Cerebral stroke risk diagnosis and early warning system and method based on millimeter wave radar fusion

Abstract

The invention relates to the technical field of intelligent medical monitoring, and discloses a cerebral apoplexy risk diagnosis and early warning system and method based on millimeter wave radar fusion. The system comprises a sensor module used for carrying out data acquisition on a target object; the micro-sign decoupling module is used for acquiring a radio frequency echo signal, executing complex feature reconstruction, determining a spatial reflection anchor point and stripping a physiological feature waveform; the spatial heterogeneous module is used for constructing a virtual anatomical projection plane, generating a kinetic energy distribution thermodynamic diagram and calculating an asymmetric index; and the early warning judgment module is used for executing cross-modal time sequence fusion to obtain a heterogeneous energy imbalance vector, mapping a risk prediction point in a diagnosis and evaluation plane, and executing hierarchical risk early warning according to an evolution slope. According to the method, spatial visualization mapping of thoracic cavity microscopic kinetic energy distribution is achieved, the problem that the missing report rate of single-mode monitoring is high is solved, and the early warning precision of sudden stroke proactive period pathological evolution is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical monitoring technology, and more specifically, to a stroke risk diagnosis and early warning system and method based on millimeter-wave radar fusion. Background Technology

[0002] With the rapid development of intelligent monitoring and mobile healthcare technologies, non-contact vital sign monitoring has demonstrated significant clinical value in the early warning and risk assessment of stroke. The prognosis of stroke highly depends on the timeliness of treatment, but most patients lack significant clinical symptoms before onset, leading to missed opportunities for optimal treatment. Currently, millimeter-wave radar technology, with its advantages of high precision, non-contact operation, and privacy protection, can achieve non-invasive monitoring of cardiac cycle details and respiratory patterns. However, traditional monitoring methods generally face bottlenecks related to "single dimension" and "static thresholds." Existing technologies often focus on macroscopic monitoring of single physiological parameters or rely on isolated fluctuations in vital signs for risk assessment. This ignores the complex multimodal physiological co-evolutionary effects in the prodromal period of stroke, namely, the deep coupling relationship between cardiac autonomic nervous system imbalance, respiratory pattern disorder, and neuromotor function decline. In real-world scenarios, radial distance information alone is insufficient to effectively distinguish the target thoracic cavity from surrounding static strong reflectors, leading to spatial directional ambiguity in physiological signal sources. Furthermore, due to the imperfect symmetry of human anatomy and interference from environmental background clutter, subtle physiological micro-vibrations are often masked, making it difficult to accurately quantify the unilateral neuromuscular motor symmetry disruption caused by early stroke damage. While traditional prediction models can identify some abnormalities, they often struggle to effectively decouple heterogeneous interferences such as individual body size, detection distance, and environmental attenuation, resulting in a trade-off between warning sensitivity and false alarm rate. Therefore, how to shift from static monitoring of a single parameter to dynamic deep fusion of multimodal physiological characteristics, transforming macroscopic vital sign records into cross-system collaborative analysis targeting the pathological mechanisms of stroke, thereby overcoming the limitations of individual differences and environmental noise and achieving ultra-early and accurate early warning of stroke risk, remains a key technical challenge in this field.

[0003] In the prior art, Chinese Patent No. CN118648887B discloses a non-contact real-time physiological sign monitoring system based on millimeter-wave radar. This system includes millimeter-wave radar, a clutter suppression module, and a signal separation module. It acquires intermediate-frequency signals by emitting electromagnetic waves and constructs a four-dimensional data matrix. Utilizing constant false alarm rate detection, adaptive distance unit selection, and optimized variational mode decomposition algorithms, it achieves non-contact extraction of physiological signs such as respiration and heart rate. The core solution addresses the problem of environmental clutter suppression and physiological signal separation, improving the accuracy of sign monitoring. Chinese Patent No. CN116269249B discloses a stroke risk prediction method and system. This system collects multiple physiological indicators of the human body and combines them with an algorithm model to construct a risk prediction system, achieving a quantitative assessment of stroke risk and providing data support for home and clinical stroke risk screening.

[0004] However, while the two existing technologies mentioned above have some application value in non-contact physiological monitoring and stroke risk prediction, they fail to address the core pain points of multimodal feature fusion, spatial signal localization, and individualized dynamic assessment in early stroke warning. Specifically, CN118648887B focuses only on the extraction and separation of single-dimensional cardiopulmonary signs, without addressing the bilateral symmetry analysis of chest cavity movement, thus failing to capture the energy imbalance characteristics caused by unilateral neuromuscular damage in the early stages of stroke, and lacking a deep transformation logic from physiological signals to pathological warnings. CN116269249B relies on static threshold analysis of conventional physiological indicators, failing to establish a dynamic risk target area based on individual health baselines, making it difficult to decouple heterogeneous interferences such as individual body size and detection environment. Furthermore, it fails to achieve collaborative analysis of physiological characteristics across the heart, lungs, and brain systems, and cannot identify the multimodal pathological evolution patterns in the prodromal phase of stroke. Neither technology solves the problem of ambiguous spatial directionality of physiological signal sources, nor does it construct a hierarchical risk warning mechanism, making it difficult to meet the clinical needs for ultra-early and accurate stroke warning. Summary of the Invention

[0005] This invention is applicable to non-contact monitoring scenarios for early risk warning of stroke, and can meet the precise warning needs of extracting weak physiological signals in complex electromagnetic environments. By performing complex feature reconstruction on radio frequency echo signals to determine spatial reflection anchor points, combined with phase unwrapping and physical displacement mapping, it achieves the dual goals of high-fidelity positioning of the center of mass of motion and submicron-level stripping of physiological feature waveforms. Physiological kinetic energy focusing mapping is performed in the virtual anatomical projection plane to generate a kinetic energy distribution heat map, dividing bilateral monitoring areas and converting the defect judgment caused by unilateral neuromuscular damage into a normalized quantitative calculation of the asymmetry index, accurately identifying the risk of bilateral trunk micro-tremor symmetry defect. Cross-modal temporal fusion combined with the evolution slope of risk prediction point solution in the diagnostic evaluation plane realizes the scientific triggering of graded risk warning commands, which not only adaptively defines individualized physiological fluctuation thresholds, but also provides precise capture support for sudden pathological evolution, effectively distinguishes progressive physiological fluctuations, and reduces the false alarm rate of stroke warning.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A stroke risk diagnosis and early warning system based on millimeter-wave radar fusion includes:

[0008] Micro-feature decoupling module: used to acquire radio frequency echo signals characterizing the micro-motion characteristics of the target object, perform complex feature reconstruction on the radio frequency echo signals to obtain the angular energy spectrum characterizing the spatial reflection intensity, perform peak detection and coordinate mapping on the angular energy spectrum to determine the spatial reflection anchor point, perform phase unwrapping and physical displacement mapping based on the spatial reflection anchor point to obtain the physiological displacement sequence, and perform multi-scale digital filtering on the physiological displacement sequence to extract the physiological feature waveform;

[0009] Spatial Heterogeneous Module: Used to construct a virtual anatomical projection surface based on spatial reflection anchor points, and to call physiological feature waveforms within the virtual anatomical projection surface to perform physiological kinetic energy focusing mapping in the grid dimension to obtain a kinetic energy distribution heat map. Bilateral monitoring areas are divided within the kinetic energy distribution heat map, and two-dimensional spatial discrete summation calculation is performed based on the bilateral monitoring areas to obtain the asymmetry index.

[0010] Early warning judgment module: It is used to perform cross-modal time series fusion based on physiological feature waveforms to obtain heterogeneous energy imbalance vectors, and to construct a diagnostic evaluation plane based on asymmetric index to map the heterogeneous energy imbalance vectors to obtain risk prediction points. The evolution slope is calculated through the risk prediction points, and graded risk early warning is performed based on the evolution slope to obtain a risk early warning set.

[0011] The sensor module consists of a millimeter-wave radar sensor, a weighing sensor array, and a bed frame as a supporting carrier. The millimeter-wave radar sensor is deployed at the head of the bed frame or built into the mattress. It is used to radiate frequency-modulated continuous waves and capture reflected signals after being modulated by the physiological micro-undulations of the chest cavity to generate radio frequency echo signals. The weighing sensor array consists of weighing sensors arranged at the foot of the bed frame and is used to obtain the weight parameters of the target object.

[0012] Furthermore, the complex feature reconstruction includes:

[0013] Multiple frequency-modulated continuous waves are continuously radiated into the monitored area containing the target object by a millimeter-wave radar sensor at a preset pulse period. The reflected signal is captured after being phase-modulated by the physiological micro-fluctuations of the thoracic cavity. The reflected signal is subjected to analog multiplication and low-pass filtering to extract the intermediate frequency signal and then quantized and sampled to obtain the radio frequency echo signal.

[0014] Perform a fast Fourier transform on the radio frequency echo signal to generate a range spectrum, and define the ranging unit with the largest energy amplitude in the range spectrum as the target ranging unit;

[0015] Based on the position of the radio frequency echo signal at the target ranging unit, a complex voltage value containing real and imaginary information is extracted, and the complex voltage values ​​are combined into a complex sampling sequence in time sequence;

[0016] Calculate and sum the instantaneous power of each complex voltage value to obtain the total energy value of a single frame. Divide the total energy value of a single frame by the number of frequency-modulated continuous waves to convert it into average energy power.

[0017] The average background energy of the monitored area under no-load conditions is obtained, and its reciprocal is taken as the environmental correction coefficient. The effective signal energy is obtained by multiplying the average energy power by the environmental correction coefficient.

[0018] Furthermore, the millimeter-wave radar sensor can be installed in two ways: the first is to fix the millimeter-wave radar sensor above the headboard, and the second is to embed the millimeter-wave radar sensor in the mattress of the bed.

[0019] Furthermore, the method for obtaining the spatial reflection anchor point includes:

[0020] The complex voltage values ​​of each independent channel of the millimeter-wave radar sensor at the target ranging unit position are obtained, and the complex voltage values ​​are sorted to construct a spatial measurement vector.

[0021] Multiple candidate angles are preset, and the theoretical phase offset of each independent channel at each candidate angle is calculated to construct an angle steering vector. The angle steering vectors form a set of reference vectors.

