A feature extraction method for vital sign anomaly based on vital sign signals

Abstract

This invention relates to the field of vital sign anomaly detection technology, specifically to a feature extraction method for vital sign anomalies based on vital sign signals: First, the time chamber size is dynamically adjusted through variational optimization to capture both transient and steady-state features, avoiding feature loss caused by a fixed window length. Second, principal component analysis, Lyapunov exponential chaotic quantization, and higher-order differential operators are combined to decouple nonlinear features from multiple dimensions in the time, frequency, and phase domains, improving feature interpretability. Furthermore, phase-velocity mapping and attention-weighted fusion are used to achieve adaptive enhancement and decoupling of spatiotemporal features, suppressing noise interference. The resulting multidimensional feature cube integrates time resolution, frequency accuracy, and phase differential information, significantly improving the robustness of anomaly detection for complex vital signs and overcoming the shortcomings of traditional methods in dynamic signal processing, such as single-dimensionality and insufficient adaptability.

Description

Technical Field

[0001] This invention relates to the field of vital sign abnormality detection technology, and in particular to a method for feature extraction of vital sign abnormalities based on vital sign signals. Background Technology

[0002] With the widespread application of FMCW (Frequency Modulated Continuous Wave) millimeter-wave radar technology in medical monitoring, non-contact vital sign detection (such as respiration, heartbeat, and body movement) has become a research hotspot. Traditional vital sign monitoring mainly relies on contact sensors (such as ECG electrodes and breathing belts), which suffer from poor comfort and susceptibility to interference. While radar-based non-contact monitoring can overcome these limitations, its signal processing faces the following technical challenges:

[0003] Transient feature loss: Fixed window length framing makes it difficult to take into account both transient and steady-state features of the signal, resulting in the smoothing or omission of key phase change information;

[0004] Nonlinear characteristics are difficult to interpret: Existing methods mostly focus on linear characteristics such as amplitude and energy, and lack the ability to effectively model and interpret the phase behavior of chaotic and nonlinear disturbances;

[0005] Limited feature fusion dimensions: Multi-scale feature fusion methods mainly target modes such as amplitude and energy, and lack structured coding methods for higher-order phase differentiation and dynamic evolution processes;

[0006] Insufficient spatiotemporal decoupling capability: Existing technologies struggle to achieve spatiotemporal separation and independent modeling of phase information, affecting the accuracy of subsequent identification and tracking.

[0007] In summary, in vital sign monitoring technology, traditional time-frequency analysis methods (such as short-time Fourier transform and wavelet transform) suffer from problems such as loss of transient features, difficulty in interpreting nonlinear features, limited feature fusion dimensions, and insufficient spatiotemporal decoupling capability when processing dynamic and nonlinear signals. Summary of the Invention

[0008] The purpose of this invention is to provide a feature extraction method for abnormal vital signs based on vital sign signals, which solves the problems of transient feature loss, difficulty in interpreting nonlinear features, limited feature fusion dimensions, and insufficient spatiotemporal decoupling capability in traditional time-frequency analysis methods for processing dynamic and nonlinear signals in vital sign monitoring technology.

[0009] To achieve the above objectives, the present invention provides a feature extraction method for abnormal vital signs based on vital sign signals. The feature extraction method for abnormal vital signs based on vital sign signals includes the following steps:

[0010] The radar echo signal reflected from the human body is collected by FMCW millimeter-wave radar and converted into a continuous phase sequence. ;

[0011] Continuous radar echo phase sequence Divided into time capsule units Each cabin carries The phase information of duration forms a spatiotemporal matrix;

[0012] The size of the time capsule is adaptively adjusted using the principle of variational optimization. ;

[0013] After solving for the optimal window length using the Lagrange multiplier method and achieving complete encapsulation of transient features, the data is then subjected to multidimensional deconstruction and phase feature decoupling.

