CN-122004824-A - Health monitoring method and device based on radar, computer equipment and storage medium

CN122004824ACN 122004824 ACN122004824 ACN 122004824ACN-122004824-A

Abstract

The invention relates to the technical field of radar signal data processing, and discloses a health monitoring method, a device, computer equipment and a storage medium based on radar. The method comprises the steps of preprocessing a thoracic radar echo signal to obtain phase time sequence data, obtaining initial time-frequency characteristics of a current time step through an initial time-frequency analysis model, further determining a joint state vector, a continuous motion vector and a time-frequency quality score of the current time step, determining dense internal rewards of the current time step according to the joint state vector, the continuous motion vector and the time-frequency quality score, iteratively updating the initial time-frequency analysis model by taking the dense internal rewards as return signals to obtain a target time-frequency analysis model based on reinforcement learning, and determining health monitoring results according to the target time-frequency characteristics obtained through the target time-frequency analysis model. The invention realizes the self-adaptive optimization of the ultra-wavelet parameters in the time-frequency analysis model, improves the robustness of time-frequency characterization and ensures the effect of health monitoring.

Inventors

LI XINGGUANG
LI JINSONG
CAI YUJIAN
WANG YONGBO
CUI WEI
XU CHUNSHENG

Assignees

长春理工大学

Dates

Publication Date: 20260512
Application Date: 20260416

Claims (10)

