Human body respiration and heartbeat signal separation method, device and equipment and storage medium

Through noise reduction and modal function decomposition technology, combined with fast Fourier transform and adaptive ICA algorithm, the problem of separation of breathing and heartbeat signals in contactless multi-person targets is solved, and the accurate separation and detection of signals is achieved.

CN118981623APending Publication Date: 2024-11-19GUANGZHOU INST OF RAILWAY TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411038353.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

In contactless multi-human target vital sign signal detection, breathing and heartbeat signals are usually weak and overlap each other, making it difficult to separate.

Method used

The noise reduction algorithm and modal function decomposition technology are used, combined with fast Fourier transform and adaptive ICA algorithm, the characteristic frequency bands of breathing and heartbeat signals are screened out, and the specific signals are separated through modal function reconstruction.

Benefits of technology

Accurate separation of breathing and heartbeat signals in mixed physiological signals is achieved, and the accuracy and reliability of detection are improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118981623A_ABST
    Figure CN118981623A_ABST
Patent Text Reader

Abstract

The invention discloses a human respiration and heartbeat signal separation method, device and equipment and a storage medium. The method comprises the following steps: acquiring a vital sign signal of a selected character object in a detection area; performing noise reduction and interference suppression on the vital sign signal to obtain a first signal; performing mode function decomposition on the first signal to obtain a plurality of first intrinsic mode functions; performing fast Fourier transform on the plurality of first intrinsic mode functions, and calculating to obtain the power spectrum density of each first intrinsic mode function; according to the characteristic frequency bands of the breathing and heartbeat signals and the power spectrum density, first intrinsic mode functions corresponding to the breathing signals and the heartbeat signals are screened out from a plurality of first intrinsic mode functions, and then the first intrinsic mode functions corresponding to the breathing signals and the heartbeat signals are reconstructed; the separated respiration signals and heartbeat signals are obtained. According to the invention, breathing and heartbeat signals of a single object can be separated.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of signal separation, and in particular to a method, device, equipment and storage medium for separating human breathing and heartbeat signals. Background Art

[0002] In recent years, non-contact vital sign signal detection methods have gradually become a research hotspot. This method uses radar, infrared, hyperspectral and other technologies to obtain vital sign signals, such as breathing and heartbeat, without contacting the person being tested. The non-contact detection method not only improves the comfort of the person being tested and avoids the risk of cross-infection, but is also suitable for a variety of occasions and can achieve continuous and uninterrupted monitoring of multiple human targets.

[0003] However, in the non-contact detection of vital signs signals of multiple human targets, the breathing and heartbeat signals are usually weak and overlap with each other.

[0004] Therefore, how to separate breathing and heartbeat signals is an urgent problem to be solved. Summary of the invention

[0005] The present invention provides a method, device, equipment and storage medium for separating human respiration and heartbeat signals, which can separate the respiration and heartbeat signals from mixed signals.

[0006] An embodiment of the present invention provides a method for separating human breathing and heartbeat signals, comprising:

[0007] Acquire vital sign signals of selected human objects in the detection area;

[0008] According to a noise reduction algorithm, the vital sign signal is subjected to noise reduction and interference suppression to obtain a first signal;

[0009] Performing modal function decomposition on the first signal to obtain a plurality of first eigenmode functions;

[0010] Performing fast Fourier transformation on the first intrinsic mode functions, and calculating the power spectrum density of each first intrinsic mode function;

[0011] According to the characteristic frequency bands of the breathing and heartbeat signals and the power spectrum density, a first intrinsic mode function corresponding to the breathing signal is selected from the plurality of first intrinsic mode functions to obtain a second intrinsic mode function, and a first intrinsic mode function corresponding to the heartbeat signal is selected from the plurality of first intrinsic mode functions to obtain a third intrinsic mode function;

[0012] The second intrinsic mode function is reconstructed to obtain a separated breathing signal, and the third intrinsic mode function is reconstructed to obtain a separated heartbeat signal.

[0013] Furthermore, the above-mentioned obtaining of the vital sign signal of the selected person object in the detection area includes:

[0014] Acquire real-time signals of multiple human objects in the detection area;

[0015] Filtering and denoising the real-time signal to obtain a third signal after filtering and denoising;

[0016] Separating the third signal according to a multi-target signal separation algorithm to obtain a plurality of separated signals and a first feature;

[0017] Inputting each separated signal into a preset multi-target separation model, so that the multi-target separation model determines the person object corresponding to each separated signal according to the first feature;

[0018] According to the person object corresponding to each separated signal, the separated signal corresponding to the selected person object is selected to obtain the vital sign signal of the selected person object.

[0019] Furthermore, the construction of the above multi-target separation model includes:

[0020] Obtaining a number of separation signal samples with real labels, wherein the real labels are used to represent the person object corresponding to each separation signal sample;

[0021] For each separated signal sample, the separated signal sample is input into the multi-target separation model to be trained, so that each separated sample is identified as a corresponding person object by the multi-target separation model, and a loss function is calculated according to the recognition result and the true label;

[0022] After each loss function value is calculated, it is determined whether the loss function converges. If so, the training of the multi-target separation model to be trained is completed; otherwise, the parameters in the multi-target separation model to be trained are adjusted and the training is continued.

[0023] Furthermore, the third signal is separated according to the multi-target signal separation algorithm to obtain a plurality of separated signals and the first feature, including:

[0024] Decomposing the third signal at multiple scales to obtain signals at different time scales and a first feature of the signal at each time scale;

[0025] The signals of the above different time scales are combined, and the combined signal is separated by an adaptive ICA algorithm to obtain the above several separated signals.

[0026] Furthermore, the above-mentioned vital sign signal is subjected to noise reduction and interference suppression according to the noise reduction algorithm to obtain the first signal, including:

[0027] Performing chest position detection on the vital sign signal according to fast Fourier transform to obtain the chest position, and then determining a fourth signal corresponding to the chest according to the chest position;

[0028] Correcting the phase of the fourth signal according to a circle tracking algorithm to obtain a corrected fourth signal;

[0029] Demodulating the corrected fourth signal according to the DACM algorithm, and performing noise elimination on the demodulated signal using singular spectrum analysis to obtain a fifth signal after noise elimination;

[0030] The fifth signal is reconstructed to obtain the first signal.

[0031] Furthermore, the first signal is subjected to modal function decomposition to obtain a plurality of first intrinsic mode functions, including:

[0032] Adding noise to the first signal to obtain a sixth signal after adding noise;

[0033] According to the CEEMDAN algorithm, the local mean is repeatedly subtracted from the sixth signal until the number of local maximum points and local minimum points is equal, and the integral average of the envelope of the local maximum value and the envelope of the local minimum value is 0, then the subtraction of the local mean is stopped to obtain the second intrinsic mode function and the first residual;

[0034] The local mean is repeatedly subtracted from the first residual and each subsequent residual to obtain an intrinsic mode function and a residual, until the number of decompositions reaches a preset value, the decomposition is stopped, and the first intrinsic mode functions are obtained.

