Ballistocardiogram signal reconstruction method and system based on improved TimeGAN

Abstract

The invention relates to a ballistocardiogram signal reconstruction method and system based on an improved TimeGAN. The method comprises the following steps: acquiring a ballistocardiogram signal BCG; preprocessing the ballistocardiogram signal BCG to obtain a preprocessed ballistocardiogram signal BCG; inputting the preprocessed ballistocardiogram signal BCG into a pre-trained improved TimeGAN model, and outputting a final new ballistocardiogram signal BCG; the improved TimeGAN model comprises an embedded network used for mapping a signal input into the improved TimeGAN model to a low-order feature space to obtain hidden features; the recovery network is used for generating a de-noised signal; the generation network is used for generating a new ballistocardiogram signal; the discrimination network is used for carrying out signal authenticity discrimination according to the de-noised signal and the new ballistocardiogram signal to obtain a discrimination result; and the output layer is used for generating a final new ballistocardiogram signal BCG according to the judgment result, the de-noised signal and the new ballistocardiogram signal.

Description

Technical Field

[0001] The present invention relates to the technical field of medical signal processing, and in particular to a method and system for reconstructing a ballistocardial signal based on an improved TimeGAN. Background Art

[0002] Ballistocardiography (BCG) is a non-invasive physiological signal that reflects the subtle mechanical vibrations of the human body caused by heartbeats. Because BCG signals can be measured through non-contact devices such as mattresses and chairs, they have broad application prospects in noninvasive cardiovascular monitoring. However, BCG signals are susceptible to interference from factors such as respiratory motion, body movement, and environmental noise, which can degrade signal quality and affect the subsequent extraction and analysis of cardiovascular parameters. Therefore, improving BCG signal quality, removing noise, and reconstructing high-quality BCG signals have become important research areas.

[0003] However, existing traditional methods are prone to losing effective features during the denoising process, affecting the quality of signal reconstruction. Although deep learning methods can automatically extract features, they still have problems such as insufficient signal pattern learning and limited authenticity of generated signals when generating high-quality BCG signals. Summary of the Invention

[0004] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a method and system for reconstructing a ballistocardial signal based on an improved TimeGAN.

[0005] In order to achieve the above objectives, the main technical solutions adopted by the present invention include:

[0006] In a first aspect, an embodiment of the present invention provides a method for reconstructing a ballistocardial signal based on an improved TimeGAN, the method comprising the following steps:

[0007] S1, obtaining a BCG (Ballistocorticoid) signal;

[0008] S2, preprocessing the ballistocardi signal BCG to obtain a preprocessed ballistocardi signal BCG;

[0009] S3, input the pre-processed cardiac ballistic signal BCG into the pre-trained improved TimeGAN model, and output the final new cardiac ballistic signal BCG;

[0010] The improved TimeGAN model includes:

[0011] The embedding network is used to map the input signal of the improved TimeGAN model to the low-level feature space to obtain the corresponding hidden features;

[0012] The recovery network is used to reconstruct the hidden features through the residual network and generate the corresponding denoised signal;

[0013] A generative network is used to generate a new cardiac signal based on hidden features and random noise;

[0014] The discriminant network is used to discriminate the authenticity of the signal based on the denoised signal and the new heartbeat signal to obtain a discrimination result;

[0015] The output layer is used to generate a final new cardiac ballistic signal BCG based on the denoised signal and the new cardiac ballistic signal according to the discrimination result of the discriminant network.

[0016] Preferably, the S2 specifically includes:

[0017] S21, using an adaptive bandpass filter to process the ballistocardiographic signal BCG to obtain a denoised ballistocardiographic signal BCG;

[0018] S22, using variational mode decomposition to extract different frequency components of the denoised BCG signal, removing false signals, and obtaining a decomposed signal;

[0019] S23 . Normalize the decomposed signal to obtain a pre-processed ballistocardiogram (BCG) signal.

