A method for monitoring life sleep signs

Abstract

The present application provides a method for monitoring vital signs during sleep, which relates to the technical field of vital sign data monitoring and includes the following steps: collecting the original echo signal in the monitoring area, and extracting the respiration signal, the heartbeat signal, and the body movement signal corresponding to the body movement event from the original echo signal; obtaining the correlation characteristics between the respiration signal and the heartbeat signal, and the adjustment characteristics of the body movement event on the respiration signal and / or the heartbeat signal; generating a biometric feature representing individual physiological differences according to the correlation characteristics and the adjustment characteristics; and obtaining the target identity probability distribution based on the biometric feature; performing interference suppression processing on the original echo signal according to the target identity probability distribution to separate the target respiration waveform and the target heartbeat waveform of the target individual from the original echo signal. This method is applicable to practical application scenarios such as multi-person co-bed monitoring and home bedroom monitoring, breaking through the technical bottleneck that the monitoring accuracy decreases as the number of targets increases in the prior art.

Description

Technical Field

[0001] This application relates to the technical field of vital sign data monitoring, and specifically to a method for monitoring vital signs during sleep. Background Art

[0002] In the field of contactless sleep monitoring technology, existing methods mainly collect human vital sign signals (such as respiration, heartbeat, and body movement signals) through biometric radar, and perform sleep state analysis based on signal processing algorithms. For example, the invention patent with the publication number CN117598664A mainly collects human physiological signals (such as respiration and heartbeat) through millimeter-wave radar, and realizes sleep parameter analysis and abnormal alarm based on signal separation and feature extraction. However, when using this method, when there are multiple target individuals in the monitoring area (such as multiple people sharing the same bed or sharing the monitoring space), the radar echo signal will be aliased and interfered with the respiration and heartbeat signals of the target individual due to the presence of multiple bodies in the monitoring space, reducing the monitoring accuracy. Summary of the Invention

[0003] In view of the above-mentioned defects or deficiencies in the prior art, this application aims to provide a method for monitoring vital signs during sleep to improve the monitoring accuracy;

[0004] The method includes the following steps:

[0005] Collect the original echo signal of the monitoring area, and extract the respiration signal, heartbeat signal, and body movement signal corresponding to the body movement event from the original echo signal;

[0006] Obtain the correlation characteristics between the respiration signal and the heartbeat signal, and the regulation characteristics of the body movement event on the respiration signal and / or the heartbeat signal;

[0007] Generate a biometric feature representing individual physiological differences based on the correlation characteristics and the regulation characteristics; and obtain the target identity probability distribution based on the biometric feature;

[0008] Perform interference suppression processing on the original echo signal according to the target identity probability distribution to separate the target respiration waveform and target heartbeat waveform of the target individual from the original echo signal.

[0009] According to the technical solution provided by this application, the obtaining of the correlation characteristics between the respiration signal and the heartbeat signal, and the regulation characteristics of the body movement event on the respiration signal and / or the heartbeat signal includes the following steps:

[0010] Calculate the heart rate variability during the respiration cycle and the respiration amplitude-heart rate correlation coefficient to obtain the correlation characteristics;

[0011] Establish a criterion for body movement-respiration coordination, analyze the stability index of the respiration signal and the heart rate recovery rate before and after the body movement event to obtain adjustment characteristics.

[0012] According to the technical solution provided by the present application, a dual-channel adaptive separation algorithm is used to extract the respiration signal and the heart rate signal from the original echo signal, including the following steps:

[0013] Construct a principal component channel of the respiration signal based on empirical mode decomposition, and retain the components in the frequency band of 0.1 - 0.8 Hz by screening the intrinsic mode functions;

[0014] Establish an auxiliary channel of the heart rate signal based on complex wavelet transform, and perform phase decoupling on the frequency band of 0.8 - 3 Hz using the Morlet wavelet basis function;

[0015] Evaluate the orthogonality of the dual-channel signals of the principal component channel and the auxiliary channel through mutual information entropy, and dynamically adjust the parameter weights between the channels to achieve signal separation, so as to obtain the respiration signal and the heart rate signal.

[0016] According to the technical solution provided by the present application, extract the body movement signal corresponding to the body movement event from the original echo signal, including the following steps:

[0017] Perform energy gradient analysis on adjacent signal frames of the original echo signal, locate the body movement candidate interval, and generate a body movement event marker;

[0018] Perform non-stationary feature decoupling on the marker signal of the body movement event, and separate the feature components in the frequency band of 2 - 15 Hz representing limb movement;

[0019] Based on the motion trajectory discrimination criterion, distinguish local body movement from global interference, and fuse the frequency band feature components to generate a body movement intensity waveform, and the body movement intensity waveform is the body movement signal.

[0020] According to the technical solution provided by the present application, the step of performing energy gradient analysis on adjacent signal frames of the original echo signal, locating the body movement candidate interval, and generating a body movement event marker includes the following steps:

[0021] Calculate the difference sequence of the Teager energy operator of adjacent signal frames of the original echo signal through a sliding time window;

[0022] Obtain a dynamic decision threshold according to the real-time signal characteristics;

[0023] According to the dynamic decision threshold, obtain the starting time point and the duration range of the body movement event;

[0024] Locate the body movement candidate interval with the starting time point and the duration range of the body movement event, and generate a body movement event marker.

[0025] According to the technical solution provided by the present application, differentiating local body movement from global interference based on the motion trajectory discrimination criterion and fusing the frequency band characteristic components to generate a body movement intensity waveform includes the following steps:

[0026] Construct a three-dimensional motion trajectory map according to the Doppler phase change, and calculate the Mahalanobis distance between each trajectory point and the historical motion reference trajectory;

[0027] When the cumulative Mahalanobis distance of consecutive trajectory points exceeds the dynamic interference threshold, it is determined as a local body movement event and the spatial positioning coordinates are output;

[0028] Perform spatio-temporal alignment on the instantaneous energy of the frequency band characteristic components and the spatial positioning coordinates to generate a body movement intensity waveform that is continuous in the time domain.

