Heartbeat interval estimation method, apparatus, device, medium and program product

Abstract

The invention provides a heartbeat interval estimation method and device, equipment, a medium and a program product, which can be applied to the field of vital sign monitoring, and the heartbeat interval estimation method comprises the following steps: obtaining a heartbeat signal; identifying a steady-state section signal in the heartbeat signal; extracting a time interval between adjacent heartbeats in the steady-state section signal to obtain a steady-state heart beat interval sequence; and reconstructing the steady-state heart beat interval sequence into a complete heart beat interval sequence through a pre-trained machine learning model. The accuracy of heartbeat interval estimation can be improved.

Description

Technical Field

[0001] The present disclosure relates to the field of vital sign monitoring, and more particularly, to a heartbeat interval estimation method, apparatus, device, medium, and program product. Background Art

[0002] Long-term, continuous vital sign monitoring is crucial for health risk assessment. Currently, a variety of sensors are available for vital sign monitoring, including contact sensors and vision-based, acoustic, and radio frequency sensing technologies. However, in practice, many sensors are not suitable for prolonged use. An ideal monitoring method should be comfortable, non-intrusive, well-adaptable to environmental changes, and able to operate continuously and stably while protecting user privacy. Radio frequency-based vital sign monitoring methods, by contrast, offer significant advantages in these areas and hold great promise for future applications.

[0003] Despite this, RF-based methods still face significant challenges, the most significant of which is motion interference. In daily life, users' daily actions such as adjusting their body position, drinking water, and eating will introduce motion artifacts, thereby interfering with the heartbeat information in the RF signal. Currently, there are two main solutions to motion artifacts. One method attempts to reconstruct the heartbeat signal from the data segments affected by motion, but the effect of directly reconstructing the heartbeat signal is limited, and it is difficult to accurately restore the true heartbeat rhythm; the other method directly discards the interfered data segments. Although this method avoids artifact interference, it leads to a large amount of data loss, causing the monitoring results to deviate from the actual physiological state. Summary of the Invention

[0004] In view of the above problems, the present disclosure provides a heartbeat interval estimation method, apparatus, device, medium and program product that can improve the accuracy of heartbeat interval estimation.

[0005] According to a first aspect of the present disclosure, a heartbeat interval estimation method is provided, comprising:

[0006] Get heartbeat signal;

[0007] Identify steady-state segment signals in heartbeat signals;

[0008] Extracting the time intervals between adjacent heartbeats in the steady-state segment signal to obtain a steady-state heartbeat interval sequence;

[0009] Through the pre-trained machine learning model, the steady-state cardiac interval sequence is reconstructed into a complete cardiac interval sequence.

[0010] According to an embodiment of the present disclosure, identifying a steady-state segment signal in a heartbeat signal includes:

[0011] Obtain the frequency domain signal of the heartbeat signal;

[0012] The motion index is obtained by summing the frequency domain signals in the frequency dimension;

[0013] The heartbeat signal corresponding to the concentrated interval of the motion index distribution is taken as the steady-state segment signal.

[0014] According to an embodiment of the present disclosure, the method further includes:

[0015] The part of the heartbeat signal that is different from the steady-state segment signal is used as the motion segment signal;

[0016] A mask is generated for the motion segment signal.

[0017] According to an embodiment of the present disclosure, obtaining a frequency domain signal of a heartbeat signal includes:

[0018] Perform short-time Fourier transform on the heartbeat signal to obtain the heartbeat Doppler spectrum as the frequency domain signal of the heartbeat signal.

[0019] According to an embodiment of the present disclosure, extracting the time interval between adjacent heartbeats in a steady-state segment signal includes:

[0020] The steady-state segment signal is input into the pre-trained neural network model based on the self-attention mechanism to obtain the recognized heartbeat segment;

[0021] splicing adjacent identified heartbeat segments to obtain a combined heartbeat segment;

[0022] With the goal of maximizing similarity, the heartbeat segments are translated and combined;

[0023] The time deviation corresponding to the translation of adjacent combined heartbeat segments is calculated to obtain the time interval between adjacent heartbeats.

[0024] According to an embodiment of the present disclosure, the steady-state segment signal is input into a pre-trained neural network model based on a self-attention mechanism, including:

[0025] Divide the steady-state segment signal into a plurality of overlapping segments of set length; the signals of adjacent overlapping segments partially overlap;

[0026] The overlapping segments are fed into a pre-trained neural network model based on the self-attention mechanism.

[0027] According to an embodiment of the present disclosure, with the goal of maximizing similarity, the combined heartbeat segments are translated, including: translating adjacent combined heartbeat segments and calculating the cosine similarity between the translated adjacent combined heartbeat segments; when the cosine similarity is maximized, completing the translation action.

[0028] According to an embodiment of the present disclosure, the machine learning model is trained based on a momentum learning comparison mechanism, and the training samples of the machine learning model are obtained by randomly adding a mask to a complete cardiac interval sequence.