[0020] Preferably, before S1, the method further includes:

[0021] S0. Using the training data set, the improved TimeGAN model is trained to obtain a trained improved TimeGAN model;

[0022] The training data set includes a plurality of ballistic heart signal samples, each of which includes: a preprocessed ballistic heart signal BCG for training, and a true state label corresponding to the preprocessed ballistic heart signal BCG.

[0023] Preferably, the loss function of the improved TimeGAN model is:

[0024] L total =αL adv +βL rec +βL sup +δL latent ;

[0025] Among them, L total To improve the loss function of the TimeGAN model;

[0026] L adv To counter the loss function;

[0027] L rec is the autoregressive loss function;

[0028] L sup is the supervision loss function;

[0029] L latent is the latent space loss function;

[0030] α is the first weight coefficient;

[0031] β is the second weight coefficient;

[0032] γ is the third weight coefficient;

[0033] δ is the fourth weight coefficient.

[0034] Preferably, the output layer generates a final new ballistocardiogram signal (BCG) based on the denoised signal and the new ballistocardiogram signal according to the discrimination result of the discriminant network, specifically including:

[0035] When the discrimination result is greater than the first preset threshold, a new ballistocardia signal BCG is generated based on the denoised signal and the new ballistocardia signal using formula (1);

[0036] The formula (1) is:

[0037]

[0038] X final The final new cardiac signal BCG;

[0039] is the denoised signal;

[0040] For new heart beat signal;

[0041] a1 is the weight parameter.

[0042] Preferably,

[0043] The weight parameter is calculated using formula (2);

[0044] The formula (2) is:

[0045]

[0046] S rec The discriminant network included in the discriminant result is used to distinguish the denoised signal The authenticity score, representing the denoised signal The degree of similarity to the real signal;

[0047] S gen The discriminant network included in the discriminant result is used to identify the new cardiac impulse signal A rating of authenticity, indicating a new cardiac impulse signal The degree of similarity to the real signal.

[0048] Preferably,

[0049] The S21 uses an adaptive bandpass filter to process the ballistic heart signal BCG, and the center frequency of the filter is f c Calculated according to formula (3);

[0050] The formula (3) is:

[0051]

[0052] Among them, f0 is the basic center frequency, which is used to provide the initial reference value of the filter;

[0053] c is the adjustment coefficient, which is used to control the response of the center frequency to the input signal;

[0054] N is the number of sampling points, that is, the total number of sampling points of the input ballistocardial signal BCG;

[0055] X i is the i-th sampling point of the ballistocardiogram (BCG) signal, which is used to calculate the average amplitude, thereby dynamically adjusting the center frequency of the bandpass filter to make it more adaptable to the physiological characteristics of different individuals.

[0056] Preferably,

[0057] The S21 processes the ballistic heart signal BCG using an adaptive bandpass filter, wherein the bandwidth BW of the filter is calculated according to formula (4);

[0058] The formula (4) is calculated as follows:

[0059] BW=BW0+d·(max(X)-min(X));

[0060] BW0 is the initial bandwidth, that is, the default bandwidth of the filter when it is not adaptively adjusted;

[0061] d is the bandwidth adjustment coefficient;

[0062] X represents the input ballistic cardioversion signal BCG, and max(X) and min(X) represent the maximum value and minimum value of the input ballistic cardioversion signal BCG, respectively.

[0063] Preferably,

[0064] The basic center frequency f0 is 2.5Hz;

[0065] The initial bandwidth BW0 is 5Hz;

[0066] The bandwidth adjustment coefficient d ranges from 0.2 to 0.5.

[0067] On the other hand, this embodiment provides a ballistocardiographic signal reconstruction system based on an improved TimeGAN, including:

[0068] at least one processor; and

[0069] At least one memory communicatively connected to the processor, wherein the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the method for reconstructing a cardiac ballistic signal based on the improved TimeGAN as described in the first aspect.