[0022] The spatial measurement vector and each angular steering vector are respectively subjected to inner product operation to obtain the spatial reflection intensity, and the corresponding candidate angles are combined to obtain the angular energy spectrum;

[0023] The local extreme point with the largest value in the search angle energy spectrum is taken as the energy peak amplitude. The energy peak amplitude is mapped to obtain the target azimuth and target elevation angles.

[0024] Based on the target ranging unit, the radial distance value is determined, and the lateral, longitudinal, and height components are calculated in the constructed radar local coordinate system in combination with the target azimuth and target elevation angles.

[0025] A millimeter-wave radar sensor is used to perform rotation operations on the lateral, longitudinal, and height components to generate the rotated lateral, longitudinal, and absolute vertical height components. The three-dimensional physical coordinate points formed by these three components are defined as spatial reflection anchor points.

[0026] Furthermore, the physiological characteristic waveform includes:

[0027] The original phase value of each complex voltage value in the complex sampling sequence is calculated using the arctangent function, and the original phase sequence is obtained by arranging them in the time sequence of the pulse period.

[0028] Calculate the phase difference between adjacent pulse periods in the original phase sequence, and perform numerical compensation operation based on the relationship between the phase difference and the preset transition threshold to eliminate phase entanglement and construct a continuous phase sequence;

[0029] The phase values ​​in the continuous phase sequence are combined with the inherent wavelength of the radio frequency echo signal to convert them into instantaneous displacement characteristic values, and then combined in time sequence to obtain the physiological displacement sequence.

[0030] Using a zero-phase shift digital bandpass filter, frequency-selective filtering is performed in parallel on the physiological displacement sequence within the preset respiratory frequency range and heart rate range, respectively, to extract respiratory and heart rate characteristic waveforms.

[0031] By combining the respiratory characteristic waveform with the heartbeat characteristic waveform, the physiological characteristic waveform is obtained.

[0032] Furthermore, the kinetic energy distribution heatmap includes:

[0033] Using the spatial reflection anchor point as the geometric reference center, a two-dimensional physical section is constructed and a rectangular region is extracted. The rectangular region is divided into multiple two-dimensional discrete grids with three-dimensional geometric coordinates. All two-dimensional discrete grids constitute a virtual anatomical projection surface.

[0034] Calculate the physical spatial straight-line distance from the two-dimensional discrete grid to each independent channel, and determine the theoretical phase delay by combining the inherent wavelength of the radio frequency echo signal;

[0035] By performing a reverse phase rotation on the complex voltage value using the theoretical phase delay, phase compensation is achieved, and the data is coherently accumulated to obtain a local time-domain reflection signal.

[0036] Within the preset respiratory rate range and heart rate range, frequency-selective filtering is performed on the local time-domain reflectance signal to obtain the local physiological micro-motion component.

[0037] The kinetic energy density points are calculated using local physiological micro-motion components. The kinetic energy density points corresponding to all two-dimensional discrete grids are combined according to their spatial topological positions to generate a kinetic energy distribution heat map.

[0038] Furthermore, the asymmetric index includes:

[0039] With the geometric reference center as the origin of the coordinate system, an anatomical sagittal projection line is drawn along the longitudinal direction in the radar local coordinate system, dividing the virtual anatomical projection plane into the left monitoring area and the right monitoring area.

[0040] The kinetic energy density points corresponding to each two-dimensional discrete grid in the left monitoring area are extracted and a two-dimensional spatial discrete summation operation is performed to obtain the total value of micro-motion energy on the left. The kinetic energy density points corresponding to each two-dimensional discrete grid in the right monitoring area are extracted and a two-dimensional spatial discrete summation operation is performed to obtain the total value of micro-motion energy on the right.

[0041] Acquire multiple frames of radio frequency echo signals of the target object during the historical health baseline period, calculate the ratio of the total left micromotion energy value to the total right micromotion energy value corresponding to each frame of radio frequency echo signal, and take the arithmetic mean of all ratios as the calibration factor.

[0042] The asymmetry index is obtained by combining the calibration factor with the total energy values ​​of the left and right micromotions.

[0043] Furthermore, the method for obtaining the risk prediction points includes:

[0044] The effective peaks of the heartbeat characteristic waveform are obtained, the time difference between adjacent effective peaks is calculated, and the heartbeat interval sequence is obtained by arranging them in time sequence.

[0045] The heart rate interval sequence is resampled and time-frequency converted to obtain the power spectral density function. Within the preset low-frequency band and high-frequency band, the power spectral density function is subjected to definite integral operation to extract the low-frequency power and high-frequency power respectively. The low-frequency power is divided by the high-frequency power to obtain the energy ratio. The heart rate variability index is obtained by combining the energy ratio and the short-term variability.

[0046] Statistical calculations were performed on the heartbeat interval sequence to obtain the atrial fibrillation load characteristics;

[0047] Statistical calculations were performed on the respiratory characteristic waveforms to obtain the cardiopulmonary physiological characteristics.

[0048] Set up a feature integration sliding window, push the asymmetric exponent and heart rate variability index into the feature integration sliding window, and obtain the asymmetric exponent sequence and the heart rate variability index sequence respectively;

[0049] The arithmetic mean of the asymmetric exponential sequence and the heart rate variability index sequence within the feature integration sliding window is calculated separately. The arithmetic mean of the asymmetric exponential sequence and the heart rate variability index sequence is then mapped onto the diagnostic evaluation plane to obtain the risk prediction point.

[0050] Furthermore, the risk warning set includes:

[0051] Obtain and calculate the standard deviation and arithmetic mean of the asymmetric index sequence and the heart rate variability index sequence during the historical health baseline period, and establish the arithmetic mean of the asymmetric index sequence and the heart rate variability index sequence as a personalized health anchor point in the diagnostic evaluation plane;

[0052] By combining the standard deviation of the asymmetric exponential sequence, the standard deviation of the heart rate variability index sequence, and the preset risk tolerance coefficient, a dynamic risk target zone with a personalized health anchor as the geometric center is constructed.

[0053] When a risk prediction point moves away from the dynamic risk target area, the change in physical distance between the risk prediction point and the personalized health anchor point is calculated and defined as the Euclidean distance increment.

[0054] Divide the Euclidean distance increment by the preset total observation duration to obtain the evolution slope;

[0055] The evolution slope is compared with a preset first slope threshold and a second slope threshold to determine the risk level and trigger a corresponding warning. The risk levels are then combined to obtain a risk warning set.

[0056] A stroke risk diagnosis and early warning method based on millimeter-wave radar fusion is applied to the aforementioned stroke risk diagnosis and early warning system based on millimeter-wave radar fusion. The method includes:

[0057] The radio frequency echo signal characterizing the microscopic motion characteristics of the target object is acquired. Complex feature reconstruction is performed on the radio frequency echo signal to obtain the angular energy spectrum characterizing the spatial reflection intensity. Peak detection and coordinate mapping are performed on the angular energy spectrum to determine the spatial reflection anchor point. Phase unwrapping and physical displacement mapping are performed based on the spatial reflection anchor point to obtain the physiological displacement sequence. Multi-scale digital filtering is performed on the physiological displacement sequence to extract the physiological feature waveform.

[0058] A virtual anatomical projection surface is constructed based on spatial reflection anchor points. Physiological feature waveforms are called within the virtual anatomical projection surface to perform physiological kinetic energy focusing mapping in the grid dimension, resulting in a kinetic energy distribution heatmap. Bilateral monitoring areas are divided within the kinetic energy distribution heatmap. Two-dimensional spatial discrete summation calculation is performed based on the bilateral monitoring areas to obtain the asymmetry index.

[0059] Cross-modal temporal fusion is performed based on physiological characteristic waveforms to obtain heterogeneous energy imbalance vectors. A diagnostic evaluation plane is constructed based on asymmetric index to map the heterogeneous energy imbalance vectors to obtain risk prediction points. Evolution slopes are calculated through risk prediction points, and graded risk warnings are performed based on the evolution slopes to obtain a risk warning set.

[0060] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0061] This invention achieves the dual goals of high-fidelity positioning of the center of mass of motion and submicron-level stripping of physiological feature waveforms through complex feature reconstruction, determination of spatial reflection anchor points, phase unwrapping, and physical displacement mapping. It addresses the shortcomings of traditional methods in complex home monitoring environments, where strong background clutter masks weak physiological signals and macroscopic body movement interferes with micro-vibration extraction. The virtual anatomical projection plane and kinetic energy distribution heatmap transform the inability of traditional single-point monitoring to perceive the differences in movement between bilateral anatomical sites into a quantitative calculation of the relative imbalance ratio of asymmetry indices. This accurately identifies the risk of unilateral neuromuscular movement symmetry disruption caused by early stroke damage, reducing the systematic interference of absolute energy values ​​due to individual body size, detection distance, and environmental attenuation. Cross-modal temporal fusion combined with the evolution slope of risk prediction points in the diagnostic evaluation plane enables graded risk warning. This adaptively defines individualized physiological fluctuation thresholds and provides precise quantitative support for sudden pathological evolution, effectively distinguishing between spontaneous slow physiological drift and sudden pathological evolution, reducing the false negative rate of stroke warnings under single vital sign monitoring modes. Attached Figure Description

[0062] 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.

[0063] Figure 1 This is a structural diagram of a stroke risk diagnosis and early warning system based on millimeter-wave radar fusion provided in an embodiment of the present invention;

[0064] Figure 2 This invention provides a schematic diagram of a non-contact monitoring scenario where the target object is in a resting state when a millimeter-wave radar sensor is installed above the bedside.

[0065] Figure 3 This invention provides a schematic diagram of a non-contact monitoring scenario where the target object is in a resting state when a millimeter-wave radar sensor is built into a mattress.

[0066] Figure 4 This invention provides a schematic diagram of spatial mapping for dividing bilateral monitoring areas based on anatomical sagittal projection lines, as provided in an embodiment of the invention.

[0067] Figure 5 This is a flowchart of a stroke risk diagnosis and early warning method based on millimeter-wave radar fusion provided in an embodiment of the present invention.