[0014] For the time capsule unit Principal component analysis was performed to extract the dominant phase mode. Principal components with a cumulative contribution rate of ≥95% are retained;

[0015] The integer frequency harmonic phase structure is analyzed using Fast Fourier Transform;

[0016] Calculate the maximum Lyapunov exponent of the nonlinear phase perturbation based on the Lyapunov exponent. The chaotic characteristics of quantification phase evolution;

[0017] Time-varying features are extracted using higher-order differential operators, and a phase gradient vector is constructed. Characterizes the time-frequency domain coupling features;

[0018] Establish phase-velocity mapping;

[0019] By integrating the features of time, frequency, and phase differential order, a multidimensional feature cube is generated, which serves as the feature basis for detecting abnormal vital signs.

[0020] Specifically, in the step "continuous radar echo phase sequence" Divided into time capsule units Each cabin carries In the phrase "phase information of duration forms a spatiotemporal matrix", the spatiotemporal matrix formed is specifically as follows:

[0021] ;

[0022] in, Representing the nth-order time derivative of the phase, construct a multidimensional phase matrix that includes time-domain differential features. Let m be the sampling interval within the time chamber, and m be the number of sampling points, satisfying m≥n+1.

[0023] Specifically, in the step of "adaptively adjusting the size of the time capsule using variational optimization principles" In this context, the objective function is specifically:

[0024] ;

[0025] in, The variance of the in-cabin phase characteristics is used to ensure feature richness; for Norms are used to constrain the smoothness of adjacent compartments; Information entropy is used to suppress redundant features.

[0026] Specifically, the steps are "for the time capsule unit". Principal component analysis was performed to extract the dominant phase mode. The core algorithm for "retaining principal components with a cumulative contribution rate of ≥95%" is as follows:

[0027] ;

[0028] Let covariance matrix be the variance matrix. These are the eigenvalues.

[0029] Specifically, the core algorithm for the step "analyzing the integer harmonic phase structure using Fast Fourier Transform" is as follows:

[0030] Harmonic Frame (FFT):

[0031] ;

[0032] Extracting harmonic components , ( Using the fundamental frequency as the base frequency, construct the harmonic phase spectrum. .

[0033] Specifically, the step "Based on the Lyapunov exponent, calculate the maximum Lyapunov exponent of the nonlinear phase perturbation" The specific content of "quantitative phase evolution chaotic characteristics" is as follows:

[0034] The Wolf algorithm is used to calculate the maximum Lyapunov exponent of the nonlinear phase perturbation. :

[0035] ;

[0036] in, Let Jacobian matrix be the phase flow. This is the tangent space vector, used to quantify the chaotic characteristics of phase evolution.

[0037] Specifically, the steps are: "Extracting time-varying features using higher-order differential operators and constructing a phase gradient vector." The specific content of "characterizing the time-frequency domain coupling features" is as follows:

[0038] Time-varying features are extracted using higher-order differential operators, and an nth-order phase gradient vector is defined:

[0039] ;

[0040] Among them, the mixed partial derivatives Characterize the time-frequency domain coupling features.

[0041] Specifically, in the step of "establishing phase-velocity mapping relationship", the target radial velocity needs to be considered. Time-varying characteristics:

[0042] ;

[0043] in, The nonlinear phase perturbation term is expressed by Taylor expansion as follows:

[0044] .

[0045] Specifically, in the step of "establishing phase-velocity mapping relationship", the nonlinear phase perturbation term is... When performing calculations, it is necessary to obtain the weights of temporal attention and frequency attention;

[0046] Calculate the temporal attention weights and use an exponential decay model to enhance recent phase features:

[0047] ;

[0048] in, The characteristic peak time, The standard deviation of the time window;

[0049] Calculate frequency attention weights based on the Gaussian kernel function to highlight feature frequency bands:

[0050] ;

[0051] in, For the target frequency, This represents the standard deviation of the frequency window.

[0052] Specifically, in the step "fusing features of time, frequency, and phase differential order to generate a multi-dimensional feature cube as the feature basis for detecting abnormal vital signs," the feature cube that fuses the time axis, frequency axis, and phase differential order is represented as follows:

[0053] ;

[0054] Where T is the number of time points, F is the number of frequency points, and N is the upper limit of the differential order.