[0070] The beneficial effects of the present invention are:

[0071] The present invention discloses a method for reconstructing a cardiac ballistic signal based on an improved TimeGAN. By adopting an improved TimeGAN model, introducing an embedding network, a recovery network, a generation network, a discriminant network, and an output layer, and combining a residual network for signal reconstruction, the method can effectively remove noise from BCG signals, improve the authenticity and stability of the signals, and achieve the technical effect of improving the quality of cardiac ballistic signals and providing more accurate data support for subsequent cardiovascular health monitoring, compared with the existing technology.

[0072] In addition, the present invention uses an adaptive bandpass filter for denoising in the signal preprocessing stage, and combines it with variational mode decomposition (VMD) to remove false signals, and performs normalization at the same time. Compared with traditional signal preprocessing methods, it can more accurately extract the effective components of the BCG signal, reduce false signal interference, enhance signal availability, and achieve the effect of improving the quality of the reconstructed signal.

[0073] Furthermore, the present invention introduces a training dataset with real-state labels during the model training phase. Compared with traditional GAN training methods without label supervision, it can more accurately learn the true pattern of BCG signals and improve the physiological rationality of the generated signals, thereby enhancing the generalization ability of the model and making the reconstructed signals more consistent with real physiological characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 This is a flow chart of a method for reconstructing a ballistocardial signal based on an improved TimeGAN according to the present invention;

[0075] Figure 2 Flowchart of a method for reconstructing a ballistocardial signal based on an improved TimeGAN in an embodiment of the present invention. DETAILED DESCRIPTION

[0076] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation methods in conjunction with the accompanying drawings.

[0077] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. Instead, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0078] Example 1

[0079] See also Figure 1 This embodiment provides a method for reconstructing a ballistocardial signal based on an improved TimeGAN, the method comprising the following steps:

[0080] S1, obtaining a BCG (Ballistocorticoid) signal;

[0081] S2, preprocessing the ballistocardi signal BCG to obtain a preprocessed ballistocardi signal BCG;

[0082] In this embodiment, S2 specifically includes:

[0083] S21, using an adaptive bandpass filter to process the ballistocardiographic signal BCG to obtain a denoised ballistocardiographic signal BCG;

[0084] S22, using variational mode decomposition to extract different frequency components of the denoised BCG signal, removing false signals, and obtaining a decomposed signal;

[0085] S23 . Normalize the decomposed signal to obtain a pre-processed ballistocardiogram (BCG) signal.

[0086] In practical applications, ballistocardiographic (BCG) signals are often affected by factors such as respiration, body movement, and environmental electromagnetic interference, resulting in a significant amount of noise and artifacts. Traditional time-domain or frequency-domain filtering methods struggle to effectively remove this complex noise. However, this embodiment employs adaptive bandpass filtering combined with VMD to separate frequency components and remove non-physiological interference signals, resulting in a clearer and more stable reconstructed BCG signal.

[0087] S3, input the pre-processed cardiac ballistic signal BCG into the pre-trained improved TimeGAN model, and output the final new cardiac ballistic signal BCG;

[0088] The improved TimeGAN model includes:

[0089] The embedding network is used to map the input signal of the improved TimeGAN model to the low-level feature space to obtain the corresponding hidden features;

[0090] The restoration network is used to reconstruct the hidden features through the residual network to generate the corresponding denoised signal. Traditional GAN has shortcomings in time series modeling. However, this embodiment adopts the TimeGAN model and introduces the residual network into the restoration network, which can better capture the timing information, make the waveform of the reconstructed signal closer to the real heart impact signal, and improve the accuracy of cardiovascular health monitoring.