[0068] Figure label:

[0069] 1. Millimeter-wave radar sensor; 2. Weighing sensor; 3. Bed frame; 4. Central processing terminal; 5. Mattress. Detailed Implementation

[0070] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] Example 1

[0072] This embodiment provides a stroke risk diagnosis and early warning system based on millimeter-wave radar fusion, including:

[0073] Micro-feature decoupling module: used to acquire radio frequency echo signals characterizing the micro-motion characteristics of the target object, perform complex feature reconstruction on the radio frequency echo signals to obtain the angular energy spectrum characterizing the spatial reflection intensity, perform peak detection and coordinate mapping on the angular energy spectrum to determine the spatial reflection anchor point, perform phase unwrapping and physical displacement mapping based on the spatial reflection anchor point to obtain the physiological displacement sequence, and perform multi-scale digital filtering on the physiological displacement sequence to extract the physiological feature waveform;

[0074] Spatial Heterogeneous Module: Used to construct a virtual anatomical projection surface based on spatial reflection anchor points, and to call physiological feature waveforms within the virtual anatomical projection surface to perform physiological kinetic energy focusing mapping in the grid dimension to obtain a kinetic energy distribution heat map. Bilateral monitoring areas are divided within the kinetic energy distribution heat map, and two-dimensional spatial discrete summation calculation is performed based on the bilateral monitoring areas to obtain the asymmetry index.

[0075] Early warning judgment module: It is used to perform cross-modal time series fusion based on physiological feature waveforms to obtain heterogeneous energy imbalance vectors, and to construct a diagnostic evaluation plane based on asymmetric index to map the heterogeneous energy imbalance vectors to obtain risk prediction points. The evolution slope is calculated through the risk prediction points, and graded risk early warning is performed based on the evolution slope to obtain a risk early warning set.

[0076] The sensor module consists of a millimeter-wave radar sensor, a weighing sensor array, and a bed frame as a supporting carrier. The millimeter-wave radar sensor is deployed at the head of the bed frame or built into the mattress. It is used to radiate frequency-modulated continuous waves and capture reflected signals after being modulated by the physiological micro-undulations of the chest cavity to generate radio frequency echo signals. The weighing sensor array consists of weighing sensors arranged at the foot of the bed frame and is used to obtain the weight parameters of the target object.

[0077] See Figure 1 This is a structural diagram of a stroke risk diagnosis and early warning system based on millimeter-wave radar fusion provided in an embodiment of the present invention. The diagram shows the overall hardware layout and logic processing center of the system, embodying a multimodal sensing architecture that combines non-contact electromagnetic sensing and physical load monitoring. The millimeter-wave radar sensor 1 is shown in two exemplary deployment methods: one is fixedly installed at the headboard of the bed frame 3, and the other is built into the mattress 5. It is used to radiate frequency-modulated continuous waves to the monitored area and capture the reflected signals modulated by the physiological micro-undulations of the chest cavity, generating radio frequency echo signals characterizing respiratory and heartbeat features. Weighing sensors 2 are arranged at the bottom of each foot of the bed frame 3, forming a weighing sensor array. The bed frame 3 serves as the mechanical structure supporting the sensing hardware and the monitored target object. The central processing terminal 4, representing the system's logic calculation and decision-making core, is connected to the sensor module and is used to integrate and run the micro-feature decoupling module, the spatial heterogeneous module, and the early warning judgment module. The mattress 5 is laid on the surface of the bed frame 3, serving as a flexible support interface for direct contact with the target object.

[0078] In the micro-feature decoupling module, the step of acquiring the radio frequency echo signal characterizing the micro-motion characteristics of the target object, and performing complex feature reconstruction on the radio frequency echo signal to obtain the angular energy spectrum characterizing the spatial reflection intensity includes:

[0079] Specifically, in non-contact monitoring scenarios for early stroke risk warning, the monitoring process is typically conducted within the monitored area where the target object is at rest. The monitored area refers to the three-dimensional physical space covered by the beam emitted by a millimeter-wave radar sensor, usually including bedding, walls, and the target object in a sleeping or resting state. A millimeter-wave radar sensor is a radio frequency sensing device that can transmit frequency-modulated continuous waves through an antenna array and receive reflected signals. During monitoring, the target object's chest cavity generates micron-level physiological undulations on the chest surface driven by heartbeats and lung respiration, which produce a phase modulation effect on the electromagnetic waves emitted by the millimeter-wave radar sensor. However, in a real home environment, strong reflective objects such as walls and bed frames within the monitored area generate extremely strong static background clutter. Simultaneously, the energy intensity generated by the target object's macroscopic turning movements or limb translations during the monitoring period is far higher than the energy intensity generated by the physiological undulations on the chest surface. To extract weak signals with pathological warning value from the complex electromagnetic environment, an acquisition framework capable of converting electromagnetic fluctuations into digital logical sequences must be established. The purpose is to acquire radio frequency echo signals through non-contact detection using a millimeter-wave radar sensor. This invention provides two independent installation modes: the first mode involves embedding the millimeter-wave radar sensor within the mattress; the second mode involves fixing the millimeter-wave radar sensor above the headboard. Regardless of the installation mode, the millimeter-wave radar sensor can acquire the microscopic physical carrier of the target object and output radio frequency echo signals characterizing the energy distribution in distance, velocity, and angular dimensions. Furthermore, it acquires the target object's basic physical parameters, including height and weight. The weight parameter is acquired in real-time by a weighing sensor array positioned at the foot of the bed, while the height parameter is obtained through auxiliary measurements, such as a height measuring instrument.

[0080] To accurately acquire radio frequency echo signals and eliminate individual heterogeneity interference, two optional sensor placement methods are provided, and each method can independently achieve full-dimensional vital sign monitoring; for details, please refer to [link to relevant documentation]. Figure 2 and Figure 3 ;like Figure 2 The diagram illustrates a non-contact monitoring scenario based on a millimeter-wave radar sensor installed above the bedside, where the target object is in a resting state, according to an embodiment of the present invention. The millimeter-wave radar sensor is fixedly installed above the bedside in the monitored area. Figure 2 The arrow pointing to the target object area in a mid-millimeter-wave radar sensor represents the transmitted signal radiated into space. When the transmitted signal contacts the chest area of ​​the target object, it is affected by the physiological micro-undulations on the surface of the chest cavity caused by the heartbeat and lung respiration. The electromagnetic waves carrying micro-motion characteristics are reflected to form a reflected signal. Figure 2 The process of the reflected signal returning to the receiving antenna array is represented by an arrow pointing from the target object to the bottom of the sensor; for example... Figure 3 The diagram illustrates a non-contact monitoring scenario where a target object is in a resting state, based on a millimeter-wave radar sensor embedded in a mattress, according to an embodiment of the present invention. The millimeter-wave radar sensor is positioned on the mattress and detects from bottom to top. Figure 3 The upward arrow represents the transmitted signal radiated into space, which is also modulated by the physiological micro-undulations of the thoracic cavity surface. The downward arrow represents the process of the reflected signal, carrying micro-motion characteristics, returning to the receiving antenna array. Because... Figure 2 and Figure 3 The millimeter-wave radar sensors described above are identical except for their installation location. The following description uses a millimeter-wave radar sensor installed above the headboard as an example. This type of sensor also includes an installation height and an installation tilt angle. The installation height refers to the vertical displacement of the phase center of the transmitting antenna of the millimeter-wave radar sensor relative to the surface of the target object's chest cavity. This height is set based on the far-field radiation characteristics and range axis resolution of the transmitting antenna, ensuring that the chest region of the target object is within the optimal detection energy gain zone of the frequency-modulated continuous wave. For example, the installation height is set to 1.2 meters. The installation tilt angle refers to the angle between the beam center axis of the millimeter-wave radar sensor and the direction perpendicular to the ground. This angle is set so that the beam center of the millimeter-wave radar sensor is angled downwards and aligned with the geometric center of the target object's torso, thereby maximizing the signal-to-noise ratio of the reflected echo. For example, this angle is set to 35 degrees. It's important to note that for millimeter-wave radar sensors mounted above the headboard, the spatial coordinate calculation depends on the installation height and tilt angle in the rotation transformation matrix. However, for millimeter-wave radar sensors built into the mattress, since they detect vertically upwards and are in close contact with the target object, their tilt angle can be considered zero degrees, and the installation height is the initial detection distance from the sensor surface to the chest cavity. By setting the tilt angle parameter in the rotation transformation matrix to zero, the mattress installation mode is logically equivalent to a special case of the headboard installation mode, ensuring that subsequent steps can be implemented equivalently regardless of the sensor's installation method.

[0081] Specifically, the frequency synthesizer inside the millimeter-wave radar sensor generates a frequency-modulated continuous wave (FM-CVT). An FM-CVT is a continuous electromagnetic wave whose signal frequency increases linearly with time. The FM-CVT is radiated to the monitored area via the transmitting antenna of the millimeter-wave radar sensor. The transmitting antenna is the radio frequency front-end unit in the millimeter-wave radar sensor that performs the conversion of high-frequency electrical signals into electromagnetic wave energy. The FM-CVT propagates in the monitored area and contacts the chest surface of the target object, causing electromagnetic reflection. The reflected electromagnetic wave carries phase shift information reflecting the physiological micro-undulations of the chest surface and is captured by the receiving antenna array of the millimeter-wave radar sensor. The receiving antenna array is an array-type electromagnetic sensing set composed of multiple independent receiving antenna units for sensing spatial reflected waves. The number of receiving antenna units is determined based on the distance between the center of the left and right chest cavities of the target object within the monitored area and the installation height of the millimeter-wave radar sensor; for example, four receiving antenna units are provided. The millimeter-wave radar sensor performs an analog multiplication operation between the FM-CVT radiated by the transmitting antenna and the reflected signal acquired by the receiving antenna array. The result of the analog multiplication operation is filtered by a low-pass filter to extract the intermediate frequency signal. Specifically, the analog multiplication operation generates a high-frequency sum component whose frequency equals the sum of the instantaneous frequency of the FM continuous wave and the instantaneous frequency of the reflected signal, and a low-frequency difference component whose frequency equals the difference between the instantaneous frequency of the FM continuous wave and the instantaneous frequency of the reflected signal. The low-pass filter has a cutoff frequency. The preset cutoff frequency is the highest frequency threshold that the low-pass filter allows the electrical signal to pass through. The cutoff frequency is set to be greater than the maximum beat frequency corresponding to the maximum expected detection distance of the target object within the monitored area, and less than the frequency value of the high-frequency sum component. The maximum beat frequency refers to the maximum difference frequency between the reflected signal and the FM continuous wave generated when the target object is at the edge of the monitored area. The low-pass filter performs frequency-selective filtering on the result of the analog multiplication operation, extracting the intermediate frequency signal by suppressing the high-frequency sum component and retaining the low-frequency difference component. Intermediate frequency (IF) signal refers to an analog low-frequency signal whose frequency is equal to the difference between the instantaneous frequency of the frequency-modulated continuous wave radiated by the transmitting antenna and the instantaneous frequency of the reflected signal acquired by the receiving antenna array. The analog-to-digital converter inside the millimeter-wave radar sensor performs quantization sampling on the IF signal at a preset sampling frequency, generating a digitized complex signal sequence, i.e., the radio frequency (RF) echo signal. The sampling frequency is set to be greater than twice the cutoff frequency to satisfy the sampling theorem and to restore the phase changes caused by the physiological micro-undulations on the surface of the chest cavity. The RF echo signal integrates range scattering characteristics, Doppler scattering characteristics, and angular scattering characteristics; specifically, the RF echo signal includes range scattering characteristics, Doppler scattering characteristics, and angular scattering characteristics.The range scattering characteristics reflect the radial distance distribution of the target object relative to the millimeter-wave radar sensor; the Doppler scattering characteristics reflect the microscopic motion velocity of the target object's surface, the physical essence of which is the reflection phase shift caused by the physiological micro-undulations of the chest cavity surface; the angular scattering characteristics reflect the spatial orientation information of the target object in the horizontal and vertical dimensions. The physiological micro-undulations of the chest cavity surface cause the phase value of the corresponding ranging unit in the radio frequency echo signal to exhibit periodic oscillations with the sampling time, forming the original data carrier characterizing the effective signal energy.