[0055] This invention discloses a feature extraction method for abnormal vital signs based on vital sign signals. First, it dynamically adjusts the time chamber size through variational optimization to capture both transient and steady-state features, avoiding feature loss caused by a fixed window length. Second, it combines principal component analysis, Lyapunov exponential chaotic quantization, and higher-order differential operators to decouple nonlinear features from multiple dimensions in the time, frequency, and phase domains, improving feature interpretability. Furthermore, it achieves adaptive enhancement and decoupling of spatiotemporal features through phase-velocity mapping and attention-weighted fusion, suppressing noise interference. The resulting multidimensional feature cube integrates time resolution, frequency accuracy, and phase differential information, significantly improving the robustness of abnormal detection of complex vital signs and overcoming the shortcomings of traditional methods in dynamic signal processing, such as single-dimensionality and insufficient adaptability. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 This is a flowchart of the steps of the feature extraction method for abnormal vital signs based on vital sign signals provided by the present invention. Detailed Implementation

[0058] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0059] Please see Figure 1 This invention provides a method for feature extraction of abnormal vital signs based on vital sign signals. The method for feature extraction of abnormal vital signs based on vital sign signals includes the following steps:

[0060] S1: Based on FMCW millimeter-wave radar, radar echo signals reflected from the human body are collected and converted into a continuous phase sequence. ;

[0061] S2: Continuous radar echo phase sequence Divided into time capsule units Each cabin carries The phase information of duration forms a spatiotemporal matrix;

[0062] S3: Adaptively adjusts the size of the time capsule using variational optimization principles. ;

[0063] S4: Solve for the optimal window length using the Lagrange multiplier method to achieve complete encapsulation of transient features, and then perform multidimensional deconstruction and phase feature decoupling on the data;

[0064] S5: For the time capsule unit Principal component analysis was performed to extract the dominant phase mode. Principal components with a cumulative contribution rate of ≥95% are retained;

[0065] S6: Analyze the integer frequency harmonic phase structure using Fast Fourier Transform;

[0066] S7: Calculate the maximum Lyapunov exponent of the nonlinear phase perturbation based on the Lyapunov exponent. The chaotic characteristics of quantification phase evolution;

[0067] S8: Extract time-varying features using higher-order differential operators and construct phase gradient vectors. Characterizes the time-frequency domain coupling features;

[0068] S9: Establish phase-velocity mapping relationship;

[0069] S10: Integrate the features of time, frequency, and phase differential order to generate a multi-dimensional feature cube, which serves as the feature basis for detecting abnormal vital signs.

[0070] In this implementation, firstly, the time chamber size is dynamically adjusted through variational optimization to capture both transient and steady-state features, avoiding feature loss caused by a fixed window length. Secondly, by combining principal component analysis, Lyapunov exponential chaotic quantization, and higher-order differential operators, nonlinear features are decoupled from multiple dimensions in the time, frequency, and phase domains, improving feature interpretability. Furthermore, through phase-velocity mapping and attention-weighted fusion, adaptive enhancement and decoupling of spatiotemporal features are achieved, suppressing noise interference. The resulting multidimensional feature cube integrates time resolution, frequency accuracy, and phase differential information, significantly improving the robustness of anomaly detection for complex vital signs and overcoming the shortcomings of traditional methods in dynamic signal processing, such as single-dimensionality and insufficient adaptability.

[0071] Furthermore, specifically, in the step of "continuous radar echo phase sequence" Divided into time capsule units Each cabin carries In the phrase "phase information of duration forms a spatiotemporal matrix", the spatiotemporal matrix formed is specifically as follows:

[0072] ;

[0073] in, Representing the nth-order time derivative of the phase, construct a multidimensional phase matrix that includes time-domain differential features. Let m be the sampling interval within the time chamber, and m be the number of sampling points, satisfying m≥n+1.

[0074] Furthermore, in the step of "adaptively adjusting the size of the time capsule using variational optimization principles"... In this context, the objective function is specifically:

[0075] ;

[0076] in, The variance of the in-cabin phase characteristics is used to ensure feature richness; for Norms are used to constrain the smoothness of adjacent compartments; Information entropy is used to suppress redundant features.

[0077] Further, the step "on the time capsule unit" Principal component analysis was performed to extract the dominant phase mode. The core algorithm for "retaining principal components with a cumulative contribution rate of ≥95%" is as follows:

[0078] ;

[0079] Let covariance matrix be the variance matrix. These are the eigenvalues.

