikes is defined as the J-peak. Heartbeat segmentation is performed based on the identification of the J-peak, with each 2.5-second segment starting from the J-peak constituting a heartbeat sample.

[0056] By analyzing the heartbeat signal q i,j Random masking and signal reconstruction are performed, and a loss function is used to make the reconstructed signal as consistent as possible with the original signal. In this process, the intermediate feature vector f of signal encoding and decoding is obtained. i,j The obtained intermediate features f i,j The feature vectors f of two different users are combined using a twin network. i,j and f k,l By performing comparative learning, the loss function is used to make the feature vectors obtained after encoding the heartbeat samples of each pair of subjects orthogonal, that is, the angle between the two vectors is 90 degrees and the inner product of the two vectors is 0.

[0057] As the number of subjects increases, although the user feature vectors are pairwise orthogonal, simply satisfying the orthogonality of feature vectors is still insufficient. When the encoded feature vectors are all near the origin, their Euclidean distances are too close, making it difficult to distinguish the identities of some users. Therefore, progressive metric learning needs to be introduced, using contrastive learning and metric learning methods in stages. Triplet loss minimizes the distance between positive sample pairs while maximizing the distance between negative sample pairs, making the feature vectors of the same user closer together and the feature vectors of different users farther apart, thus facilitating subsequent identity recognition.

[0058] After expanding the user's feature vectors, a user identity database is constructed using fingerprint recognition methods. The identity features of the user to be identified are then compared with this database for final identification. Alternatively, each user's feature vector is grouped into a cluster, with each cluster having a center. All user feature vectors constitute the target identity database. Finally, the distance from the user's feature vector to the target identity database is calculated for identification.

[0059] Optionally, S1 specifically includes:

[0060] S11. Collect the cardiac impact signal of each user, and store the cardiac impact signal data of each user separately in a CSV format file in the form of time and amplitude tuples to obtain the raw signal;

[0061] S12. Perform first-order difference processing on the original signal, and use a sliding window to calculate the mean and standard deviation of the signal from the first-order difference result. Mark segments whose window energy fluctuation exceeds 2 standard deviations as invalid signal segments.

[0062] S13. Remove invalid signal segments from the original signal.

[0063] BCG (Ballistocardiogram) signals are acquired using a data acquisition device, for example, a pressure sensor; they are stored as CSV (Comma-Separated Values) files in binary pairs of time and energy values. Each user's cardiac impact signal data is stored separately in a CSV file, numbered starting from sequence 1. Let the initial value of i be 1, and q... i This represents the cardiac impact signal data of the i-th user. Signal preprocessing primarily involves removing invalid signal segments (i.e., those containing large motion artifacts and excessive external noise), reading the stored file, and segmenting the file for subsequent processing. Data slices are created for each subject starting from index 1, with initial values ​​of j and q... i,j Let j be the signal segment of the i-th user. In order to identify such invalid segments due to intense physical movement or excessive external noise interference, the signal is subjected to first-order difference processing. The average energy of the first-order difference result is calculated using a sliding window, and the mean and standard deviation are calculated. Segments with energy fluctuations exceeding 2 standard deviations will be marked as invalid segments.

[0064] Optionally, S2 specifically includes:

[0065] S21. Perform sequence decomposition on the preprocessed signal X. The decomposition formula is as follows:

[0066] X br =AvgPool(Padding(X))

[0067] X s =XX br

[0068] Where AvgPool represents the average pooling layer, Padding represents padding, and X br X represents the separated respiratory component. sIncludes heart rate component and noise component;

[0069] S22. Use a sliding window of size L1 from X s Extract the first sample point Use a sliding window with a window size of L2 from X s Second sampling point Where L1 is the window size corresponding to 0.5s to 0.7s, L2 is the window size corresponding to 0.04s to 0.08s, and g is the median coordinate of the signal within the sliding window;

[0070] S23. Calculate the energy of the first sample point respectively. Energy of the second sample point The first and second sample points are input into a multilayer perceptron (MLP) for prediction, and the heartbeat features Y = MLP(X) are obtained respectively. s [g-1],…,X s [g-L2]) and noise characteristics Z=MLP(X s [g-1],…,X s [g-L1]), by The standardized heartbeat signal segment is obtained; the loss function of the multilayer perceptron (MLP) is: When inputting the first sample point, V in the loss function adopts... When inputting the second sample point, V in the loss function adopts...

