Physiological parameter detection method and system based on joint domain analysis

Abstract

The invention belongs to the technical field of artificial intelligence, human body physiological data processing and computer data processing, and particularly relates to a physiological parameter detection method and system based on joint domain analysis. The problems that in the prior art, due to the fact that the signal strength is weakened, the detection range is shortened, the signal-to-noise ratio is reduced, and the multipath effect is enhanced in the detection process, the accuracy and stability of physiological parameter detection are affected are solved, the accuracy and stability of physiological parameter detection are improved, and the detection efficiency is improved. In addition, noise interference is effectively reduced in the data processing process, and the analyzability and reliability of signals are enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical fields of artificial intelligence, human physiological data processing and computer data processing, and in particular to a physiological parameter detection method and system based on joint domain analysis. Background Art

[0002] With the continuous development of society and the continuous change of population structure, the problem of population aging is becoming more and more serious, and "empty nesters" have become the focus of global attention. The health of the elderly has become particularly severe due to the decline of cognitive, motor and sensory functions. More than 78% of the elderly suffer from at least one chronic disease, and the number of disabled elderly people is also increasing year by year. However, compared with the growing health needs of the elderly, the institutions and equipment related to healthy aging are obviously insufficient. Timely detection and treatment are the most effective strategies to combat chronic diseases. For cardiovascular and cerebrovascular diseases, respiratory characteristics and heartbeat characteristics are important reference indicators. At present, common respiratory characteristic detection methods include electrode chest impedance scanning, and the main method for human heartbeat characteristic detection is electrocardiogram. However, these methods are not only costly and require contact with the human body during data collection, but also require the participation of professionals. For the elderly whose cognitive, motor and sensory functions are gradually declining, it is difficult to complete these data collection tasks independently. In addition, many elderly people have a rejection of the use of instruments for physiological parameter detection and find it difficult to accept it, which further increases the difficulty of chronic disease prevention in the elderly. Therefore, designing a method that can detect human physiological parameters through walls without contact has important academic and application significance. The wall-mounted physiological parameter detection technology can continuously and in real time detect key health indicators such as heart rate, respiratory rate, activity level, etc. without disturbing the daily life of the elderly. This technology is particularly important for the elderly living in "empty nests" because it can provide non-intrusive monitoring throughout the day to ensure timely detection of health abnormalities. For example, sudden cardiovascular and cerebrovascular events, breathing difficulties, etc. can be quickly detected and medical staff or family members can be notified immediately, so that intervention measures can be taken quickly. This not only improves the safety and quality of life of the elderly, but also reduces hospitalization rates and medical costs. In addition, this non-intrusive detection method reduces the burden of wearing equipment on the elderly, enhances their comfort and acceptance, and makes chronic disease management more efficient and humane. In post-disaster rescue, the wall physiological parameter detection technology is also of great significance. After natural disasters such as earthquakes and floods, many victims may be buried in the rubble. Traditional rescue methods require a lot of manpower and time. With the wall physiological parameter detection technology, rescuers can quickly and accurately locate the survivors and determine their vital signs by detecting their heartbeat, breathing and other physiological parameters. This greatly improves the rescue efficiency and increases the survival rate of survivors. At the same time, this technology does not require direct contact with the victims during the detection process, avoiding secondary damage to them. This non-contact rescue method shows great advantages and potential, especially when the rescue environment is complex and the victims cannot move. Due to the high penetration of UWB signals, radar can be used to penetrate buildings or obstacles to achieve detection, positioning, imaging and tracking, and can be used to find and prioritize rescue of trapped people after natural disasters. When UWB signals penetrate walls, the signal attenuates severely due to the dielectric constant. This includes: First, the signal strength is weakened: the signal strength decreases rapidly when passing through walls and human tissue. Weaker signals may be difficult to detect at the receiving end, affecting the reliability and accuracy of detection. Second, the detection range is shortened: the available distance of the system in detecting physiological parameters through walls will be reduced, and it may not be able to effectively cover targets in the entire room or a larger space. Third, the signal-to-noise ratio is reduced: signal attenuation causes the received signal to weaken, while the environmental noise remains relatively unchanged or changes less, which reduces the signal-to-noise ratio. A lower signal-to-noise ratio will affect the sensitivity and accuracy of the detection system, making the detection of physiological parameters more susceptible to noise interference. Fourth, the multipath effect is enhanced: when passing through walls and other obstacles, more reflections and scattering paths may be triggered. These multipath effects will cause the received signal to contain multiple delayed and attenuated components, resulting in signal distortion and increased detection errors. If the signal processing method is not applicable to these situations, it cannot be used in wall detection and post-disaster rescue applications. In view of the problem that the echo signal is affected by the wall and the signal-to-noise ratio is reduced, the heartbeat signal is weak and easily affected by respiratory harmonics, and traditional contact vital signs detection devices are limited in certain environments and have difficulty detecting human vital signs. In summary, the prior art has many problems during the detection process, such as weakened signal strength, shortened detection range, reduced signal-to-noise ratio, and enhanced multipath effect, which affect the accuracy and stability of physiological parameter detection. Summary of the invention

[0003] The present invention provides a physiological parameter detection method and system based on joint domain analysis to solve the above problems existing in the above background technology. Specifically, the technical problem solved by the present invention is achieved by adopting the following technical solutions: Physiological parameter detection method based on joint domain analysis, including: Noise reduction processing: Based on the original vital signs signals of different time series, the original vital signs signals are denoised and optimized into reconstructed vital signs signals through adaptive and soft threshold wavelet denoising, least squares fitting denoising and singular value decomposition denoising SVD; Data acquisition: Based on the reconstructed vital sign signal, the corresponding signal feature information is obtained through time domain and frequency domain analysis and wavelet transform processing. Based on the signal feature information, the average final solid mode function IMF is converted into a complete reconstructed vital sign signal through the ensemble empirical mode decomposition algorithm EEMD and weight combination reconstruction. Based on the time-frequency diagram of the reconstructed vital sign signal based on the fast Fourier transform FFT, the heart rate and respiratory rate are obtained from the identified high-frequency components and low-frequency components through the heartbeat and breathing confirmation method.

[0004] Preferably, the adaptive and soft threshold wavelet denoising includes: Based on the vital sign signals of different time series after Butterworth low-pass filtering, the vital sign signals are divided into low-frequency vital sign signals and high-frequency noise signals of different frequencies through adaptive wavelet denoising method. The low-frequency vital sign signals and high-frequency noise signals are segmented based on different frequencies. The high-frequency noise signals and low-frequency vital sign signals are removed by the soft threshold wavelet denoising method to obtain the wavelet denoised vital sign signals.

[0005] Preferably, the adaptive wavelet denoising method comprises: Based on the vital sign signal, the vital sign signal is analyzed at different frequencies through the original wavelet transform function, and finally decomposed into low-frequency vital sign signals and high-frequency noise signals of different frequencies; The soft threshold wavelet denoising method comprises: If the absolute value of the original wavelet coefficient is less than the preset soft threshold, the optimized wavelet coefficient is set to zero through the soft threshold optimization function; If the absolute value of the original wavelet coefficient is greater than or equal to the preset soft threshold, the optimized wavelet coefficient is set to the result of subtracting the original wavelet coefficient from the soft threshold through the soft threshold optimization function; The soft threshold wavelet denoising method also includes: Based on high-frequency noise signals of different frequencies, high-frequency noise signals are converted into zero-noise signals by optimizing wavelet transform functions to remove high-frequency noise signals; Based on low-frequency vital sign signals of different frequencies, the low-frequency vital sign signals are converted into noise-reduced low-frequency vital sign signals by optimizing the wavelet transform function to reduce the influence of noise on the signal; Finally, the zero noise signal and the denoised low-frequency vital sign signal are formed into a wavelet denoised vital sign signal; The optimized wavelet transform function is: ; Among them, the To optimize the wavelet coefficients; is the wavelet basis function; is a vital sign signal subjected to wavelet denoising; are different frequencies corresponding to different time series; Different positions correspond to different time series.

[0006] Preferably, the least squares fitting noise reduction includes: Based on the wavelet denoising, the vital signs signal is processed by removing the linear trend to form a nonlinear vital signs signal; Based on the nonlinear sign signal, a nonlinear flat wave sign signal is formed by eliminating the stationary clutter. Based on the nonlinear flat wave vital sign signal, a weighted nonlinear flat wave vital sign signal is formed through weighted fitting processing.