[0082] The time interval between the start times of two adjacent frequency-modulated continuous waves (FM-CLL) is defined as the pulse period. The millimeter-wave radar sensor continuously generates M FM-CLLs according to the pulse period, where the value of M is determined based on the Doppler frequency resolution required for stroke risk warning. Specifically, to distinguish between the respiratory rate and heart rate of a target object, the total observation time of each frame of the radio frequency echo signal is set, for example, to 128 FM-CLLs. Within each pulse period, the analog-to-digital converter (ADC) performs discretization processing on the intermediate frequency signal within the sweep duration, which refers to the time window for effective data acquisition within the FM-CLL. The discrete values ​​generated by the ADC during the sampling process within the sweep duration are defined as sampling points. The total number of sampling points generated within each pulse period is determined by the product of the sampling frequency and the sweep duration. Each frame of the radio frequency echo signal consists of all sampling points within the M pulse periods. A fast Fourier transform is performed on the radio frequency echo signal stream to convert the sampling points into a range spectrum reflecting the energy distribution. The range spectrum consists of multiple equally spaced frequency indices, each frequency index being defined as a ranging unit. The number of ranging units is determined based on the maximum detection range of the millimeter-wave radar sensor and the range resolution determined by the sweep bandwidth, to achieve equally spaced discretization of the monitored area in the range dimension. The ranging unit with the largest energy amplitude is selected from the range spectrum as the target ranging unit. The target ranging unit represents the radial distance position of the target object's chest cavity in three-dimensional physical space. The radial distance value is based on the frequency index position of the target ranging unit in the range spectrum. The radial distance value relative to the millimeter-wave radar sensor is calculated using the sweep bandwidth of the frequency-modulated continuous wave and the speed of light. Specifically, the radial distance value is equal to the product of the beat frequency corresponding to the target ranging unit and the speed of light, divided by twice the ratio of the sweep bandwidth to the sweep duration. Values ​​located at the index position of the target ranging unit are extracted from the pulse period. Since the intermediate frequency signal retains the phase and amplitude characteristics of the reflected wave after analog-to-digital conversion, the obtained value contains real and imaginary information and is defined as a complex voltage value. M complex voltage values ​​are combined into a one-dimensional vector according to the temporal order of the pulse period, defined as a complex sampling sequence. The complex sampling sequence reflects the continuous phase modulation pattern of electromagnetic waves generated by the physiological micro-undulations on the surface of the target object's chest cavity within the total observation time. The square of the modulus of each complex voltage value in the complex sampling sequence is calculated, i.e., the sum of the squares of the real and imaginary parts, to obtain the instantaneous power generated at the chest cavity location of the target object in each pulse period. All instantaneous powers in the complex sampling sequence are summed to obtain the total energy value of a single frame. This total energy value is then divided by the number of frequency-modulated continuous waves to obtain the average energy power. An environmental correction coefficient is introduced. This environmental correction coefficient is used to correct for the weight of the contribution of non-physiological background clutter such as walls and bed boards to the energy distribution within the monitored area.By acquiring the average background energy output of the millimeter-wave radar sensor when the monitored area is in an unloaded state, and taking the reciprocal of the average background energy as the environmental correction coefficient, for each complex voltage value in the complex sampling sequence, the sum of the square of the real part information and the square of the imaginary part information is calculated to obtain the instantaneous power value corresponding to each pulse period. The instantaneous power values ​​corresponding to all complex voltage values ​​in the complex sampling sequence are accumulated to obtain the total energy value. The total energy value is divided by the number of frequency-modulated continuous waves to obtain the average energy power value. The average energy power value is multiplied by the environmental correction coefficient to finally generate the effective signal energy.

[0083] After acquiring effective signal energy, to further address the issue of spatial aliasing between the target object's chest cavity and surrounding static strong reflectors due to insufficient radial distance information, and to resolve the ambiguity in the spatial directionality of physiological signal sources caused by single-dimensional distance detection, a spatial calculation logic is established to convert the phase difference between channels of the receiving antenna array into three-dimensional geometric coordinates. The aim is to leverage the multi-channel aperture advantage of millimeter-wave radar sensors to capture the unique physical mapping of the target object's chest cavity surface in the spatial dimension, i.e., the spatial reflection anchor point.

[0084] Specifically, adjacent receiving antenna elements in the receiving antenna array have a fixed physical spacing, and the array contains multiple independent channels. Each channel refers to a dedicated radio frequency (RF) front-end circuit and analog-to-digital conversion path for each receiving antenna element, ensuring that the reflected signal sensed by each element can be independently and with high fidelity quantized into digital values. To extract the spatial geometric deflection angle of the target object relative to the millimeter-wave radar sensor, the complex voltage value at the target ranging unit index position of each independent channel is extracted from the RF echo signal. Due to slight differences in the electromagnetic wave propagation path from the target object's chest cavity to each receiving antenna element, phase delay and amplitude fluctuation characteristics are generated between the independent channels when the reflected signal reaches different elements. The complex voltage values ​​containing phase delay and amplitude fluctuation characteristics are combined according to the spatial geometric order of the receiving antenna elements to construct a spatial measurement vector. The spatial measurement vector encapsulates information reflecting the spatial phase distribution of the reflected signal in the spatial dimension. The phase delay value refers to the difference in phase angle between the complex voltage values ​​acquired by each independent channel on the complex plane. Physically, it characterizes the wavefront arrival time difference caused by the difference in the propagation path length from the chest cavity of the target object to different receiving antenna elements. The amplitude fluctuation characteristic refers to the change in the magnitude of the complex voltage values ​​acquired by each independent channel with spatial position distribution. Physically, it characterizes the spatial attenuation of electromagnetic waves during propagation and the uneven energy distribution on the receiving antenna array.

[0085] To reveal the hidden geometric orientation features in the spatial phase distribution information and address the technical deficiencies of decreased spatial resolution and physiological signal source location drift caused by phase aliasing, an angle analysis logic based on the principle of coherent interferometry is established. A set of reference vectors for spatial matching is constructed. Specifically, the independent channel at the head of the receiving antenna array is set as the reference phase center; for each candidate angular direction, the geometric projection of the physical distance between two adjacent independent channels in the incident direction of the reflected signal is calculated using trigonometric relationships. This geometric projection is defined as the spatial displacement, and the candidate angle refers to the spatial observation direction formed by the candidate horizontal azimuth angle and the candidate vertical elevation angle. Specifically, based on the detection field of view of the millimeter-wave radar sensor, discretization is performed in both the horizontal azimuth and vertical elevation dimensions. Discrete angle sampling points generated in the horizontal azimuth dimension are defined as candidate horizontal azimuth angles, and discrete angle sampling points generated in the vertical elevation dimension are defined as candidate vertical elevation angles. The unique spatial vector determined by a pair of candidate horizontal azimuth and candidate vertical elevation angles constitutes the candidate angle. Since electromagnetic waves generate a phase rotation period of one wavelength per propagation, the phase rotation radians generated by each independent channel relative to the reference phase center are determined by calculating the ratio of the spatial displacement to the wavelength of the radio frequency echo signal. These phase rotation radians are then used as the theoretical phase offset for the corresponding candidate angle. The theoretical phase offset corresponding to each candidate angle is encapsulated as an angle steering vector, and the angle steering vectors corresponding to all candidate angles together constitute a reference vector set. This reference vector set physically forms a mapping database of spatial angles and phase characteristics. The reference vector set is traversed, and the inner product operation is performed between the spatial measurement vector and the angle steering vector corresponding to each candidate angle in the reference vector set. The inner product operation refers to the process of performing conjugate multiplication and accumulation of the complex voltage values ​​corresponding to each independent channel in the spatial measurement vector with the theoretical phase offset of the corresponding channel in the angle steering vector. Since the spatial measurement vector carries the true phase delay value of the target object's chest cavity reflection, when the true phase delay in the spatial measurement vector perfectly matches the angle steering vector corresponding to the candidate angle in the reference vector set, the signal phases between different independent channels are aligned, thus generating directional coherent superposition in the complex domain. The physical effect of directional coherent superposition is that the magnitude of the complex result output by the inner product operation reaches its maximum value in the matched candidate angle direction, thereby refocusing the discrete spatial phase distribution information into a digital beam with high spatial gain. The magnitude of the output result of each inner product operation is extracted and defined as the spatial reflection intensity. By matching and traversing all angle steering vectors in the reference vector set, the angle energy spectrum is finally generated.The angular energy spectrum refers to a two-dimensional energy distribution matrix constructed using candidate horizontal azimuth and candidate vertical elevation angles as dual-axis coordinate indices, and the spatial reflection intensity corresponding to each candidate angle as its numerical value. The local extremum point with the largest value is searched within the angular energy spectrum. The value of this local extremum point is defined as the peak energy amplitude, and the row and column index of this local extremum point in the angular energy spectrum is defined as the peak point coordinates. The candidate horizontal azimuth angle corresponding to the peak point coordinates is determined as the target azimuth angle, and the candidate vertical elevation angle corresponding to the peak point coordinates is determined as the target elevation angle.