[0091] A generative network is used to generate a new cardiac signal based on hidden features and random noise;

[0092] The discriminant network is used to discriminate the authenticity of the signal based on the denoised signal and the new heartbeat signal to obtain a discrimination result;

[0093] In this embodiment, the generative network + discriminative network collaborative optimization is used to generate high-quality BCG signals. Even low-quality input signals can be reconstructed into more reliable data through the model, thereby improving the accuracy of medical diagnosis.

[0094] The output layer is used to generate a final new cardiac ballistic signal BCG based on the denoised signal and the new cardiac ballistic signal according to the discrimination result of the discriminant network.

[0095] This embodiment improves the TimeGAN model and combines it with adaptive signal preprocessing, which not only enhances the noise resistance and improves the timing consistency, but also generates high-quality personalized BCG signals, which has significant technical advantages and practical value.

[0096] Example 2

[0097] See also Figure 2 In the second embodiment, the method for reconstructing a ballistocardial signal based on the improved TimeGAN includes:

[0098] Step 000: Using the training data set, the improved TimeGAN model is trained to obtain a trained improved TimeGAN model;

[0099] The training data set includes a plurality of ballistic heart signal samples, each of which includes: a preprocessed ballistic heart signal BCG for training, and a true state label corresponding to the preprocessed ballistic heart signal BCG.

[0100] In this example, supervised learning of the model using a training dataset (containing preprocessed cardiac ballistomic signals and their true state labels) enables TimeGAN to more accurately generate cardiac ballistomic signals that conform to real physiological characteristics. This improved TimeGAN not only learns the true distribution of time series signals but also incorporates true state labels, resulting in higher physiological plausibility and improved data quality.

[0101] Step 100: Obtaining a ballistocardiogram (BCG) signal.

[0102] Step 200: pre-processing the ballistocardiogram signal (BCG) to obtain a pre-processed ballistocardiogram signal (BCG);

[0103] In this embodiment, step 200 specifically includes:

[0104] Step 201: Process the ballistocardiographic signal (BCG) using an adaptive bandpass filter to obtain a denoised ballistocardiographic signal (BCG);

[0105] Step 201 uses an adaptive bandpass filter to process the ballistocardial signal BCG, and the center frequency of the filter is f c Calculated according to formula (3);

[0106] The formula (3) is:

[0107]

[0108] Among them, f0 is the basic center frequency, which is used to provide the initial reference value of the filter; c is the adjustment coefficient, which is used to control the response degree of the center frequency to the input signal; N is the number of sampling points, that is, the total number of sampling points of the input heart ballistic signal BCG; X i The i-th sampling point of the BCG signal is used to calculate the average amplitude, thereby dynamically adjusting the center frequency of the bandpass filter to make it more suitable for the physiological characteristics of different individuals. The basic center frequency f0 is 2.5Hz;

[0109] Step 201 uses an adaptive bandpass filter to process the ballistocardial signal BCG, and the bandwidth BW of the filter is calculated according to formula (4);

[0110] The formula (4) is calculated as follows:

[0111] BW=BW0+d·(max(X)-min(X));

[0112] BW0 is the initial bandwidth, that is, the default bandwidth of the filter when it is not adaptively adjusted; d is the bandwidth adjustment coefficient; X represents the input ballistic cardioversion signal BCG, and max(X) and min(X) represent the maximum and minimum values of the input ballistic cardioversion signal BCG, respectively.

[0113] The initial bandwidth BW0 is 5 Hz; the bandwidth adjustment coefficient d ranges from 0.2 to 0.5.

[0114] Step 202: Using variational mode decomposition to extract different frequency components of the denoised BCG ballistocardial signal, remove the spurious signal, and obtain a decomposed signal;

[0115] Step 203: Normalize the decomposed signal to obtain the preprocessed ballistocardiogram (BCG) signal. Traditional filtering methods generally use a fixed frequency range, while this method dynamically adjusts the center frequency and bandwidth to optimize signal quality based on individual differences. By combining the basic center frequency, adjustment coefficient, and signal amplitude information, the filtering process becomes more intelligent, improving applicability across diverse populations.