[0007] Preferably, the linear trend removal process comprises: Based on the linear trend signal, the slope and intercept of the linear trend are estimated by minimizing the weighted error, and a linear trend equation is formed; Based on the wavelet denoising of the vital sign signal, a nonlinear vital sign signal is obtained through a nonlinear fitting function to remove the linear trend in the vital sign signal; The nonlinear fitting function is: ;

[0008] Among them, the is a vital sign signal subjected to wavelet denoising; is a nonlinear sign signal; is a linear trend equation, where a is the slope, b is the intercept, and is a linear data set; The stationary clutter elimination process comprises: Based on the wavelet denoised vital sign signal, a function is formulated through the estimated value of stationary clutter to obtain the estimated value of stationary clutter; Based on the nonlinear sign signal, the sign signal after wavelet denoising is subtracted from the estimated value of the stationary clutter through the clutter elimination function to obtain the nonlinear flat wave sign signal, so as to remove the stationary clutter in the signal; The clutter elimination function is: ;

[0009] Among them, the is a nonlinear flat wave sign signal; is a nonlinear sign signal; is the estimated value of stationary clutter; The weighted fitting process includes: Based on the nonlinear flat wave sign signal, a nonlinear flat wave sign signal of a flat wave is obtained through a weighted fitting function, and a weight is assigned to each data point of the nonlinear flat wave sign signal so that high-quality data in the signal can make a high contribution to the fitting result; The weighted fitting function is: ;

[0010] Among them, the is the weight of each data point of the nonlinear flat wave vital sign signal; is a nonlinear flat wave sign signal; is the weighted nonlinear flat wave vital sign signal.

[0011] Preferably, the noise reduction process further includes: Singular value decomposition denoising SVD: The weighted nonlinear flat wave vital sign signals are formed into a vital sign signal matrix; Based on the vital sign signal matrix, through singular value decomposition, it is decomposed into: Orthogonal left singular vector matrix; Diagonal singular value matrix; Orthogonal right singular vector matrix; Based on the diagonal singular value matrix, the current maximum singular value is determined according to the maximum stretching effect in the key direction; Based on the current maximum singular value, the current maximum contribution energy of the current maximum singular value is obtained by calculating the current contribution energy; Based on all the maximum singular values, the maximum contribution energy sum of all the maximum singular values ​​is obtained by calculating the contribution energy sum; Based on each current maximum contribution energy and the total maximum contribution energy, the maximum contribution energy ratio of each current maximum singular value is obtained by energy ratio calculation; Based on the maximum contribution energy ratio, through energy ratio selection, the maximum singular value whose maximum contribution energy ratio exceeds 95% is selected to form the maximum singular value group; Based on several maximum singular value members with the largest contribution of the maximum singular value group, a low-rank approximate matrix is ​​generated through singular value reconstruction; A reconstructed vital sign signal is generated based on the reconstructed vital sign signal matrix.

[0012] Preferably, the singular value decomposition denoising SVD further includes: The current contribution energy calculation includes a current contribution energy calculation function; The current contribution energy calculation function is: ;

[0013] Among them, the is the current maximum contribution energy of the current largest singular value; is the energy contribution corresponding to the current largest singular value; the contribution energy sum calculation includes a contribution energy sum calculation function; The contribution energy sum calculation function is: ;

[0014] Among them, the is the sum of the maximum contribution energies of all the largest singular values; for The largest singular value; The energy proportion calculation includes an energy proportion calculation function; The energy proportion calculation function is: h e = ;

[0015] Among them, the h e is the maximum contribution energy ratio of each current maximum singular value; The energy proportion selection function is: ;

[0016] Among them, the is the number of maximum singular value members with the largest contribution; the singular value reconstruction includes a singular value reconstruction function; The singular value reconstruction function comprises: ;

[0017] Among them, the is the k largest contributing physical sign signal matrix; is an orthogonal left singular vector matrix of k largest contributions, wherein the left singular vector matrix includes left singular vectors; is a diagonal singular value matrix of k largest contributions, wherein the singular value matrix includes several typical singular values; is an orthogonal right singular vector matrix of k largest contributions, wherein the right singular vector matrix includes right singular vectors; The low-rank approximation matrix is: ; ; ; ; ; Among them, the is the reconstructed vital sign signal matrix; is a reconstructed orthogonal left singular vector matrix, wherein the left singular vector matrix includes left singular vectors; is a reconstructed diagonal singular value matrix, wherein the singular value matrix includes several typical singular values; is a reconstructed orthogonal right singular vector matrix, wherein the right singular vector matrix includes right singular vectors.

[0018] Preferably, the data acquisition further includes: Based on the reconstructed vital sign signal, the signal characteristic information of the time domain information and the frequency domain information of the reconstructed vital sign signal is identified through time domain and frequency domain analysis; Based on the reconstructed vital sign signal, signal characteristic information of the spectrum information and frequency information of the reconstructed vital sign signal is obtained through wavelet transform processing; Based on the signal feature information, each group of decomposed solid mode functions IMF is averaged into the final solid mode function IMF through the ensemble empirical mode decomposition algorithm EEMD; Based on each group of solid state mode functions IMF, according to the weights corresponding to each group of solid state mode functions IMF, a complete reconstructed vital sign signal is obtained by weight combination reconstruction; Based on the reconstructed vital sign signal, the amplitude spectrum and phase spectrum in the reconstructed vital sign signal are obtained through fast Fourier transform FFT; Based on the amplitude spectrum and the phase spectrum, the high-frequency component and the low-frequency component in the reconstructed vital sign signal are determined by signal frequency recognition; Based on the high-frequency component and the low-frequency component in the reconstructed vital sign signal, the heart rate and the respiratory rate are obtained through the first heartbeat and breathing confirmation method or the second heartbeat and breathing confirmation method.

[0019] Preferably, the data acquisition further includes: The weighted combination reconstruction includes a weighted combination reconstruction function; The weight combination reconstruction function is: ; Among them, the is the weighting coefficient of each IMF component; To reconstruct the vital signs signal again; The first heartbeat and breathing confirmation method comprises: Based on the high-frequency components in the reconstructed vital sign signal, the high-frequency components are used to determine the heartbeat vital sign signal range through frequency component fitting; Based on the low-frequency component in the reconstructed vital sign signal, the low-frequency component is determined to determine the range of the respiratory vital sign signal through frequency component fitting; Based on the range of the heartbeat vital sign signal, determine the high-frequency region of the spectrum, determine the peak with the largest amplitude in the region based on the high-frequency region of the spectrum, and determine the frequency corresponding to the peak with the largest amplitude as the heartbeat frequency; Based on the range of the respiratory vital sign signal, determine the low-frequency region of the spectrum, determine the peak with the largest amplitude in the region based on the low-frequency region of the spectrum, and determine the frequency corresponding to the peak with the largest amplitude as the respiratory frequency; The frequency component fitting includes a frequency component fitting function; The frequency component fitting function is: ; Among them, the The high-frequency component and the low-frequency component of the frequency domain signal are respectively identified by the amplitude spectrum; are high frequency components and low frequency components in the frequency domain signal; said f is the frequency domain signal; The second heartbeat and breathing confirmation method comprises: If the final solid state mode function IMF is in a state of horizontal stripes with a high degree of compactness, it is inferred that the high frequency component of the current final solid state mode function IMF is in the range of the heartbeat vital sign signal; Based on the estimated high-frequency component of the IMF as the heartbeat vital sign signal range, the peak with the largest amplitude is identified, and the peak with the largest amplitude is determined as the heartbeat frequency; If the final solid-state modal function IMF is in a state of low compactness and gradual smoothness, it is inferred that the low-frequency component of the current final solid-state modal function IMF is a respiratory sign signal; Based on the inferred high-frequency component of the IMF as the respiratory sign signal range, the peak with the largest amplitude is identified, and the peak with the largest amplitude is determined as the respiratory frequency; The ensemble empirical mode decomposition algorithm EEMD includes: Based on the signal characteristic information, a white noise with a constant power spectrum density is generated through the definition of white noise and constant Gaussian distribution; Based on the reconstructed vital sign signal, by adding white noise, multiple groups of white noise signals with constant power spectrum density are added to the reconstructed vital sign signal to form multiple groups of noise auxiliary signals; Based on each group of noise auxiliary signals, the ensemble empirical mode decomposition algorithm EMD is used to decompose each group of solid mode functions IMF; Based on each group of solid-state modal functions IMF, by averaging the modal functions, in multiple experiments, the IMF corresponding to each group of noise auxiliary signals is averaged multiple times to form the final solid-state modal function IMF; The white noise addition includes a white noise addition function; The white noise adding function comprises: ; Among them, the is the i-th solid state mode function; N is the number of modes; is the noise auxiliary signal.

[0020] At the same time, the present invention also provides a physiological parameter detection system based on joint domain analysis, including: The physiological parameter detection system implements the above-mentioned physiological parameter detection method, and the physiological parameter detection system includes: Noise reduction processing module: used for optimizing the original vital sign signals into reconstructed vital sign signals based on the original vital sign signals of different time series through adaptive and soft threshold wavelet denoising, least squares fitting denoising and singular value decomposition denoising SVD; Data acquisition module: used to obtain corresponding signal feature information based on the reconstructed vital sign signal through time domain and frequency domain analysis and wavelet transform processing; based on the signal feature information, through the ensemble empirical mode decomposition algorithm EEMD and weight combination reconstruction, the average final solid mode function IMF is converted into a complete reconstructed vital sign signal; based on the time-frequency diagram of the reconstructed vital sign signal based on the fast Fourier transform FFT, the heart rate and respiratory rate are obtained from the identified high-frequency components and low-frequency components through the heartbeat and breathing confirmation method.