[0071] Let the original one-dimensional BCG signal be X = {X_1, X_2, ..., X_n}, where n represents the number of discrete points, i.e., the length of the one-dimensional signal. The original BCG signal X can be decomposed into three beat components X_1, X_2, ..., X_n. hb Respiratory component X br And noise component ε, X = X hb +X br +ε.

[0072] To extract the pure heartbeat component, the BCG signal is first decomposed into a sequence, as follows:

[0073] X br =AvgPool(Padding(X))

[0074] X s =XX br

[0075] Among them, X br X represents the separated respiratory component. s It includes heart rate components and noise components.

[0076] In this embodiment, sliding windows with window sizes L2=7 and L1=101 are used sequentially from X... s The first sample point was obtained by mid-sampling. Second sampling point Calculate the energy of the first sample point respectively. Energy of the second sample point The signal is input into a multilayer perceptron (MLP) for prediction, resulting in heartbeat features Y and noise features Z. Then, the loss function Loss1 is used to verify whether the combined heartbeat features Y and noise features Z constitute the signal X. s This allows for data correction. (Reference) Figure 2 The diagrams of each component can be seen. Figure 3 The heatmaps generated when the sliding window values ​​are L2=7 and L1=101 show that the effect is best under these values.

[0077] The feedforward neural network used in this embodiment consists of multiple layers, including an input layer, hidden layers, and an output layer. Each layer comprises trainable neurons connected by weights. Multiprocessor Learning (MLP) is one of the most commonly used supervised learning algorithms, widely applied to classification and regression problems. The training process of MLP involves adjusting the weights using backpropagation and gradient descent to minimize the loss function. This process typically requires a large amount of data and computational resources, and may require adjusting hyperparameters such as the learning rate, regularization term, number of hidden layers, and number of neurons. When data is limited, the loss function Loss1 is used to address the potential overfitting problem in MLP.

[0078] Optionally, S3 specifically includes:

[0079] S31, By analyzing the heartbeat signal segment q i,j Perform random masking, and then encode the processed signal to obtain the intermediate feature vector f. i,j Based on the intermediate feature vector f i,j Decoding yields a reconstructed signal, which is then compared with the heartbeat signal segment q. i,j Consistent;

[0080] S32, the intermediate feature vector f i,j The input is fed into a Siamese network model with self-attention as its core, and the contrastive features r are obtained. i,j The loss function Loss2 of the Siamese network model is:

[0081]

[0082] Where k represents the k-th user and l represents the l-th signal segment;

[0083] S33, the comparison feature r i,j Inputting the data into a multilayer perceptron (MLP) for dimensionality reduction enhances the contrastive features r corresponding to different users. i,j The difference between them yields the metric feature s. i,j ; metric features s corresponding to different users i,j The distance between them is greater than or equal to a preset value m, and the metric feature s i,j As the user's feature vector; the loss function Loss3 expression of the multilayer perceptron (MLP) is:

[0084] Loss3=∑ i,j,k,l I i＝k (‖s i,j -s k,l ‖)+I i≠k (ReLU(m-‖s i,j -s k,l ||)).

[0085] Heartbeat signal segment q i,j Random masking is performed by dividing the signal into 6 patches, randomly setting the content of 3 patches to 0, and adding position embedding information to each patch. These patches are then input into the encoder and decoder layers of the Transformer, resulting in intermediate feature vectors f. i,j Reconstructed signal p i,j The loss function is Loss = ∑ i,j ||p i,j -q i,j This stage is mainly used to extract the intermediate feature vector f of the signal segment. i,j .

[0086] The obtained intermediate feature vector f i,j The input is fed into a Siamese Network architecture centered on self-attention, that is, f i,j Q, K, and V are obtained through linear layer encoding respectively, and contrastive features are calculated. Where d represents the vector dimension.

[0087] To ensure greater diversity in the feature space among samples from different users, a metric learning approach is introduced. Building upon contrastive learning, this approach further enhances the differences between user features. Specifically, a preset value *m* is set, and the metric features *s* are obtained through dimensionality reduction using a multilayer perceptron (MLP). i,j =MLP(r i,j ).

[0088] Measure the features s of the dimensionality reduction resulti,j Construct a distance function to enhance appropriate features, such that the distance between samples from different users is greater than or equal to a preset value m, and use triples to calculate the loss function Loss3.