[0029] According to the technical solution provided by the present application, after obtaining the correlation characteristics of the respiration signal and the heartbeat signal, and the regulation characteristics of the body movement event on the respiration signal and / or the heartbeat signal, before generating a biometric characteristic representing individual physiological differences according to the correlation characteristics and the regulation characteristics, the following steps are further included:

[0030] Judge the size of the duration range of the body movement event and a preset threshold, and judge whether there is an overlap of multi-target reflection cross-sections;

[0031] Generating a biometric characteristic representing individual physiological differences according to the correlation characteristics and the regulation characteristics includes the following steps:

[0032] If the duration range of the body movement event is less than or equal to the preset threshold and there is no overlap of multi-target reflection cross-sections, then generate an initial characteristic representing individual physiological differences according to the correlation characteristics and the regulation characteristics, and use the initial characteristic as the biometric characteristic.

[0033] According to the technical solution provided by the present application, after judging the size of the duration range of the body movement event and a preset threshold, and judging whether there is an overlap of multi-target reflection cross-sections, the following steps are further included:

[0034] If the duration range of the body movement event is greater than the preset threshold, or there is an overlap of multi-target reflection cross-sections, then adopt a two-channel adaptive separation algorithm to perform secondary decoupling processing on the respiration signal and the heartbeat signal;

[0035] Based on the three-dimensional motion trajectory map, exclude the overlapping area of the reflection cross-section corresponding to the global interference, and re-extract the processed respiration signal and the processed heartbeat signal;

[0036] Based on the processed respiration signal, obtain the processed correlation features of the target individual; based on the processed heartbeat signal, obtain the processed adjustment features; the processed correlation features and the processed adjustment features are used to optimize the initial features.

[0037] According to the technical solution provided by the present application, after the secondary decoupling process of the respiration signal and the heartbeat signal, the following steps are further included:

[0038] Verify the stability of the signal energy distribution after the secondary decoupling process;

[0039] If the verification is successful, based on the processed correlation features and the processed adjustment features, correct the initial features to obtain the processed features, and use the processed features as biometric features.

[0040] According to the technical solution provided by the present application, the interference suppression process of the original echo signal according to the target identity probability distribution to separate the target respiration waveform and the target heartbeat waveform of the target individual from the original echo signal includes the following steps:

[0041] Perform weighted filtering on the original echo signal according to the target identity probability distribution to suppress the respiration and heartbeat interference components of non-target individuals and obtain the filtered signal;

[0042] Perform variational mode decomposition on the filtered signal to extract the target respiration waveform and the target heartbeat waveform.

[0043] Compared with the prior art, the beneficial effects of the present application are as follows: By extracting the correlation features of respiration and heartbeat signals and combining the dynamic adjustment features of body movement events on physiological signals, the present application constructs a multi-dimensional biometric recognition system to solve the problem of physiological signal aliasing interference in the scenario of multiple people coexisting. Among them, the correlation features can capture the unique respiration-heartbeat coupling rules of different individuals, while the body movement adjustment features can reflect the adaptive characteristics of physiological signals under movement interference. The two work together to significantly improve the recognition rate of multi-target physiological features. In addition, by converting the individual differences of physiological features into probability distribution parameters, accurate suppression of interference signals from non-target individuals can be achieved. Compared with traditional signal separation methods that rely on spatial positioning or spectral analysis, signal separation in the same frequency band can be realized. In summary, while ensuring continuous monitoring of the vital signs of the target individual, this method can also adaptively process dynamic changes of people in the monitoring area (such as complex scenarios such as body posture flipping and position movement), improve the separation accuracy of the target respiration / heartbeat waveform, and is applicable to practical application scenarios such as multi-person co-bed monitoring and home bedroom monitoring, breaking through the technical bottleneck that the monitoring accuracy decreases as the number of targets increases in the prior art. Description of the Drawings

[0044] Figure 1 Flow chart of steps of the method for monitoring vital signs during sleep provided by this application. Detailed implementation mode

[0045] The following further elaborates on this application in conjunction with the accompanying drawings and embodiments. It can be understood that the specific embodiments described herein are merely used to explain the relevant invention, rather than limiting the invention. Additionally, it should be noted that for ease of description, only parts related to the invention are shown in the drawings.

[0046] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will elaborate on this application in detail with reference to the drawings and embodiments.

[0047] Embodiment 1

[0048] Just as mentioned in the background art, in response to the problems in the prior art, this application proposes a method for monitoring vital signs during sleep, as Figure 1 shown, including the following steps:

[0049] S1. Collect the original echo signal of the monitoring area, and extract the respiration signal, heartbeat signal, and body movement signal corresponding to the body movement event from the original echo signal;

[0050] Specifically, the original echo signal is a signal containing various information reflected from the monitoring area after the monitoring device emits a signal. It is received by the receiving device of the monitoring device (such as a radar, etc.), and the body movement event is an event of body movement, such as turning over, etc.

[0051] Furthermore, the method for extracting the respiration signal and heartbeat signal from the original echo signal by using a two-channel adaptive separation algorithm includes the following steps:

[0052] Construct the principal component channel of the respiration signal based on empirical mode decomposition, and retain the frequency band components of 0.1 - 0.8 Hz by screening the intrinsic mode functions;

[0053] Specifically, empirical mode decomposition (EMD for short) is a method for decomposing a complex signal into multiple intrinsic mode functions (IMF for short). IMF is a function that satisfies two conditions: one is that the number of extreme points and the number of zero-crossing points must be equal or differ by at most one over the entire data length; the other is that at any moment, the mean value of the upper envelope line formed by the local maximum points and the lower envelope line formed by the local minimum points is zero.

[0054] The specific implementation process is as follows: Perform EMD decomposition on the original echo signal to obtain a series of IMF components. For example, assume that the original echo signal is a complex signal containing respiration, heartbeat, and other noises. After EMD decomposition, multiple components such as IMF1, IMF2, IMF3, etc. will be obtained. Screen these IMF components and retain the components with frequencies in the 0.1 - 0.8 Hz frequency band. Because the frequency of the respiration signal is usually within this range. For example, if the frequency range of IMF2 is 0.15 - 0.7 Hz and the frequency range of IMF3 is 0.2 - 0.6 Hz, then retain IMF2 and IMF3 and combine them to construct the principal component channel of the respiration signal.