[0021] Beneficial technical effects: This scheme adopts methods such as noise reduction processing and data acquisition, among which the noise reduction processing adopts the original vital sign signals based on different time series, and optimizes the original vital sign signals into reconstructed vital sign signals through adaptive and soft threshold wavelet noise reduction, least squares fitting noise reduction and singular value decomposition noise reduction SVD; since there are many problems in the detection process, such as weakened signal strength, shortened detection range, reduced signal-to-noise ratio, enhanced multipath effect, etc., which affect the accuracy and stability of physiological parameter detection; In view of the above problems, this scheme proposes a data processing method for physiological parameter detection of partition walls based on EEMD and fast and slow time diagrams, and fully verifies the effectiveness of EEMD and SVD algorithms in detecting human vital signs through partition walls from the data preprocessing stage, simulation experiment stage, and experimental verification stage. This method not only improves the accuracy and stability of physiological parameter detection, but also effectively reduces noise interference in the data processing process, and enhances the signal's resolvability and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a main flow chart of Embodiment 1 of the physiological parameter detection method of the present invention; Figure 2 is a system flow chart of Embodiment 1 of the physiological parameter detection method of the present invention; Figure 3 is a flow chart of a method for noise reduction processing in Embodiment 1 of the physiological parameter detection method of the present invention; Figure 4 is a flow chart of a method for data acquisition in Embodiment 1 of the physiological parameter detection method of the present invention; Figure 5 is a module structure diagram of the second embodiment of the physiological parameter detection method of the present invention; Figure 6 It is a module flow chart of the second embodiment of the physiological parameter detection method of the present invention. DETAILED DESCRIPTION

[0023] The present invention is further described below in conjunction with the accompanying drawings: In the figure: S10-Noise reduction processing: Based on the original vital sign signals of different time series, the original vital sign signals are denoised and optimized into reconstructed vital sign signals through adaptive and soft threshold wavelet denoising, least squares fitting denoising and singular value decomposition denoising SVD; S20-data acquisition: based on the reconstructed vital sign signal, decompose it into a certain intrinsic mode function IMF through the ensemble empirical mode decomposition algorithm EEMD, and based on the intrinsic mode function IMF and time-frequency diagram analysis, determine the heart rate and respiratory rate in the reconstructed vital sign signal through frequency component fitting; S301-Adaptive and soft threshold wavelet noise reduction; S302-Least squares fitting noise reduction; S303-Singular value decomposition denoising SVD; S401- Based on the reconstructed vital sign signal, the signal feature information of the time domain information and the frequency domain information of the reconstructed vital sign signal is identified through time domain and frequency domain analysis; S402-Acquire signal characteristic information of spectrum information and frequency information of the reconstructed vital sign signal through wavelet transform processing based on the reconstructed vital sign signal; S403-Based on the signal feature information, the EEMD algorithm is used to average each group of decomposed solid mode functions IMF to obtain the final solid mode function IMF; S404-Based on each group of solid mode functions IMF, according to the weights corresponding to each group of solid mode functions IMF, a complete reconstructed vital sign signal is obtained by weight combination reconstruction; S405-Based on the reconstructed vital sign signal, obtain the amplitude spectrum and phase spectrum in the reconstructed vital sign signal through fast Fourier transform FFT; S406-determine the high-frequency component and the low-frequency component in the reconstructed vital sign signal through signal frequency recognition based on the amplitude spectrum and the phase spectrum; S407 - Based on the high-frequency components and low-frequency components in the reconstructed vital sign signal, the heart rate and respiratory rate are obtained through the first heartbeat and breathing confirmation method or the second heartbeat and breathing confirmation method.

[0024] Embodiment 1: like Figure 1 , 2 As shown, the physiological parameter detection method based on joint domain analysis includes: Noise reduction processing: Based on the original vital sign signals of different time series, the original vital sign signals are denoised and optimized into reconstructed vital sign signals S10 through adaptive and soft threshold wavelet denoising S301, least squares fitting denoising S302 and singular value decomposition denoising SVDS303; Data acquisition: Based on the reconstructed vital sign signal, the corresponding signal feature information is obtained through time domain and frequency domain analysis and wavelet transform processing. Based on the signal feature information, the average final solid mode function IMF is converted into a complete reconstructed vital sign signal through the ensemble empirical mode decomposition algorithm EEMD and weight combination reconstruction. Based on the time-frequency diagram of the reconstructed vital sign signal based on the fast Fourier transform FFT, the heart rate and respiratory rate S20 are obtained from the identified high-frequency components and low-frequency components through the heartbeat and breathing confirmation method.

[0025] This scheme adopts noise reduction processing, data acquisition and other methods, wherein the noise reduction processing adopts the original vital sign signal based on different time series, and optimizes the original vital sign signal into a reconstructed vital sign signal through adaptive and soft threshold wavelet noise reduction and least square fitting noise reduction and singular value decomposition noise reduction SVD, wherein the data acquisition adopts the reconstructed vital sign signal, and decomposes it into a certain intrinsic mode function IMF through the ensemble empirical mode decomposition algorithm EEMD, and based on the intrinsic mode function IMF and time-frequency diagram analysis, the heart rate and respiratory rate in the reconstructed vital sign signal are determined through frequency component fitting; currently, in practice, due to It is impossible to provide non-intrusive detection solutions for all-weather or special environments, which leads to the failure to timely detect health abnormalities or affect rescue judgment. More specifically, due to the weakening of signal strength, shortened detection range, reduced signal-to-noise ratio, enhanced multipath effect and many other problems in the detection process, the accuracy and stability of physiological parameter detection are affected. In view of the above problems, this paper proposes a data processing method for physiological parameter detection of partition walls based on EEMD and fast and slow time diagrams. Specifically, we mainly focus on three main points: First, in the data preprocessing stage, we make full use of linear least squares fitting and SVD decomposition. Remove the stationary clutter and non-stationary clutter generated by environmental noise, so as to obtain relatively pure vital sign detection signals, which provides a basis for subsequent data processing and analysis; second, in the simulation experiment stage, use EEMD decomposition to obtain the intrinsic mode function (IMF) component, reconstruct the respiratory and heartbeat signals according to the fast and slow time diagrams, analyze the energy intensity characteristics of the fast and slow time diagrams, and roughly determine the frequency range of heartbeat and breathing. Finally, the specific respiratory and heartbeat vital signs values ​​were obtained through Fourier transform; Third, in the experimental verification stage, the simulation experimental data was collected by DW1000 radar through the wall, which fully verified the effectiveness of EEMD and SVD algorithms in detecting human vital signs through the wall. To be more specific, the original data collected by the broadband radar DW000 was a 1000*100 matrix. After simple preprocessing of the data using Butterworth low-pass filtering, a 1000*98 time series was obtained. After visualizing the filtering results, it was found that although the high-frequency noise in the data was initially removed, there were still stationary clutter, non-stationary clutter and white noise in the low-frequency part; in order to further remove these noises in the signal, linear least squares fitting, singular value decomposition and other processing were performed, and the preprocessed data was decomposed by EEMD, and the appropriate components were found by constructing fast and slow time diagrams to estimate the respiratory frequency and heart rate; This method not only improves the accuracy and stability of physiological parameter detection, but also effectively reduces noise interference during data processing, and enhances the signal's resolvability and reliability.

[0026] like Figure 3 As shown, adaptive and soft threshold wavelet denoising S301: Based on the vital sign signals of different time series after Butterworth low-pass filtering, the vital sign signals are divided into low-frequency vital sign signals and high-frequency noise signals of different frequencies through adaptive wavelet denoising method. The low-frequency vital sign signals and high-frequency noise signals are segmented based on different frequencies. The high-frequency noise signals and low-frequency vital sign signals are removed by the soft threshold wavelet denoising method to obtain the wavelet denoised vital sign signals.

[0027] This scheme uses adaptive and soft threshold wavelet denoising to obtain wavelet denoised vital sign signals. Specifically, this scheme includes (1) based on the vital sign signals of different time series after Butterworth low-pass filtering, the vital sign signals are divided into low-frequency vital sign signals and high-frequency noise signals of different frequencies through an adaptive wavelet denoising method; (2) based on the segmentation of low-frequency vital sign signals and high-frequency noise signals of different frequencies, the high-frequency noise signals are removed and the low-frequency vital sign signals are denoised through a soft threshold wavelet denoising method to obtain wavelet denoised vital sign signals; in order to remove high-frequency noise in the data, an adaptive wavelet denoising method is used. The adaptive wavelet denoising method is a noise suppression technology based on wavelet transform, which can effectively denoise signals at different scales and frequency resolutions. Wavelet transform is a method based on signal decomposition, which analyzes the signal at different scales (frequencies). For signals that are , its discrete wavelet transform can be expressed as: ;

[0028] Among them, in the above formula, are the coefficients of the wavelet transform; is the wavelet basis function, Indicates scale, Indicates location.