[0029] Another aspect of the present disclosure provides a heartbeat interval estimation device, the device comprising:

[0030] Heartbeat signal acquisition module, used to obtain heartbeat signals;

[0031] A steady-state signal recognition module is used to identify steady-state segment signals in heartbeat signals;

[0032] a steady-state cardiac interval sequence acquisition module, configured to extract the time intervals between adjacent heartbeats in the steady-state segment signal to obtain a steady-state cardiac interval sequence; and

[0033] The reconstruction module is used to reconstruct the steady-state cardiac interval sequence into a complete cardiac interval sequence through a pre-trained machine learning model.

[0034] Another aspect of an embodiment of the present disclosure provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the above method.

[0035] Another aspect of an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the processor to perform the above method.

[0036] Another aspect of an embodiment of the present disclosure provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0037] One or more of the above embodiments have the following beneficial effects: The present disclosure provides a robust, non-contact, sustainable, and real-world scenario-applicable long-term cardiac activity monitoring solution. By identifying heartbeat information in the steady-state interval, extracting its inter-beat interval (IBI) features, and using this information to accurately reconstruct the missing IBI during motion interference, accurate estimation of the heartbeat interval is achieved, and the impact of motion artifacts on heartbeat information in the non-steady-state state can be effectively avoided. Compared with existing signal processing or deep learning methods based on peak detection and filtering, the present disclosure exhibits higher accuracy and robustness in both steady-state and motion (non-steady-state) intervals. At the same time, the present disclosure exhibits high stability and generalization capabilities under different individuals, environments, time spans, and different radar deployment conditions, demonstrating strong application potential in long-term, non-contact cardiac activity monitoring tasks, and promoting the practical application of radio frequency sensing technology in health monitoring, smart healthcare, and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The above contents and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0039] Figure 1 The flowchart of the heartbeat interval estimation method according to an embodiment of the present disclosure is schematically shown;

[0040] Figure 2 The following schematically shows the overall framework of the heartbeat interval estimation method according to an embodiment of the present disclosure;

[0041] Figure 3 Schematically shows a flow chart for identifying a steady-state section signal according to an embodiment of the present disclosure;

[0042] Figure 4 The figure schematically shows a processing diagram of identifying a steady-state segment signal according to an embodiment of the present disclosure;

[0043] Figure 5 The flowchart of the process of extracting the time interval between adjacent heartbeats in the steady-state segment signal according to an embodiment of the present disclosure is schematically shown;

[0044] Figure 6 The figure schematically shows a process diagram of extracting the time intervals between adjacent heartbeats in a steady-state segment signal according to an embodiment of the present disclosure;

[0045] Figure 7 The following schematically illustrates a training process diagram of a machine learning model according to an embodiment of the present disclosure;

[0046] Figure 8 Schematic diagram showing the effect of reconstructing tachycardia and bradycardia signals by the heartbeat interval estimation method according to an embodiment of the present disclosure, wherein (a) corresponds to a tachycardia signal, and (b) corresponds to a bradycardia signal;

[0047] Figure 9 The schematic diagram shows the effect of reconstructing the IBI sequence in a steady-state scenario according to the heartbeat interval estimation method of an embodiment of the present disclosure, wherein (a) corresponds to the true value of the IBI sequence, and (b) corresponds to the estimated value of the IBI sequence;

[0048] Figure 10 Schematic diagram showing the effect of reconstructing an IBI sequence in the presence of motion using a heartbeat interval estimation method according to an embodiment of the present disclosure; wherein (a) corresponds to the true value of the IBI sequence, (b) corresponds to the estimated value of the IBI sequence in the steady-state interval, and (c) corresponds to the reconstructed value of the complete IBI sequence;

[0049] Figure 11 The following schematically shows a structural block diagram of a heartbeat interval estimation device according to an embodiment of the present disclosure;

[0050] Figure 12 The figure schematically shows a block diagram of an electronic device suitable for implementing a heartbeat interval estimation method according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0051] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the detailed description below, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0052] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise," "include," etc. used herein indicate the presence of the features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0053] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0054] When expressions such as "at least one of A, B, and C, etc." are used, they should generally be interpreted in accordance with the meaning commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0055] Figure 1 The figure schematically shows a flow chart of a heartbeat interval estimation method according to an embodiment of the present disclosure.

[0056] like Figure 1 As shown, this embodiment includes operations S110 to S140:

[0057] Operation S110: Acquire a heartbeat signal.

[0058] Operation S120: Identify a steady-state segment signal in the heartbeat signal.

[0059] Operation S130: extract the time intervals between adjacent heartbeats in the steady-state segment signal to obtain a steady-state heartbeat interval sequence.

[0060] Operation S140: Reconstruct the steady-state cardiac interval sequence into a complete cardiac interval sequence through a pre-trained machine learning model.

[0061] In operation S110, for example, the heartbeat signal Obtained by preprocessing the radar echo signal. Figure 2 , a radar device transmits a signal to the human body and receives an echo signal, and pre-processes the received echo signal to obtain a heartbeat signal that can be used to analyze the heartbeat.

[0062] For example, radar equipment uses Frequency Modulated Continuous Wave Radar (FMCW) radar and combines it with Multiple Input Multiple Output (MIMO) array technology. A 12-channel virtual antenna array is constructed based on a two-dimensional MIMO array, and 3D beamforming is used to achieve high-resolution spatial perception. The formula for 3D beamforming is:

[0063] ;

[0064] in, The output of three-dimensional beamforming, which represents the radar system's response to a certain position The spatial response intensity of , where r represents the distance, represents the pitch angle, represents the direction angle, is the distance-oriented vector, is the Kronecker product of the transmit and receive steering vectors, used for spatial pointing enhancement, is the intermediate frequency signal matrix.