[0035] Further, according to the characteristic frequency bands of the respiratory and heartbeat signals and the power spectral density, the first intrinsic mode function corresponding to the respiratory signal is screened out from the plurality of first intrinsic mode functions to obtain the second intrinsic mode function, and the first intrinsic mode function corresponding to the heartbeat signal is screened out from the plurality of first intrinsic mode functions to obtain the third intrinsic mode function, including:

[0036] According to the characteristic frequency band of the respiratory signal, a first eigenmode function containing the characteristic frequency is selected from the power spectrum density to obtain a second eigenmode function;

[0037] According to the characteristic frequency band of the heartbeat signal, the first intrinsic mode function containing the characteristic frequency is screened out from the power spectrum density to obtain the third intrinsic mode function.

[0038] Based on the above method embodiment, the present invention provides a corresponding device embodiment;

[0039] The present invention provides a human breathing and heartbeat signal separation device, comprising:

[0040] Signal acquisition module, signal noise reduction module, signal decomposition module, power spectrum density calculation module, signal screening module and signal reconstruction module;

[0041] The signal acquisition module is used to acquire the vital sign signal of the selected person object in the detection area;

[0042] The signal noise reduction module is used to perform noise reduction and interference suppression on the vital sign signal according to a noise reduction algorithm to obtain a first signal;

[0043] The signal decomposition module is used to perform modal function decomposition on the first signal to obtain a plurality of first intrinsic mode functions;

[0044] The power spectrum density calculation module is used to perform fast Fourier transform on the first intrinsic mode functions and calculate the power spectrum density of each first intrinsic mode function;

[0045] The signal screening module is used to screen out the first intrinsic mode function corresponding to the breathing signal from the plurality of first intrinsic mode functions according to the characteristic frequency bands of the breathing and heartbeat signals and the power spectrum density to obtain the second intrinsic mode function, and to screen out the first intrinsic mode function corresponding to the heartbeat signal from the plurality of first intrinsic mode functions to obtain the third intrinsic mode function;

[0046] The signal reconstruction module reconstructs the second intrinsic mode function to obtain a separated breathing signal, and reconstructs the third intrinsic mode function to obtain a separated heartbeat signal.

[0047] Based on the above method embodiment, the present invention provides a terminal device embodiment;

[0048] The present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the method for separating human respiration and heartbeat signals described in any embodiment of the present invention.

[0049] Based on the above method embodiment, the present invention provides a storage medium embodiment;

[0050] The present invention provides a storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute a method for separating human respiration and heartbeat signals described in any embodiment of the present invention.

[0051] The embodiments of the present invention have the following beneficial effects:

[0052] The present invention provides a method, device, equipment and storage medium for separating breathing and heartbeat signals of multiple human targets. The method comprises:

[0053] First, the vital sign signal of the selected human object in the detection area is obtained; then, according to the noise reduction algorithm, the above vital sign signal is subjected to noise reduction and interference suppression to obtain the first signal; then, the above first signal is subjected to modal function decomposition to obtain a plurality of first intrinsic mode functions; then, the above first intrinsic mode functions are subjected to fast Fourier transform, and the power spectrum density of each first intrinsic mode function is calculated; then, according to the characteristic frequency band of the respiratory and heartbeat signals and the above power spectrum density, the first intrinsic mode function corresponding to the respiratory signal is screened out from the above first intrinsic mode functions to obtain the second intrinsic mode function, and the first intrinsic mode function corresponding to the heartbeat signal is screened out from the above first intrinsic mode functions to obtain the third intrinsic mode function; finally, the above second intrinsic mode function is reconstructed to obtain the separated respiratory signal, and the above third intrinsic mode function is reconstructed to obtain the separated heartbeat signal. Therefore, the present invention can accurately obtain the intrinsic mode function corresponding to the frequency of the respiratory and heartbeat signals by performing the process of modal function decomposition and reconstruction on the above-mentioned first signal. After reconstructing the above-mentioned intrinsic mode function, the purpose of extracting and separating the respiratory and heartbeat signals from the mixed physiological signals is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 The figure is a flow chart of a method for separating human respiration and heartbeat signals provided in one embodiment of the present invention.

[0055] Figure 2 It is a schematic diagram of radar signal collection provided by an embodiment of the present invention.

[0056] Figure 3 The figure is a schematic diagram of a signal processing flow in a radar according to an embodiment of the present invention.

[0057] Figure 4 The present invention is a schematic diagram of a device for separating human respiration and heartbeat signals according to an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] like Figure 1 As shown, a method for separating human breathing and heartbeat signals provided by an embodiment of the present invention includes:

[0060] Step S101: Acquire the vital sign signal of a selected person object in the detection area;

[0061] In a preferred embodiment, the above-mentioned obtaining of the vital sign signal of the selected person object in the detection area includes:

[0062] Acquire real-time signals of multiple human objects in the detection area;

[0063] Filtering and denoising the real-time signal to obtain a third signal after filtering and denoising;

[0064] Separating the third signal according to a multi-target signal separation algorithm to obtain a plurality of separated signals and a first feature;

[0065] Inputting each separated signal into a preset multi-target separation model, so that the multi-target separation model determines the person object corresponding to each separated signal according to the first feature;

[0066] According to the person object corresponding to each separated signal, the separated signal corresponding to the selected person object is selected to obtain the vital sign signal of the selected person object.

[0067] Specifically, the first feature includes signal amplitude and signal period, and the real-time signal includes a real-time motion signal and a real-time non-motion signal.

[0068] Specifically, each separated signal is input into a preset multi-target separation model, and finally the vital sign signal of the selected person object is obtained through the built-in multi-scale attention mechanism layer, convolution layer and pooling layer.

[0069] Preferably, non-contact sensors, such as infrared cameras, radar sensors, etc., are used to collect real-time signals from the detection area.

[0070] Specifically, in order to obtain the required signal, obtaining the distance information of the target through the radar system is the key first step. This can be achieved by measuring the time difference between the radar transmitting signal and the target reflecting back, thereby calculating the distance of the target. This step is crucial for subsequent state judgment, because targets at different distances may require different signal processing and analysis methods. Next, the system needs to analyze the radar signal clusters emitted by the human body. These signal clusters contain the reflection information of the target. By analyzing their characteristics (such as amplitude, frequency, phase, etc.), it can be further determined whether the target is stationary or moving. For example, a moving target may produce a Doppler effect, causing the frequency of the signal cluster to change; while a stationary target will not produce such a change.