[0029] Through discrete wavelet transform, the signal can be decomposed into low-frequency parts (approximation coefficients) and high-frequency parts of different scales; using the soft threshold method, a threshold is applied to the wavelet coefficients , so that the absolute value is less than The coefficients of are set to zero, and the absolute values ​​are greater than The coefficient of The formula for soft threshold is as follows: ;

[0030] Among them, in the above formula, is the coefficient after denoising;

[0031] like Figure 3As shown, the adaptive wavelet denoising method includes analyzing the vital sign signal at different frequencies through the original wavelet transform function based on the vital sign signal, and finally decomposing it into low-frequency vital sign signals and high-frequency noise signals of different frequencies; The soft threshold wavelet denoising method comprises: If the absolute value of the original wavelet coefficient is less than the preset soft threshold, the optimized wavelet coefficient is set to zero through the soft threshold optimization function; If the absolute value of the original wavelet coefficient is greater than or equal to the preset soft threshold, the optimized wavelet coefficient is set to the result of subtracting the original wavelet coefficient from the soft threshold through the soft threshold optimization function; The soft threshold wavelet denoising method also includes: Based on high-frequency noise signals of different frequencies, high-frequency noise signals are converted into zero-noise signals by optimizing wavelet transform functions to remove high-frequency noise signals; Based on low-frequency vital sign signals of different frequencies, the low-frequency vital sign signals are converted into noise-reduced low-frequency vital sign signals by optimizing the wavelet transform function to reduce the influence of noise on the signal; Finally, the zero noise signal and the denoised low-frequency vital sign signal are formed into a wavelet denoised vital sign signal; The optimized wavelet transform function is: ;

[0032] Among them, the To optimize the wavelet coefficients; is the wavelet basis function; is a vital sign signal subjected to wavelet denoising; are different frequencies corresponding to different time series; Different positions correspond to different time series.

[0033] This scheme obtains a wavelet-denoised vital sign signal through adaptive and soft threshold wavelet denoising. Specifically, this scheme also includes (1) based on high-frequency noise signals of different frequencies, by optimizing the wavelet transform function, the high-frequency noise signal is converted into a zero-noise signal to remove the high-frequency noise signal; (2) based on low-frequency vital sign signals of different frequencies, by optimizing the wavelet transform function, the low-frequency vital sign signal is converted into a denoised low-frequency vital sign signal to reduce the influence of noise on the signal; (3) finally, the zero-noise signal and the denoised low-frequency vital sign signal are formed into a wavelet-denoised vital sign signal; since for the high-frequency part (usually noise), when the absolute value of its coefficient is less than the threshold, it will be set to zero, thereby removing the noise; and for the important signal part (i.e., the coefficient greater than the threshold), the influence of noise on the signal is reduced by subtracting the threshold.

[0034] like Figure 3 As shown, the least squares fitting noise reduction S302: Based on the wavelet denoising, the vital signs signal is processed by removing the linear trend to form a nonlinear vital signs signal; Based on the nonlinear sign signal, a nonlinear flat wave sign signal is formed by eliminating the stationary clutter. Based on the nonlinear flat wave vital sign signal, a weighted nonlinear flat wave vital sign signal is formed through weighted fitting processing.

[0035] This solution uses the least squares fitting method to reduce noise and obtain weighted nonlinear flat wave sign signals, including: During the wall detection process, the signal will be interfered by many factors, among which the gravel in the wall and the stable clutter caused by the background environment are one of the main sources of interference. At the same time, due to the instability of the amplitude during the data acquisition process, the signal amplitude may produce a linear trend in time. In order to solve these problems, the linear least squares fitting method can be used to estimate and remove stable clutter and linear trends; stable clutter refers to random noise in the signal that does not change with time. This noise may come from the structural heterogeneity inside the wall, such as gravel, uneven wall materials, etc. These stable clutters will form a fixed background noise layer in the signal, affecting the detection and analysis of useful signals. Linear trend refers to the continuous change of signal amplitude in time. This change is usually caused by the instability of the amplitude of the equipment during the data acquisition process. If the signal presents a linear and continuous change trend on the time axis, then this trend will mask the nonlinear characteristics of the actual signal, causing difficulties in data analysis. In order to eliminate stationary clutter and remove linear trends, we can use linear fitting methods. In order to remove "stationary clutter", we use (1) wavelet denoising-based vital sign signals to remove linear trends to form nonlinear vital sign signals; (2) based on nonlinear vital sign signals, by eliminating stationary clutter processing to form nonlinear flat wave vital sign signals; (3) based on nonlinear flat wave vital sign signals, by weighted fitting processing to form weighted nonlinear flat wave vital sign signals.

[0036] like Figure 3 As shown, the least squares fitting noise reduction S302 also includes: The linear trend removal process includes: Based on the linear trend signal, the slope and intercept of the linear trend are estimated by minimizing the weighted error, and a linear trend equation is formed; Based on the wavelet denoising of the vital sign signal, a nonlinear vital sign signal is obtained through a nonlinear fitting function to remove the linear trend in the vital sign signal; The nonlinear fitting function is: ;

[0037] Among them, the is a vital sign signal subjected to wavelet denoising; is a nonlinear sign signal; is a linear trend equation, where a is the slope, b is the intercept, and is a linear data set; The stationary clutter elimination process comprises: Based on the wavelet denoised vital sign signal, a function is formulated through the estimated value of stationary clutter to obtain the estimated value of stationary clutter; Based on the nonlinear sign signal, the sign signal after wavelet denoising is subtracted from the estimated value of the stationary clutter through the clutter elimination function to obtain the nonlinear flat wave sign signal, so as to remove the stationary clutter in the signal; The clutter elimination function is: ;

[0038] Among them, the is a nonlinear flat wave sign signal; is a nonlinear sign signal; is the estimated value of the stationary clutter; the weighted fitting process includes: Based on the nonlinear flat wave sign signal, a nonlinear flat wave sign signal of a flat wave is obtained through a weighted fitting function, and a weight is assigned to each data point of the nonlinear flat wave sign signal so that high-quality data in the signal can make a high contribution to the fitting result; The weighted fitting function is: ;

[0039] Among them, the is the weight of each data point of the nonlinear flat wave vital sign signal; is a nonlinear flat wave sign signal; is the weighted nonlinear flat wave vital sign signal.

[0040] This solution uses the least squares method to fit and reduce noise, and obtains a weighted nonlinear flat wave vital sign signal. It also includes: (1) removing linear trend processing; (2) eliminating stationary clutter processing; (3) weighted fitting processing; In this case, the respiratory and heartbeat signals are more important than other parts, so the weighted least squares method can be used to assign a weight to each data point so that the higher quality part of the signal has a greater impact on the fitting result. The goal of the weighted least squares method is to minimize the following loss function: ;

[0041] Among them, in the above formula, is the weight of each data point; ; For cases where there is a linear trend in the signal, the goal is to fit a linear function by weighted least squares , where a is the slope and b is the intercept. The weighted least squares method estimates these two parameters by minimizing the weighted error. Once the estimated values ​​a and b of the linear trend are obtained, this trend can be subtracted from the original signal. For each data point, the signal after removal is: ;

[0042] Among them, in the above formula, is the signal after the linear trend is removed. When there is stationary clutter in the signal, the stationary clutter usually appears as a fixed background noise layer in the signal, affecting the accuracy of the signal. Assuming that the signal consists of useful signal and stationary clutter (noise), the weighted least squares method is used to fit the clutter part. If the stationary clutter is a constant background noise, it can be fitted with the following linear model: ;

[0043] Where c is the estimate of the stationary clutter. The sum of squared errors of the weighted least squares method is: ;

[0044] Taking the partial derivative of c and setting it to zero, we can get an estimate of the stationary clutter: ;

[0045] After calculating the estimated value c of the stationary clutter, subtract the constant value from the original signal, that is: ;

[0046] In this way, the stationary clutter part of the signal can be removed to obtain the signal after clutter removal. By performing linear fitting on the data, the stationary clutter and linear trend are estimated, and the stationary clutter and linear trend are removed by subtracting the fitting value from the data.

[0047] like Figure 3 As shown, the singular value decomposition denoising SVDS303: The weighted nonlinear flat wave vital sign signals are formed into a vital sign signal matrix; Based on the vital sign signal matrix, through singular value decomposition, it is decomposed into: Orthogonal left singular vector matrix; Diagonal singular value matrix; Orthogonal right singular vector matrix; Based on the diagonal singular value matrix, the current maximum singular value is determined according to the maximum stretching effect in the key direction; Based on the current maximum singular value, the current maximum contribution energy of the current maximum singular value is obtained by calculating the current contribution energy; Based on all the maximum singular values, the maximum contribution energy sum of all the maximum singular values ​​is obtained by calculating the contribution energy sum; Based on each current maximum contribution energy and the total maximum contribution energy, the maximum contribution energy ratio of each current maximum singular value is obtained by energy ratio calculation; Based on the maximum contribution energy ratio, through energy ratio selection, the maximum singular value whose maximum contribution energy ratio exceeds 95% is selected to form the maximum singular value group; Based on several maximum contributing maximum singular value members of the maximum singular value group, a low-rank approximate matrix is ​​generated through singular value reconstruction.