1. A radar-based health monitoring method, comprising: Acquiring a thoracic radar echo signal, and preprocessing the thoracic radar echo signal to obtain phase time sequence data; Performing ultra-wavelet transformation processing on the phase time sequence data through an initial time-frequency analysis model to obtain initial time-frequency characteristics of a current time step, wherein the time step refers to the iteration update round of the initial time-frequency analysis model; determining a joint state vector of the current time step according to the phase time sequence data and the initial time-frequency characteristics, determining a continuous motion vector of the current time step according to the joint state vector, and determining a time-frequency quality score of the current time step according to the initial time-frequency characteristics; Determining a dense internal reward for a current time step based on the joint state vector, the continuous motion vector, and the time-frequency quality score; Iteratively updating the initial time-frequency analysis model by taking the dense internal rewards as return signals, and obtaining a target time-frequency analysis model based on reinforcement learning after the iterative updating is completed; And performing ultra-wavelet transformation processing on the phase time sequence data through the target time-frequency analysis model to obtain target time-frequency characteristics, and determining a health monitoring result according to the target time-frequency characteristics.
2. The radar-based health monitoring method according to claim 1, wherein the performing the ultra-wavelet transform processing on the phase time series data by using an initial time-frequency analysis model to obtain an initial time-frequency characteristic of a current time step comprises: Acquiring a reference period parameter, a modulation factor parameter and an order index parameter of the current time step through the initial time-frequency analysis model; Obtaining ultra-wavelet transformation parameters under each center frequency according to preset frequency interval parameters, the reference period parameters, the modulation factor parameters and the order index parameters; Performing convolution processing on the ultra-wavelet transformation parameters and the phase time sequence data to obtain wavelet convolution results under each center frequency; And carrying out aggregation treatment on the wavelet convolution results of each order under the same center frequency to obtain the initial time-frequency characteristic of the current time step.
3. The radar-based wellness monitoring method of claim 1, wherein the determining a joint state vector for a current time step based on the phase time series data and the initial time-frequency characteristic comprises: performing Fourier transformation and normalization processing on the phase time sequence data to obtain a power spectrum feature vector; performing envelope transformation on the phase time sequence data to obtain an envelope statistical vector; performing entropy analysis on the initial time-frequency characteristic according to a preset order to obtain a renyi entropy; and generating a joint state vector of the current time step according to the initial time-frequency characteristic, the power spectrum characteristic vector, the envelope statistical vector and the renyi entropy.
4. The radar-based wellness monitoring method of claim 1, wherein the determining a time-frequency quality score for a current time step based on the initial time-frequency characteristic comprises: Calculating a time-frequency quality index value according to the initial time-frequency characteristic, wherein the time-frequency quality index value comprises time-frequency energy concentration degree, time resolution, frequency resolution and ridge line definition; And carrying out weighted calculation on the time-frequency energy concentration degree, the time resolution, the frequency resolution and the ridge line definition according to preset weight parameters to obtain a time-frequency quality score.
5. The radar-based wellness monitoring method of claim 1, wherein the joint state vector comprises an initial time-frequency characteristic component, a power spectrum characteristic component, an envelope statistics component, and a renyi entropy component; said determining a dense internal prize for a current time step based on said joint state vector, said continuous motion vector, and said time-frequency quality score, comprising: Performing convolution feature extraction on the initial time-frequency feature component to obtain a first state feature vector; carrying out full-connection feature extraction on the power spectrum feature component, the envelope statistical component and the renyi entropy component to obtain a second state feature vector; fusing the first state feature vector and the second state feature vector to obtain a state embedded vector; coding the continuous motion vector to obtain a motion embedded vector; determining an initial dense internal prize based on the state embedded vector and the action embedded vector; And determining a state value according to the state embedding vector, determining a round sparse evaluation value according to the time-frequency quality score, and correcting the initial dense internal rewards according to the round sparse evaluation value and the state value to obtain the dense internal rewards.
6. The radar-based health monitoring method according to claim 1, wherein the performing the ultra-wavelet transform processing on the phase time series data through the target time-frequency analysis model to obtain target time-frequency characteristics includes: extracting a target continuous motion vector from the target time-frequency analysis model, wherein the target continuous motion vector comprises a first motion component, a second motion component and a third motion component; Mapping the first action component according to a preset period upper limit value and a preset period lower limit value to obtain a target reference period parameter; mapping the second action component according to a preset modulation upper limit value and a preset modulation lower limit value to obtain a target modulation factor parameter; mapping the third motion component according to a preset order upper limit value and a preset order lower limit value to obtain a target order index parameter; And determining a target time-frequency characteristic according to the preset frequency interval parameter, the target reference period parameter, the target modulation factor parameter and the target order index parameter.
7. The radar-based wellness monitoring method of claim 1, wherein the wellness monitoring results comprise heart rate variability monitoring results; the determining the health monitoring result according to the target time-frequency characteristic comprises the following steps: Identifying a heart beat time frequency main ridge line according to the target time frequency characteristic, and determining the time interval between adjacent heart beat events according to the heart beat time frequency main ridge line to obtain a heart beat interval sequence; and calculating a heart rate variability index value according to the heart beat interval sequence, and determining a heart rate variability monitoring result according to the heart rate variability index value.
8. A radar-based health monitoring device, comprising: the preprocessing module is used for acquiring a thoracic radar echo signal, preprocessing the thoracic radar echo signal and obtaining phase time sequence data; The initial model analysis module is used for performing ultra-wavelet transformation processing on the phase time sequence data through an initial time-frequency analysis model to obtain initial time-frequency characteristics of a current time step, wherein the time step refers to the round of iterative updating of the initial time-frequency analysis model; the characteristic analysis module is used for determining a joint state vector of the current time step according to the phase time sequence data and the initial time-frequency characteristic, determining a continuous motion vector of the current time step according to the joint state vector, and determining a time-frequency quality score of the current time step according to the initial time-frequency characteristic; the rewards analysis module is used for determining dense internal rewards of the current time step according to the joint state vector, the continuous motion vector and the time-frequency quality score; The model updating module is used for iteratively updating the initial time-frequency analysis model by taking the dense internal rewards as return signals and obtaining a target time-frequency analysis model based on reinforcement learning after the iterative updating is completed; And the monitoring result determining module is used for performing ultra-wavelet transformation processing on the phase time sequence data through the target time-frequency analysis model to obtain target time-frequency characteristics, and determining health monitoring results according to the target time-frequency characteristics.
9. A computer device comprising a memory, a processor, and computer readable instructions stored in the memory and executable on the processor, wherein the processor, when executing the computer readable instructions, implements the radar-based wellness monitoring method of any one of claims 1-7.
10. A computer-readable storage medium storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the radar-based wellness monitoring method of any one of claims 1-7.