[0071] Specifically, after determining the moving or stationary state of the target, the system also needs to combine the distance information between multiple sensors and multiple targets, as well as the moving position and stationary state (such as sitting, standing, etc.) of the target to make a comprehensive judgment. This can be achieved by fusing the data of multiple radars and using spatial geometric relationships and signal processing technology. For example, by comparing the distance and angle measurements of the same target by different radars, the position and state of the target can be further confirmed. Due to the occlusion problem of multiple IR-UWB (infrared ultra-wideband) radars and the limited information dimensions they can provide, they can only monitor distance information, so they are limited in posture monitoring. Specifically, when the target is in different postures (such as sitting or standing), the distance change caused by the change in posture may be confused with the actual distance change of the target's position, which makes it difficult for IR-UWB radar to distinguish between the two situations. On the other hand, FMCW (frequency modulated continuous wave) radars will be severely interfered with by signals in a multi-person environment. Due to the mutual reflection and signal superposition between multiple people, it is difficult for FMCW radars to accurately distinguish the signals of each individual, so it is not enough to meet the conditions for multi-person posture monitoring.

[0072] In order to solve the above problems, a multi-radar Internet of Things monitoring system is proposed. This system combines the advantages of different radar types and realizes efficient and accurate monitoring of multiple targets through reasonable layout and collaborative work. Specifically, the high resolution and anti-interference ability of IR-UWB radar can be used to make up for the shortcomings of FMCW radar in multi-person environments; at the same time, the accuracy and reliability of the system can be further improved through data fusion and collaborative processing between multiple radars.

[0073] Specifically, first, the frequency modulated continuous wave radar is placed on a fixed side of the detection area, facing the center of the area, to ensure that it covers the entire monitoring area and the angle gain of the left and right antennas is the same. Then, multiple pulse ultra-wideband radars are placed at the three vertices of an equilateral triangle in the detection area, and one of the edge points is located at the same monitoring position and angle as the frequency modulated continuous wave radar, and the radar node faces the center of the area. Secondly, multiple pulse ultra-wideband radars are used to collect signals in the area to obtain multiple radar signal matrices. At the same time, the frequency modulated continuous wave radar also collects signals, and the collected binary data stream is converted to a binary number and a complex number to obtain a four-dimensional radar matrix signal. Due to the large differences in the principles and characteristics of pulse ultra-wideband radar and frequency modulated continuous wave radar, it is difficult to develop an automatic fusion algorithm directly using the point cloud output of the two through an algorithm. Therefore, it is necessary to deploy them in a specific area, and obtain the relative position relationship between the two through actual measurement and calibration. In a multi-target scenario, the signal peak in the ultra-wideband radar signal is extracted by dividing the signal into multiple coherent distance intervals, and there is only one local maximum peak in each coherent interval. These local maximum peaks are regarded as effective peaks, and the position of each effective peak corresponds to the position of the individual. In this way, accurate detection and localization of multiple targets can be achieved.

[0074] Schematically, the radar signal acquisition diagram is as follows Figure 2As shown in the figure, when collecting signals through radar, the transmitting antenna (TX) plays an important role in emitting electromagnetic waves outward, which is the first step in the detection process. At the same time, the receiving antenna (RX) is responsible for capturing the electromagnetic waves reflected from the target object, providing a key channel for signal reception. In order to ensure the strength and quality of signal transmission, the power amplifier (PA) and the low noise amplifier (LNA) play a role at the transmitting and receiving ends respectively, the former enhancing the signal power and the latter reducing noise interference. The power divider ensures that the signal is evenly distributed in the transmitting and receiving paths, maintaining the overall performance of the system. The local oscillator (LO) generates a stable frequency reference signal, which is mixed with the received signal in the mixer to generate a difference frequency signal, providing a basis for subsequent signal processing. The low-pass filter further filters out unwanted high-frequency components and retains the low-frequency signal containing the target information. Finally, the analog-to-digital converter (ADC) converts the analog signal into a digital signal. This step is crucial because it enables the signal to be processed and analyzed by the digital system, laying the foundation for further processing of the deep learning algorithm and extraction of vital signs signals. The deployment of the entire radar system is the basis for non-contact vital sign detection. Each component plays an indispensable role, jointly ensuring the effective operation of the system and high-precision detection capabilities. Specific implementation: electromagnetic waves are emitted outward through the antenna, and the electromagnetic waves are reflected when they touch the object. The radar obtains target information by processing the received echo signal. Usually, the FMCW radar system consists of a transmitting antenna TX, a receiving antenna RX, a power amplifier (PA), a low-noise amplifier (LNA), a power divider, a local oscillator (LO), a mixer, a low-pass filter, an analog to digital converter (ADC), and a digital signal processor. First, LO generates the linear frequency modulation pulse sent by the radar; secondly, the signal encounters an object during propagation and is reflected (received signal) and mixed with the transmitted signal to generate a mixed signal; then, the mixed signal generates an IF signal through a low-pass filter; finally, the IF signal is converted into a digital signal after passing through the analog-to-digital converter.

[0075] Specifically, when the system needs to detect a single target, the radar system uses sinusoidal wave modulation. This modulation method optimizes the transmission and reception of the signal by adjusting the frequency deviation of the signal, thereby achieving accurate detection of a single target; for multi-target detection, the system usually uses sawtooth wave modulation or triangle wave modulation. These modulation methods can provide a wider frequency range, which helps to distinguish and identify multiple targets; the radar system's transmitting antenna (TX) emits modulated electromagnetic waves, and when the electromagnetic waves encounter human targets, they are reflected. The reflected electromagnetic waves carry the target's information, including distance, speed, etc. Using the correspondence between distance, time and speed, the distance of the target can be obtained as:

[0076]

[0077] In the formula, R represents the distance, c represents the speed, and t d Indicates time.

[0078] Specifically, the receiving antenna (RX) receives the reflected electromagnetic waves. The received signal will undergo frequency changes due to the Doppler effect, and these changes are related to the target's motion state. In this process, the radar system uses a carefully designed modulation strategy to ensure the effective transmission and reception of the signal. Sine wave modulation is suitable for the detection of a single target due to its frequency stability, while sawtooth wave or triangle wave modulation is more suitable for multi-target environments because they can provide rich frequency information to help distinguish adjacent or overlapping target signals. The emitted electromagnetic waves are reflected after contacting targets such as the human body, and these reflected signals contain key information such as the distance and speed of the target. By receiving these reflected signals through the receiving antenna and analyzing the frequency changes using the Doppler effect, the system can monitor the target's motion state, including micro-movements, vital signs such as breathing and heartbeat. This non-contact detection method not only improves the accuracy and efficiency of detection, but also provides a safe, convenient and real-time solution for monitoring the vital signs of multiple human targets.