[0086] In the micro-feature decoupling module, the step of performing peak detection and coordinate mapping on the angular energy spectrum to determine the spatial reflection anchor point includes:

[0087] Specifically, a radar local coordinate system is constructed with the center of the transmitting antenna of the millimeter-wave radar sensor as the origin. This local coordinate system is a three-dimensional Cartesian coordinate system following the right-hand rule. Its Z-axis is defined as perpendicular to the receiving antenna array plane and pointing towards the monitored area; the X-axis is defined as the axis along the horizontal arrangement direction of the receiving antenna elements, used to characterize the target's lateral orientation; and the Y-axis is defined as the axis within the receiving antenna array plane and perpendicular to the X-axis, used to characterize the target's longitudinal orientation. The lateral, longitudinal, and height components are calculated using this radar local coordinate system. Specifically, the lateral component is determined by the sinusoidal projection of the radial distance value onto the target's azimuth direction, used to characterize the target object's displacement along the X-axis; the longitudinal component is determined by the sinusoidal projection of the radial distance value onto the target's elevation direction, used to characterize the target object's displacement along the Y-axis; and the height component is determined by the projection of the radial distance value onto the Z-axis normal direction, used to characterize the target object's vertical depth relative to the millimeter-wave radar sensor plane. The coordinate points defined by the lateral, longitudinal, and height components in the radar local coordinate system are defined as three-dimensional mounting points. A rotation transformation matrix is ​​used to rotate the lateral, longitudinal, and height components in the radar's local coordinate system around the X-axis. The rotation angle is equal to the installation tilt angle to correct for the tilt of the observation axis caused by the elevation installation of the millimeter-wave radar sensor. Specifically, the lateral component value is kept constant and defined as the rotated lateral component. The longitudinal component is scaled using the cosine of the installation tilt angle, and the sinusoidal projection of the height component in the installation tilt direction is superimposed to generate the rotated longitudinal component. The height component is scaled using the cosine of the installation tilt angle, and the sinusoidal projection of the longitudinal component in the installation tilt direction is subtracted to generate the rotated height component. The rotated height component and the installation height are then vector-displaced and superimposed. Specifically, the absolute vertical height of the target object relative to the ground of the monitored area is determined by calculating the algebraic sum of the installation height and the rotated height component. The three-dimensional physical coordinate point formed by the rotated lateral component, the rotated longitudinal component, and the absolute vertical height is defined as a spatial reflection anchor point. The spatial reflection anchor point represents the center of mass of motion of the target object's chest surface in three-dimensional physical space.

[0088] In the micro-feature decoupling module, the phase unwrapping and physical displacement mapping based on spatial reflection anchors are performed to obtain a physiological displacement sequence. Multi-scale digital filtering is then applied to the physiological displacement sequence to extract the physiological feature waveforms, including:

[0089] Specifically, after obtaining the spatial reflection anchor point, in order to transform the microscopic phase modulation effect into a physiological waveform with clinical diagnostic value, and to solve the problems of waveform distortion caused by phase entanglement and the overlapping of respiratory and heartbeat signal spectra in non-contact sensing, a demodulation logic from time-domain complex sequence to physical displacement time history is established.

[0090] Specifically, since the spatial reflection anchor point represents the center of mass of motion on the chest cavity surface of the target object, for the complex voltage value corresponding to each pulse cycle in the complex sampling sequence, the real and imaginary parts are extracted, and I and Q are used to represent the real and imaginary parts respectively. The original phase value of each pulse cycle is then calculated using the arctangent function. The original phase value reflects the instantaneous electrical length of the reflecting interface relative to the phase center of the millimeter-wave radar sensor, and its calculation formula is as follows: ,in, (·) denotes the arctangent function. The original phase values ​​corresponding to all pulse periods are arranged and combined in time sequence into a one-dimensional vector to obtain the original phase sequence. This original phase sequence physically represents the trajectory of the electrical length evolution of the reflecting interface relative to the phase center of the millimeter-wave radar sensor over the total observation time. To recreate the true continuous motion, the original phase sequence is unwrapped. Specifically, the displacement of the target object's chest cavity may cause a step jump in the original phase values ​​within the range of [-π, π] radians, i.e., phase entanglement. To reconstruct the continuous physical motion trajectory, the phase difference between two adjacent pulse cycles is detected, and a jump threshold is set. According to the phase sampling theorem, to ensure the single-valued mapping of phase evolution on the complex plane, the phase change between adjacent sampling points must be limited to half the length of the principal phase value interval [-π, π]. If the phase difference is greater than π, it is determined that the radial displacement of the target object has caused phase overlap across the boundary, rather than a true instantaneous displacement change. A numerical compensation operation of subtracting 2π is performed on the original phase value of all subsequent pulse cycles of the radio frequency echo signal of the current frame. The reason is that when the positive phase difference is greater than π, it indicates that the current measurement value has crossed the -π boundary and entered the previous observation cycle, producing a 2π-radian advance artifact, which needs to be canceled by subtraction. Conversely, if the phase difference is less than -π, a 2π-accumulated numerical compensation operation is performed on the original phase values ​​of all subsequent pulse cycles of the radio frequency echo signal in the current frame. This is because when the negative phase difference exceeds π, it indicates that the current measurement value has crossed the π boundary into the next observation cycle, generating a 2π-radian hysteresis artifact. By performing consistency compensation on all pulse cycles, a continuous phase sequence with topological continuity is constructed, realistically restoring the physiological micro-vibration trajectory of the thoracic cavity surface. This continuous phase sequence characterizes the cumulative phase value sequence of the topological continuous evolution of the electrical length of the thoracic cavity surface relative to the phase center of the millimeter-wave radar sensor over time. Using the linear mapping relationship between phase and radial distance, the continuous phase sequence is mapped to a physical displacement value with units of length. Specifically, for each pulse cycle in the continuous phase sequence, the product of the wavelength of the radio frequency echo signal and the corresponding phase value is calculated and divided by a constant operator of 4π to obtain the instantaneous displacement characteristic value. This instantaneous displacement characteristic value is used to quantify the numerical characteristics of the submicron-level mechanical vibration displacement amplitude of the target object's thoracic cavity surface relative to the initial reference position under the corresponding pulse cycle. The calculation basis for obtaining the instantaneous displacement characteristic value is based on the principle of linear mapping between electromagnetic wave phase and wavelength and the two-way propagation mechanism of millimeter-wave radar. A constant operator is introduced to counteract the double phase accumulation effect generated during the round trip of electromagnetic waves.This method achieves high-fidelity reconstruction of phase values ​​from a continuous phase sequence into instantaneous displacement feature values ​​that quantify the amplitude of microscopic deformation on the surface of the chest cavity of a target object, ensuring sub-micron level accuracy in displacement reconstruction. The instantaneous displacement feature values ​​corresponding to all pulse cycles within a frame are then arranged and combined in temporal order to obtain a physiological displacement sequence. This physiological displacement sequence physically realizes the concrete transformation of an abstract electrical length trajectory into a sub-micron level mechanical vibration displacement trajectory, fully preserving the minute displacement features generated by the heartbeat and lung respiration of the target object.

[0091] Multi-scale digital filtering is performed on the physiological displacement sequence to remove the physiological feature components generated by heartbeat and lung respiration on the surface of the target object's chest cavity. Two sets of frequency-selective filtering operations are performed in parallel using a zero-phase-shift digital bandpass filter. This zero-phase-shift digital bandpass filter refers to a digital algorithm logic preset within the millimeter-wave radar sensor, designed to cancel out the group delay generated by the digital filter, ensuring that the extracted waveform components are physically aligned with the sampling time of the physiological displacement sequence on the time axis. Specifically, the first set of filtering operations extracts the displacement features generated by lung respiration. A respiratory frequency range is set based on the medical statistical characteristics of human respiratory rate at rest, for example, between 0.1 and 0.5 Hz. Within this respiratory frequency range, the physiological displacement sequence is filtered... Frequency-selective filtering is performed to suppress frequency components outside the respiratory frequency range, outputting a waveform representing the macroscopic periodic expansion and contraction trajectory of the thoracic cavity surface driven by lung respiration, defined as the respiratory characteristic waveform. The second set of filtering operations extracts displacement characteristics driven by heartbeats. A heartbeat frequency range is set based on the medical statistical characteristics of the human heartbeat frequency in a resting state, for example, between 0.8 and 2 Hz. Within the heartbeat frequency range, frequency-selective filtering is performed on the physiological displacement sequence, suppressing frequency components outside the range to extract the weak tremor component formed by the mechanical pulsation of the heart, defined as the heartbeat characteristic waveform. The heartbeat characteristic waveform and the respiratory characteristic waveform together constitute the physiological characteristic waveform representing the microscopic dynamic state of the target object's thoracic cavity surface.

[0092] The micro-feature decoupling module addresses the technical challenges of strong background clutter masking weak physiological signals, macroscopic bodily motion interference in micro-vibration extraction, and waveform distortion caused by phase entanglement in complex home monitoring environments. It achieves high-fidelity, sub-micron-level extraction of respiratory and heartbeat characteristic waveforms from mixed radio frequency echoes. Specifically, complex feature reconstruction transforms physical layer signals into original carriers with energy distribution characteristics; the spatial reflection anchor point, leveraging multi-channel aperture advantages, locates the center of mass of motion, eliminating spatial directional ambiguity of physiological signal sources caused by radial detection; phase untangling and displacement mapping utilize a two-way electromagnetic wave sensing mechanism to visualize abstract phase evolution as a continuous mechanical vibration trajectory; and multi-scale digital filtering employs zero-phase translation logic to ensure the physical alignment and pure extraction of multi-dimensional physiological components along the time axis.