[0065] Then along The last two dimensions are summed to obtain the distance profile, and then the constant false alarm rate (CFAR) is used to detect the distance of the human target in the distance profile. The pitch and azimuth angle space are statistically analyzed based on the autocorrelation function. The autocorrelation function can help identify the periodic components in the signal, which facilitates the analysis of the heartbeat signal. The pitch and azimuth perspectives with the strongest periodicity are selected as the best observation perspectives for subsequent heartbeat signal extraction. With the determined best observation perspective as the center, a three-dimensional receptive field matching the human chest structure is constructed, and phase sequences are extracted from different perspectives to form a time series matrix. Finally, the time series matrix is divided into segments according to a fixed time window, and the signal perspective with the smallest variance in each time period is selected to help remove noise and irrelevant motion effects, thereby obtaining a purer heartbeat signal. The selected signals are spliced along the time dimension to form the final heartbeat signal for analysis. .

[0066] In operation S120, the obtained heartbeat signal is divided to distinguish between steady-state and motion intervals, and the steady-state segment signal and the motion segment signal are separated, so as to obtain steady-state heart beat interval data based on the steady-state segment signal, avoiding the influence of interference signals in the motion segment on the heart beat interval calculation.

[0067] For example, by performing a frequency domain transformation on the heartbeat signal, combining frequency domain analysis with statistical modeling, the signal can be divided into steady-state and motion intervals based on spectral energy distribution and histogram strategies. Furthermore, this division can be achieved using bandpass filtering, wavelet transform, and other methods. Figure 3 Schematically shows a flow chart for identifying a steady-state section signal according to an embodiment of the present disclosure, Figure 4 Schematic diagram of the process of identifying steady-state segment signals according to an embodiment of the present disclosure, see Figure 3 、 Figure 4 In an embodiment of the present disclosure, identifying a steady-state segment signal in a heartbeat signal includes:

[0068] In operation S310, a frequency domain signal of the heartbeat signal is acquired. For example, in the disclosed embodiment, acquiring the frequency domain signal of the heartbeat signal is accomplished by performing a short-time Fourier transform (STFT) on the heartbeat signal, thereby obtaining a heartbeat Doppler spectrogram. The STFT provides resolution in both time and frequency, making it particularly suitable for processing heartbeat signals containing periodic components. Furthermore, in other embodiments, methods such as wavelet transforms may be employed to acquire the frequency domain signal.

[0069] In operation S320, the frequency domain signal is accumulated in the frequency dimension to obtain the motion index. The motion index indicates the individual's motion intensity. The larger the motion index, the more intense the individual's motion. The motion index can clearly distinguish the steady-state interval from the motion interval, so as to better extract information from the steady-state interval and thus better reconstruct the IBI information of the motion interval. The specific operation is to accumulate the heartbeat Doppler spectrogram in the frequency dimension to obtain the motion index. , the formula is:

[0070] ;

[0071] ;

[0072] in, represents the short-time Fourier transform spectrum of the heartbeat signal, The n in the The frequency index in , N represents the total number of frequency points of the short-time Fourier transform, Represents short-time Fourier transform.

[0073] The advantage of using the motion index to distinguish between steady-state and moving intervals is that slight body movements, posture adjustments, and even changes in breathing patterns can cause changes in the spectral characteristics of the Doppler spectrum. By accumulating data in the frequency dimension, this method can sensitively capture these subtle perturbations, which may not be sufficiently visible to be effectively detected by other time-domain statistical methods or energy thresholding methods. This makes the "motion index" more advantageous than other methods in distinguishing between "absolute stillness" and "minor perturbations."

[0074] In operation S330, the signal corresponding to the concentrated interval of the motion index distribution is used as the steady-state segment signal. The motion index is used to measure the intensity of motion disturbances in a time segment. By analyzing the distribution of the motion index, the most stable heartbeat frequency band is determined. In this embodiment, a motion index histogram is obtained by statistically analyzing the motion index. The motion index histogram is a frequency statistic of a specific interval of the motion index and can be used to more accurately distinguish between steady-state and motion intervals. The interval containing the main peak in the motion index histogram is used as the most stable heartbeat frequency band, and the corresponding heartbeat signal is used as the steady-state segment signal.

[0075] In some embodiments, after completing the steady-state segment signal identification, the method further includes:

[0076] In operation S340 , a portion of the heartbeat signal that is different from the steady-state segment signal is used as a motion segment signal, thereby dividing the heartbeat signal into a steady-state segment signal and a motion segment signal according to the motion index histogram.

[0077] In operation S350, a mask is generated for the motion segment signal. Based on the steady-state segment signal and the motion segment signal divided in operation S240, the steady-state segment signal is marked as 1 and the motion segment signal is marked as 0, and a mask is generated for the motion segment signal. In subsequent processing, only the steady-state signal segment is processed and analyzed to avoid the influence of the motion segment signal. The process of generating the mask can be expressed as follows:

[0078] ;

[0079] in, represents a mask sequence, , represents the lower threshold boundary of the histogram, represents the upper threshold boundary of the histogram, A generator of bounding indices representing the highest peaks of the histogram, Represents the operation of generating a histogram.