[0079] Schematically, the signal processing flow chart in the radar is as follows Figure 3As shown, Frame Sequence: Multiple frames are listed in the image, from Frame 1 to Frame N, which shows that the radar system collects data at continuous time intervals. Frame Periodicity: The time interval between each frame is 50 milliseconds. This means that the radar system sends a signal and receives the reflected signal every 50 milliseconds. Range: In each frame, the radar system measures the distance of the object. "Range-" may represent a specific distance value measured by the radar system. Fast Fourier Transform (FFT): In each frame, the radar signal is processed by FFT, which is a mathematical transformation used to convert the signal from the time domain to the frequency domain. Target: In radar signal processing, the target refers to the object to be detected and tracked. Range-Bins: After FFT processing, the signal is decomposed into different frequency components, each component corresponds to a specific range binary, which can be understood as the strength of the signal at different distances. Further Processing: After FFT, the signal needs to be further processed to extract useful information. Extract Phase and unwrap: Further processing may include extracting the phase information of the signal and unwrapping it to remove phase discontinuities. Vital Signs Estimation: For specific range bins, further processing is performed to estimate vital signs such as heart rate, breathing rate, etc. Slow Time Axis: This may refer to the radar system collecting data on a slower time scale for analysis and tracking of targets.

[0080] Specifically, radar systems play an important role in the field of vital signs monitoring. Their working principle is to detect and track targets by emitting electromagnetic waves and receiving reflected signals. In the above process, the radar system continuously collects data at periodic intervals of 50 milliseconds to form a series of frames. Each frame of data is processed by fast Fourier transform (FFT) to convert the signal from the time domain to the frequency domain, thereby obtaining the signal intensity distribution at different distances, that is, range-binary. Further processing includes extracting the phase information of the signal and performing phase unpacking to eliminate phase discontinuities. For a specific range binary, the system will perform vital sign estimation, analysis and extraction of key physiological parameters such as heart rate and respiratory rate. Throughout the process, the use of a slow time axis allows the system to track and analyze targets over a longer time scale, thereby providing stable and reliable vital signs monitoring data.

[0081] In this preferred embodiment, the real-time signals of multiple human objects collected in the detection area are separated by a target separation algorithm and a multi-person target separation model to obtain the vital sign signal of a single human body.

[0082] In another preferred embodiment, the construction of the above multi-target separation model includes:

[0083] Obtaining a number of separation signal samples with real labels, wherein the real labels are used to represent the person object corresponding to each separation signal sample;

[0084] For each separated signal sample, the separated signal sample is input into the multi-target separation model to be trained, so that each separated sample is identified as a corresponding person object by the multi-target separation model, and a loss function is calculated according to the recognition result and the true label;

[0085] After each loss function value is calculated, it is determined whether the loss function converges. If so, the training of the multi-target separation model to be trained is completed; otherwise, the parameters in the multi-target separation model to be trained are adjusted and the training is continued.

[0086] In this preferred embodiment, a trained multi-target separation model is obtained by training a multi-target separation model to be trained using separation signal samples.

[0087] In another preferred embodiment, the third signal is separated according to the multi-target signal separation algorithm to obtain a plurality of separated signals and the first feature, including:

[0088] Decomposing the third signal at multiple scales to obtain signals at different time scales and the first feature of each time scale signal;

[0089] The signals of the above different time scales are combined, and the combined signal is separated by an adaptive ICA algorithm to obtain the above several separated signals.

[0090] In this preferred embodiment, the third signal is first decomposed into multiple scales to obtain signals of different time scales and the above-mentioned first feature of each time scale signal, and then the signals of different time scales are merged, and then the merged signal is separated by an adaptive ICA (independent component analysis) algorithm to obtain several separated signals.

[0091] Preferably, an adaptive ICA algorithm is used to preliminarily separate the real-time signals of multiple human objects. Then, in the multi-object separation model, the spatiotemporal features of the signals are automatically extracted through multiple convolutional layers and pooling layers.

[0092] Specifically, the automatic extraction of the spatiotemporal features of the signal through multiple convolutional layers and pooling layers is mainly reflected in the following aspects: Hierarchical feature learning: Convolutional neural networks learn the multi-scale features of the signal through different levels of convolutional layers. The bottom convolutional layer may capture the basic pattern of the signal, such as local features of periodic changes, while the high-level convolutional layer may capture more complex combination features, which may be related to the specific behavior or state of human movement. Local connection in the time dimension: In a one-dimensional convolutional neural network, the convolution operation maintains the local connectivity of the signal in the time dimension. This means that when the network processes the signal, it only focuses on the local area near the current time point, so that it can capture the local temporal characteristics of the signal. Parameter sharing: The convolution kernels in the convolutional layer share the same parameters when sliding over the entire signal, which enables the network to learn features that are generally applicable at different time points and enhances the model's generalization ability for the spatiotemporal features in the signal. Temporal abstraction of the pooling layer: The pooling layer reduces the temporal dimension of the signal by downsampling the feature map output by the convolutional layer while retaining the most important information. Max pooling can highlight the main change points in the signal, while average pooling provides a smooth representation of the signal. Multi-scale feature fusion: In some advanced network structures, feature pyramid networks (FPN) or multi-scale fusion techniques may be used, which can capture the characteristics of the signal at different scales and perform feature fusion at different stages of the network to more comprehensively understand the spatiotemporal characteristics of the signal. Attention mechanism: Integrating the attention mechanism layer can make the model pay more attention to the key parts of the signal. For example, the self-attention mechanism can learn the relationship between different time points in the signal, while the position attention can emphasize the importance of a specific time point in the signal. Combination of sequence models: For signals that need to consider long-term dependent information, recurrent neural networks (RNN) or long short-term memory networks (LSTM) can be combined with convolutional layers. This combination allows the model to extract local spatiotemporal features while also considering the long-term temporal information of the signal. Depth and width expansion: By increasing the depth (more convolutional layers) and width (the number or size of convolution kernels) of the network, the model can learn more complex spatiotemporal features, thereby improving the ability to understand and separate signals. End-to-end training: Under the deep learning framework, these layers can be trained end-to-end, that is, directly from the original signal to the separated signal, without manually designing the feature extractor, and the model automatically learns the most useful feature representation.

[0093] In the integrated attention mechanism layer, by learning the key features in the signal, the model's ability to focus on the target signal is improved, thereby more accurately separating and identifying the signals of different human targets.