[0089] Optionally, determining the center of each cluster includes: taking the feature vector s of each user... i,j The average value is used as the center of the corresponding cluster.

[0090] After the above alignment operation, feature clusters are obtained in the latent space, and corresponding target identity databases are established based on these clusters.

[0091] Based on the above embodiments, a method for identifying the identity of the user to be tested can be implemented using a target identity database, specifically including:

[0092] The cardiac impulse signal of the user to be identified is acquired and stored as a pair of time and amplitude to obtain the original signal to be identified; the original signal to be identified is preprocessed to remove invalid signal segments.

[0093] The preprocessed raw signal to be identified is decomposed to extract the heartbeat signal fragment q. j q j This represents the j-th signal segment of the user; the value of j ranges from 1 to N2, where N2 is the number of signal segments;

[0094] By analyzing the heartbeat signal segment q j After random masking, signal reconstruction is performed to obtain the heartbeat signal segment q. i,j A consistent reconstructed signal is obtained by acquiring the intermediate feature vector f from the signal encoding and decoding processes during signal reconstruction. j Based on the intermediate feature vector f j Comparative learning and metric learning are performed to obtain the metric features of the user to be identified;

[0095] Calculate the Euclidean distances from the metric features of the user to be identified to the centers of all clusters in the target identity database, and obtain the identity features F of the user to be identified. i The identity feature F i The input is fed into a multi-head self-attention module encoder, and the result is fed into a softmax layer to obtain the identity feature F. i The probability of being authenticated as the user corresponding to each cluster;

[0096] When the maximum probability exceeds the preset threshold, the identity feature F will be... i The identity information is identified as having the highest probability.

[0097] The identity features F of the user to be identified are obtained by comparing the distances from the metric features of the user to each cluster point. iIdentification is performed through the encoding layer of the TRM. Specifically, the method for constructing each individual's ECG identity involves calculating core features: The static ECG database, i.e., the target identity database R = [R1, R2, ...], is obtained. The method for identifying signal segments of the user to be identified is as follows: calculate the Euclidean distance from the identity feature corresponding to the signal segment to each feature (the center of all clusters) in the target identity database, i.e., F... i =[∑(s x -R i ) 2 ,for i in range(len(R))]. The above identity features F i The input is fed into the multi-head self-attention module encoder, and the result is fed into the Softmax layer to obtain the probability of the identity feature being authenticated as each user. When the probability of the highest probability of being authenticated as each user exceeds a preset threshold, the user information corresponding to the cluster corresponding to that probability is determined as the identity information of the signal segment, thereby obtaining the identity information of the user to be identified; where the preset threshold is 0.6.

[0098] This solution provides an efficient, accurate, and stable method for assessing vital signs, enabling the accurate construction of a user database for rapid and precise user identification. Specifically, it employs a progressive, cardiac impulse signal-based identification method applicable to multiple fields, including professional medical diagnosis, smart home systems, and personal health care, providing users with a more reliable vital sign monitoring solution. The method's potential applications include, but are not limited to: in medical environments, combining cardiac impulse signals with patient identification to ensure the security and accuracy of medical information; in mobile devices such as smartphones, utilizing heart rate sensors for identity verification, providing a contactless and secure authentication method; and in sensitive and confidential positions, continuously identifying individuals through cardiac impulse signal acquisition embedded in chairs.

[0099] In a specific embodiment, pressure sensors are installed on the mattress to acquire each user's cardiac impact signal for continuous authentication. This prevents unauthorized access and potential security threats, reduces the frequency of user logins, and provides a seamless experience. After successful authentication, personalized settings can be accessed to modify the mattress or monitor the user's status based on those settings. The Matplotlib toolkit in Python was used to generate the image; the horizontal axis represents time, and the vertical axis represents body posture characteristics. Figure 4As can be seen, at the beginning and end, the user was in an out-of-bed state (state1), with two periods of physical activity (state3), and the remaining time was in a calm state (state2). It can be clearly observed that the algorithm accurately calculated the user's different vital signs, which corresponded to the user's actual vital signs. This demonstrates that the algorithm of this invention can effectively and accurately calculate the user's vital signs. Figure 4 This is a heartbeat pattern diagram of a person before and after exercise, with different markers indicating that our method extracts the specificity of each person's heartbeat, unaffected by heart rate. Figure 5 (a) in the graph is the original cluster graph of each user's identity features. Figure 5 (b) in the figure shows the result obtained after differentiating different users through contrastive learning. Figure 5 (c) in the diagram represents the result obtained by widening the distance between users through metric learning. Furthermore, on mobile devices, users can securely access applications without entering a password.