[0048] This scheme uses singular value decomposition denoising SVD to obtain a low-rank approximate matrix, including: singular value decomposition and other denoising operations to generate a low-rank approximate matrix. In order to remove "unsteady clutter", in the process of detecting physiological parameters of the partition wall, non-stationary clutter will be generated due to the environment, the human target being measured, and the characteristics of the radar itself; in the process of partition wall detection, the heterogeneity and unevenness of the wall, as well as the small moving objects in the environment, the non-periodic and irregular movement of the human body, etc. will introduce non-stationary clutter; at the same time, the radar signal will be reflected by objects such as walls and furniture during propagation, resulting in multipath effects, causing the amplitude of the received signal to change, and also causing non-stationary clutter. In addition, since the ultra-wideband radar has a large bandwidth, although the large bandwidth helps to improve the resolution of the signal, it will also make the radar more sensitive to small changes in the environment and the target, resulting in the inevitable presence of non-stationary clutter in the received signal.

[0049] like Figure 3 As shown, the singular value decomposition denoising SVDS303 also includes: The current contribution energy calculation includes a current contribution energy calculation function; The current contribution energy calculation function is: ;

[0050] Among them, the is the current maximum contribution energy of the current largest singular value; is the energy contribution corresponding to the current largest singular value.

[0051] The contribution energy sum calculation includes a contribution energy sum calculation function; The contribution energy sum calculation function is: ;

[0052] Among them, the is the sum of the maximum contribution energies of all the largest singular values; for The largest singular value; The energy proportion calculation includes an energy proportion calculation function; The energy proportion calculation function is: h e = ;

[0053] Among them, the h e is the maximum contribution energy ratio of each current maximum singular value; The energy proportion selection includes an energy proportion selection function; The energy proportion selection function is: ;

[0054] Among them, the The number of members with the largest singular values ​​that contribute the most; The singular value reconstruction includes a singular value reconstruction function; The singular value reconstruction function comprises: ;

[0055] Among them, the is the k largest contributing physical sign signal matrix; is an orthogonal left singular vector matrix of k largest contributions, wherein the left singular vector matrix includes left singular vectors; is a diagonal singular value matrix of k largest contributions, wherein the singular value matrix includes several typical singular values; is an orthogonal right singular vector matrix of k largest contributions, wherein the right singular vector matrix includes right singular vectors.

[0056] The low-rank approximation matrix is: ; ; ; ; ; Among them, the is the reconstructed vital sign signal matrix; is a reconstructed orthogonal left singular vector matrix, wherein the left singular vector matrix includes left singular vectors; is a reconstructed diagonal singular value matrix, wherein the singular value matrix includes several typical singular values; is a reconstructed orthogonal right singular vector matrix, wherein the right singular vector matrix includes right singular vectors.

[0057] Among them, the is the reconstructed vital sign signal matrix; is a reconstructed orthogonal left singular vector matrix, wherein the left singular vector matrix includes left singular vectors; is a reconstructed diagonal singular value matrix, wherein the singular value matrix includes several typical singular values; Said is a reconstructed orthogonal right singular vector matrix, wherein the right singular vector matrix includes right singular vectors.

[0058] This scheme uses functions such as contribution energy calculation function and contribution energy sum calculation function to finally generate a low-rank approximate matrix; in the data processing process, in order to remove non-stationary clutter, the singular value decomposition method is used. Through singular value decomposition, the data matrix A that removes stationary clutter and linear trends is decomposed into the product of three matrices, an orthogonal left singular vector matrix U, a diagonal singular value matrix Σ, and an orthogonal right singular vector matrix V. Among them, the data matrix A is an m*n matrix, and the U obtained after SVD decomposition is an m*m matrix, Σ is an m*n matrix, and V is an n*n matrix. The largest singular value corresponds to the maximum stretching effect of the matrix in the most important direction, which is usually associated with the main information or energy of the matrix. Therefore, finding the singular value of the signal can find the key to the signal energy. The principle of singular value selection is that according to the physical meaning, the main singular value is selected based on the energy of the selected singular value in the total energy is greater than 95%, and then a low-rank approximate matrix is ​​constructed based on the found singular value and visualized. The specific process is as follows: The data matrix A (size is ) performs singular value decomposition and obtains the product of three matrices: ; in, yes An orthogonal matrix containing left singular vectors; yes A diagonal matrix of containing singular values; yes An orthogonal matrix containing right singular vectors. The size of the singular values ​​reflects the "energy" or importance of the matrix in a particular direction. Energy for The contribution of is usually defined as: ; Total Energy is the sum of all singular values: ; The energy proportion of each singular value is: Energy ratio = ; According to the energy proportion of each singular value, select the main singular value whose cumulative energy proportion is greater than 95%. The energy proportion of the singular values ​​satisfies: ; The selected The energy contained in the first singular value basically covers most of the information of the signal, while the subsequent singular values ​​contribute less and can be removed. After selecting the main singular values, according to these singular values ​​and the singular vectors of pairs, the following is constructed: ;

[0059] After SVD decomposition, the main change trend of the echo signal is well preserved. In addition, the echo signal becomes smoother, the noise component is effectively suppressed, and the signal noise reduction and data compression processing are achieved. The low-rank approximation matrix is ​​as follows: ; ; ; ; ; like Figure 4 As shown, the data acquisition S20 also includes: Based on the reconstructed vital sign signal matrix, restore it to a reconstructed vital sign signal S401; Based on the reconstructed vital sign signal, identifying signal feature information of the time domain information and the frequency domain information of the reconstructed vital sign signal through time domain and frequency domain analysis S402; Based on the reconstructed vital sign signal, signal characteristic information of the spectrum information and frequency information of the reconstructed vital sign signal is obtained through wavelet transformation processing S403; Based on the signal feature information, each group of decomposed solid mode functions IMF is averaged to obtain the final solid mode function IMFS404 through the ensemble empirical mode decomposition algorithm EEMD; Based on each group of solid state modal functions IMF, according to the weights corresponding to each group of solid state modal functions IMF, a complete reconstructed vital sign signal is obtained by weight combination reconstruction S405; Based on the reconstructed vital sign signal, the amplitude spectrum and phase spectrum of the reconstructed vital sign signal are obtained by fast Fourier transform (FFT) S406; Based on the amplitude spectrum and the phase spectrum, the high-frequency component and the low-frequency component in the reconstructed vital sign signal are determined by signal frequency recognition S407; Based on the high-frequency component and the low-frequency component in the reconstructed vital sign signal, the heart rate and the respiratory rate are obtained by the first heartbeat and breathing confirmation method or the second heartbeat and breathing confirmation method S408.

[0060] This scheme obtains the heart rate and respiratory rate through time domain and frequency domain analysis, wavelet transform processing, etc. In order to decompose and reconstruct the signal and extract physiological parameters, that is, to extract the respiratory rate and heart rate from the data matrix, the empirical mode decomposition algorithm (EEMD) is selected in this paper to decompose the relevant data. The empirical mode decomposition algorithm is an adaptive decomposition algorithm for analyzing non-stationary nonlinear signals. It targets problems such as mode aliasing in the ensemble empirical mode decomposition algorithm (EMD).

[0061] like Figure 4 As shown, the data acquisition S20 also includes: The weighted combination reconstruction includes a weighted combination reconstruction function; The weight combination reconstruction function is: ; Said is the weighting coefficient for each IMF component; Said To reconstruct the vital signs signal again; The first heartbeat and breathing confirmation method comprises: Based on the high-frequency components in the reconstructed vital sign signal, the high-frequency components are used to determine the heartbeat vital sign signal range through frequency component fitting; Based on the low-frequency component in the reconstructed vital sign signal, the low-frequency component is determined to determine the range of the respiratory vital sign signal through frequency component fitting; Based on the range of the heartbeat vital sign signal, determine the high-frequency region of the spectrum, determine the peak with the largest amplitude in the region based on the high-frequency region of the spectrum, and determine the frequency corresponding to the peak with the largest amplitude as the heartbeat frequency; Based on the range of the respiratory vital sign signal, determine the low-frequency region of the spectrum, determine the peak with the largest amplitude in the region based on the low-frequency region of the spectrum, and determine the frequency corresponding to the peak with the largest amplitude as the respiratory frequency; The frequency component fitting includes a frequency component fitting function; The frequency component fitting function is: ; Said The high-frequency component and low-frequency component of the frequency domain signal are respectively identified by the amplitude spectrum; Said are the high-frequency components and low-frequency components in the frequency domain signal; The f is a frequency domain signal; The second heartbeat and breathing confirmation method comprises: If the final solid state mode function IMF is in a state of horizontal stripes with a high degree of compactness, it is inferred that the high frequency component of the current final solid state mode function IMF is in the range of the heartbeat vital sign signal; Based on the estimated high-frequency component of the IMF as the heartbeat vital sign signal range, the peak with the largest amplitude is identified, and the peak with the largest amplitude is determined as the heartbeat frequency; If the final solid-state modal function IMF is in a state of low compactness and gradual smoothness, it is inferred that the low-frequency component of the current final solid-state modal function IMF is a respiratory sign signal; Based on the inferred high-frequency component of the IMF as the respiratory sign signal range, the peak with the largest amplitude is identified, and the peak with the largest amplitude is determined as the respiratory frequency; The ensemble empirical mode decomposition algorithm EEMD includes: Based on the signal characteristic information, a white noise with a constant power spectrum density is generated through the definition of white noise and constant Gaussian distribution; Based on the reconstructed vital sign signal, by adding white noise, multiple groups of white noise signals with constant power spectrum density are added to the reconstructed vital sign signal to form multiple groups of noise auxiliary signals; Based on each group of noise auxiliary signals, the ensemble empirical mode decomposition algorithm EMD is used to decompose each group of solid mode functions IMF; Based on each group of solid-state modal functions IMF, by averaging the modal functions, in multiple experiments, the IMFs corresponding to each group of noise auxiliary signals are correspondingly averaged multiple times to form the final solid-state modal function IMF.