[0055] Establish an auxiliary channel for the heartbeat signal based on complex wavelet transform, and use Morlet wavelet basis function to perform phase decoupling on the 0.8 - 3 Hz frequency band;

[0056] Specifically, complex wavelet transform (Complex Wavelet Transform, abbreviated as CWT) is a transformation method developed on the basis of wavelet transform. Compared with real wavelet transform, it has better direction selectivity and phase information retention ability. Morlet wavelet basis function is a commonly used wavelet basis function, and its shape is similar to a complex sine wave modulated by a Gaussian. In this step, use Morlet wavelet basis function to analyze a specific frequency band. Phase decoupling means separating the phase information mixed in the signal to more accurately extract a specific signal.

[0057] The specific implementation process is as follows: Perform complex wavelet transform on the original echo signal and select Morlet wavelet basis function. For example, set parameters such as the center frequency and bandwidth of the Morlet wavelet basis function to make it suitable for the analysis of the 0.8 - 3 Hz frequency band. For the transformed signal, perform phase decoupling operation within the 0.8 - 3 Hz frequency band. Since the frequency of the heartbeat signal is usually within this frequency band, through phase decoupling, the heartbeat signal can be more clearly separated from other mixed signals, thereby establishing an auxiliary channel for the heartbeat signal.

[0058] Evaluate the orthogonality of the dual-channel signals of the principal component channel and the auxiliary channel through mutual information entropy, and dynamically adjust the parameter weights between channels to achieve signal separation, so as to obtain the respiration signal and the heartbeat signal.

[0059] Specifically, mutual information entropy is used in this scenario to evaluate the independence between the principal component channel of the respiration signal and the signal of the auxiliary channel of the heartbeat signal. If the signals of the two channels are completely orthogonal (independent), the mutual information entropy is zero; the larger the mutual information entropy, the stronger the correlation between the signals of the two channels. The parameter weights refer to assigning different weight values to the signals of the principal component channel and the auxiliary channel when separating the signals. By adjusting these weight values, the finally separated respiration signal and heartbeat signal can be made more accurate.

[0060] The specific implementation process is as follows: Calculate the mutual information entropy of the dual-channel signals of the principal component channel and the auxiliary channel. For example, using a specific mutual information entropy calculation algorithm, taking the signal data of the principal component channel and the signal data of the auxiliary channel as inputs, a value of mutual information entropy is obtained. Suppose this value is 0.2 (the larger this value, the stronger the correlation). Dynamically adjust the parameter weights between the channels according to the value of the mutual information entropy. If the value of the mutual information entropy is large, it indicates that there is a strong correlation between the signals of the two channels, and the weights need to be adjusted to make the signals of the two channels as independent as possible. For example, through an optimization algorithm such as the gradient descent algorithm, continuously adjust the weights of the principal component channel and the auxiliary channel to gradually reduce the mutual information entropy. Suppose the initial weights are the weight w1 = 0.5 of the principal component channel and the weight w2 = 0.5 of the auxiliary channel. After multiple iterative adjustments, w1 = 0.6 and w2 = 0.4 are obtained. At this time, the mutual information entropy is reduced to a relatively small value, such as 0.05. Based on the adjusted weights, perform weighted combination on the signals of the principal component channel and the auxiliary channel to separate the respiration signal and the heartbeat signal. For example, respiration signal = principal component channel signal × w1, heartbeat signal = auxiliary channel signal × w2, and finally the accurately separated respiration signal and heartbeat signal are obtained.

[0061] Exemplarily, in a smart home sleep monitoring device, when the device receives the original echo signal, it uses the dual-channel adaptive separation algorithm. For example, in a home bedroom environment, there may be electromagnetic interference generated by electrical appliances such as TVs and computers. This algorithm can effectively separate the respiration signal and the heartbeat signal of the human body from the complex original echo signal, providing accurate sleep monitoring data for the user.

[0062] Furthermore, extracting the body movement signal corresponding to the body movement event from the original echo signal includes the following steps:

[0063] Perform energy gradient analysis on adjacent signal frames of the original echo signal to locate the body movement candidate interval and generate body movement event markers;

[0064] Specifically, the continuous original echo signal is divided according to a certain time length, and each segment is called a signal frame. For example, if the original echo signal is divided in units of 1 second, then the signal of each 1 second is a signal frame. Energy gradient analysis is to calculate the change rate of energy between adjacent signal frames. The energy of a signal can be calculated by the sum of the squares of the amplitudes of all sampling points within the signal frame. For example, for the nth signal frame Sn and the (n + 1)th signal frame Sn+1, first calculate their energies En and En+1 respectively, and the energy gradient G = (En+1 - En) / Δt, where Δt is the time interval between the two signal frames. When the energy gradient exceeds a certain threshold, the time period where the adjacent signal frames are located is considered to possibly have body movement, and this time period is the body movement candidate interval. Assume that the energy gradient threshold is set as T. If G > T, then the time period from the start of the nth signal frame to the end of the (n + 1)th signal frame is a body movement candidate interval. Marking is performed on the body movement candidate interval for subsequent further analysis. For example, in an original echo signal containing multiple body movement candidate intervals, the signal part corresponding to each body movement candidate interval is marked as "possibly having body movement".

[0065] Perform non-stationary feature decoupling on the marked signal of the body movement event to separate the feature components in the 2 - 15 Hz frequency band representing limb movement;

[0066] Specifically, the body movement signal in the original echo signal often has non-stationary characteristics, that is, the statistical characteristics of the signal (such as mean, variance, etc.) change with time. Non-stationary feature decoupling is to decompose this complex non-stationary signal into components with different characteristics. Some signal processing methods can be used, such as time-frequency analysis methods (short-time Fourier transform, wavelet transform, etc.). For example, use the short-time Fourier transform to convert the signal marked with the body movement event from the time domain to the frequency domain, and observe the frequency component distribution at different time points. The signal frequency generated by human limb movement usually concentrates in the frequency band range of 2 - 15 Hz. By analyzing the signal after non-stationary feature decoupling, the feature components within this frequency band are extracted. For example, after the short-time Fourier transform, it is observed in the spectrogram that there is obvious energy distribution in the 2 - 15 Hz frequency band, and the signal components corresponding to this part of the frequency are extracted as the frequency band feature components.