[0094] Specifically, in the integrated attention mechanism layer, by learning the key features in the signal, information can be extracted from multiple dimensions, including but not limited to the following aspects: Time-dependent features: Time series characteristics of the signal, such as periodicity, frequency and phase information, which are crucial for identifying breathing and heartbeat patterns. Local pattern features: Local changes in the signal, such as the starting point, peak or mutation point of the movement, which can reflect the subtle differences in human body movement. Frequency features: The frequency components of the signal, including the main frequency and the distribution of frequencies, these features help to distinguish different types of physiological signals. Amplitude features: The amplitude changes of the signal, such as the fluctuation of the peak height and amplitude, can reflect the strength and stability of the signal. Energy features: The energy distribution of the signal, such as the total energy of the signal or the energy of a specific frequency band, can be used to evaluate the activity level of the signal. Entropy features: The complexity and uncertainty of the signal, such as information entropy, can reflect the randomness and complexity of the signal. Time-frequency features: Combining time and frequency analysis, such as wavelet transform coefficients, can provide the characteristics of the signal at different time and frequency scales. Nonlinear features: The nonlinear dynamic characteristics of the signal, such as chaotic characteristics, can reveal the nonlinear behavior and interaction of the signal. Phase features: The phase information of the signal, such as phase difference or phase consistency, can reflect the synchronization and phase locking phenomenon of the signal. Spatial features (if it is a multi-channel signal): The characteristics of the signal at different sensors or spatial locations can reflect the spatial distribution and propagation characteristics of the signal. Contextual features: The contextual information of the signal, such as the contextual relationship of the signal, can provide the situational information of the signal, which helps to understand the background and environment of the signal. Multimodal features: If the signal is multimodal, integrating features from different sensors or data sources can provide a more comprehensive description of the signal. Attention weight: The attention mechanism can identify the most important parts of the signal and assign higher weights, which helps the model focus on the most discriminative features. Sequence position features: In sequence data, signals at different positions may have different meanings, and the attention mechanism can identify and emphasize these key positions. By integrating these key features, the attention mechanism layer can significantly improve the model's ability to understand the signal, thereby achieving more accurate separation and recognition in a variety of applications such as human signal processing, behavior recognition, and health monitoring. With the continuous development of deep learning technology, the attention mechanism has become an important tool to improve model performance.

[0095] Preferably, when performing multi-scale decomposition, the bottleneck residual structure is improved by replacing the 3×3 convolution kernel of the full channel with a group of smaller filter groups, and connecting these grouped filters in a residual-like hierarchical manner to increase the number of scales that the output features can represent.

[0096] Step S102: performing noise reduction and interference suppression on the vital sign signal according to a noise reduction algorithm to obtain a first signal;

[0097] In a preferred embodiment, the above-mentioned denoising and interference suppression are performed on the above-mentioned vital sign signal according to the denoising algorithm to obtain the first signal, including:

[0098] Performing chest position detection on the vital sign signal according to fast Fourier transform to obtain the chest position, and then determining a fourth signal corresponding to the chest according to the chest position;

[0099] Correcting the phase of the fourth signal according to a circle tracking algorithm to obtain a corrected fourth signal;

[0100] Demodulating the corrected fourth signal according to the DACM algorithm, and performing noise elimination on the demodulated signal using singular spectrum analysis to obtain a fifth signal after noise elimination;

[0101] The fifth signal is reconstructed to obtain the first signal.

[0102] Preferably, the phase of the fourth signal is corrected by a circle tracking algorithm, which can eliminate the error caused by the position change. The DC offset and phase ambiguity of the signal can be solved by demodulating the corrected fourth signal according to the DACM algorithm.

[0103] In this preferred embodiment, the chest position of the vital sign signal is first detected based on the fast Fourier transform, and then the phase of the signal is corrected using a circle tracking algorithm. The corrected signal is demodulated using a DACM (extended differential cross multiplication) algorithm and finally reconstructed to achieve noise reduction and interference suppression of the vital sign signal.

[0104] Step S103: performing modal function decomposition on the first signal to obtain a plurality of first intrinsic mode functions;

[0105] Specifically, each of the first eigenmode functions mentioned above includes different signal frequencies.

[0106] In a preferred embodiment, the first signal is subjected to modal function decomposition to obtain a plurality of first eigenmode functions, including:

[0107] Adding noise to the first signal to obtain a sixth signal after adding noise;

[0108] According to the CEEMDAN algorithm, the local mean is repeatedly subtracted from the sixth signal until the number of local maximum points and local minimum points is equal, and the integral average of the envelope of the local maximum value and the envelope of the local minimum value is 0, then the subtraction of the local mean is stopped to obtain the second intrinsic mode function and the first residual;

[0109] The local mean is repeatedly subtracted from the first residual and each subsequent residual to obtain an intrinsic mode function and a residual, until the number of decompositions reaches a preset value, the decomposition is stopped, and the first intrinsic mode functions are obtained.

[0110] Specifically, the above noise is relatively small to avoid excessive disturbance of the original signal. After the above noise is added to the first signal as initial noise, it helps to start the iterative process of the CEEMDAN (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise) algorithm, so that the algorithm can decompose the signal more effectively, and can also prevent the algorithm from overfitting to the random fluctuations or noise components in the signal in the initial stage; the initial noise helps to generate a more stable intrinsic mode function in the algorithm's loop process, thereby improving the effect of signal decomposition; the selection of the initial noise level can be combined with the algorithm's stopping criterion to help determine when to stop the loop process to ensure the accuracy of the decomposition; the initial noise helps to improve the algorithm's adaptability to different types of signals and enhance the robustness of the algorithm.

[0111] Preferably, it can also be determined when to stop decomposing the residual based on conditions such as the stability of the IMF, the energy ratio or the predetermined number of decomposition levels.

[0112] In this preferred embodiment, the first signal is subjected to modal function decomposition using the CEEMDAN algorithm to obtain a plurality of first intrinsic mode functions.

[0113] Step S104: performing fast Fourier transform on the first intrinsic mode functions, and calculating the power spectrum density of each first intrinsic mode function;

[0114] In this step, the first intrinsic mode function is subjected to fast Fourier transform to convert the signal from the time domain to the frequency domain, and then the power spectrum density of each first intrinsic mode function in the frequency domain is calculated.

[0115] Step S105: According to the characteristic frequency bands of the respiratory and heartbeat signals and the power spectrum density, a first intrinsic mode function corresponding to the respiratory signal is selected from the plurality of first intrinsic mode functions to obtain a second intrinsic mode function, and a first intrinsic mode function corresponding to the heartbeat signal is selected from the plurality of first intrinsic mode functions to obtain a third intrinsic mode function;

[0116] In a preferred embodiment, according to the characteristic frequency bands of the respiratory and heartbeat signals and the power spectrum density, the first intrinsic mode function corresponding to the respiratory signal is selected from the plurality of first intrinsic mode functions to obtain the second intrinsic mode function, and the first intrinsic mode function corresponding to the heartbeat signal is selected from the plurality of first intrinsic mode functions to obtain the third intrinsic mode function, including:

[0117] According to the characteristic frequency band of the respiratory signal, a first eigenmode function containing the characteristic frequency is selected from the power spectrum density to obtain a second eigenmode function;

[0118] According to the characteristic frequency band of the heartbeat signal, the first intrinsic mode function containing the characteristic frequency is screened out from the power spectrum density to obtain the third intrinsic mode function.