[0062] The white noise addition includes a white noise addition function; The white noise adding function comprises: ; Among them, the is the i-th solid state mode function; N is the number of modes; is the noise auxiliary signal; This scheme obtains the heart rate and respiratory rate through time domain and frequency domain analysis, wavelet transform processing, etc. The specific process is as follows: For the original signal Analyze and identify its time domain or frequency domain characteristics. Use wavelet transform to obtain the spectrum information of the signal and understand the frequency range of the signal. Based on the signal analysis results, multiple noise auxiliary signals are formed by adding multiple independent groups of white noise to the original signal. White noise has a flat spectrum and can cover all frequency components. It is suitable for a wide frequency range of the signal. The definition of white noise is: ; in, is a white noise signal, The mean is 0 and the variance is Gaussian distribution, the power spectral density of white noise is constant: ; The generated white noise will be added to the original signal to form a noise auxiliary signal: ; Perform EMD decomposition on each noise auxiliary signal to obtain a set of solid mode functions (IMFs). ;

[0063] in, is the ith solid state mode function, and N is the number of modes.

[0064] Multiple noise auxiliary signals are generated using different white noise signals and EMD decomposition is performed on each of them. Finally, each IMF obtained is averaged in multiple experiments to reduce the impact of noise on the decomposition results. Then the IMFs of all noise auxiliary signals are averaged to obtain the final IMFs to reduce the problem of modal aliasing and improve the stability and robustness of the decomposition.

[0065] ; There are many obvious horizontal stripes in the time-frequency diagram of the IMF component. The IMF components are distinguished by the density and color changes of the stripes. Among them, the IMF components with high density and obvious horizontal stripes represent the existence of high-frequency components, from which it can be inferred that the component is a heartbeat signal. The horizontal stripes are wider, the intervals are larger, and the color gradient is smoother. These features suggest low-frequency components, and it can be inferred that the IMF component may be a respiratory signal. In order to extract the respiratory frequency and heart rate from the reconstructed signal and generate a respiratory and heart rate graph, the IMF components need to be reconstructed. The reconstructed signal is usually obtained by recombining all meaningful IMF components. The selected IMF components are combined into a reconstructed signal according to certain weights, as shown in the following formula: ; in, is the weighting coefficient of each IMF component.

[0066] By reconstructing the IMF components containing the breathing and heartbeat signals, the complete breathing and heartbeat signals can be obtained. Fourier transform can convert the time domain signal to the frequency domain, thereby extracting the frequency components of the signal. Perform fast Fourier transform (FFT) on the reconstructed signal to obtain the frequency components: ; Among them, in the above formula, is a frequency domain signal; is the Fourier transform. The amplitude spectrum and phase spectrum are obtained after the Fourier transform. The main frequency components in the signal can be identified by the amplitude spectrum: ; Typically, the frequency range of the heartbeat signal is between 0.5 Hz and 2 Hz, while the frequency range of the breathing signal is between 0.1 Hz and 0.5 Hz. The breathing frequency usually appears in the low-frequency region of the spectrum (0.1 Hz to 0.5 Hz), and the peak with the largest amplitude in the spectrum can be found as the breathing frequency. The heartbeat frequency usually appears in the high-frequency region of the spectrum (0.5 Hz to 2 Hz), and the peak with the largest amplitude in the spectrum can also be found as the heartbeat frequency. By extracting the peak value of the spectrum graph, the specific values ​​of breathing and heartbeat can be determined.

[0067] At the same time, the present invention also provides a physiological parameter detection system based on joint domain analysis, including: The physiological parameter detection system implements the above-mentioned physiological parameter detection method, and the physiological parameter detection system includes: Noise reduction processing module: used for optimizing the original vital sign signals into reconstructed vital sign signals S101 based on the original vital sign signals of different time series through adaptive and soft threshold wavelet noise reduction S301, least squares fitting noise reduction S302, and singular value decomposition noise reduction SVDS303; Data acquisition module: used to obtain corresponding signal feature information based on the reconstructed vital sign signal through time domain and frequency domain analysis and wavelet transform processing; based on the signal feature information, through the ensemble empirical mode decomposition algorithm EEMD and weight combination reconstruction, the average final solid mode function IMF is converted into a complete reconstructed vital sign signal; based on the time-frequency diagram of the reconstructed vital sign signal based on the fast Fourier transform FFT, the heart rate and respiratory rate are obtained from the identified high-frequency components and low-frequency components through the heartbeat and breathing confirmation method S201.

[0068] The present invention also provides a physiological parameter detection system based on joint domain analysis, which includes a noise reduction processing module and a data acquisition module. At the same time, through the function of the noise reduction processing module: through adaptive and soft threshold wavelet noise reduction and least squares fitting noise reduction and singular value decomposition noise reduction SVD, the original vital sign signal is denoised and optimized to be a reconstructed vital sign signal; and through the function of the data acquisition module: based on the reconstructed vital sign signal, it is decomposed into a certain intrinsic mode function IMF through the ensemble empirical mode decomposition algorithm EEMD, based on the intrinsic mode function IMF and time-frequency diagram analysis, through frequency component fitting, the heart rate and respiratory rate in the reconstructed vital sign signal are determined, thereby finally determining the heart rate and respiratory rate, thereby improving the accuracy and stability of physiological parameter detection.

[0069] Embodiment 2: like Figure 5 , 6 As shown, the experimental verification phase includes: The acquisition of original vital signs signals of different time series includes: Contactless sensing means that the sensor does not need to directly contact the object being measured, the sensor and the object being measured are completely separated, the object being measured is not disturbed, the accuracy of sensing does not change due to the influence of the object being measured, and more software is allowed to regulate the sensing accuracy. The realization of contactless sensing depends on a certain medium such as sound, light, and electromagnetism. Therefore, contactless sensing is divided into sound sensing, light sensing, and electric (magnetic) sensing. Common methods in light sensing include visible light and infrared light, which have strict requirements on light conditions, limited field of view, and privacy issues. Although sound sensing has high detection accuracy, it has a small sensing range and poor universality. Common technologies for electric (magnetic) sensing include wireless radio frequency RFID, WiFi, and radar. RFID-based sensing requires a short distance between the sensor and the tag, so it is suitable for short-distance communication. WiFi-based sensing is usually limited to indoors, has a short sensing range, and is closely related to the surrounding environment, and has poor universality. Radar is widely used in contactless human physiological parameter detection due to its high accuracy, high sensitivity, and high adaptability.

[0070] Commonly used radar technologies include continuous wave radar, frequency modulated continuous wave radar (FMCW) and ultra-wideband radar (UWB). Although continuous wave radar has a simple structure and low power consumption, it has poor distance resolution and is difficult to calibrate. Compared with UWB radar, FMCW radar does not have advantages in distance resolution and anti-interference. At the same time, due to the small signal bandwidth, the signal penetration ability is weak. UWB radar signals have the characteristics of wide bandwidth, strong penetration ability, high distance resolution ability, good anti-interference ability, low power consumption and low cost. It is a key technology for non-contact detection of human physiological parameters through walls. In summary, contactless sensing technology achieves high-precision physiological parameter detection without disturbing the object being measured. Its specific application and development prospects are particularly significant in the health management of the elderly and post-disaster rescue.

[0071] About the ultra-wideband channel impulse response signal acquisition module: The core function of the power module in the ultra-wideband device is to provide sufficient voltage and current requirements for the system. In this embodiment, the LD1117 series chip is selected as the voltage stabilizing chip, and the corresponding main power switch, battery interface and power indicator light are designed. The communication interface module provides an interface for downloading the single-chip program and transmitting the ultra-wideband channel impulse response data. The single-chip module is used for ultra-wideband chip signal reception and control and ultra-wideband channel impulse response data collection. In this embodiment, the STM32F105 series RCT6 chip is selected as the data processing chip. The ultra-wideband module mainly includes the DWM1000 chip and its peripheral circuits, which are the core parts for realizing the reception and transmission of ultra-wideband signals. The overall hardware structure diagram of the ultra-wideband device is as follows: Figure 5 shown.