[0067] Distinguish local body movement and global interference based on the motion trajectory discrimination criterion, and fuse the frequency band feature components to generate a body movement intensity waveform, and the body movement intensity waveform is the body movement signal.

[0068] Specifically, the motion trajectory discrimination criterion is to judge whether the body movement is local (various movements during individual sleep, such as arm swinging) or global interference (such as overall human body movement) based on some characteristics of the body movement signal, such as the amplitude change law, frequency change law and duration of the signal. For example, if the duration of the body movement signal is short and the amplitude change is more drastic, it may be local body movement; if the duration is long and the amplitude change is relatively gentle, it may be global interference. The 2-15Hz frequency band feature components separated previously are combined according to certain rules to obtain the fused frequency band feature components. For example, for the 2-15Hz frequency band feature components corresponding to multiple body movement event markers, the amplitude of each component can be weighted and superimposed according to the time order of their occurrence. The weight can be determined according to the credibility of the body movement event in which the component is located (for example, the body movement event with a larger energy gradient has a higher corresponding component weight). The body movement intensity waveform is a waveform that changes over time after fusing the frequency band feature components, and its amplitude indicates the intensity of the body movement. For example, after weighted superposition, a time-amplitude waveform is obtained. This waveform is the body motion intensity waveform, which can intuitively reflect the intensity changes of body motion. This body motion intensity waveform is the body motion signal that is finally extracted.

[0069] For example, during sleep monitoring, individuals will have various movements during sleep. By performing energy gradient analysis on the original echo signal, when an individual turns over, energy changes will appear in the original echo signal, thereby locating and marking the candidate intervals of body movement. Then, through non-stationary feature decoupling and motion trajectory discrimination criteria, it can be determined that this is the individual's local body movement, and a body movement intensity waveform can be generated.

[0070] This implementation can accurately identify body movement events, distinguish between local body movement and global interference, generate a waveform reflecting the intensity of body movement, and provide accurate data for subsequent analysis of the impact of body movement on vital signs.

[0071] S2, acquiring the correlation characteristics between the respiratory signal and the heartbeat signal, and the regulation characteristics of the body motion event on the respiratory signal and / or the heartbeat signal;

[0072] Specifically, the correlation feature is a feature that reflects the correlation between the respiratory signal and the heartbeat signal. The regulation feature is a feature that the body motion event regulates the respiratory signal and / or the heartbeat signal.

[0073] Further, the acquiring of the correlation characteristics between the respiratory signal and the heartbeat signal, and the regulation characteristics of the body motion event on the respiratory signal and / or the heartbeat signal, comprises the following steps:

[0074] Calculate the heart rate rhythm variability and the respiratory amplitude-heart rate correlation coefficient within the respiratory cycle to obtain the correlation characteristics;

[0075] Specifically, the respiratory cycle is the time interval from the start of one inspiration to the start of the next inspiration. It can be determined by analyzing the previously extracted respiratory signal. For example, on the waveform diagram of the respiratory signal, find the time interval between two adjacent peaks (or troughs), and this interval is one respiratory cycle. Heart rate variability (HRV) refers to the time variance between successive heartbeats, which reflects the activity of the cardiac autonomic nervous system and its regulatory effect on the sinoatrial node of the heart. A common calculation method is to calculate the standard deviation of the adjacent heartbeats interval (RR interval). By calculating the HRV within each respiratory cycle, the changes in the heart rate rhythm during the respiratory process can be understood. The respiratory amplitude-heart rate correlation coefficient is used to measure the degree of association between the changes in respiratory amplitude and heart rate. The respiratory amplitude can be represented by the amplitude of the respiratory signal, and the heart rate is calculated from the heartbeat signal. The Pearson correlation coefficient can be used to calculate the correlation coefficient between them. By calculating this coefficient, it can be known to what extent the changes in respiratory amplitude are related to the changes in heart rate, thereby obtaining the correlation characteristics of the respiratory signal and the heartbeat signal.

[0076] Establish a body movement-respiration coordination criterion, and analyze the stability index of the respiratory signal and the heart rate recovery rate before and after the body movement event to obtain the regulation characteristics.

[0077] Specifically, the body movement-respiration coordination criterion is a standard for judging whether body movement and respiration are coordinated. It can be established according to the changes in respiratory frequency and respiratory amplitude when body movement events occur. For example, if the respiratory frequency suddenly increases or decreases when body movement occurs, and the respiratory amplitude also changes significantly, which is quite different from the breathing pattern in the normal state, it is considered that the coordination between body movement and respiration is poor; on the contrary, if the changes in respiratory frequency and amplitude are relatively small and remain within a certain normal range, it is considered that the coordination between body movement and respiration is good. Some specific thresholds can be set to quantify this coordination, such as if the respiratory frequency changes by more than ±20% of the normal mean, and the respiratory amplitude changes by more than ±30% of the normal mean, it is determined that body movement and respiration are not coordinated. The stability index of the respiratory signal is used to measure the stability of the respiratory signal over a period of time. It can be obtained by calculating the standard deviation, coefficient of variation and other statistics of the respiratory signal. By comparing the stability index of the respiratory signal before and after the body movement event, the influence of body movement on the stability of the respiratory signal can be understood. If the stability index of the respiratory signal after the body movement event increases significantly, it means that the body movement has caused a great interference to the stability of the respiratory signal. The heart rate recovery rate is the speed at which the heartbeat recovers from the disturbed state to the normal state after a body motion event occurs. It can be measured by calculating the time required for the heart rate to recover to the pre-body motion level after a period of time after the body motion event ends. For example, the heart rate before the body motion event is 60 beats / minute, the heart rate rises to 80 beats / minute during the body motion event, and after 30 seconds after the body motion event ends, the heart rate recovers to 60 beats / minute. Then the heart rate recovery rate can be expressed as (80-60) beats / minute ÷ 30 seconds = 0.67 beats / second. By analyzing the heart rate recovery rate, we can obtain the regulatory characteristics of the body motion event on the heartbeat signal and understand the heart's ability to recover normal rhythm after body motion interference.