Description

Health monitoring method and device based on radar, computer equipment and storage medium Technical Field The present invention relates to the field of radar signal data processing technologies, and in particular, to a radar-based health monitoring method, a radar-based health monitoring device, a computer device, and a storage medium. Background In the field of smart medicine and non-contact physiological monitoring, frequency modulated continuous wave (Frequency Modulated Continuous Wave, FMCW) radar can be used for health monitoring of heart rate, cardiac intervals, respiratory rate and the like. In the prior art, a classical time-frequency analysis method (such as short-time fourier transform) has an inherent trade-off between time resolution and frequency resolution, and cannot guarantee the precision of the time resolution and the frequency resolution at the same time. Although the time-frequency analysis method based on the ultra-wavelet can improve the time-frequency energy concentration to a certain extent, the method relies on experience to select and fix key parameters (such as wavelet cycle number/order related parameters, frequency sampling density, scale control parameters and the like). When the amplitude-frequency structure of the phase micro-motion signal is changed due to individual difference, body movement interference, monitoring distance change or heart rate fluctuation, the fixed parameters are easy to cause time-frequency energy diffusion, side lobe enhancement or ridge line fracture, so that generalization and robustness of radar signal analysis results under complex application scenes (such as heart rate fluctuation after exercise and long-distance monitoring) are reduced, and further analysis reliability of health state indexes such as heart rate and heart beat interval is affected. Disclosure of Invention Based on the foregoing, it is necessary to provide a health monitoring method, apparatus, computer device and storage medium based on radar, so as to solve the problem of poor robustness of radar signal analysis during health monitoring based on radar. A radar-based health monitoring method, comprising: Acquiring a thoracic radar echo signal, and preprocessing the thoracic radar echo signal to obtain phase time sequence data; Performing ultra-wavelet transformation processing on the phase time sequence data through an initial time-frequency analysis model to obtain initial time-frequency characteristics of a current time step, wherein the time step refers to the iteration update round of the initial time-frequency analysis model; determining a joint state vector of the current time step according to the phase time sequence data and the initial time-frequency characteristics, determining a continuous motion vector of the current time step according to the joint state vector, and determining a time-frequency quality score of the current time step according to the initial time-frequency characteristics; Determining a dense internal reward for a current time step based on the joint state vector, the continuous motion vector, and the time-frequency quality score; Iteratively updating the initial time-frequency analysis model by taking the dense internal rewards as return signals, and obtaining a target time-frequency analysis model based on reinforcement learning after the iterative updating is completed; And performing ultra-wavelet transformation processing on the phase time sequence data through the target time-frequency analysis model to obtain target time-frequency characteristics, and determining a health monitoring result according to the target time-frequency characteristics. A radar-based health monitoring device, comprising: the preprocessing module is used for acquiring a thoracic radar echo signal, preprocessing the thoracic radar echo signal and obtaining phase time sequence data; The initial model analysis module is used for performing ultra-wavelet transformation processing on the phase time sequence data through an initial time-frequency analysis model to obtain initial time-frequency characteristics of a current time step, wherein the time step refers to the round of iterative updating of the initial time-frequency analysis model; the characteristic analysis module is used for determining a joint state vector of the current time step according to the phase time sequence data and the initial time-frequency characteristic, determining a continuous motion vector of the current time step according to the joint state vector, and determining a time-frequency quality score of the current time step according to the initial time-frequency characteristic; the rewards analysis module is used for determining dense internal rewards of the current time step according to the joint state vector, the continuous motion vector and the time-frequency quality score; The model updating module is used for iteratively updating the initial time-frequency analysis model by taking the dense in