[0119] Specifically, the characteristic frequency band of a respiratory signal is usually 0.12-0.5 Hz, and the characteristic frequency band of a heartbeat signal is usually 1-3 Hz.

[0120] In this preferred embodiment, the first intrinsic mode function containing the characteristic frequency is screened out from the power spectrum density according to the characteristic frequency bands of the respiratory signal and the heartbeat signal, respectively, to obtain the intrinsic mode functions containing the characteristic frequency bands of the respiratory signal and the heartbeat signal, respectively.

[0121] Step S106: reconstructing the second intrinsic mode function to obtain a separated breathing signal, and reconstructing the third intrinsic mode function to obtain a separated heartbeat signal.

[0122] In this step, the second intrinsic mode function and the third intrinsic mode function are respectively merged and reconstructed to obtain the separated breathing signal and heartbeat signal.

[0123] Preferably, the obtained breathing signal and heartbeat signal can also be used for real-time monitoring of human breathing and heartbeat, and the monitoring results are displayed on a user interaction interface so that the user can perform relevant analysis based on the monitoring results in daily life or clinical practice.

[0124] Preferably, after obtaining the separated respiratory signal and heartbeat signal, the accuracy of the obtained respiratory and heartbeat signals can be verified by using the frequency, period and amplitude of the respiratory and heartbeat signals through autocorrelation analysis.

[0125] Specifically, before performing autocorrelation analysis, ensure that the reconstructed signal is pure and minimize the influence of noise. It may be necessary to bandpass filter the signal to isolate the frequency bands of breathing and heartbeat signals. Subtracting the signal from its mean to ensure that the signal fluctuates around zero mean will help the accuracy of the autocorrelation analysis. The autocorrelation function is calculated by the following formula:

[0126]

[0127] Where R(τ) represents the autocorrelation function, N represents the signal length, τ represents the time delay, μ represents the signal mean, and x(t) represents the signal at time t.

[0128] Specifically, the autocorrelation function is a measure of the similarity between a signal at zero delay and itself at other time delays. The autocorrelation function is normalized to the normalized formula:

[0129]

[0130] In the formula, R(τ) normalized represents the normalized autocorrelation function, and (R(0)) represents the autocorrelation function of the signal at zero delay.

[0131] Specifically, its value is 1 at zero delay to facilitate comparison of autocorrelation characteristics of different signals. Peaks are detected in the autocorrelation function, which indicate the periodic components of the signal. According to the peak position of the autocorrelation function, the period and frequency of breathing and heartbeat are estimated. The frequency resolution is determined to distinguish adjacent peaks. If the signal is filtered or modified during preprocessing or reconstruction, the frequency estimate needs to be calibrated to match the characteristics of the original signal. If there are multiple peaks, further analysis may be required to determine which peaks correspond to breathing or heartbeat signals. Statistical analysis is performed on multiple estimates of the autocorrelation function to improve the accuracy and reliability of frequency estimation. Visualize the autocorrelation function and frequency estimation results for easy analysis and interpretation. Verify whether the frequency, amplitude and periodicity obtained by the autocorrelation analysis are consistent with expectations to ensure the accuracy and reliability of signal characteristics. Through the above steps of autocorrelation analysis, the characteristics of breathing and heartbeat signals can be effectively verified, the quality of signal reconstruction can be ensured, and accurate signal characteristics can be provided for subsequent analysis and monitoring; determine which feature parameters are important, such as breathing frequency and heartbeat frequency; and combine the selected feature parameters into a feature vector for subsequent analysis or machine learning models. The construction of feature vectors is an important step in data analysis and machine learning, which involves extracting useful information from raw data and organizing it into a format suitable for model input.

[0132] In another preferred embodiment, in addition to using the above method to obtain the separated respiratory signal and heartbeat signal, a deep learning neural network model is also established to obtain the separated respiratory signal and heartbeat signal.

[0133] Specifically, the data preprocessing module is first used to perform bandpass filtering on the input raw physiological signal to isolate the frequency range of the breathing and heartbeat signals, and then the filtered signal is standardized so that the signal value is in the [0,1] interval.

[0134] Specifically, the above deep learning model includes convolutional layers, recurrent layers, attention mechanism layers, and LSTM layers. Multiple convolutional layers are designed to extract features of different scales. A one-dimensional convolutional layer is used to extract local time domain features of the signal. A one-dimensional convolutional layer can capture local patterns, and stacking multiple convolutional layers can extract more complex features; the recurrent layer can process the sequence data output by the convolutional layer and extract time series features. LSTM or recurrent layers are used to process long-term dependencies in sequence data; the attention mechanism layer is used to introduce the self-attention mechanism in the Transformer architecture to enhance the model's ability to recognize key features of BCG signals; in addition, a feature fusion module is designed to merge the outputs of the convolutional layer and the LSTM layer; a fully connected layer is used at the end of the model to map features to the classification of breathing and heartbeat signals.

[0135] Specifically, the classification cross entropy loss function was used to optimize the model so that it could accurately distinguish signal components; the Adam optimizer was used to update parameters, and the learning rate was set to 0.001; Dropout and L2 regularization were applied to reduce overfitting; the model was trained using a labeled training dataset, and accuracy, precision, recall, and F1 score were used as performance indicators to evaluate the model performance on the validation set; model pruning and quantization techniques were applied to optimize the model to meet the needs of real-time applications; the algorithm was validated in a clinical setting to ensure its effectiveness and accuracy in practical applications; clinical data were collected and professional medical personnel were worked with to evaluate the performance of the algorithm.

[0136] Specifically, the model is trained using the training data set to learn the patterns and features in the data. During the training process, the model performance is regularly evaluated on the validation set (a portion of the data that has not participated in the training). Performance indicators such as accuracy, precision, recall, and F1 score are used to measure the model performance. The performance of the model on the validation set is analyzed to identify the deficiencies of the model, such as overfitting, underfitting, or bias problems. Based on the results of the performance analysis, hyperparameters such as learning rate, batch size, number of iterations, and choice of optimizer are adjusted. Hyperparameters are systematically adjusted using methods such as Grid Search, Random Search, or Bayesian optimization. The model structure is adjusted as needed, such as adding or reducing the number of layers, adjusting the parameters of the layers (such as convolution kernel size, number of filters, etc.). Consider introducing regularization techniques (such as Dropout, L2 regularization) to prevent overfitting. The training and validation process is repeated until the model performance reaches a satisfactory level or further improvement becomes negligible. In order to more comprehensively evaluate the stability and generalization ability of the model, a cross-validation method can be used. If the model will be used in real-time applications, it is also necessary to optimize the model to meet the needs of real-time processing, such as model pruning, quantization, etc. Record the performance changes after each adjustment, compare the model performance under different configurations, and select the best model. Evaluate the adjusted model on the final test set to ensure that the model has good generalization ability. Once the model adjustment is completed and the performance is satisfactory, prepare the model to be deployed in actual applications. Through this process, it can be ensured that the model has good generalization ability and adaptability while maintaining high accuracy to cope with various situations in actual applications.