[0078] This implementation can provide an in-depth understanding of the intrinsic connection between breathing and heartbeat and the impact of body movement on them, providing more basis for accurately assessing the physiological state of an individual, and collaboratively distinguishing target individuals from non-target individuals through correlation features and regulation features, with greater accuracy.

[0079] S3, generating a biological feature characterizing individual physiological differences according to the correlation feature and the adjustment feature; and obtaining a target identity probability distribution based on the biological feature;

[0080] Specifically, the above-mentioned correlation features and adjustment features can be integrated in the form of a feature vector. HRV, the correlation coefficient between respiratory amplitude and heart rate, the score of body movement-respiration coordination criterion (recorded as 1 if it meets the coordination standard and 0 if it does not), the change value of the respiratory signal stability index (the stability index after a body movement event minus the stability index before the body movement event), the heart rate recovery rate, etc. can be used as different dimensions of the vector. For example, for individual A, their biometric vector may be [0.05, 0.6, 1, 0.02, 0.8], corresponding to an HRV of 0.05 (the unit is obtained according to actual calculation), a correlation coefficient between respiratory amplitude and heart rate of 0.6, body movement-respiration coordination (score of 1), a change value of the respiratory signal stability index of 0.02, and a heart rate recovery rate of 0.8 beats per second. Such a vector comprehensively represents the biometric features related to the physiological differences of individual A.

[0081] Specifically, the target identity probability distribution represents the likelihood of a given biometric feature belonging to different target identities. For example, in a database containing 10 known identity individuals, it is necessary to determine the probability that the currently acquired biometric feature belongs to each of these 10 identities.

[0082] The specific execution process is as follows: Establish a biometric database: First, collect a large number of biometric vectors of different individuals. For each individual, obtain their feature vectors according to the above method of generating biometric features, and store these vectors together with the corresponding individual identity information in the database. Suppose there are biometric vectors and identity information of 1000 individuals in the database. Taking the Naive Bayes classifier as an example, it calculates the probability of belonging to each target identity based on Bayes' theorem and the assumption of feature conditional independence given the known biometric vector. Train the model: Use the biometric vectors and corresponding identity information in the database to train the selected classification algorithm. During the training process, the algorithm will learn the association patterns between different biometric vectors and identities. For example, for the Naive Bayes classifier, it will calculate the probability distribution parameters of each feature dimension under each identity category. Calculate the target identity probability distribution: When there is a new biometric vector whose identity needs to be determined, input this vector into the trained model. The model will calculate the probability that this biometric vector belongs to each target identity in the database according to the learned patterns. For example, after model calculation, for the new biometric vector, the probability that it belongs to identity 1 is 0.05, the probability that it belongs to identity 2 is 0.1, and so on, obtaining a probability distribution [0.05, 0.1, 0.08, 0.03, 0.2, 0.15, 0.07, 0.09, 0.1, 0.03] for 10 target identities. This probability distribution represents the matching possibility of this biometric feature with each target identity.

[0083] S4. Perform interference suppression processing on the original echo signal according to the target identity probability distribution to separate the target respiration waveform and the target heartbeat waveform of the target individual from the original echo signal.

[0084] Exemplarily, in the sleep monitoring ward of a hospital, the monitoring device is installed in the ward, emits signals to the hospital bed area, and receives the original echo signal reflected from the patient's body. By this method, even if there are interferences from other devices or movements of other people in the ward, the respiration and heartbeat waveforms of the patient can be accurately obtained, providing a basis for doctors to judge the patient's health status during sleep.

[0085] In a preferred embodiment, the energy gradient analysis of adjacent signal frames of the original echo signal, locating the body movement candidate interval, and generating body movement event markers include the following steps:

[0086] Calculate the Teager energy operator difference sequence of adjacent signal frames of the original echo signal through a sliding time window;

[0087] Specifically, the Teager energy operator is an operator for calculating the energy of a signal. It measures the energy of the signal at a certain moment based on information such as the amplitude and derivative of the signal. For example, for a discrete signal x(n), its Teager energy operator calculation formula can be expressed as: , and the Teager energy operator difference sequence is a sequence composed of the differences of the Teager energy operator calculation results of adjacent signal frames. The acquisition method is to take the difference of the Teager energy operator calculation results of adjacent signal frames, that is, for the Teager energy operator values of the i-th and i + 1-th signal frames and , the element in the difference sequence is .

[0088] Obtain a dynamic decision threshold according to the real-time signal characteristics;

[0089] Specifically, the real-time signal characteristics include the mean and variance characteristics of the real-time signal. For example, let the mean of the real-time signal be μ and the variance be , the dynamic decision threshold T can be expressed as , where and are coefficients determined according to experiments or experience.

[0090] Obtain the starting time point and duration range of the body movement event according to the dynamic decision threshold;

[0091] Specifically, when an element in the difference sequence first exceeds the dynamic determination threshold, the corresponding time point is the starting time point. Starting from the starting time point and ending at the time point when the elements in the Teager energy operator difference sequence are continuously lower than the dynamic determination threshold, the duration between these two time points is the duration range.

[0092] Locate the body movement candidate interval based on the starting time point and duration range of the body movement event, and generate a body movement event marker.

[0093] Specifically, a sleep monitoring experiment can be first conducted on all individuals in the monitoring space to obtain the change patterns of the respiratory signal and heart rate signal after body movement for each individual, and then the corresponding change patterns of each individual are stored in the pattern judgment database to facilitate subsequent determination of the target individual according to the adjustment characteristics.

[0094] Specifically, calculate the Teager energy operator difference sequence of adjacent signal frames of the original echo signal through a sliding time window to capture the change of signal energy. Since the body movement event will cause a change in the energy of the original echo signal, the abnormal points of the energy change can be found by analyzing the difference sequence. Calculate the dynamic determination threshold according to the real-time signal characteristics, and this threshold can adapt to the changes of signal characteristics in different monitoring environments. Use this threshold to determine the starting time point and duration range of the body movement event, and then locate the body movement candidate interval and generate a body movement event marker. These body movement event-related information (starting time point, duration range, etc.) is an important basis for subsequent analysis of the adjustment characteristics of the body movement event on the respiratory and heart rate signals. By analyzing the changes in the respiratory and heart rate signals before and after the body movement event, such as adjustment characteristics such as the stability index of the respiratory signal and the heart rate recovery rate, it can provide key data support for generating biometric features representing individual physiological differences and help distinguish the target individual from non-target individuals. For example, after a body movement event occurs, the adjustment characteristics of the respiratory and heart rate signals of different individuals may be different. The target individual may have a specific range of changes in the respiratory stability index and a heart rate recovery rate pattern. By accurately capturing the body movement event and subsequent analysis of the adjustment characteristics, the target individual can be distinguished from other individuals.