[0072] In this embodiment, the specific process of collecting the ultra-wideband channel impulse response signal is as follows: Step 1: The person being tested is located in the room facing the wall, the ultra-wideband device is close to the outside of the wall, and the pulse signal is transmitted outside the wall. During the detection process, the radar samples the UWB channel at a sampling frequency of 1 nanosecond, and intercepts about 1000 sample points before and after the received signal into the memory. The data is composed of 16-bit imaginary numbers and 16-bit real numbers. By taking the modulus of the sample point, the amplitude data of the sample point can be obtained. The amplitude data can be used to intuitively observe the performance of the UWB signal in the time domain. The data acquisition process is as follows: Figure 6 shown.

[0073] Step 2: In the communication system, the signal will encounter various reflection, refraction and scattering phenomena during the propagation process. These phenomena will cause the signal to reach the receiving end through different paths, forming multipath propagation. This multipath effect will cause problems such as signal attenuation and phase shift, affecting the correct decoding of the signal by the receiving end. In the ultra-wideband (UWB) communication system, due to its extremely wide bandwidth, it can penetrate obstacles and reach the receiving end at the same time. In order to more accurately extract useful signals in the process of contactless measurement of breathing and heartbeat, and accurately identify the tiny signals of heartbeat and breathing in a complex environment with more multipath effects and noise interference, the ultra-wideband channel impulse response signal model is used as follows: ; In the above formula, the left side Indicates the amplitude of the CIR signal corresponding to time t. The right side of the formula Indicates the transmission signal. It indicates the number of multipath components of the transmitted signal after mirror reflection. represents the complex amplitude of the multipath component, represents the departure angle of the multipath component, represents the transmission delay of the component, The variance is , Gaussian white noise with a mean of 0.

[0074] Step 1: Use the following model to represent the ultra-wideband channel impulse response signal ; The left side of the formula Indicates the amplitude of the CIR signal corresponding to time t. The right side of the formula Indicates the transmission signal. It indicates the number of multipath components of the transmitted signal after mirror reflection. represents the complex amplitude of the multipath component, represents the departure angle of the multipath component, represents the transmission delay of the component, The variance is , Gaussian white noise with a mean of 0.

[0075] Step 3: In order to discretize and model the collected channel impulse response signal, extract the tiny displacement information caused by human breathing and heartbeat, and thus achieve high-precision long-distance detection. The discretization model is as follows: ; ; The above formula represents discretizing the continuous UWB signal into a vector, where represents the sampling interval, Represents the number of sampling points. Such discretization processing can convert a continuous signal into a set of discrete sampling points. ; The above formula represents a matrix containing multipath effect information, which depends on the phase of the signal and delay Each row corresponds to a different time point, and each column corresponds to a different multipath component. This matrix construction method helps to describe the characteristics of different reflection paths in the channel, and then describe the subtle displacement caused by human breathing and heartbeat.

[0076] ; in, They are all discretized multipath components, reflecting the attenuation and delay of the signal on different reflection paths.

[0077] ;

[0078] The vector in the above formula represents the noise term, which includes system noise and interference from the external environment. Modeling the noise as a vector and adding it to the signal model allows us to better suppress noise in subsequent signal processing.

[0079] ; In non-contact measurement, the displacement caused by human breathing and heartbeat is very small and has a certain periodicity. This tiny movement will appear as a periodic vibration signal in the multipath reflection signal. The vector represents the amplitude of each reflection path. By analyzing the amplitude, we can observe the small changes caused by the periodic movement of the human body.

[0080] Step 2: The sampled ultra-wideband channel impulse response signal can be expressed by a discretized model: ; ; ; ; ; ;

[0081] The sampling frequency is The number of samples collected is .

[0082] In this embodiment, the person being tested is located in the room facing the wall, the ultra-wideband device is close to the outside of the wall, and the pulse signal is transmitted outside the wall. During the detection process, the radar samples the UWB channel at a sampling frequency of 1 nanosecond, and intercepts about 1000 sample points before and after the received signal into the memory. The data is composed of 16-bit imaginary numbers and 16-bit real numbers. By taking the modulus of the sample points, the amplitude data of the sample points can be obtained. The performance of the UWB signal in the time domain can be intuitively observed through the amplitude data.

[0083] Utilizing the technical solution of the present invention, or those skilled in the art designing similar technical solutions inspired by the technical solution of the present invention, and achieving the above technical effects, all fall within the protection scope of the present invention. It should be pointed out that the specific implementations in the embodiments of the present invention can be at least partially combined with each other, and the combination and use of the above embodiments, or their different combinations, should also be regarded as falling within the protection scope of the present invention.

Claims

1. A physiological parameter detection method based on joint domain analysis, characterized in that: include: Noise reduction processing: Based on the original vital signs signals of different time series, the original vital signs signals are denoised and optimized into reconstructed vital signs signals through adaptive and soft threshold wavelet denoising, least squares fitting denoising and singular value decomposition denoising SVD; Data acquisition: Based on the reconstructed vital sign signal, the corresponding signal feature information is obtained through time domain and frequency domain analysis and wavelet transform processing. Based on the signal feature information, the average final solid mode function IMF is converted into a complete reconstructed vital sign signal through the ensemble empirical mode decomposition algorithm EEMD and weight combination reconstruction. Based on the time-frequency diagram of the reconstructed vital sign signal based on the fast Fourier transform FFT, the heart rate and respiratory rate are obtained from the identified high-frequency components and low-frequency components through the heartbeat and breathing confirmation method.

2. The physiological parameter detection method according to claim 1, characterized in that: The adaptive and soft threshold wavelet denoising includes: Based on the vital sign signals of different time series after Butterworth low-pass filtering, the vital sign signals are divided into low-frequency vital sign signals and high-frequency noise signals of different frequencies through adaptive wavelet denoising method. The low-frequency vital sign signals and high-frequency noise signals are segmented based on different frequencies. The high-frequency noise signals and low-frequency vital sign signals are removed by the soft threshold wavelet denoising method to obtain the vital sign signals with wavelet denoising.

3. The physiological parameter detection method according to claim 2, characterized in that: The adaptive wavelet denoising method comprises: Based on the vital sign signal, the vital sign signal is analyzed at different frequencies through the original wavelet transform function, and finally decomposed into low-frequency vital sign signals and high-frequency noise signals of different frequencies; The soft threshold wavelet denoising method comprises: If the absolute value of the original wavelet coefficient is less than the preset soft threshold, the optimized wavelet coefficient is set to zero through the soft threshold optimization function; If the absolute value of the original wavelet coefficient is greater than or equal to the preset soft threshold, the optimized wavelet coefficient is set to the result of subtracting the original wavelet coefficient from the soft threshold through the soft threshold optimization function; The soft threshold wavelet denoising method also includes: Based on high-frequency noise signals of different frequencies, high-frequency noise signals are converted into zero-noise signals by optimizing wavelet transform functions to remove high-frequency noise signals; Based on low-frequency vital sign signals of different frequencies, the low-frequency vital sign signals are converted into noise-reduced low-frequency vital sign signals by optimizing the wavelet transform function to reduce the influence of noise on the signal; Finally, the zero noise signal and the denoised low-frequency vital sign signal are formed into a wavelet denoised vital sign signal; The optimized wavelet transform function is: ； Among them, the To optimize the wavelet coefficients; is the wavelet basis function; is a vital sign signal subjected to wavelet denoising; are different frequencies corresponding to different time series; Different positions correspond to different time series.

4. The physiological parameter detection method according to claim 1, characterized in that: The least squares fitting noise reduction method includes: Based on the wavelet denoising, the vital signs signal is processed by removing the linear trend to form a nonlinear vital signs signal; Based on the nonlinear sign signal, a nonlinear flat wave sign signal is formed by eliminating the stationary clutter. Based on the nonlinear flat wave vital sign signal, a weighted nonlinear flat wave vital sign signal is formed through weighted fitting processing.

5. The physiological parameter detection method according to claim 4, characterized in that: The linear trend removal process includes: Based on the linear trend signal, the slope and intercept of the linear trend are estimated by minimizing the weighted error, and a linear trend equation is formed; Based on the wavelet denoising of the vital sign signal, a nonlinear vital sign signal is obtained through a nonlinear fitting function to remove the linear trend in the vital sign signal; The nonlinear fitting function is: ； Among them, the is a vital sign signal subjected to wavelet denoising; is a nonlinear sign signal; is a linear trend equation, where a is the slope, b is the intercept, and is a linear data set; The stationary clutter elimination process comprises: Based on the wavelet denoised vital sign signal, a function is formulated through the estimated value of stationary clutter to obtain the estimated value of stationary clutter; Based on the nonlinear sign signal, the sign signal after wavelet denoising is subtracted from the estimated value of the stationary clutter through the clutter elimination function to obtain the nonlinear flat wave sign signal, so as to remove the stationary clutter in the signal; The clutter elimination function is: ； Among them, the is a nonlinear flat wave sign signal; is a nonlinear sign signal; is the estimated value of stationary clutter; The weighted fitting process includes: Based on the nonlinear flat wave sign signal, a nonlinear flat wave sign signal of a flat wave is obtained through a weighted fitting function, and a weight is assigned to each data point of the nonlinear flat wave sign signal so that high-quality data in the signal can make a high contribution to the fitting result; The weighted fitting function is: ； Among them, the is the weight of each data point of the nonlinear flat wave vital sign signal; is a nonlinear flat wave sign signal; is the weighted nonlinear flat wave vital sign signal.