[0080] After obtaining the steady-state segment signal in operation S130, the inter-beat interval (IBI) sequence corresponding to the steady-state segment signal needs to be accurately estimated. The IBI estimation process involves identifying each heartbeat segment in the steady-state segment signal and calculating the time interval between adjacent heartbeats, that is, the time difference between corresponding points in adjacent heartbeat segments.

[0081] Figure 5 The flowchart of the process of extracting the time interval between adjacent heartbeats in the steady-state segment signal according to an embodiment of the present disclosure is schematically shown. Figure 6 The schematic diagram of the process of extracting the time interval between adjacent heartbeats in the steady-state segment signal according to an embodiment of the present disclosure is shown. Figure 5 、 Figure 6 In the embodiment of the present disclosure, a neural network model is used to identify heartbeat segments and a correlation-based pattern matching method is used to calculate the time interval between two adjacent heartbeats. In the embodiment of the present disclosure, the time interval between adjacent heartbeats in the steady-state segment signal is extracted, including:

[0082] In operation S510, the steady-state segment signal is input into a pre-trained neural network model based on the self-attention mechanism to obtain an identified heartbeat segment. The self-attention mechanism can capture global dependencies within the entire signal sequence, so that it can still recognize slightly changed or distorted heartbeat signals, thereby improving the robustness of the algorithm to different users and physiological states. Exemplarily, in this embodiment, the neural network model adopts a Transformer model based on a multi-head self-attention mechanism. Different attention heads in the multi-head self-attention mechanism can focus on different components in the signal respectively. This mechanism can automatically separate interference and effective information during the feature extraction process, thereby improving the discrimination accuracy. In some embodiments, other models based on the self-attention mechanism, such as BERT, can also be used for heartbeat segment recognition.

[0083] In some embodiments, the steady-state segment signal is further divided into multiple overlapping segments of a predetermined length. These overlapping segments are then fed into a pre-trained neural network model based on a self-attention mechanism for heartbeat segment recognition. The signals between adjacent overlapping segments partially overlap, allowing the neural network to fully extract features for recognition and facilitate subsequent splicing to produce a combined heartbeat segment.

[0084] The following is an example of the Transformer model. First, it is divided into overlapping segments with a segment length of 400ms and a step size of 100ms. , that is, there is a 300ms signal overlap between adjacent overlapping segments, where Represents the number of overlapping segments. This is mapped into an embedding vector through a multi-layer perceptron network, and a learnable positional encoding of the same dimension and size as the embedding vector is added. This token is then fed into the Transformer encoder. The Transformer encoder employs a multi-head self-attention mechanism, which captures global dependencies across the entire signal sequence. This allows it to recognize even slightly altered or distorted heartbeat signals, improving the algorithm's robustness to different users and physiological states. Furthermore, different attention heads can focus on different components of the signal. This mechanism automatically separates interference from valid information during feature extraction, improving discrimination accuracy.

[0085] In operation S520, adjacent identified heartbeat segments are spliced together to obtain a combined heartbeat segment set. ,in represents the number of combined heartbeat segments. In the embodiment of the present disclosure, the length of each combined heartbeat segment is fixed at 600 ms, and the maximum similarity offset calculation is performed on each combined heartbeat segment signal.

[0086] In operation S530, the combined heartbeat segments are shifted with the goal of maximizing similarity. Time alignment and interval calculation are performed on the identified heartbeat segments. Specifically, adjacent detected heartbeat segments are concatenated into a new sequence set (combined heartbeat segments). A maximum similarity offset calculation is then performed on the signals of each combined heartbeat segment. The cosine similarity between the shifted adjacent combined heartbeat segments is calculated. When the cosine similarity is maximized, the shifting operation is completed, thereby maximizing the cosine similarity of the signals between the segments.

[0087] In operation S540, the time deviation corresponding to the translation of adjacent combined heartbeat segments is calculated to obtain the time interval between adjacent heartbeats. For ease of calculation, the time interval between adjacent combined heartbeat segments is calculated, for example, using the time difference between the center points of adjacent combined heartbeat segments. After completing the offset calculated based on the index corresponding to the maximum similarity of the signal, the time difference between the center points of each adjacent heartbeat segment is finally calculated to obtain the time interval between two adjacent heartbeats. This achieves steady-state IBI extraction without the need for peak values or specific templates, significantly improving the accuracy and reliability of non-contact heart rate measurement. It is particularly suitable for monitoring environments in real life where physiological states frequently change and exercise is frequent.

[0088] In operation S140 , the missing interbeat interval (IBI) sequence is reconstructed under the influence of motion interference, addressing the problem of discontinuities in steady-state IBI data caused by motion artifacts. The core concept is to learn the underlying variation pattern from the valid IBI sequence obtained during the steady-state interval. Using a contrastive learning mechanism, the model learns the dynamic distribution of IBI under the condition of mask information, ultimately reconstructing the complete IBI sequence.

[0089] Specifically, this embodiment adopts a machine learning model to learn the dynamic distribution law of the IBI sequence, adopts an encoder-decoder structure to reconstruct the IBI sequence, and introduces a momentum learning comparison mechanism. By adding a momentum encoder to work in conjunction with the original main encoder, feature stability is ensured through momentum updates.