[0095] In a preferred embodiment, the method for distinguishing local body movement from global interference based on the motion trajectory discrimination criterion and fusing the frequency band feature components to generate a body movement intensity waveform includes the following steps:

[0096] Construct a three-dimensional motion trajectory map according to the Doppler phase change, and calculate the Mahalanobis distance between each trajectory point and the historical motion reference trajectory;

[0097] Specifically, the three-dimensional motion trajectory map is a visualized map of the three-dimensional space motion path constructed through Doppler phase changes, which is obtained by performing spatial coordinate mapping on the phase change data within the monitoring area. The historical motion reference trajectory is a pre-established normal body motion trajectory model, which is obtained by performing cluster analysis on the historical body motion data of the monitoring object. The Mahalanobis distance is a statistic that measures the distance between a data point and a data distribution.

[0098] When the cumulative Mahalanobis distance of consecutive trajectory points exceeds the dynamic interference threshold, it is determined as a local body motion event and the spatial positioning coordinates are output.

[0099] Specifically, the dynamic interference threshold is a threshold dynamically adjusted according to the real-time environmental interference, which is determined by calculating the noise level of the current monitoring signal. The spatial positioning coordinates are the three-dimensional coordinates where the body motion event occurs, which are calculated through triangulation combined with multi-sensor data fusion.

[0100] Align the instantaneous energy of the frequency band characteristic component with the spatial positioning coordinates in space-time to generate a body motion intensity waveform that is continuous in the time domain.

[0101] Exemplarily, when the user turns over (local body motion), the system identifies that this action occurs in the bed area through the three-dimensional trajectory map, and the Mahalanobis distance exceeds the threshold. At the same time, the energy in the 2 - 15 Hz frequency band increases significantly, and the system associates this energy change with the positioning coordinates to generate a body motion intensity waveform. When a pet enters the monitoring area (global interference), the system determines it as global interference through trajectory analysis and automatically filters the relevant signals.

[0102] This embodiment can achieve the spatial positioning and intensity quantification of body motion events, can effectively distinguish local body motion (such as turning over) and global interference (such as device vibration), and provides accurate data support for subsequent adjustment feature analysis.

[0103] In a preferred embodiment, after obtaining the correlation characteristics of the respiration signal and the heartbeat signal, and the adjustment characteristics of the body motion event on the respiration signal and / or the heartbeat signal, before generating the biometric characteristics representing individual physiological differences according to the correlation characteristics and the adjustment characteristics, the following steps are further included:

[0104] Judge the size relationship between the duration range of the body motion event and a preset threshold, and judge whether there is an overlap of multi-target reflection cross-sections;

[0105] Specifically, the preset threshold is a critical value of body motion duration set according to clinical experience, which is determined by performing statistical analysis on sleep medicine research data. The overlap of multi-target reflection cross-sections is a phenomenon where the radar reflection waves of multiple monitoring objects overlap with each other, which can be determined through the spatial density analysis of the three-dimensional trajectory map.

[0106] Generating biometric features characterizing individual physiological differences based on the correlation features and the adjustment features includes the following steps:

[0107] If the duration range of the body movement event is less than or equal to a preset threshold and there is no overlap of multi-target reflection cross-sections, initial features characterizing individual physiological differences are generated according to the correlation features and the adjustment features, and the initial features are used as biometric features.

[0108] Specifically, the initial features are biometric features directly composed of correlation features and adjustment features, which are obtained by splicing feature vectors. First, it is judged whether the body movement duration exceeds a preset threshold (such as 3 seconds), and whether there is multi-target reflection overlap is detected. If the body movement time is short and there is no overlap, the original features are directly used to generate biometric features. This is because the influence of short-term body movement on breathing and heartbeat has individual specificity and can be directly used for identity recognition.

[0109] Exemplarily, in a single-person bedroom monitoring scenario, when the user accidentally turns over at night (duration 1.2 seconds) and the system detects no other reflection overlap, the initial biometric features are directly generated using the respiratory amplitude-heart rate correlation coefficient (0.85) and the heart rate recovery rate (3.2 seconds) for subsequent identity verification.

[0110] This embodiment can reduce unnecessary complex processing, quickly generate biometric features in a simple scenario, and improve the system response speed and recognition efficiency.

[0111] In a preferred embodiment, after judging the size of the duration range of the body movement event and the preset threshold and judging whether there is an overlap of multi-target reflection cross-sections, the following steps are further included:

[0112] If the duration range of the body movement event is greater than the preset threshold or there is an overlap of multi-target reflection cross-sections, a dual-channel adaptive separation algorithm is used to perform secondary decoupling processing on the respiratory signal and the heartbeat signal;

[0113] Specifically, the dual-channel adaptive separation algorithm is re-executed on the basis of the initial separation for the spatial region where multiple target reflection waves overlap. The acquisition method is determined by the spatial density analysis of the three-dimensional trajectory map.

[0114] Based on the three-dimensional motion trajectory map, the reflection cross-section overlap region corresponding to the global interference is excluded, and the processed respiratory signal and the processed heartbeat signal are re-extracted;

[0115] According to the processed respiratory signal, the processed correlation features of the target individual are obtained; according to the processed heartbeat signal, the processed adjustment features are obtained; the processed correlation features and the processed adjustment features are used to optimize the initial features.

[0116] Specifically, when the body movement duration is too long or there are overlapping multi-targets, the signal separation algorithm is applied twice, and the interference area is excluded by combining with the three-dimensional motion trajectory atlas, and the respiration and heartbeat signals are re-extracted. Through the above method of obtaining the correlation features and adjustment features, the new correlation features and adjustment features are recalculated to optimize the initial biometric features.