[0137] Specifically, the following are the steps to implement the construction of feature vectors: Select features Determine feature importance: Determine which signal features are important for the analysis task based on domain knowledge. For example, in physiological signal analysis, respiratory rate and heart rate are key features; Apply statistical tests (such as t-test, ANOVA) to evaluate the correlation between different features and target variables (such as health status); Select the most relevant features based on test results and the opinions of domain experts; If the number of features is large, dimensionality reduction techniques such as principal component analysis (PCA) or linear discriminant analysis (LDA) can be used to reduce the number of features while retaining as much information as possible. Dimensionality reduction techniques are very useful in machine learning and data analysis, especially when dealing with high-dimensional data sets. They can help reduce computational complexity, improve model training efficiency, and help avoid overfitting. Through dimensionality reduction, the most important information can be extracted from the original features to form a new, lower-dimensional feature space that can largely retain the variability of the original data. Create new features from the original features, such as calculating the power spectral density, energy, entropy, etc. of the signal; ensure that the data type and scale of the selected features are consistent, and perform necessary normalization or standardization; combine all selected features of each sample into a vector; encapsulate the feature vector in a data structure suitable for data processing and model training, such as an array, matrix, or data frame.

[0138] Specifically, DataFrame features: can store data columns of different types (numeric, categorical, Boolean, etc.); supports row and column indexes for easy access and manipulation of data; provides a wealth of data manipulation functions, such as sorting, filtering, grouping, merging, etc.; supports missing data processing, can automatically process or fill missing values; can be easily integrated with other data processing libraries (such as NumPy, SciPy, Scikit-learn). In DataFrame, you can perform the following operations: data cleaning: processing missing values, outliers, etc.; feature engineering: creating new features or transforming existing features; feature scaling: standardizing or normalizing feature values; data set partitioning: dividing data into training sets, validation sets, and test sets.

[0139] Specifically, the encapsulation process is as follows: add the feature vector of each sample as a row of data to the DataFrame; each column represents a feature, such as breathing rate, heart rate, etc.; if there are multiple samples, the feature vector of each sample will become a row in the DataFrame;

[0140] Specifically, the encapsulated DataFrame can be directly used for training machine learning models. Before training, it is usually necessary to separate the feature data (columns of DataFrame) from the target variable (label) and ensure that they have the same row index; if it is a supervised learning task, it is necessary to ensure that each feature vector has a corresponding label or result. In DataFrame, these labels are usually stored as a separate column and used for model training together with the feature data; by using DataFrame as a data structure, the method of the present invention can efficiently process and organize feature data, which provides convenience for the training and evaluation of machine learning models. This encapsulation method not only improves the efficiency of data processing, but also enhances the accuracy of model training and prediction; if it is a supervised learning task, ensure that each feature vector has a corresponding label or result; divide the data set into a training set, a validation set, and a test set to evaluate the model performance; according to the performance during the model training process, it may be necessary to return and optimize the feature selection; design an algorithm to achieve real-time signal processing and analysis to meet the needs of real-time monitoring. Feedback adjustment: According to the real-time monitoring results, dynamically adjust the algorithm parameters to optimize the signal processing effect; if necessary, the millimeter wave radar data can be fused with other sensor data (such as a contact heart rate monitor) to improve the accuracy of signal detection.

[0141] Based on the above method embodiment, the present invention provides a corresponding device embodiment.

[0142] like Figure 4 As shown, an embodiment of the present invention provides a human breathing and heartbeat signal separation device, including: a signal acquisition module, a signal noise reduction module, a signal decomposition module, a power spectrum density calculation module, a signal screening module and a signal reconstruction module;

[0143] The signal acquisition module is used to acquire the vital sign signal of the selected person object in the detection area;

[0144] The signal noise reduction module is used to perform noise reduction and interference suppression on the vital sign signal according to a noise reduction algorithm to obtain a first signal;

[0145] The signal decomposition module is used to perform modal function decomposition on the first signal to obtain a plurality of first intrinsic mode functions;

[0146] The power spectrum density calculation module is used to perform fast Fourier transform on the first intrinsic mode functions and calculate the power spectrum density of each first intrinsic mode function;

[0147] The signal screening module is used to screen out the first intrinsic mode function corresponding to the breathing signal from the plurality of first intrinsic mode functions according to the characteristic frequency bands of the breathing and heartbeat signals and the power spectrum density to obtain the second intrinsic mode function, and to screen out the first intrinsic mode function corresponding to the heartbeat signal from the plurality of first intrinsic mode functions to obtain the third intrinsic mode function;

[0148] The signal reconstruction module reconstructs the second intrinsic mode function to obtain a separated breathing signal, and reconstructs the third intrinsic mode function to obtain a separated heartbeat signal.

[0149] It should be noted that the device embodiments described above are merely illustrative, wherein the modules described above as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art may understand and implement it without paying creative labor. The above schematic diagram is only an example of a human respiratory and heartbeat signal separation device, and does not constitute a limitation on the human respiratory and heartbeat signal separation device, and may include more or fewer components than shown in the figure, or combine certain components, or different components.

[0150] Based on the above method item embodiments, the present invention provides corresponding terminal device item embodiments.

[0151] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the above memory and configured to be executed by the above processor, and when the above processor executes the above computer program, a method for separating human respiration and heartbeat signals described in any embodiment of the present invention is implemented.

[0152] Exemplarily, in this embodiment, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to implement the present invention. The one or more module elements may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program in the device;

[0153] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The device may include, but is not limited to, a processor, a memory;

[0154] The processor referred to herein may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The above-mentioned processor is the control center of the above-mentioned device, and uses various interfaces and lines to connect various parts of the entire device;

[0155] The above-mentioned memory can be used to store the above-mentioned computer programs and / or modules. The above-mentioned processor realizes various functions of the above-mentioned device by running or executing the computer programs and / or modules stored in the above-mentioned memory, and calling the data stored in the memory. The above-mentioned memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; in addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0156] Based on the above method item embodiments, the present invention provides a corresponding storage medium item embodiment.