[0090] Figure 7 A schematic diagram of the training process of the machine learning model according to an embodiment of the present disclosure is shown schematically, wherein the pre-trained machine learning model is trained based on the momentum learning comparison mechanism, and the training samples of the machine learning model are obtained by randomly adding masks to the complete cardiac interval sequence.

[0091] First, data from different signal modalities, such as ECG (electrocardiogram), SCG (seismocardiogram), and PPG (photoplethysmogram), are uniformly processed into a standard IBI time series format. For each modality, the interbeat interval (i.e., the time difference between two consecutive heartbeats) is calculated by detecting R waves or other equivalent feature points. This constructs a cross-modality consistent IBI training sample, resulting in an IBI sequence library for model training and testing.

[0092] In order to enable the model to process mask data, this embodiment introduces a dynamic mask generation function , for IBI sequence Perform mask processing, where R is the real number field and L is the length of the IBI sequence. For any complete IBI sequence , randomly generate two versions of mask views , the generation process can be expressed as:

[0093] ;

[0094] Masks are generated using a geometric distribution control mechanism to simulate the temporal structure of occlusion in real scenes. Specifically, the average segment length and mask ratio of each mask are controlled. The average segment length varies between 10 and 40, and the mask ratio varies between 0.2 and 0.8. Furthermore, masks are generated randomly during the training phase, while masks during the testing phase are generated using operations S310 to S350. In principle, these two sets of masks are disjoint.

[0095] In order to enhance the adaptability of the model to different mask structures and improve its robust modeling ability for IBI temporal features, a momentum contrast learning mechanism is introduced. This mechanism constructs two sets of encoder networks with the same structure but different parameter update methods: encoder The parameters are updated by backpropagation, while the momentum encoder It does not participate in back propagation, and its parameters are inherited from the main encoder through sliding average. The update rules are as follows:

[0096] ;

[0097] in, and are the parameters of the encoder and momentum encoder respectively, is the momentum coefficient. The encoder outputs the latent vector , represents the temporal representation of the same IBI sequence under different mask conditions. The goal is to maximize the similarity between different mask views of the same original IBI in the embedding space, while maintaining differentiation from mask views of other sequences.

[0098] The loss function of a sample with index i in the dataset The standard InfoNCE contrast loss form is used:

[0099] ;

[0100] in, Represents the normalized vector and The dot product similarity between them; is the temperature parameter; is one of the mask views of the i-th sample Through the encoder The potential vector obtained, the corresponding positive sample is , represents another mask view of the i-th sample Through the momentum encoder The resulting latent vector; represents the i-th latent vector The corresponding feature of the j-th negative sample; K represents the number of negative samples provided for each query.

[0101] Since contrastive learning works better when a large number of negative samples are introduced, this embodiment introduces a feature queue that is much larger than the batch size and controls its size as an independent hyperparameter. , only keep the main encoder The main encoder and decoder Together they form a complete encoder-decoder architecture, the reconstruction loss of the IBI sequence It can be expressed as:

[0102] ;

[0103] in, Represents the reconstructed IBI sequence results, is the dataset size.

[0104] Finally, the loss function The definition is as follows:

[0105] ;

[0106] in, is a hyperparameter used to balance reconstruction error and contrastive learning loss, is the dataset size, It means calculating the contrast loss of each latent vector and summing them up.

[0107] After training the machine learning model, its reconstruction of the IBI sequence was verified. The IBI estimation results with motion artifact signals were compared with the true value results to measure the IBI reconstruction effect.

[0108] For example, the hardware platform of this embodiment consists of a millimeter-wave radar module and a data transmission module. The millimeter-wave radar module operates in the 60–64 GHz frequency range, has a maximum bandwidth of 4 GHz, and is equipped with three transmitting antennas and four receiving antennas. To achieve efficient data transmission, an ESP32-based module is designed as an intermediate communication unit, responsible for receiving data from the RF front-end module and transmitting it to the main control computer. In addition, to obtain a high-precision electrocardiogram as a true reference, the system is simultaneously connected to an electrocardiogram acquisition board.

[0109] In the IBI estimation process, Adam is used for optimization with a batch size of 1024 and a learning rate of 0.001.

[0110] During the IBI mask reconstruction process, the encoder contains five convolutional layers with feature channels of 64, 128, 256, 512, and 1024, respectively. Each layer consists of two layers: a 1D convolution followed by a ReLU activation function and a batch normalization layer. The decoder structure is symmetrical to the encoder, but uses deconvolution instead of standard convolution. Adam optimization is also used with a batch size of 256 and a learning rate of 0.03.

[0111] For the determination of abnormal heart rhythm under the condition of motion artifact interference, Figure 8 The diagram schematically shows the effect of reconstructing tachycardia and bradycardia signals by the heartbeat interval estimation method according to an embodiment of the present disclosure, where (a) corresponds to a tachycardia signal and (b) corresponds to a bradycardia signal, and the data source is a public dataset.