[0093] In the aforementioned spatial heterogeneous module, the construction of a virtual anatomical projection surface based on spatial reflection anchor points, and the invocation of physiological feature waveforms within the virtual anatomical projection surface to perform grid-dimensional physiological kinetic energy focusing mapping to obtain a kinetic energy distribution heatmap, includes:

[0094] Specifically, the three-dimensional physical coordinates of the spatial reflection anchor point, consisting of the rotated lateral component, the rotated longitudinal component, and the absolute vertical height, are extracted as the geometric reference center. Within the three-dimensional physical space of the monitored area, a two-dimensional physical cross-section is constructed, passing through the geometric reference center and parallel to the plane containing the X and Y axes of the radar local coordinate system. Based on the statistical dimensions of human thoracic cavity anatomy, lateral and longitudinal boundary thresholds are set; for example, the lateral boundary threshold is set to 0.3 meters, and the longitudinal boundary threshold is set to 0.25 meters. Using the geometric reference center as the origin, a rectangular area within the range of the lateral and longitudinal boundary thresholds is extracted from the two-dimensional physical cross-section. The rectangular area is divided into multiple two-dimensional discrete grids of equal size according to a preset grid side length, and each two-dimensional discrete grid is assigned a unique three-dimensional geometric coordinate. The grid side length is determined based on the minimum spatial resolution achievable by the receiving antenna array at the spatial reflection anchor point; for example, it is set to 0.02 meters. The spatial geometric surface formed by all the two-dimensional discrete grids is defined as the virtual anatomical projection surface. The virtual anatomical projection plane physically simulates a parallel observation section covering the surface of the target object's thoracic cavity. The three-dimensional geometric coordinates refer to the absolute positioning coordinates of the two-dimensional discrete mesh in three-dimensional physical space, composed of lateral, longitudinal, and height positioning values. Specifically, the lateral and longitudinal positioning values ​​are determined by superimposing the rotated lateral and longitudinal components of the three-dimensional physical coordinate points, along with the lateral and longitudinal offsets of the two-dimensional discrete mesh relative to the geometric reference center within the rectangular region. The height positioning value is always equal to the absolute vertical height. The lateral and longitudinal offsets are determined based on the mesh side length and the spatial arrangement order of the two-dimensional discrete mesh within the rectangular region. For example, with the geometric reference center as the index zero point, each two-dimensional discrete mesh is assigned an integer-order lateral and longitudinal mesh index. The lateral offset is equal to the product of the lateral mesh index and the mesh side length, and the longitudinal offset is equal to the product of the longitudinal mesh index and the mesh side length.

[0095] To achieve the spatial dimension mapping extension of the physiological feature extraction logic, spatial energy focusing calculation is performed for each two-dimensional discrete grid within the virtual anatomical projection plane: the three-dimensional installation coordinates of the receiving antenna element corresponding to each independent channel in the receiving antenna array of the millimeter-wave radar sensor are obtained in the radar local coordinate system; the physical spatial straight-line distance from the three-dimensional geometric coordinates of the two-dimensional discrete grid to the three-dimensional installation point of each receiving antenna element is calculated using the three-dimensional spatial distance formula; specifically, the physical spatial straight-line distance is equal to the square root of the sum of the squares of the following three terms: the difference between the lateral positioning value of the two-dimensional discrete grid and the lateral component in the three-dimensional installation point, the difference between the longitudinal positioning value of the two-dimensional discrete grid and the longitudinal component in the three-dimensional installation point, and the difference between the height positioning value of the two-dimensional discrete grid and the height component in the three-dimensional installation point. The calculation of the physical spatial straight-line distance is based on the distance measurement principle of three-dimensional Euclidean geometry. Under the physical premise that electromagnetic waves propagate in a uniform straight line within the monitored area, the absolute one-way propagation path length between the phase center of the receiving antenna of the RF front-end and a specific micro-grid on the virtual anatomical projection surface is accurately quantified by calculating the sum of the squares of the coordinate differences in each orthogonal dimension of the three-dimensional space and taking the arithmetic square root. This provides a rigid geometric scale benchmark for subsequent phase compensation. The physical spatial straight-line distance is combined with the wavelength of the RF echo signal, and the theoretical phase delay for each independent channel is determined based on the phase evolution law of electromagnetic wave two-way propagation. Specifically, the theoretical phase delay is the phase value formed by multiplying twice the physical spatial straight-line distance by the wavelength of the RF echo signal by a constant operator of 2π. The calculation of the theoretical phase delay is based on the electromagnetic wave propagation theory and the two-way sensing mechanism of millimeter-wave radar. For every wavelength of physical distance that an electromagnetic wave travels in free space, its electrical phase evolution in the complex plane is manifested as a rotation of 2π radians. Since the local time-domain reflected signal captured by the millimeter-wave radar is a round-trip signal radiated from the transmitting antenna to a specific two-dimensional discrete grid surface and reflected back to the receiving antenna array, the total physical propagation path length it experiences is actually twice the straight-line distance of a single physical space. Therefore, by multiplying the straight-line distance of a single physical space by 2 and dividing by the wavelength to calculate the total number of wavelengths, and then multiplying by the constant operator of 2π, the total cumulative phase offset generated by the electromagnetic wave on the two-way round-trip path can be accurately calculated. This total cumulative phase offset constitutes the theoretical phase compensation benchmark required for reverse cancellation when performing digital beamforming and spatial coherent energy focusing. The complex voltage values ​​obtained from each independent channel in the radio frequency echo signal are then subjected to inverse phase rotation on the complex plane using the theoretical phase delay of the corresponding independent channel for phase compensation. The complex voltage values ​​of each independent channel after phase compensation are then coherently accumulated to obtain the local time-domain reflection signal.

[0096] The respiratory and heart rate ranges associated with physiological characteristic waveforms are used to filter out non-target frequency band interference caused by strong background reflectors and accurately extract targeted physiological signals. Specifically, the local time-domain reflection signal is input into a zero-phase shift digital bandpass filter, and frequency-selective filtering is performed in parallel within the respiratory and heart rate ranges to extract local respiratory and heart rate micro-motion waveforms at a two-dimensional discrete grid. The local respiratory and heart rate micro-motion waveforms are then superimposed and reconstructed in the time domain to obtain local physiological micro-motion components. These components refer to the physical trajectory of the microscopic mechanical displacement generated by the combined mechanical pulsation of the heart and lung respiration at a local physical location corresponding to the two-dimensional discrete grid on the surface of the target object's chest cavity, which evolves continuously over time. According to the theory of mechanical vibration dynamics and the signal energy theorem, the total vibration energy released by a mechanical vibration system within a specific time window is strictly proportional to the absolute integral of the square of its instantaneous vibration displacement amplitude over that time window. The time integral of the squared amplitude of the local physiological micro-motion component within the total observation time of a single frame of radio frequency echo signal is calculated to obtain the numerical value that quantifies the absolute magnitude of the micro-mechanical vibration energy at the two-dimensional discrete grid, defined as the kinetic energy density point. All two-dimensional discrete grids within the virtual anatomical projection plane are traversed, and the kinetic energy density points corresponding to all extracted two-dimensional discrete grids are combined into a two-dimensional energy matrix according to their respective horizontal and vertical grid indices, generating a kinetic energy distribution heatmap. This kinetic energy distribution heatmap, through the spatial variation of the numerical gradient in the two-dimensional energy matrix, faithfully recreates the spatial morphology of the micro-kinetic energy distribution at various anatomical sites on the thoracic surface of the target object under the drive of heartbeat and lung respiration.

[0097] In the aforementioned spatial heterogeneous module, the process of dividing the kinetic energy distribution heatmap into dual-sided monitoring zones and performing two-dimensional spatial discrete summation calculations based on these dual-sided monitoring zones to obtain the asymmetric index includes:

[0098] Specifically, a dual-sided monitoring zone is divided within the kinetic energy distribution heatmap, and a two-dimensional spatial discretization and summation calculation is performed based on the dual-sided monitoring zone to obtain the asymmetric index.

[0099] The geometric reference center, set during the construction of the virtual anatomical projection plane, is extracted and combined with spatial reflection anchor points to characterize the physical properties of the target object's thoracic cavity surface's center of mass. This geometric reference center serves as the anatomical reference for the human thoracic cavity. Using the geometric reference center as the origin, a vertical dividing line with a horizontal grid index always equal to zero is drawn along a direction parallel to the Y-axis in the radar local coordinate system. This vertical dividing line is defined as the anatomical sagittal projection line. This anatomical sagittal projection line divides the virtual anatomical projection plane into two spatial sub-regions with strictly symmetrical physical dimensions. Based on the spatial arrangement order of the two-dimensional discrete grids in the two-dimensional energy matrix of the kinetic energy distribution heatmap, the set of two-dimensional discrete grids with negative horizontal grid indices is defined as the left monitoring area, and the set of two-dimensional discrete grids with positive horizontal grid indices is defined as the right monitoring area. The left and right monitoring areas spatially map the left and right thoracic cavity surfaces of the target object, respectively. The polarity of the horizontal grid index is determined by the physical position of the two-dimensional discrete grid relative to the geometric reference center on the X-axis. Specifically, the horizontal grid indices of the two-dimensional discrete grid extending in the opposite direction of the X-axis with the geometric reference center as the origin are assigned decreasing negative integers and spatially correspond to the left thoracic surface of the target object; the horizontal grid indices of the two-dimensional discrete grid extending in the positive direction of the X-axis are assigned increasing positive integers and spatially correspond to the right thoracic surface of the target object. The left and right monitoring areas are combined into a dual-sided monitoring area.