[0116] In practical applications, ballistocardiographic (BCG) signals are often affected by factors such as respiration, body movement, and environmental electromagnetic interference, resulting in a significant amount of noise and artifacts. Traditional time-domain or frequency-domain filtering methods struggle to effectively remove this complex noise. However, this embodiment employs adaptive bandpass filtering combined with VMD to separate frequency components and remove non-physiological interference signals, resulting in a clearer and more stable reconstructed BCG signal.

[0117] Traditional filtering methods struggle to effectively remove complex noise, but VMD decomposes the signal into distinct frequency components and removes non-physiological interference, resulting in more stable data. VMD, combined with a bandpass filter, achieves more precise noise removal, resolving the inability of conventional time-domain or frequency-domain filtering methods to effectively separate complex noise. This method preserves the true physiological characteristics of the ballistocardial signal and improves signal reconstruction quality.

[0118] Step 300: Input the pre-processed ballistocardiographic signal (BCG) into the pre-trained improved TimeGAN model, and output the final new ballistocardiographic signal (BCG);

[0119] The improved TimeGAN model includes:

[0120] The embedding network is used to map the input signal of the improved TimeGAN model to a low-dimensional feature space to obtain the corresponding hidden features; the low-dimensional feature mapping reduces data complexity, improves the learning efficiency of the model, and avoids the computational burden brought by high-dimensional data.

[0121] The restoration network is used to reconstruct the hidden features through the residual network to generate the corresponding denoised signal. Traditional GAN has shortcomings in time series modeling. However, this embodiment adopts the TimeGAN model and introduces the residual network into the restoration network, which can better capture the timing information, make the waveform of the reconstructed signal closer to the real heart impact signal, and improve the accuracy of cardiovascular health monitoring.

[0122] The generative network is used to generate new cardiac shock signals based on hidden features and random noise; random noise and hidden features are combined to generate richer cardiac shock signal data and increase data diversity.

[0123] The discriminant network is used to discriminate the authenticity of the signal based on the denoised signal and the new heartbeat signal to obtain a discrimination result;

[0124] In this embodiment, the generative network + discriminative network collaborative optimization is used to generate high-quality BCG signals. Even low-quality input signals can be reconstructed into more reliable data through the model, thereby improving the accuracy of medical diagnosis.

[0125] The output layer generates a new BCG (Ballistic Cardiac Signal) based on the denoised signal and the new BCG signal, according to the discriminant network's results. Weights are calculated based on the discriminant network's results, and the denoised signal and the newly generated signal are fused to generate the final BCG signal, improving signal quality.

[0126] The loss function of the improved TimeGAN model is:

[0127] L total =αL adv +βl rec +γL sup +δL latent ;

[0128] Among them, L total To improve the loss function of the TimeGAN model; L adv is the adversarial loss function; L rec is the autoregressive loss function; L sup is the supervision loss function; L latent is the latent space loss function; α is the first weight coefficient; β is the second weight coefficient; γ is the third weight coefficient; δ is the fourth weight coefficient.

[0129] In this embodiment, adversarial loss + autoregressive loss + supervised loss + latent space loss are used to comprehensively optimize the model and improve the signal reconstruction quality.

[0130] The output layer generates the final new cardiac ballistometry signal (BCG) based on the denoised signal and the new cardiac ballistometry signal according to the discrimination result of the discriminant network. Specifically, it includes:

[0131] When the discrimination result is greater than the first preset threshold, a new ballistocardia signal BCG is generated based on the denoised signal and the new ballistocardia signal using formula (1);

[0132] The formula (1) is:

[0133]

[0134] X final The final new cardiac signal BCG; is the denoised signal; is the new heartbeat signal; a1 is the weight parameter.