[0112] First, the IBI sequence affected by motion is masked (as shown by the blue scattered dots in the figure, "IBI mask sequence"). Then, the pre-trained machine learning model obtained in operation S140 of this embodiment is used to reconstruct the masked segment. The model adopts a one-dimensional encoder-decoder structure combined with a contrastive learning mechanism. The reconstruction process is completed by the following steps:

[0113] (1) Temporally mask the original IBI sequence to form training samples with missing regions;

[0114] (2) The encoder extracts the latent feature representation of the IBI mask sequence;

[0115] (3) The decoder reconstructs the original IBI curve and generates a complete estimation result;

[0116] (4) Calculate the average IBI value of the reconstructed sequence and determine the rhythm type.

[0117] See also Figure 8 In (a), the average IBI of the original mask sequence is 603ms, which is incorrectly judged as "normal". However, the average IBI after restoration by the mask reconstruction method of this embodiment is 589ms, which is closer to the true value (as shown by the black dotted line) and is successfully identified as "tachycardia". Figure 8 In (b), the average value of the original mask sequence is 998ms, which may lead to misjudgment as "normal"; while the average IBI after reconstruction is 1001ms, accurately reflecting the true physiological state of "bradycardia".

[0118] It can be seen that this embodiment can not only effectively restore the overall fluctuation trend of the IBI curve, but also assist in realizing heart rhythm classification and abnormality identification in medical applications, and has broad practical value in remote health monitoring and smart medical equipment.

[0119] For the IBI estimation effect of steady-state scenarios such as typing, Figure 9 The figure schematically shows the effect of reconstructing an IBI sequence in a steady-state scenario according to the heartbeat interval estimation method of an embodiment of the present disclosure. Figure 9 The figure shows a heartbeat signal collected by a millimeter-wave radar from a subject in a static state, and a comparison between the IBI sequence estimated by the method provided by this embodiment and the true value. (a) is the true value of the IBI sequence, calculated from the RR interval of a synchronously collected electrocardiogram, and serves as an evaluation reference standard. (b) is the IBI estimation result of the heartbeat interval estimation method of this embodiment for heartbeat data in the same time period. The comparison shows that the estimated IBI value (the red line in (b)) is highly consistent with the true IBI value (the black dotted line in (a)) in terms of trend change, oscillation amplitude, and overall rhythm. This verifies the estimation accuracy and time series continuity of the method of this embodiment within the steady-state range.

[0120] The estimation process includes the following steps:

[0121] (1) Signal preprocessing: First, convert the original radar signal into a heartbeat signal;

[0122] (2) Heartbeat pattern recognition: The preprocessed heartbeat signal is divided into several overlapping segments and embedded into the feature space; through the multi-head self-attention mechanism, the model automatically identifies the valid segments containing heartbeat activities.

[0123] (3) Heartbeat pattern matching and alignment: For segments identified as containing heartbeats, time alignment and similarity matching methods are used to derive the final IBI sequence by calculating the time difference between the centers of adjacent segments.

[0124] This method, unlike traditional approaches that rely on peak detection, is particularly suitable for scenarios where radar signals lack stable waveform markers. It demonstrates excellent generalization and application value. Experiments have shown that this method can maintain an IBI estimation error within 10ms within the steady-state range, sufficient for clinical and health monitoring needs.

[0125] For the IBI estimation effect of large body movements such as adjusting posture, Figure 10 A schematic diagram illustrates the effect of reconstructing an IBI sequence in the presence of motion using the heartbeat interval estimation method according to an embodiment of the present disclosure. (a) corresponds to the true value of the IBI sequence, which is the subject's actual IBI sequence derived from the RR interval extracted from the electrocardiogram and serves as an evaluation benchmark. The red curve in (b) corresponds to the estimated value of the IBI sequence in the steady-state interval. The red curve in (c) corresponds to the reconstructed value of the complete IBI sequence, i.e., the complete IBI sequence obtained after masking and reconstructing the motion-affected area.

[0126] The estimation process includes the following complete steps: (1) signal preprocessing; (2) motion detection to generate mask; (3) IBI estimation; (4) IBI mask reconstruction.

[0127] As can be seen, even when a large number of IBI segments are missing due to strenuous exercise such as adjusting posture, the method provided in this embodiment can still effectively reconstruct the missing heart rhythm and ensure the continuity and accuracy of the IBI timing, thereby providing a stable and reliable data foundation for subsequent heart rhythm analysis, HRV calculation, and health monitoring. It has good robustness, wide adaptability, and practical value, and is suitable for heart rate variability assessment tasks in real-world scenarios such as remote health monitoring, home medical care, and sleep monitoring.

[0128] Based on the above-mentioned heartbeat interval estimation method, the present disclosure also provides a heartbeat interval estimation device. Figure 11 The device is described in detail.

[0129] Figure 11 The figure schematically shows a structural block diagram of a heartbeat interval estimation device according to an embodiment of the present disclosure.

[0130] like Figure 11 As shown, the heartbeat interval estimation device 1100 of this embodiment includes a heartbeat signal acquisition module 1110 , a steady-state signal recognition module 1120 , a steady-state heartbeat interval sequence acquisition module 1130 and a reconstruction module 1140 .

[0131] The heartbeat signal acquisition module 1110 is used to acquire a heartbeat signal. In one embodiment, the heartbeat signal acquisition module 1110 can be used to perform the operation S110 described above, which will not be described in detail here.