[0117] Exemplarily, in a collective dormitory in a nursing home, an old person continuously turns over (duration of 5 seconds), and at the same time, the movement of the person in the adjacent bed causes the reflection waves to overlap. The system separates the respiration signal twice, excludes the overlapping area, and then recalculates the heart rate variability within the respiration cycle (increased from 0.08 Hz to 0.12 Hz). The optimized biometric features improve the accuracy of identity recognition.

[0118] This embodiment can maintain the accuracy of biometric features in a complex monitoring environment, solve the problem of multi-target interference, and improve the robustness of the system.

[0119] Further, after the secondary decoupling process of the respiration signal and the heartbeat signal, the following steps are further included:

[0120] Verify the stability of the signal energy distribution after the secondary decoupling process;

[0121] Specifically, the signal energy distribution stability is the consistency of the signal energy distribution in the time-frequency domain, which is obtained by calculating the energy entropy value.

[0122] If the verification is successful, based on the processed correlation features and the processed adjustment features, the initial features are corrected to obtain the processed features, and the processed features are used as biometric features.

[0123] Specifically, the feature correction is to optimize the feature vector through weighted fusion, which is dynamically adjusted using the Kalman filter.

[0124] Specifically, the energy distribution stability of the signal after secondary decoupling is verified. If the entropy value is lower than the threshold (such as 0.5), the signal is considered stable. The initial features are corrected using the optimal fusion strategy to ensure the reliability of the biometric features, avoid misjudgment caused by unstable signals, and improve the credibility of the biometric features.

[0125] In a preferred embodiment, the interference suppression process of the original echo signal according to the target identity probability distribution to separate the target respiration waveform and the target heartbeat waveform of the target individual from the original echo signal includes the following steps:

[0126] Perform weighted filtering on the original echo signal according to the target identity probability distribution to suppress the respiration and heartbeat interference components of non-target individuals and obtain the filtered signal;

[0127] Specifically, in this scenario of weighted filtering, the weights are determined according to the target identity probability distribution. For the original echo signal, we assume that it contains a mixed component of respiratory and heartbeat signals from multiple individuals (including the target individual and non-target individuals). First, the original echo signal is divided into multiple small signal segments in chronological order, for example, every 10 milliseconds as a signal segment. Then, for each signal segment, weights are assigned according to the target identity probability distribution. Suppose the target identity is Identity 3, and its probability in the probability distribution is 0.2. For the current signal segment, we assign a higher weight, such as 0.8, to the signal components related to the target identity, and a lower weight to the signal components related to other non-target identities. For the signal components related to non-target identities such as Identity 1 (probability 0.05) and Identity 2 (probability 0.1), the total weight is set to 0.2 (1 - 0.8), and then they are distributed according to their probability ratios. The weight corresponding to Identity 1 may be 0.2 * (0.05 / (0.05 + 0.1)) ≈ 0.067, and the weight corresponding to Identity 2 is 0.2 * (0.1 / (0.05 + 0.1)) ≈ 0.133. Next, for each frequency component within each signal segment (respiratory and heartbeat signals are in different frequency ranges), weighted processing is performed according to the assigned weights. For example, if a certain frequency component within the signal segment is closer to the respiratory and heartbeat characteristic frequencies of target Identity 3, it is multiplied by a weight of 0.8; if it is closer to the characteristic frequency of non-target Identity 1, it is multiplied by a weight of 0.067. Finally, all the weighted signal segments are recombined to obtain the filtered signal after suppressing the interference components of the respiratory and heartbeat of non-target individuals.

[0128] Perform variational mode decomposition on the filtered signal to extract the target respiratory waveform and the target heartbeat waveform.

[0129] Specifically, Variational Mode Decomposition (VMD) decomposes complex signals into a series of mode functions with different center frequencies. These mode functions are called Intrinsic Mode Functions (IMFs), and each IMF represents the characteristics of the signal in a specific frequency band. The execution process is as follows: The filtered signal is input into the VMD algorithm. The VMD algorithm first initializes a series of parameters, including the number of mode functions K to be decomposed (usually set according to experience or prior knowledge. For example, for respiratory and heartbeat signals, K can be set to 2, corresponding to the respiratory and heartbeat modes respectively), the penalty factor α (used to balance the decomposition accuracy and computational efficiency), etc. Through continuous iterative optimization, the algorithm decomposes the filtered signal into K mode functions. During the iteration process, it adjusts parameters such as the center frequency and bandwidth of each mode function so that these mode functions can best fit the different frequency components of the signal. After multiple iterations, K mode functions are obtained. At this time, it is necessary to determine which mode function corresponds to the target respiratory waveform and which corresponds to the target heartbeat waveform according to the frequency ranges of the respiratory and heartbeat signals. Because the frequency of the respiratory signal is generally between 0.1 - 0.8 Hz, and the frequency of the heartbeat signal is between 0.8 - 3 Hz. For example, through frequency analysis, if it is found that the main frequency components of one mode function are between 0.2 - 0.6 Hz, then this mode function is very likely to be the target respiratory waveform; if the main frequency components of another mode function are between 1 - 2 Hz, it corresponds to the target heartbeat waveform. Finally, the mode functions corresponding to the target respiratory waveform and the target heartbeat waveform are extracted, thus achieving precise signal separation in a multi-target environment, effectively suppressing cross-interference, and improving the accuracy of single-sign feature extraction.

[0130] In this article, specific examples are used to elaborate on the principles and implementation methods of this application. The descriptions of the above embodiments are only for helping to understand the method and its core idea of this application. The above is only the preferred implementation method of this application. It should be noted that due to the limited nature of written expression and the objectively infinite specific structures, for those of ordinary skill in the art in this technical field, without departing from the principle of this invention, several improvements, refinements or changes can be made, or the above technical features can be combined in an appropriate manner; these improvements, refinements, changes or combinations, or directly applying the concept and technical solution of the invention to other occasions without improvement, should all be regarded as the protection scope of this application.

Claims

1. A method for monitoring vital signs of sleep, characterized in that: The following steps are involved: Collecting the original echo signal of the monitoring area, and extracting the breathing signal, heartbeat signal and body motion signal corresponding to the body motion event from the original echo signal; Acquire the correlation characteristics between the respiratory signal and the heartbeat signal, and the regulation characteristics of the body motion event on the respiratory signal and / or the heartbeat signal; Generating a biological feature characterizing individual physiological differences according to the correlation feature and the regulation feature; and based on the biometrics, obtaining a probability distribution of the target identity; The original echo signal is subjected to interference suppression processing according to the target identity probability distribution, so as to separate the target breathing waveform and the target heartbeat waveform of the target individual from the original echo signal.