[0157] Another embodiment of the present invention provides a storage medium, the storage medium includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute a method for separating human respiration and heartbeat signals as described in any embodiment of the present invention.

[0158] In this embodiment, the storage medium is a computer-readable storage medium, the computer program includes computer program code, and the computer program code may be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0159] Compared with the prior art, by implementing the above-mentioned embodiments of the present invention, the respiration and heartbeat signals can be separated.

[0160] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for separating human breathing and heartbeat signals, characterized in that: include: Acquire vital sign signals of selected human objects in the detection area; According to a noise reduction algorithm, noise reduction and interference suppression are performed on the vital sign signal to obtain a first signal; Performing modal function decomposition on the first signal to obtain a plurality of first intrinsic mode functions; Performing fast Fourier transform on the plurality of first intrinsic mode functions, and calculating the power spectrum density of each first intrinsic mode function; According to the characteristic frequency bands of the breathing and heartbeat signals and the power spectral density, a first intrinsic mode function corresponding to the breathing signal is selected from the plurality of first intrinsic mode functions to obtain a second intrinsic mode function, and a first intrinsic mode function corresponding to the heartbeat signal is selected from the plurality of first intrinsic mode functions to obtain a third intrinsic mode function; The second intrinsic mode function is reconstructed to obtain a separated breathing signal, and the third intrinsic mode function is reconstructed to obtain a separated heartbeat signal.

2. A method for separating human breathing and heartbeat signals according to claim 1, characterized in that: The step of obtaining the vital sign signal of the selected person object in the detection area includes: Acquire real-time signals of multiple human objects in the detection area; Filtering and denoising the real-time signal to obtain a third signal after filtering and denoising; Separating the third signal according to a multi-target signal separation algorithm to obtain a plurality of separated signals and a first feature; Inputting each separated signal into a preset multi-target separation model, so that the multi-target separation model determines the human object corresponding to each separated signal according to the first feature; According to the person object corresponding to each separated signal, the separated signal corresponding to the selected person object is selected to obtain the vital sign signal of the selected person object.

3. A method for separating human breathing and heartbeat signals according to claim 2, characterized in that: The construction of the multi-target separation model includes: Acquire a number of separation signal samples with real labels; wherein the real labels are used to represent the person object corresponding to each separation signal sample; For each separated signal sample, the separated signal sample is input into the multi-target separation model to be trained, so that each separated sample is identified as a corresponding person object by the multi-target separation model, and a loss function is calculated according to the recognition result and the true label; After each loss function value is calculated, it is determined whether the loss function converges. If so, the training of the multi-target separation model to be trained is completed; otherwise, the parameters in the multi-target separation model to be trained are adjusted and the training continues.

4. A method for separating human breathing and heartbeat signals according to claim 3, characterized in that: The step of separating the third signal according to the multi-target signal separation algorithm to obtain a plurality of separated signals and the first feature includes: Decomposing the third signal at multiple scales to obtain signals at different time scales and the first feature of each time scale signal; The signals of different time scales are merged, and the merged signal is separated by an adaptive ICA algorithm to obtain the plurality of separated signals.

5. A method for separating human breathing and heartbeat signals according to claim 4, characterized in that: The step of performing noise reduction and interference suppression on the vital sign signal according to a noise reduction algorithm to obtain a first signal includes: Performing chest position detection on the vital sign signal according to fast Fourier transform to obtain the chest position, and then determining a fourth signal corresponding to the chest according to the chest position; Correcting the phase of the fourth signal according to a circle tracking algorithm to obtain a corrected fourth signal; Demodulating the corrected fourth signal according to the DACM algorithm, and performing noise elimination on the demodulated signal using singular spectrum analysis to obtain a fifth signal after noise elimination; The fifth signal is reconstructed to obtain the first signal.

6. A method for separating human breathing and heartbeat signals according to claim 5, characterized in that: The step of performing modal function decomposition on the first signal to obtain a plurality of first intrinsic mode functions includes: adding noise to the first signal to obtain a sixth signal after the noise is added; According to the CEEMDAN algorithm, the local mean is repeatedly subtracted from the sixth signal until the number of local maximum points and local minimum points is equal, and the integral average of the envelope of the local maximum value and the envelope of the local minimum value is 0, then the subtraction of the local mean is stopped to obtain the second intrinsic mode function and the first residual; The local mean is repeatedly subtracted from the first residual and each subsequent residual to obtain an intrinsic mode function and a residual, until the number of decompositions reaches a preset value, the decomposition is stopped, and the first intrinsic mode functions are obtained.

7. A method for separating human breathing and heartbeat signals according to claim 6, characterized in that: The method comprises: filtering out a first intrinsic mode function corresponding to the respiratory signal from the plurality of first intrinsic mode functions according to the characteristic frequency bands of the respiratory and heartbeat signals and the power spectrum density to obtain a second intrinsic mode function; filtering out a first intrinsic mode function corresponding to the heartbeat signal from the plurality of first intrinsic mode functions to obtain a third intrinsic mode function, comprising: According to the characteristic frequency band of the respiratory signal, a first intrinsic mode function containing the characteristic frequency is selected from the power spectrum density to obtain a second intrinsic mode function; According to the characteristic frequency band of the heartbeat signal, a first intrinsic mode function containing the characteristic frequency is screened out from the power spectrum density to obtain a third intrinsic mode function.

8. A human breathing and heartbeat signal separation device, characterized in that: include: Signal acquisition module, signal noise reduction module, signal decomposition module, power spectrum density calculation module, signal screening module and signal reconstruction module; The signal acquisition module is used to acquire the vital sign signal of the selected human object in the detection area; The signal noise reduction module is used to perform noise reduction and interference suppression on the vital sign signal according to a noise reduction algorithm to obtain a first signal; The signal decomposition module is used to perform modal function decomposition on the first signal to obtain a plurality of first intrinsic mode functions; The power spectrum density calculation module is used to perform fast Fourier transform on the plurality of first intrinsic mode functions and calculate the power spectrum density of each first intrinsic mode function; The signal screening module is used to screen out a first intrinsic mode function corresponding to the breathing signal from the plurality of first intrinsic mode functions according to the characteristic frequency bands of the breathing and heartbeat signals and the power spectrum density to obtain a second intrinsic mode function, and to screen out a first intrinsic mode function corresponding to the heartbeat signal from the plurality of first intrinsic mode functions to obtain a third intrinsic mode function; The signal reconstruction module reconstructs the second intrinsic mode function to obtain a separated breathing signal, and reconstructs the third intrinsic mode function to obtain a separated heartbeat signal.

9. A terminal device, characterized in that: It comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, it implements a method for separating human respiration and heartbeat signals as described in any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute a method for separating human respiration and heartbeat signals as described in any one of claims 1 to 7.