[0135] In this embodiment, the formula adopts a weighted fusion strategy to combine the denoised signal and the newly generated signal The signals are combined according to the weight a1. By adjusting the value of a1, a balance can be found between preserving the original signal information and enhancing the smoothness of the signal generated by the model. This fusion method reduces the information loss that may be caused by using the denoised signal or the generated signal alone, improving the quality of the final signal.

[0136] Among them, the denoised signal After preprocessing such as filtering and variational mode decomposition (VMD), most of the noise is removed, but some useful signals may be lost. Generated by the improved TimeGAN, it can supplement the information that may be lost in the denoising process and maintain signal integrity. By adjusting the weight a1, this method can adaptively fuse the denoised signal and the generated signal based on the results of the discriminator, avoiding the limitations of a single method.

[0137] The weight parameter is calculated using formula (2);

[0138] The formula (2) is:

[0139]

[0140] S rec The discriminant network included in the discriminant result is used to distinguish the denoised signal The authenticity score, representing the denoised signal The degree of similarity to the real signal; S gen The discriminant network included in the discriminant result is used to identify the new cardiac impulse signal A rating of authenticity, indicating a new cardiac impulse signal The degree of similarity to the real signal.

[0141] Denoised signal It may be affected by the denoising method, resulting in the loss of some cardiac ballistic features, but if it is still closer to the real signal than the generated signal, then S rec Higher, more weight is given. Generate signal It is possible that some of the lost information is recovered through data-driven methods, but if it has many artifacts or fails to fully fit the real signal, then S gen Formula (2) makes the final signal fusion scheme more intelligent and more dependent on objective evaluation rather than artificially set weights.

[0142] In this embodiment, adaptive bandpass filtering + VMD is used, which is more effective in removing noise than traditional filtering methods and adaptively adjusts to individual physiological characteristics. Combining TimeGAN and residual networks improves the ability to reconstruct time series signals, making the signal closer to the actual heartbeat signal. Combining multiple loss functions (adversarial, autoregressive, supervised, and latent space) comprehensively improves signal generation quality. Dynamic weight adjustment is used to dynamically adjust the signal fusion ratio based on the discrimination results, making the final output signal more stable and reliable.

[0143] Example 3

[0144] This third embodiment provides a method for reconstructing a cardiac ballistocardiogram (BBO) signal based on an improved TimeGAN. This method is suitable for use in healthcare monitoring devices, telemedicine platforms, and other applications requiring high-quality cardiac ballistocardiogram (BBO) signal analysis. This method can be integrated into wearable devices (such as smart bracelets and smart watches) or medical-grade monitoring systems to improve signal quality and enhance diagnostic accuracy.

[0145] First, collect raw signal data from different types of ballistocardiogram (BCG) sensors (e.g., photoplethysmography (PPG), electrocardiogram (ECG), and ballistocardiogram (BCG). Data acquisition devices include, but are not limited to, wearable sensors, portable medical monitoring devices, and hospital ECG monitoring systems.

[0146] All collected data were preprocessed, including outlier removal, normalization, and smoothing, to reduce the influence of noise and baseline drift.

[0147] After data collection is completed, the data is divided into training and test sets. In order to ensure the generalization ability of the model, the training data needs to cover cardiac shock signals under different physiological states, such as:

[0148] Movement state, resting state, sleeping state, interference state (such as environmental noise, signal loss);

[0149] In addition, data enhancement techniques, such as adding simulated noise and random signal loss, are used to improve the model's adaptability to complex environments.

[0150] This example optimizes the original TimeGAN, with the following improvements:

[0151] Generator optimization: Adjust the network structure to make it more consistent with the timing characteristics of the cardiac shock signal, and improve the coherence and authenticity of signal reconstruction.

[0152] Discriminator enhancement: A dual discriminator structure is used to distinguish short-term and long-term patterns respectively, improving model stability.

[0153] Loss function adjustment: Add constraints on the spectral characteristics of the real signal to make the spectral characteristics of the generated signal closer to the real heartbeat signal.