2. The method for monitoring vital signs of sleep according to claim 1, characterized in that: The acquiring of the correlation characteristics between the respiratory signal and the heartbeat signal, and the adjustment characteristics of the body motion event on the respiratory signal and / or the heartbeat signal, comprises the following steps: Calculate the heartbeat rhythm variability and the respiratory amplitude-heart rate correlation coefficient within the respiratory cycle to obtain correlation characteristics; A body movement-respiration coordination criterion is established, and the stability index of the respiratory signal and the heart rate recovery rate before and after the body movement event are analyzed to obtain the regulation characteristics.

3. The method for monitoring vital signs of sleep according to claim 1, characterized in that: A dual-channel adaptive separation algorithm is used to extract a respiratory signal and a heartbeat signal from the original echo signal, comprising the following steps: Constructing a principal component channel of the respiratory signal based on empirical mode decomposition, and retaining a 0.1-0.8 Hz frequency band component through intrinsic mode function screening; Establishing an auxiliary channel of the heartbeat signal based on complex wavelet transform, and using Morlet wavelet basis function to perform phase decoupling on the 0.8-3 Hz frequency band; The orthogonality of the dual-channel signals of the main component channel and the auxiliary channel is evaluated by mutual information entropy, and the parameter weights between the channels are dynamically adjusted to achieve signal separation, so as to obtain the respiratory signal and the heartbeat signal.

4. The method for monitoring vital signs of sleep according to claim 1, characterized in that: Extracting a body motion signal corresponding to a body motion event from the original echo signal comprises the following steps: Performing energy gradient analysis on adjacent signal frames of the original echo signal, locating body motion candidate intervals, and generating body motion event markers; Performing non-stationary feature decoupling on the marker signal of the body motion event marker to separate the 2-15 Hz frequency band feature component representing the limb motion; Based on the motion trajectory discrimination criterion, local body motion and global interference are distinguished, and the frequency band characteristic components are fused to generate a body motion intensity waveform, which is a body motion signal.

5. The method for monitoring vital signs of sleep according to claim 4, characterized in that: The step of performing energy gradient analysis on adjacent signal frames of the original echo signal, locating body motion candidate intervals, and generating body motion event markers comprises the following steps: Calculating a Teager energy operator difference sequence of adjacent signal frames of the original echo signal through a sliding time window; According to the real-time signal characteristics, a dynamic judgment threshold is obtained; According to the dynamic determination threshold, the starting time point and duration range of the body movement event are obtained; The body movement candidate interval is located according to the starting time point and duration range of the body movement event, and a body movement event marker is generated.

6. The method for monitoring vital signs of sleep according to claim 4, characterized in that: The method of distinguishing local body motion from global interference based on the motion trajectory discrimination criterion and fusing the frequency band characteristic components to generate a body motion intensity waveform includes the following steps: Construct a three-dimensional motion trajectory map based on Doppler phase changes, and calculate the Mahalanobis distance between each trajectory point and the historical motion reference trajectory; When the cumulative Mahalanobis distance of continuous trajectory points exceeds the dynamic interference threshold, it is determined as a local body motion event and the spatial positioning coordinates are output; The instantaneous energy of the frequency band characteristic component is aligned with the spatial positioning coordinates in time and space to generate a time-domain continuous body motion intensity waveform.

7. The method for monitoring vital signs of sleep according to claim 1, characterized in that: After acquiring the correlation feature between the respiratory signal and the heartbeat signal, and the adjustment feature of the body motion event on the respiratory signal and / or the heartbeat signal, and before generating the biological feature characterizing the individual physiological difference according to the correlation feature and the adjustment feature, the following steps are also included: Determine the duration range of the body motion event and the size of a preset threshold, and determine whether there is overlap of multiple target reflection cross sections; The step of generating a biological feature characterizing individual physiological differences according to the correlation feature and the regulation feature comprises the following steps: If the duration range of the body motion event is less than or equal to a preset threshold and there is no overlap of multiple target reflection cross sections, an initial feature characterizing individual physiological differences is generated based on the correlation feature and the adjustment feature, and the initial feature is used as a biological feature.

8. The method for monitoring vital signs of sleep according to claim 7, characterized in that: After determining the duration range of the body motion event and the size of the preset threshold, and determining whether there is an overlap of multiple target reflection cross sections, the following steps are also included: If the duration range of the body motion event is greater than a preset threshold, or there is an overlap of multiple target reflection cross sections, a dual-channel adaptive separation algorithm is used to perform secondary decoupling processing on the respiratory signal and the heartbeat signal; Based on the three-dimensional motion trajectory map, the overlapping area of ​​the reflection cross section corresponding to the global interference is excluded, and the processed breathing signal and the processed heartbeat signal are re-extracted; According to the processed breathing signal, a processed correlation feature of the target individual is obtained; according to the processed heartbeat signal, a processed adjustment feature is obtained; the processed correlation feature and the processed adjustment feature are used to optimize the initial feature.

9. The method for monitoring vital signs of sleep according to claim 8, characterized in that: After the secondary decoupling process is performed on the respiratory signal and the heartbeat signal, the following steps are also included: Verify the stability of signal energy distribution after secondary decoupling processing; If the verification is successful, the initial feature is corrected based on the processed correlation feature and the processed adjustment feature to obtain the processed feature, and the processed feature is used as the biometric feature.

10. The method for monitoring vital signs of sleep according to claim 1, characterized in that: The interference suppression processing is performed on the original echo signal according to the target identity probability distribution to separate the target breathing waveform and the target heartbeat waveform of the target individual from the original echo signal, comprising the following steps: Performing weighted filtering on the original echo signal according to the target identity probability distribution, suppressing the respiratory and heartbeat interference components of non-target individuals, and obtaining a filtered signal; The filtered signal is subjected to variational modal decomposition to extract a target breathing waveform and a target heartbeat waveform.

Images