[0080] Furthermore, the core algorithm for the step "analyzing the integer harmonic phase structure through fast Fourier transform" is as follows:

[0081] Harmonic Frame (FFT):

[0082] ;

[0083] Extracting harmonic components , ( Using the fundamental frequency as the base frequency, construct the harmonic phase spectrum. .

[0084] Further, the step "Calculates the maximum Lyapunov exponent of the nonlinear phase perturbation based on the Lyapunov exponent" The specific content of "quantitative phase evolution chaotic characteristics" is as follows:

[0085] The Wolf algorithm is used to calculate the maximum Lyapunov exponent of the nonlinear phase perturbation. :

[0086] ;

[0087] in, Let Jacobian matrix be the phase flow. This is the tangent space vector, used to quantify the chaotic characteristics of phase evolution.

[0088] Furthermore, the step "extracts time-varying features through higher-order differential operators and constructs a phase gradient vector" The specific content of "characterizing the time-frequency domain coupling features" is as follows:

[0089] Time-varying features are extracted using higher-order differential operators, and an nth-order phase gradient vector is defined:

[0090] ;

[0091] Among them, the mixed partial derivatives Characterize the time-frequency domain coupling features.

[0092] Furthermore, in the step of "establishing phase-velocity mapping relationship", the target radial velocity needs to be considered. Time-varying characteristics:

[0093] ;

[0094] in, The nonlinear phase perturbation term is expressed by Taylor expansion as follows:

[0095] .

[0096] Furthermore, in the step of "establishing phase-velocity mapping relationship", the nonlinear phase disturbance term is... When performing calculations, it is necessary to obtain the weights of temporal attention and frequency attention;

[0097] Calculate the temporal attention weights and use an exponential decay model to enhance recent phase features:

[0098] ;

[0099] in, The characteristic peak time, The standard deviation of the time window;

[0100] Calculate frequency attention weights based on the Gaussian kernel function to highlight feature frequency bands:

[0101] ;

[0102] in, For the target frequency, This represents the standard deviation of the frequency window.

[0103] Furthermore, in the step "fusing the features of time, frequency, and phase differential order to generate a multi-dimensional feature cube as the feature basis for detecting abnormal vital signs," the feature cube that fuses the time axis, frequency axis, and phase differential order is represented as follows:

[0104] ;

[0105] Where T is the number of time points, F is the number of frequency points, and N is the upper limit of the differential order.

[0106] In summary, this technical solution, through the aforementioned multi-level "time capsule" framing and multi-dimensional phase feature decoupling and fusion method, ultimately obtains structured and interpretable phase feature data. This feature data can comprehensively characterize the time-varying, frequency-varying, and higher-order differential characteristics in radar echo signals, providing a robust and high-resolution feature foundation for subsequent diagnosis of respiratory distress syndrome, chest movement analysis, and detection of abnormal vital signs.

[0107] Specifically, this technical solution has the following beneficial effects:

[0108] Transient features are fully preserved: The adaptive "time capsule" framing method can dynamically adjust the analysis window length, accurately capture transient changes in the signal, and avoid feature loss caused by the traditional fixed window length method;

[0109] Modeling of nonlinear and chaotic characteristics: By using chaotic analysis methods such as the Lyapunov exponent, nonlinear perturbations in phase sequences are effectively extracted and quantified, providing theoretical support for complex action recognition;

[0110] Multi-dimensional feature fusion and decoupling: By combining principal component analysis, harmonic analysis and higher-order differential operators, multi-modal feature fusion in the time domain, frequency domain and differential domain is realized, which improves the feature expression capability;

[0111] Spatiotemporal attention mechanism: Introducing time and frequency attention weights highlights key moments and feature frequency bands, enhancing the discriminativeness and robustness of features.