[0100] For the left monitoring area, the kinetic energy density points corresponding to all two-dimensional discrete grids are extracted and subjected to two-dimensional spatial discrete summation to obtain the total micro-motion energy value of the left side. Specifically, the horizontal grid index of the two-dimensional energy matrix is ​​defined as a column variable, and the vertical grid index is defined as a row variable. The value set of the column variable is determined to be all negative integers, and the value set of the row variable is all integer indices of the virtual anatomical projection plane in the vertical dimension. All grid coordinate pairs formed by the intersection of the value sets of the column variables and the value sets of the row variables are generated in sequence. Based on each grid coordinate pair, the corresponding unique kinetic energy density point value on the two-dimensional discrete grid is located and extracted in the two-dimensional energy matrix. All extracted kinetic energy density point values ​​are algebraically summed to obtain the total micro-motion energy value of the left side. Similarly, for the right-side monitoring area, extract the kinetic energy density points corresponding to all two-dimensional discrete grids, perform two-dimensional spatial discrete summation operations, determine the value set of the column variables as all positive integers, use the same grid coordinate pair traversal logic to generate all grid coordinate pairs corresponding to the right-side region, locate and extract all kinetic energy density point values ​​on the corresponding two-dimensional discrete grids of the grid coordinate pairs, and perform algebraic accumulation to obtain the total micro-motion energy value on the right side. The total micro-motion energy values ​​on the left and right sides quantify the total macroscopic mechanical kinetic energy released by the left and right halves of the target object's pleural cavity under the dual physiological activities of cardiac pulsation and lung respiration within the total observation time of a single frame of radio frequency echo signal. For example... Figure 4 The image shown is a schematic diagram of spatial mapping for dividing bilateral monitoring areas based on anatomical sagittal plane projection lines, provided by an embodiment of the present invention; as shown... Figure 4 The bottom layer of the image uses gray dashed lines to depict the physical outline projection of the target object's thoracic cavity, in order to help present the physical correspondence between the abstract two-dimensional discrete mesh and the actual anatomical parts of the human body. Figure 4 The solid red dot at the center represents the extracted geometric reference center, serving as the origin of the coordinate system for spatial geometric division. The thick red vertical dashed line passing through this geometric reference center is the anatomical sagittal projection line, while the horizontally penetrating gray solid line with an arrow represents the X-axis in the radar local coordinate system. Figure 4 The green shaded area on the left is the left monitoring area, which contains a set of meshes extending in the opposite direction of the X-axis and spatially corresponds to the left thoracic surface of the target object; accordingly, Figure 4 The orange shaded area on the right side represents the right-side monitoring area. This area contains a set of meshes extending along the positive X-axis and spatially corresponds to the right thoracic surface of the target object. Furthermore, Figure 4 The numerical values ​​provided in the examples are used to vividly illustrate negative and positive integers.

[0101] Based on the pathological evolution of unilateral motor neuron damage in the early stages of stroke leading to weak muscle retardation, when the target object is at rest, the energy of minute physiological displacements on both sides of its chest cavity will exhibit a non-random symmetry disruption. To quantify the degree of this symmetry disruption, the asymmetry index K between the total energy of the left and right sides of the micro-motion is calculated. Specifically, the formula for calculating the asymmetry index is: ,in, The calibration factor is introduced based on the anatomical bias of the human heart being located on the left side of the chest cavity, and the slight physical bias of the millimeter-wave radar sensor's installation tilt angle. The total micro-motion energy values ​​on the left and right sides of the target object in an absolutely healthy state also exhibit inherent background bias. By extracting the total micro-motion energy values ​​on the left and right sides corresponding to multiple frames of radio frequency echo signals from the target object within its historical healthy baseline period, and calculating the ratio of the total micro-motion energy values ​​on the left to the right in each frame, the arithmetic mean of this ratio across all frames is determined as the basic deviation coefficient. Simultaneously, by calling the weight and height parameters and dividing the weight parameter by the square of the height parameter, a body mass index (BMI) characterizing the thickness of the chest wall medium of the target object is obtained. A standard reference index is set, based on the median BMI of healthy adults in clinical medical statistics; for example, it is set to 22. The BMI is divided by the standard reference index to obtain a dimensionless body shape compensation coefficient. This body shape compensation coefficient is used to perform a personalized weighted operation on the basic deviation coefficient, and the product is determined as the calibration factor. The asymmetry index calculation formula is based on the principles of differential measurement and relative normalization of physiological signals. The numerator extracts the absolute difference between the total micro-motion energy values ​​on the left and right sides, aiming to directly quantify the absolute energy gradient of the bilateral thoracic cavities in micro-mechanical vibration, thereby capturing the objective physical representation of physiological symmetry disruption caused by unilateral motor neuron damage. The denominator introduces the algebraic sum of the total micro-motion energy values ​​on the left and right sides as the divisor, constructing a ratio normalization mechanism. Its physical significance lies in completely decoupling the strong correlation between absolute energy values ​​and individual heterogeneous factors, such as absolute body size, basal tidal volume, and electromagnetic wave spatial attenuation factors, such as the absolute detection range of radar, transforming absolute energy differences into dimensionless relative imbalance ratios. The asymmetry index physically eliminates the systematic interference of absolute body size and absolute radar detection range. An abnormal increase in its value indicates a subclinical weakening of the coordination of bilateral trunk micro-tremors under neuromuscular control, providing a two-dimensional pathological mapping benchmark with strong spatial topological correlation for subsequent analysis.

[0102] The spatial heterogeneous module addresses the challenges of traditional single-point monitoring, which struggles to detect bilateral anatomical differences in movement and whose absolute energy values ​​are severely affected by individual body size, detection distance, and environmental attenuation, making it difficult to quantify pathological characteristics. This is achieved by constructing a virtual anatomical projection surface, performing physiological kinetic energy focusing mapping, generating a kinetic energy distribution heatmap, and calculating an asymmetry index. The module enables spatial visualization of the microscopic kinetic energy distribution in the human thoracic cavity and provides a normalized measure of the symmetry disruption in unilateral neuromuscular movement caused by early stroke damage. Specifically, the virtual anatomical projection surface simulates a parallel observation section covering the thoracic cavity surface, providing a unified spatial index for energy analysis across anatomical sites. Physiological kinetic energy focusing mapping utilizes three-dimensional Euclidean geometry to perform inverse phase compensation, achieving precise focusing of vibrational energy in the grid dimension. Bilateral monitoring area division, combined with anatomical sagittal projection lines, establishes a mapping benchmark from signal space to human anatomical structures. The asymmetry index decouples system interference through differential measurement and ratio normalization mechanisms, extracting energy imbalance features with strong pathological orientation.

[0103] In the aforementioned early warning judgment module, the cross-modal temporal fusion based on physiological characteristic waveforms is performed to obtain a heterogeneous energy imbalance vector. A diagnostic evaluation plane is then constructed based on an asymmetric exponent to map the heterogeneous energy imbalance vector, resulting in risk prediction points. This includes:

[0104] Specifically, the heartbeat and respiratory waveforms from the physiological characteristic waveforms are invoked. For the heartbeat waveform, which is essentially a weak tremor component filtered and extracted from the physiological displacement sequence within the heartbeat frequency range, the values ​​corresponding to this weak tremor component at each sampling time are extracted, and all values ​​are arranged in chronological order of sampling time to form a one-dimensional discrete data sequence. To accurately extract the effective peaks generated by the mechanical pulsation of the heart, an extreme value search window is constructed. The extreme value search window contains a fixed number of continuous sampling points in its data structure. This fixed number is determined by the product of the time length of the extreme value search window and the sampling frequency of the analog-to-digital converter; for example, it is set to 15 continuous sampling points. Inside the extreme value search window, the single sampling point located at the exact middle of the time sequence is defined as the center data point of the window, and all other sampling points before and after the center data point within the extreme value search window are collectively defined as the surrounding data point set. The time length of the extreme value search window is set based on the shortest time interval between two adjacent cardiac mechanical beats under the physical limits of the human body, thereby avoiding repeated misjudgments of peaks within a single cardiac cycle. For example, it is set to 0.3 seconds. In the one-dimensional discrete data sequence, the extreme value search window is shifted point by point according to the time sequence. At each shift pause, the value of the central data point of the window is extracted and algebraically compared with the values ​​of all data points in the surrounding data point set. If the value of the central data point of the window is greater than all the values ​​in the surrounding data point set, the central data point of the window is determined to be a valid peak generated by a cardiac mechanical beat, and the timestamp corresponding to the valid peak on the time axis is extracted. After traversing the entire heartbeat feature waveform to extract the timestamps of all valid peaks, the timestamps of two adjacent valid peaks are differentially calculated, that is, the timestamp of the latter peak is subtracted from the timestamp of the former peak, resulting in a series of discrete time differences. All the above time differences are arranged in chronological order of the corresponding peaks to generate a cardiac interval sequence. For the cardiac interbeat sequence, the sum of squares of the differences between two adjacent time differences is calculated. This sum is then divided by the total number of time differences, and the arithmetic square root is taken to obtain the root mean square of the differences between adjacent cardiac interbeats, which is defined as the short-time variability and used to quantify short-time cardiac rhythm variability. Simultaneously, since the cardiac interbeat sequence is a non-equidistant sequence on the time axis, it is resampled into an equidistant time sequence using a cubic spline interpolation algorithm, defined as an equidistant cardiac temporal sequence. To extract frequency domain features, a time-frequency transformation is performed on this equidistant cardiac temporal sequence to generate a power spectral density function, for example, by mapping the discrete time domain signal to a discrete Fourier transform of the frequency domain energy distribution.Based on international medical statistical benchmarks for assessing the cardiac autonomic nervous system, low-frequency and high-frequency bands are predefined. For example, the low-frequency band is set to 0.04 to 0.15 Hz to characterize the frequency range of the joint regulatory effect of the sympathetic and parasympathetic nervous systems, and the high-frequency band is set to 0.15 to 0.40 Hz to characterize the frequency range of parasympathetic activity. Within the predefined low-frequency and high-frequency bands, mathematical definite integral operations are performed on the power spectral density function to extract low-frequency power and high-frequency power. Specifically, the lower and upper frequency limits of low-frequency power and high-frequency power are used as integration intervals, and numerical definite integral calculations are performed on the curve containing the power spectral density function to obtain the enclosing areas under the two integration intervals, i.e., low-frequency power and high-frequency power. The low-frequency power is divided by the high-frequency power to obtain the energy ratio, which is used to quantify the balance between the sympathetic and parasympathetic nervous systems in the frequency domain. The calculated short-term variability is combined with the energy ratio to obtain a heart rate variability index that quantifies the autonomic nervous system regulation ability of the target object's heart. The standard deviation of the intercardiac interval sequence is calculated, and the standard deviation is divided by the arithmetic mean of the intercardiac interval sequence to obtain the absolute coefficient of variation. A rhythm analysis sliding window is set to extract the intercardiac interval sequence. The duration of the rhythm analysis sliding window is set based on the shortest clinically defined duration of paroxysmal atrial fibrillation, for example, 30 seconds, and the sliding step size is set to 10 seconds. The rhythm analysis sliding window is sequentially translated on the intercardiac interval sequence according to the sliding step size; within each local window generated by the translation, all time differences falling within the local window are extracted, and the standard deviation of the time differences is divided by the arithmetic mean of the time differences to obtain the local absolute coefficient of variation. An irregular rhythm threshold is set, which is set based on the clinical statistical cutoff benchmark for cardiac rhythm variability between healthy individuals and atrial fibrillation patients, for example, 0.15. Within each local window, if the local absolute coefficient of dispersion is greater than the irregular rhythm threshold, an atrial fibrillation event is determined to have occurred within the corresponding time period of that local window. The absolute time span of the entire heartbeat feature waveform being processed is extracted on the time axis and defined as the total observation duration. All local windows within the total observation duration that are determined to have caused atrial fibrillation events are extracted. The time segments covered by these local windows on the time axis are overlapped, deduplicated, and summed to obtain the total duration of atrial fibrillation. This total duration of atrial fibrillation is then divided by the total observation duration to obtain the atrial fibrillation load characteristics.