6. The physiological parameter detection method according to claim 1, characterized in that: The noise reduction process further includes: Singular value decomposition denoising SVD: The weighted nonlinear flat wave vital sign signals are formed into a vital sign signal matrix; Based on the vital sign signal matrix, through singular value decomposition, it is decomposed into: Orthogonal left singular vector matrix; Diagonal singular value matrix; Orthogonal right singular vector matrix; Based on the diagonal singular value matrix, the current maximum singular value is determined according to the maximum stretching effect in the key direction; Based on the current maximum singular value, the current maximum contribution energy of the current maximum singular value is obtained by calculating the current contribution energy; Based on all the maximum singular values, the maximum contribution energy sum of all the maximum singular values ​​is obtained by calculating the contribution energy sum; Based on each current maximum contribution energy and the total maximum contribution energy, the maximum contribution energy ratio of each current maximum singular value is obtained by energy ratio calculation; Based on the maximum contribution energy ratio, through energy ratio selection, the maximum singular value whose maximum contribution energy ratio exceeds 95% is selected to form the maximum singular value group; Based on several maximum singular value members with the largest contribution of the maximum singular value group, a low-rank approximate matrix is ​​generated through singular value reconstruction; A reconstructed vital sign signal is generated based on the reconstructed vital sign signal matrix.

7. The physiological parameter detection method according to claim 6, characterized in that: The singular value decomposition denoising SVD also includes: The current contribution energy calculation includes a current contribution energy calculation function; The current contribution energy calculation function is: ； Among them, the is the current maximum contribution energy of the current largest singular value; is the energy contribution corresponding to the current largest singular value; the contribution energy sum calculation includes a contribution energy sum calculation function; The contribution energy sum calculation function is: ； Among them, the is the sum of the maximum contribution energies of all the largest singular values; for The largest singular value; The energy proportion calculation includes an energy proportion calculation function; The energy proportion calculation function is: h e = ； Among them, the h e is the maximum contribution energy ratio of each current maximum singular value; The energy proportion selection function is: ； Among them, the is the number of maximum singular value members with the largest contribution; the singular value reconstruction includes a singular value reconstruction function; The singular value reconstruction function comprises: ； Among them, the is the k largest contributing physical sign signal matrix; is an orthogonal left singular vector matrix of k largest contributions, wherein the left singular vector matrix includes left singular vectors; is a diagonal singular value matrix of k largest contributions, wherein the singular value matrix includes several typical singular values; is an orthogonal right singular vector matrix of k largest contributions, wherein the right singular vector matrix includes right singular vectors; The low-rank approximation matrix is: ； ； ； ； ； Among them, the is the reconstructed vital sign signal matrix; is a reconstructed orthogonal left singular vector matrix, wherein the left singular vector matrix includes left singular vectors; is a reconstructed diagonal singular value matrix, wherein the singular value matrix includes several typical singular values; is a reconstructed orthogonal right singular vector matrix, wherein the right singular vector matrix includes right singular vectors.

8. The physiological parameter detection method according to claim 1, characterized in that: The data acquisition also includes: Based on the reconstructed vital sign signal, the signal characteristic information of the time domain information and the frequency domain information of the reconstructed vital sign signal is identified through time domain and frequency domain analysis; Based on the reconstructed vital sign signal, signal characteristic information of the spectrum information and frequency information of the reconstructed vital sign signal is obtained through wavelet transform processing; Based on the signal feature information, each group of decomposed solid mode functions IMF is averaged into the final solid mode function IMF through the ensemble empirical mode decomposition algorithm EEMD; Based on each group of solid state mode functions IMF, according to the weights corresponding to each group of solid state mode functions IMF, a complete reconstructed vital sign signal is obtained by weight combination reconstruction; Based on the reconstructed vital sign signal, the amplitude spectrum and phase spectrum in the reconstructed vital sign signal are obtained through fast Fourier transform FFT; Based on the amplitude spectrum and the phase spectrum, the high-frequency component and the low-frequency component in the reconstructed vital sign signal are determined by signal frequency recognition; Based on the high-frequency component and the low-frequency component in the reconstructed vital sign signal, the heart rate and the respiratory rate are obtained through the first heartbeat and breathing confirmation method or the second heartbeat and breathing confirmation method.

9. The physiological parameter detection method according to claim 8, characterized in that: The data acquisition also includes: The weighted combination reconstruction includes a weighted combination reconstruction function; The weight combination reconstruction function is: ； Among them, the is the weighting coefficient of each IMF component; To reconstruct the vital signs signal again; The first heartbeat and breathing confirmation method comprises: Based on the high-frequency components in the reconstructed vital sign signal, the high-frequency components are used to determine the heartbeat vital sign signal range through frequency component fitting; Based on the low-frequency component in the reconstructed vital sign signal, the low-frequency component is determined to determine the range of the respiratory vital sign signal through frequency component fitting; Based on the range of the heartbeat vital sign signal, determine the high-frequency region of the spectrum, determine the peak with the largest amplitude in the region based on the high-frequency region of the spectrum, and determine the frequency corresponding to the peak with the largest amplitude as the heartbeat frequency; Based on the range of the respiratory vital sign signal, determine the low-frequency region of the spectrum, determine the peak with the largest amplitude in the region based on the low-frequency region of the spectrum, and determine the frequency corresponding to the peak with the largest amplitude as the respiratory frequency; The frequency component fitting includes a frequency component fitting function; The frequency component fitting function is: ； Among them, the The high-frequency component and the low-frequency component of the frequency domain signal are respectively identified by the amplitude spectrum; are high frequency components and low frequency components in the frequency domain signal; said f is the frequency domain signal; The second heartbeat and breathing confirmation method comprises: If the final solid state mode function IMF is in a state of horizontal stripes with a high degree of compactness, it is inferred that the high frequency component of the current final solid state mode function IMF is in the range of the heartbeat vital sign signal; Based on the estimated high-frequency component of the IMF as the heartbeat vital sign signal range, the peak with the largest amplitude is identified, and the peak with the largest amplitude is determined as the heartbeat frequency; If the final solid-state modal function IMF is in a state of low compactness and gradual smoothness, it is inferred that the low-frequency component of the current final solid-state modal function IMF is a respiratory sign signal; Based on the inferred high-frequency component of the IMF as the respiratory sign signal range, the peak with the largest amplitude is identified, and the peak with the largest amplitude is determined as the respiratory frequency; The ensemble empirical mode decomposition algorithm EEMD includes: Based on the signal characteristic information, a white noise with a constant power spectrum density is generated through the definition of white noise and constant Gaussian distribution; Based on the reconstructed vital sign signal, by adding white noise, multiple groups of white noise signals with constant power spectrum density are added to the reconstructed vital sign signal to form multiple groups of noise auxiliary signals; Based on each group of noise auxiliary signals, the ensemble empirical mode decomposition algorithm EMD is used to decompose each group of solid mode functions IMF; Based on each group of solid-state modal functions IMF, by averaging the modal functions, in multiple experiments, the IMF corresponding to each group of noise auxiliary signals is averaged multiple times to form the final solid-state modal function IMF; The white noise addition includes a white noise addition function; The white noise adding function comprises: ； Among them, the is the i-th solid state mode function; N is the number of modes; is the noise auxiliary signal.

10. A physiological parameter detection system based on joint domain analysis, characterized in that: include: The physiological parameter detection system implements the physiological parameter detection method according to any one of claims 1 to 9, and the physiological parameter detection system comprises: Noise reduction processing module: used for optimizing the original vital sign signals into reconstructed vital sign signals based on the original vital sign signals of different time series through adaptive and soft threshold wavelet denoising, least squares fitting denoising and singular value decomposition denoising SVD; Data acquisition module: used to obtain corresponding signal feature information based on the reconstructed vital sign signal through time domain and frequency domain analysis and wavelet transform processing; based on the signal feature information, through the ensemble empirical mode decomposition algorithm EEMD and weight combination reconstruction, the average final solid mode function IMF is converted into a complete reconstructed vital sign signal; based on the time-frequency diagram of the reconstructed vital sign signal based on the fast Fourier transform FFT, the heart rate and respiratory rate are obtained from the identified high-frequency components and low-frequency components through the heartbeat and breathing confirmation method.

Images