[0100] This invention uses BCG signals as test signals, eliminating the need for ECG signals for comparison. It accurately extracts the heartbeat component from the composite signal, generating a refined heartbeat waveform. Progressive metric learning (contrastive learning and metric learning) projects heartbeat segments into a latent space to obtain identity feature information and promotes the separation of unique features for each individual. Combined with a fingerprint recognition strategy, it creates accurate individual matching identifiers, constructs an identity database, and performs identity recognition. This solves the technical problems of existing technologies, such as high costs in acquiring BCG and ECG signals, low specificity leading to inaccurate identity databases, and inaccurate recognition results. It achieves efficient, accurate, and stable vital sign monitoring, constructs an accurate user identity database, and thus enables accurate and efficient identity recognition.

[0101] Example 2

[0102] The present invention also provides a progressive identity recognition device based on cardiac impulse signals, comprising:

[0103] The signal decomposition module is used to decompose the preprocessed raw signal and extract the heartbeat signal fragment q. i,j q i,j This represents the j-th signal segment of the i-th user; i ranges from 0 to N1, where N1 is the number of users, and j ranges from 1 to N2, where N2 is the number of signal segments.

[0104] The training module is used to train the heartbeat signal segment q i,j After random masking, signal reconstruction is performed to obtain the heartbeat signal segment q. i,jA consistent reconstructed signal is obtained by acquiring the intermediate feature vector f from the signal encoding and decoding processes during signal reconstruction. i,j Based on the intermediate feature vector f i,j Comparative learning and metric learning are performed to obtain the user's feature vector;

[0105] The identity database construction module is used to treat each user's feature vector set as a cluster, determine the center of each cluster, and construct the target identity database.

[0106] The progressive identity recognition device based on cardiac impulse signals provided in this embodiment of the invention is used to execute the progressive identity recognition method based on cardiac impulse signals provided in any embodiment of the invention, and has corresponding beneficial effects.

[0107] Example 3

[0108] The present invention also provides a computer-readable storage medium storing computer instructions for causing a processor to execute and implement the progressive identity recognition method based on cardiac impulse signals as described in any one of Embodiments 1.

[0109] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A progressive method for constructing an identity database based on cardiac impact signals, characterized in that, include: S1. Collect cardiac impact signals from multiple users and store them as a pair of time and amplitude values ​​to obtain the raw signals; Preprocess the original signal to remove invalid signal segments; S2. Perform signal decomposition on the preprocessed raw signal to extract heartbeat signal segments. , Indicates the first The first user's Segment of signal; The value ranges from 0 to N1, where N1 is the number of users. The value range is from 1 to N2, where N2 is the number of signal segments; S31, By analyzing the heartbeat signal segment Perform random masking and then encode the processed signal to obtain an intermediate feature vector. Based on the intermediate feature vector Decoding yields a reconstructed signal, which is then compared with a segment of the heartbeat signal. Consistent; S32, the intermediate feature vector The input is fed into a Siamese network model with self-attention as its core, and contrastive features are obtained. The loss function of the twin network model for: in, Indicates the first One user, Indicates the first Segment of signal; S33, The comparison features Dimensionality reduction is performed by inputting into a multilayer perceptron (MLP) to enhance the contrast features corresponding to different users. The differences between them yield the metric features. ; Measurement features corresponding to different users The distance between them is greater than or equal to a preset value m, and the measurement feature The feature vector of the user; the loss function of the multilayer perceptron (MLP) The expression is: ； S4. Treat each user's feature vector set as a cluster, determine the center of each cluster, and construct the target identity database.

2. The method as described in claim 1, characterized in that, The preprocessing of the original signal includes: performing first-order difference processing on the original signal, calculating the mean and standard deviation of the signal using a sliding window on the first-order difference result, and marking segments with window energy fluctuations exceeding 2 standard deviations as invalid signal segments.