[0154] Time consistency constraint: Introduce time series smoothing loss to avoid unreasonable mutation points in the signal reconstruction process.

[0155] The embedding network of TimeGAN is trained using real heartbeat signals, enabling the model to accurately learn the basic patterns of time series.

[0156] The generator and discriminator of TimeGAN are trained using training data, and the authenticity of the generated signal is improved through adversarial training.

[0157] A supervision mechanism is introduced to compare the generated signal with the real signal and adjust the hyperparameters to optimize the signal quality.

[0158] The quality of the generated signal is evaluated on the test dataset, and signal similarity indicators (such as mean square error MSE and signal correlation) are used for evaluation.

[0159] After training, the model can be used in two main application scenarios: When the cardiac ballistometry signal collected by the device is missing or interfered with by noise, TimeGAN is used to generate a complete signal sequence to ensure data continuity and integrity. In medical research or artificial intelligence training, high-quality cardiac ballistometry signal samples are generated to improve the training effect of deep learning models while protecting patient privacy.

[0160] This example demonstrates that the improved TimeGAN can effectively improve the reconstruction quality of ballistocardial signals. Experimental results on multiple test datasets demonstrate that, compared to traditional methods, this method significantly improves signal similarity, signal timing consistency, and noise removal capabilities, providing more reliable data support for cardiac health monitoring and telemedicine.

[0161] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0162] In the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," "connect," "fixed," etc. should be understood broadly. For example, they may refer to fixed connection, detachable connection, or integration; mechanical connection or electrical connection; direct connection or indirect connection through an intermediate medium; and internal communication between two components or interaction between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0163] In the present invention, unless otherwise expressly specified or limited, when a first feature is "above" or "below" a second feature, it may mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediate medium. Furthermore, when a first feature is "above," "above," or "above" a second feature, it may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is at a higher level than the second feature. When a first feature is "below," "below," or "below" a second feature, it may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is at a lower level than the second feature.

[0164] In the description of this specification, the terms "one embodiment", "some embodiments", "embodiments", "examples", "specific examples" or "some examples" refer to the specific features, structures, materials or characteristics described in conjunction with the embodiment or example and included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and features of different embodiments or examples, unless they are mutually inconsistent.

[0165] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may alter, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for reconstructing cardiac ballistic signals based on improved TimeGAN, characterized in that: The method comprises the following steps: S1, obtaining a BCG (Ballistocorticoid) signal; S2, preprocessing the ballistocardi signal BCG to obtain a preprocessed ballistocardi signal BCG; S3, input the pre-processed cardiac ballistic signal BCG into the pre-trained improved TimeGAN model, and output the final new cardiac ballistic signal BCG; The improved TimeGAN model includes: The embedding network is used to map the input signal of the improved TimeGAN model to the low-level feature space to obtain the corresponding hidden features; The recovery network is used to reconstruct the hidden features through the residual network and generate the corresponding denoised signal; A generative network is used to generate a new cardiac signal based on hidden features and random noise; The discriminant network is used to discriminate the authenticity of the signal based on the denoised signal and the new heartbeat signal to obtain a discrimination result; The output layer is used to generate a final new cardiac ballistic signal BCG based on the denoised signal and the new cardiac ballistic signal according to the discrimination result of the discriminant network.

2. The method for reconstructing the ballistocardial signal based on the improved TimeGAN according to claim 1, characterized in that: The S2 specifically includes: S21, using an adaptive bandpass filter to process the ballistocardiographic signal BCG to obtain a denoised ballistocardiographic signal BCG; S22, using variational mode decomposition to extract different frequency components of the denoised BCG signal, removing false signals, and obtaining a decomposed signal; S23 . Normalize the decomposed signal to obtain a pre-processed ballistocardiogram (BCG) signal.