[0112] The above description discloses only one preferred embodiment of the present invention, and should not be construed as limiting the scope of the present invention. Those skilled in the art will understand that all or part of the processes of the above embodiments can be implemented, and equivalent changes made in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A method for feature extraction of abnormal vital signs based on vital sign signals, characterized in that, Includes the following steps: The radar echo signal reflected from the human body is collected by FMCW millimeter-wave radar and converted into a continuous phase sequence. ; Continuous radar echo phase sequence Divided into time capsule units Each cabin carries The phase information of duration is used to form a spatiotemporal matrix, which is specifically: ； In the formula, Representing the nth-order time derivative of the phase, construct a multidimensional phase matrix that includes time-domain differential features. Let m be the sampling interval within the time chamber, and m be the number of sampling points, satisfying m≥n+1; The size of the time capsule is adaptively adjusted using the principle of variational optimization. Its objective function is: ； in, The variance of the in-cabin phase characteristics is used to ensure feature richness; for Norms are used to constrain the smoothness of adjacent compartments; Information entropy is used to suppress redundant features; After solving for the optimal window length using the Lagrange multiplier method and achieving complete encapsulation of transient features, the data is then subjected to multidimensional deconstruction and phase feature decoupling. For the time capsule unit Principal component analysis was performed to extract the dominant phase mode. Principal components with a cumulative contribution rate of ≥95% are retained; The integer frequency harmonic phase structure is analyzed using Fast Fourier Transform; Calculate the maximum Lyapunov exponent of the nonlinear phase perturbation based on the Lyapunov exponent. The chaotic characteristics of quantification phase evolution; Time-varying features are extracted using higher-order differential operators, and a phase gradient vector is constructed. Characterizes the time-frequency domain coupling features; Establish a phase-velocity mapping relationship, and consider the target radial velocity. Time-varying characteristics: ； in, The nonlinear phase perturbation term is expressed by Taylor expansion as follows: ； By integrating the features of time, frequency, and phase differential order, a multidimensional feature cube is generated, which serves as the feature basis for detecting abnormal vital signs.

2. The feature extraction method for abnormal vital signs based on vital sign signals as described in claim 1, characterized in that, Steps for the time capsule unit Principal component analysis was performed to extract the dominant phase mode. The core algorithm for retaining principal components with a cumulative contribution rate of ≥95% is as follows: ； Let covariance matrix be the variance matrix. These are the eigenvalues.

3. The feature extraction method for abnormal vital signs based on vital sign signals as described in claim 2, characterized in that, The core algorithm for analyzing the integer harmonic phase structure using Fast Fourier Transform is as follows: Harmonic chamber: ； Extracting harmonic components , Using the fundamental frequency, construct the harmonic phase spectrum. .

4. The feature extraction method for abnormal vital signs based on vital sign signals as described in claim 3, characterized in that, The steps are based on the Lyapunov exponent, calculating the maximum Lyapunov exponent of the nonlinear phase perturbation. The specific content of the chaotic characteristics of quantified phase evolution is as follows: The Wolf algorithm is used to calculate the maximum Lyapunov exponent of the nonlinear phase perturbation. : ； in, Let Jacobian matrix be the phase flow. This is the tangent space vector, used to quantify the chaotic characteristics of phase evolution.

5. The feature extraction method for abnormal vital signs based on vital sign signals as described in claim 4, characterized in that, The steps involve extracting time-varying features using higher-order differential operators and constructing a phase gradient vector. The specific content describing the time-frequency domain coupling characteristics is as follows: Time-varying features are extracted using higher-order differential operators, and an nth-order phase gradient vector is defined: ； Among them, the mixed partial derivatives Characterize the time-frequency domain coupling features.

6. The feature extraction method for abnormal vital signs based on vital sign signals as described in claim 5, characterized in that, In the step of establishing the phase-velocity mapping relationship, the nonlinear phase perturbation term is... When performing calculations, it is necessary to obtain the weights of temporal attention and frequency attention; Calculate the temporal attention weights and use an exponential decay model to enhance recent phase features: ； in, The characteristic peak time, The standard deviation of the time window; Calculate frequency attention weights based on the Gaussian kernel function to highlight feature frequency bands: ； in, For the target frequency, This represents the standard deviation of the frequency window.

7. The feature extraction method for abnormal vital signs based on vital sign signals as described in claim 6, characterized in that, In the step of fusing time, frequency, and phase differential order features, a multidimensional feature cube is generated as the feature basis for detecting abnormal vital signs. The feature cube fusing time axis, frequency axis, and phase differential order is represented as follows: ； Where T is the number of time points, F is the number of frequency points, and N is the upper limit of the differential order.

Images