3. The method as described in claim 1, characterized in that, S2 specifically includes: S21, process the preprocessed signal Sequence decomposition is performed using the following formula: in, Indicates the average pooling layer. Indicates filling in the blanks. This indicates the separated respiratory component. Includes heart rate component and noise component; S22, Use window size as The sliding window from Extract the first sample point Use window size The sliding window from Second sampling point ;in, The value range is the window size corresponding to 0.5s to 0.7s. The value ranges from 0.04s to 0.08s, corresponding to the window size. The median coordinates of the signal within the sliding window; S23. Calculate the energy of the first sample point respectively. Energy of the second sample point The first and second sample points are then input into a multilayer perceptron (MLP) for prediction to obtain heartbeat features. and noise characteristics ,Depend on The standardized heartbeat signal segment is obtained; the loss function of the multilayer perceptron (MLP) is: When the first sample point is input, the loss function contains... use When inputting the second sample point, the loss function in use .

4. The method as described in claim 1, characterized in that, Determining the center of each cluster includes: taking the feature vector of each user... The average value is used as the center of the corresponding cluster.

5. A progressive identity recognition method based on cardiac impulse signals, using a target identity database constructed according to any one of claims 1-4, characterized in that, include: The cardiac impulse signal of the user to be identified is acquired and stored as a pair of time and amplitude to obtain the original signal to be identified; the original signal to be identified is preprocessed to remove invalid signal segments. The preprocessed raw signal to be identified is decomposed to extract heartbeat signal segments. , Indicates the user's first Segment of signal; The value range is from 1 to N2, where N2 is the number of signal segments; By analyzing heartbeat signal segments Perform random masking and then encode the processed signal to obtain an intermediate feature vector. Based on the intermediate feature vector Decoding yields a reconstructed signal, which is then compared with a segment of the heartbeat signal. Consistent; The intermediate feature vector The input is fed into a Siamese network model with self-attention as its core, and contrastive features are obtained. The loss function of the twin network model for: in, Indicates the first One user, Indicates the first Segment of signal; The contrast features Dimensionality reduction is performed by inputting into a multilayer perceptron (MLP) to enhance the contrast features corresponding to different users. The differences between them yield the metric features. ; Measurement features corresponding to different users The distance between them is greater than or equal to a preset value m, and the measurement feature The feature vector of the user; the loss function of the multilayer perceptron (MLP) The expression is: ； Calculate the Euclidean distances from the metric features of the user to be identified to the centers of all clusters in the target identity database to obtain the identity features of the user to be identified. The identity features The input is fed into the multi-head self-attention module encoder, and the result is input into... Layers, to obtain identity characteristics The probability of being authenticated as the user corresponding to each cluster; When the maximum probability exceeds a preset threshold, the identity features will be... The identity information is identified as having the highest probability.

6. A progressive identity database construction device based on cardiac impact signals, characterized in that, include: The data acquisition and preprocessing module is used to collect cardiac impact signals from multiple users and store them as a pair of time and amplitude values ​​to obtain the raw signals. Preprocess the original signal to remove invalid signal segments; The signal decomposition module is used to decompose the preprocessed raw signal and extract heartbeat signal segments. , Indicates the first The first user's Segment of signal; The value ranges from 0 to N1, where N1 is the number of users. The value range is from 1 to N2, where N2 is the number of signal segments; The training module is used to perform the following: By analyzing heartbeat signal segments Perform random masking and then encode the processed signal to obtain an intermediate feature vector. Based on the intermediate feature vector Decoding yields a reconstructed signal, which is then compared with a segment of the heartbeat signal. Consistent; The intermediate feature vector The input is fed into a Siamese network model with self-attention as its core, and contrastive features are obtained. The loss function of the twin network model for: in, Indicates the first One user, Indicates the first Segment of signal; The contrast features Dimensionality reduction is performed by inputting into a multilayer perceptron (MLP) to enhance the contrast features corresponding to different users. The differences between them yield the metric features. ; Measurement features corresponding to different users The distance between them is greater than or equal to a preset value m, and the measurement feature The feature vector of the user; the loss function of the multilayer perceptron (MLP) The expression is: ； The identity database construction module is used to treat each user's feature vector set as a cluster, determine the center of each cluster, and construct the target identity database.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute and implement the progressive identity database construction method based on cardiac impulse signals as described in any one of claims 1-4, or the progressive identity recognition method based on cardiac impulse signals as described in claim 5.

Images