[0132] The steady-state signal identification module 1120 is used to identify the steady-state segment signal in the heartbeat signal. In one embodiment, the steady-state signal identification module 1120 can be used to perform the operation S120 described above, which will not be repeated here.

[0133] The steady-state cardiac interval sequence acquisition module 1130 is used to extract the time intervals between adjacent heartbeats in the steady-state segment signal to obtain a steady-state cardiac interval sequence. In one embodiment, the steady-state cardiac interval sequence acquisition module 1130 can be used to perform the operation S130 described above, which will not be repeated here.

[0134] The reconstruction module 1140 is used to reconstruct the steady-state cardiac interval sequence into a complete cardiac interval sequence through a pre-trained machine learning model. In one embodiment, the reconstruction module 1140 can be used to perform the operation S140 described above, which will not be repeated here.

[0135] In some embodiments, the steady-state signal identification module 1120 is specifically used to: obtain the frequency domain signal of the heartbeat signal; accumulate the frequency domain signal in the frequency dimension to obtain a motion index histogram; and use the signal corresponding to the interval where the main peak in the motion index histogram is located as the steady-state segment signal.

[0136] In some embodiments, the steady-state signal identification module 1120 is further configured to: use the signal corresponding to the interval where the non-main peak in the motion index histogram is located as the motion segment signal; and generate a mask for the motion segment signal.

[0137] In some embodiments, the steady-state cardiac interval sequence acquisition module 1130 is specifically used to: input the steady-state segment signal into a pre-trained neural network model based on the self-attention mechanism to obtain an identified heartbeat segment; splice adjacent identified heartbeat segments to obtain a combined heartbeat segment; shift the combined heartbeat segment with the goal of maximizing similarity; calculate the time deviation corresponding to the shift amount of adjacent combined heartbeat segments to obtain the time interval between adjacent heartbeats.

[0138] In some embodiments, the steady-state cardiac interval sequence acquisition module 1130 is specifically configured to: translate adjacent combined heartbeat segments and calculate the cosine similarity between the translated adjacent combined heartbeat segments; and complete the translation when the cosine similarity is the largest.

[0139] In some embodiments, in the reconstruction module 1140, the machine learning model is trained based on a momentum learning comparison mechanism, and the training samples of the machine learning model are obtained by randomly adding a mask to the complete cardiac interval sequence.

[0140] For the parts not mentioned in the apparatus part, they can be understood with reference to the various embodiments of the above-mentioned method. That is, the apparatus part includes modules for executing the various steps of any one of the method embodiments described above. In addition, the implementation methods, technical problems solved, functions achieved, and technical effects achieved of each module / unit / subunit, etc. in the apparatus part embodiment are respectively the same or similar to the implementation methods, technical problems solved, functions achieved, and technical effects achieved of each corresponding step in the method part embodiment, and will not be repeated here.

[0141] According to an embodiment of the present disclosure, any multiple modules among the heartbeat signal acquisition module 1110, the steady-state signal identification module 1120, the steady-state cardiac interval sequence acquisition module 1130, and the reconstruction module 1140 can be combined into a single module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in a single module.

[0142] According to an embodiment of the present disclosure, at least one of the heartbeat signal acquisition module 1110, the steady-state signal identification module 1120, the steady-state cardiac interval sequence acquisition module 1130, and the reconstruction module 1140 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented by hardware or firmware through any other reasonable means of integrating or packaging circuits, or implemented in any one of the three implementation methods of software, hardware, and firmware, or in an appropriate combination of any of them. Alternatively, at least one of the heartbeat signal acquisition module 1110, the steady-state signal identification module 1120, the steady-state cardiac interval sequence acquisition module 1130, and the reconstruction module 1140 can be at least partially implemented as a computer program module, which can perform the corresponding function when executed.

[0143] Figure 12 A block diagram of an electronic device suitable for implementing a heartbeat interval estimation method according to an embodiment of the present disclosure is schematically shown.

[0144] like Figure 12As shown, the electronic device 1200 according to an embodiment of the present disclosure includes a processor 1201, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1202 or a program loaded from a storage portion 1208 into a random access memory (RAM) 1203. The processor 1201 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1201 may also include onboard memory for caching purposes. The processor 1201 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0145] Various programs and data required for the operation of the electronic device 1200 are stored in the RAM 1203. The processor 1201, the ROM 1202, and the RAM 1203 are connected to each other via a bus 1204. The processor 1201 performs various operations of the method flow according to the embodiment of the present disclosure by executing the programs in the ROM 1202 and / or the RAM 1203. It should be noted that the programs may also be stored in one or more memories other than the ROM 1202 and the RAM 1203. The processor 1201 may also perform various operations of the method flow according to the embodiment of the present disclosure by executing the programs stored in the one or more memories.

[0146] According to an embodiment of the present disclosure, electronic device 1200 may further include an input / output (I / O) interface 1205, which is also connected to bus 1204. Electronic device 1200 may also include one or more of the following components connected to I / O interface 1205: an input section 1206 including a keyboard, mouse, etc.; an output section 1207 including devices such as a cathode ray tube (CRT), liquid crystal display (LCD), and speakers; a storage section 1208 including a hard disk; and a communication section 1209 including a network interface card such as a LAN card or modem. Communication section 1209 performs communication processing via a network such as the Internet. A drive 1210 is also connected to I / O interface 1205 as needed. Removable media 1211, such as a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed in drive 1210 as needed, so that computer programs read from the removable media can be installed into storage section 1208 as needed.