3. The method for reconstructing the ballistocardial signal based on the improved TimeGAN according to claim 1, characterized in that: Before S1, the method further includes: S0. Using the training data set, the improved TimeGAN model is trained to obtain a trained improved TimeGAN model; The training data set includes a plurality of ballistic heart signal samples, each of which includes: a preprocessed ballistic heart signal BCG for training, and a true state label corresponding to the preprocessed ballistic heart signal BCG.

4. The method for reconstructing the ballistocardial signal based on the improved TimeGAN according to claim 3, characterized in that: The loss function of the improved TimeGAN model is: L total =αL adv +βL rec +γL sup +δL latent ; Among them, L total To improve the loss function of the TimeGAN model; L adv To counter the loss function; L rec is the autoregressive loss function; L sup is the supervision loss function; L latent is the latent space loss function; α is the first weight coefficient; β is the second weight coefficient; γ is the third weight coefficient; δ is the fourth weight coefficient.

5. The method for reconstructing the ballistocardial signal based on the improved TimeGAN according to claim 4, characterized in that: The output layer generates the final new cardiac ballistometry signal (BCG) based on the denoised signal and the new cardiac ballistometry signal according to the discrimination result of the discriminant network. Specifically, it includes: When the discrimination result is greater than the first preset threshold, a new ballistocardia signal BCG is generated based on the denoised signal and the new ballistocardia signal using formula (1); The formula (1) is: X final The final new cardiac signal BCG; is the denoised signal; For new heart beat signal; a1 is the weight parameter.

6. The method for reconstructing the ballistocardial signal based on the improved TimeGAN according to claim 5, characterized in that: The weight parameter is calculated using formula (2); The formula (2) is: S rec The discriminant network included in the discriminant result is used to distinguish the denoised signal The authenticity score, representing the denoised signal The degree of similarity to the real signal; S gen The discriminant network included in the discriminant result is used to identify the new cardiac impulse signal A rating of authenticity, indicating a new cardiac impulse signal The degree of similarity to the real signal.

7. The method for reconstructing the ballistocardial signal based on the improved TimeGAN according to claim 6, characterized in that: The S21 uses an adaptive bandpass filter to process the ballistic heart signal BCG, and the center frequency of the filter is f c Calculated according to formula (3); The formula (3) is: Among them, f0 is the basic center frequency, which is used to provide the initial reference value of the filter; c is the adjustment coefficient, which is used to control the response of the center frequency to the input signal; N is the number of sampling points, that is, the total number of sampling points of the input ballistocardial signal BCG; X i is the i-th sampling point of the ballistocardiogram (BCG) signal, which is used to calculate the average amplitude, thereby dynamically adjusting the center frequency of the bandpass filter to make it more adaptable to the physiological characteristics of different individuals.

8. The method for reconstructing the ballistocardial signal based on the improved TimeGAN according to claim 7, characterized in that: The S21 processes the ballistic heart signal BCG using an adaptive bandpass filter, wherein the bandwidth BW of the filter is calculated according to formula (4); The formula (4) is calculated as follows: BW=BW0+d·(max(X)-min(X)); BW0 is the initial bandwidth, that is, the default bandwidth of the filter when it is not adaptively adjusted; d is the bandwidth adjustment coefficient; X represents the input ballistic cardioversion signal BCG, and max(X) and min(X) represent the maximum value and minimum value of the input ballistic cardioversion signal BCG, respectively.

9. The method for reconstructing the ballistocardial signal based on the improved TimeGAN according to claim 8, characterized in that: The basic center frequency f0 is 2.5Hz; The initial bandwidth BW0 is 5Hz; The bandwidth adjustment coefficient d ranges from 0.2 to 0.

5.

10. A ballistocardial signal reconstruction system based on improved TimeGAN, characterized in that: include: at least one processor; as well as At least one memory communicatively connected to the processor, wherein the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the cardiac ballistic signal reconstruction method based on the improved TimeGAN as described in any one of claims 1-9.

Images