[0147] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments, or may exist independently and not be incorporated into the device / apparatus / system. The computer-readable storage medium carries one or more programs, and when executed, implements the method according to the embodiments of the present disclosure.

[0148] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, and may include, for example, but not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present disclosure, a computer-readable storage medium may include the ROM 1202 and / or RAM 1203 described above, and / or one or more memories other than ROM 1202 and RAM 1203.

[0149] The embodiments of the present disclosure also include a computer program product, which includes a computer program containing program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the method provided by the embodiments of the present disclosure.

[0150] The computer program executes the above functions defined in the system / device of the embodiment of the present disclosure when the processor 1201 executes the computer program. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0151] In one embodiment, the computer program may be stored on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may be transmitted and distributed in the form of a signal on a network medium, downloaded and installed via the communication portion 1209, and / or installed from the removable medium 1211. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0152] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 1209 and / or installed from the removable medium 1211. When the computer program is executed by the processor 1201, the above-described functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the systems, devices, means, modules, units, etc. described above can be implemented by computer program modules.

[0153] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages include, but are not limited to, languages such as Java, C++, Python, "C" or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect via the Internet).

[0154] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0155] Those skilled in the art will appreciate that the features described in the various embodiments of the present disclosure may be combined and / or coupled in various ways, even if such combinations or couplings are not explicitly described in the present disclosure. In particular, the features described in the various embodiments of the present disclosure may be combined and / or coupled in various ways without departing from the spirit and teachings of the present disclosure. All such combinations and / or couplings fall within the scope of the present disclosure.

[0156] The embodiments of the present disclosure are described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although each embodiment has been described separately above, this does not mean that the measures in each embodiment cannot be used in combination to advantage. The scope of the present disclosure is defined by the appended claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A heartbeat interval estimation method, characterized in that: The method comprises: Get heartbeat signal; identifying a steady-state segment signal in the heartbeat signal; Extracting the time intervals between adjacent heartbeats in the steady-state segment signal to obtain a steady-state heartbeat interval sequence; The steady-state cardiac interval sequence is reconstructed into a complete cardiac interval sequence through a pre-trained machine learning model.

2. The method according to claim 1, characterized in that The identifying of the steady-state segment signal in the heartbeat signal comprises: Acquire a frequency domain signal of the heartbeat signal; Accumulating the frequency domain signal in the frequency dimension to obtain a motion index histogram; The signal corresponding to the interval where the main peak in the motion index histogram is located is used as the steady-state segment signal.

3. The method according to claim 2, characterized in that Also includes: The signal corresponding to the interval where the non-main peak in the motion index histogram is located is used as the motion segment signal; A mask is generated for the motion segment signal.

4. The method according to claim 2, characterized in that The acquiring of the frequency domain signal of the heartbeat signal comprises: Performing short-time Fourier transform on the heartbeat signal to obtain a heartbeat Doppler spectrogram as a frequency domain signal of the heartbeat signal.

5. The method according to claim 1, wherein The extracting the time interval between adjacent heartbeats in the steady-state segment signal includes: Inputting the steady-state segment signal into a pre-trained neural network model based on a self-attention mechanism to obtain a recognized heartbeat segment; splicing adjacent identified heartbeat segments to obtain a combined heartbeat segment; translating the combined heartbeat segment with the goal of maximizing similarity; The time deviation corresponding to the translation of adjacent combined heartbeat segments is calculated to obtain the time interval between adjacent heartbeats.

6. The method according to claim 5, characterized in that The step of inputting the steady-state segment signal into a pre-trained neural network model based on a self-attention mechanism comprises: Dividing the steady-state segment signal into a plurality of overlapping segments of set length; wherein the signals of adjacent overlapping segments partially overlap; The overlapping segments are input into a pre-trained neural network model based on the self-attention mechanism.

7. The method according to claim 5, characterized in that The translating of the combined heartbeat segments with the goal of maximizing similarity includes: translating adjacent combined heartbeat segments and calculating cosine similarities between the translated adjacent combined heartbeat segments; and completing the translation when the cosine similarity is maximized.

8. The method according to claim 1, characterized in that The machine learning model is trained based on a momentum learning comparison mechanism, and the training samples of the machine learning model are obtained by randomly adding a mask to a complete cardiac interval sequence.

9. A heartbeat interval estimation device, characterized in that: The device comprises: Heartbeat signal acquisition module, used to obtain heartbeat signals; A steady-state signal recognition module, configured to recognize a steady-state segment signal in the heartbeat signal; a steady-state cardiac interval sequence acquisition module, configured to extract the time intervals between adjacent heartbeats in the steady-state segment signal to obtain a steady-state cardiac interval sequence; and A reconstruction module is used to reconstruct the steady-state cardiac interval sequence into a complete cardiac interval sequence through a pre-trained machine learning model.

10. An electronic device comprising: one or more processors; a memory for storing one or more computer programs, It is characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

12. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images