[0105] The respiratory characteristic waveform represents the macroscopic periodic expansion and contraction trajectory of the thoracic cavity surface driven by lung respiration. The values ​​corresponding to each consecutive sampling time of this waveform component are extracted and arranged in chronological order to form a real part data sequence. The corresponding analytical signal is then constructed to extract the envelope features. Specifically, a Hilbert transform is performed on the real part data sequence to generate an imaginary part data sequence with a 90-degree phase shift. The values ​​of the real and imaginary part data sequences at the same sampling time are extracted point-by-point. The sum of the squares of the values ​​of the real and imaginary part data sequences at the same sampling time is calculated and its square root is taken to obtain the instantaneous amplitude envelope. The discrete values ​​on the instantaneous amplitude envelope reflect the macroscopic expansion and contraction amplitude of the thoracic cavity at the corresponding sampling time. The discrete values ​​on the instantaneous amplitude envelope are combined in time sequence to obtain the instantaneous tidal volume sequence. The arithmetic mean of the instantaneous tidal volume sequence of the target object during its historical healthy baseline period is obtained and defined as the baseline tidal volume threshold. An airflow limitation duration threshold is also set, based on the minimum duration standard for clinically diagnosing sleep-disordered breathing events; for example, it is set to 10 seconds. The amplitudes in the real-time generated instantaneous tidal volume sequence are sequentially compared with the baseline tidal volume threshold. When the amplitudes of H consecutive sampling points in the instantaneous tidal volume sequence are all less than 30% of the baseline tidal volume threshold, the target object is determined to be in an airflow-limited state. The 30% percentage is set based on the medically defined critical standard for tidal volume decline in hypoventilation events. The number of H is determined by the product of the airflow limitation duration threshold and the sampling frequency of the analog-to-digital converter; for example, it is set to 500 consecutive sampling points. If the duration of the airflow limitation state is greater than the airflow limitation duration threshold, it is identified as a respiratory limitation event. The total number of respiratory limitation events within the total observation time is counted and divided by the total observation time to obtain the respiratory disturbance characteristics used to characterize the density of respiratory disturbances. Heart rate variability, atrial fibrillation load characteristics, and respiratory disturbance characteristics are collectively defined as cardiopulmonary physiological characteristics.

[0106] To deeply fuse spatial motion asymmetry features with temporal cardiopulmonary electrophysiological features, cross-modal temporal fusion is performed. This cross-modal temporal fusion refers to cross-dimensional correlation mining and temporal feature integration processing of spatial reflection features representing the spatial dimension and physiological feature signals representing the temporal dimension. Specifically, a feature integration sliding window is constructed. The duration of the feature integration sliding window is set based on the minimum observation time required to capture the evolution trend of physiological parameters in the prodromal period of stroke, for example, 5 minutes. The sliding step size of the feature integration sliding window is also set, which is equal to the total observation time of a single frame of radio frequency echo signal. During monitoring, the acquisition and processing of each frame of radio frequency echo signal is defined as a monitoring cycle. The asymmetry index calculated in each monitoring cycle and the obtained heart rate variability index are pushed into the buffer queue of the feature integration sliding window as the input values ​​at the current moment. When the total duration of monitoring periods stored in the cache queue reaches the length of the feature integration sliding window, all values ​​within the same feature integration sliding window at the current moment are extracted: Asymmetric indices belonging to multiple consecutive monitoring periods within the same feature integration sliding window are arranged in chronological order to obtain an asymmetric index sequence; heart rate variability indicators belonging to multiple consecutive monitoring periods within the same feature integration sliding window are arranged in chronological order to obtain a heart rate variability indicator sequence; to eliminate dimensional differences between heterogeneous physical quantities and achieve semantic alignment, standardization processing is performed on the asymmetric index sequence and the heart rate variability indicator sequence respectively: For the asymmetric index sequence, the asymmetric index order is calculated... The arithmetic mean and standard deviation of all values ​​in the asymmetric exponential sequence are calculated. Each value in the asymmetric exponential sequence is subtracted from the arithmetic mean and then divided by the standard deviation to obtain a standardized asymmetric exponential sequence with a mean of zero and a variance of one. For the heart rate variability index sequence, the arithmetic mean and standard deviation of all values ​​in the heart rate variability index sequence are calculated. Each value in the heart rate variability index sequence is subtracted from the arithmetic mean and then divided by the standard deviation to obtain a standardized heart rate variability index sequence with a mean of zero and a variance of one. The values ​​of the standardized asymmetric exponential sequence and the standardized heart rate variability index sequence at the same moment are extracted and multiplied. All products with the same feature are accumulated and averaged within a sliding window to obtain the covariance bias. The covariance bias is combined with cardiopulmonary physiological features to form a heterogeneous energy imbalance vector. This heterogeneous energy imbalance vector is a high-dimensional feature vector used to characterize the multi-system physiological dysfunction state of the target object in the prodromal period of stroke.Simultaneously, a diagnostic evaluation plane is established with the asymmetric index as the horizontal axis and the heart rate variability index as the vertical axis. The arithmetic mean of the asymmetric index sequence within the same feature integrated sliding window is extracted as the horizontal axis coordinate value, and the arithmetic mean of the heart rate variability index sequence within the same feature integrated sliding window is extracted as the vertical axis coordinate value. The coordinate point determined by the horizontal axis coordinate value and the vertical axis coordinate value is defined as the risk prediction point. The risk prediction point realizes the dimensionality upgrade from single vital sign monitoring to a cross-system collaborative analysis mode of heart, lung, and brain, and provides a high-dimensional feature carrier to solve the technical defects of high false negative rate of single-modality early warning.

[0107] In the aforementioned early warning judgment module, the evolution slope is calculated by solving for risk prediction points, and a graded risk early warning is executed based on the evolution slope to obtain a risk early warning set, including:

[0108] Specifically, the asymmetric exponential sequence accumulated by the target object during its historical health baseline period is retrieved, and the arithmetic mean of all values ​​in the asymmetric exponential sequence is calculated, which is defined as the first baseline mean. Simultaneously, the accumulated heart rate variability index sequence during the historical healthy baseline period is invoked, and the arithmetic mean of all values ​​in the heart rate variability index sequence is calculated, which is defined as the second baseline mean. Within the diagnostic evaluation plane, the coordinate point determined by the first baseline mean as the abscissa and the second baseline mean as the ordinate is defined as the personalized health anchor point D. That is, the coordinates of the personalized health anchor point D are D(…). , The personalized health anchor point physically represents the distribution center of the physiological characteristics of the target object in an absolutely healthy state.

[0109] Using a personalized health anchor as the geometric center, a dynamic risk target zone is constructed within the diagnostic evaluation plane. Specifically, for the asymmetric index sequence accumulated during the historical health baseline period, the standard deviation of the asymmetric index sequence is calculated and defined as the first standard deviation of fluctuation. For the cumulative heart rate variability index series during the historical healthy baseline period, the standard deviation of the heart rate variability index series is calculated and defined as the second standard deviation of fluctuation. 2; The purpose of introducing the first and second standard deviations of fluctuation is to quantify the spontaneous fluctuation intensity of the target object under healthy conditions due to individual physiological differences; a horizontal risk tolerance coefficient is set. With longitudinal risk tolerance factor The horizontal and vertical risk tolerance coefficients are set based on the confidence interval principle of the statistical normal distribution, used to establish a dynamic balance between early warning sensitivity and false alarm rate; for example, both the horizontal and vertical risk tolerance coefficients are set to 2.5; the dynamic risk target area consists of four vertices, namely, the upper vertex, right vertex, lower vertex, and left vertex connected end-to-end in sequence; wherein, the coordinates of the upper vertex are... The coordinates are set based on the consideration that the prodromal period of stroke is usually accompanied by an imbalance in cardiac autonomic regulation, leading to the evolution of heart rate variability indicators toward extreme values. By superimposing a second fluctuation standard deviation of a specific multiple on the second baseline mean, the aim is to capture abnormal rhythms that exceed the random fluctuation range of the target object's own physiological regulation. The ordinate of this vertex defines the highest discrete boundary allowed for the target object in the dimension of cardiac autonomic regulation. Once real-time data exceeds this point, it indicates a significant decrease in the ability of cardiac pacing to be regulated by the central nervous system; the coordinates of the lower vertex are... The coordinates are set to identify rhythmic over-steadiness caused by abnormally increased parasympathetic activity or sluggish autonomic responses. Due to significant differences in baseline heart rate variability among individuals, a second baseline mean is used, minus a tolerance term calculated based on its own standard deviation. This adaptively defines the critical value for maintaining the lower limit of physiological function of the target object, thus physically characterizing the lowest permissible steady-state boundary in the cardiac autonomic regulation dimension; the coordinates of the left vertex are... The coordinates are set based on the fact that unilateral motor neuron damage caused by stroke directly leads to slight contraction and relaxation delays in targeted muscles, such as the pectoralis major and intercostal muscles, which manifests as a unilateral decrease in kinetic energy on the kinetic energy distribution heatmap. By applying a negative bias based on the first baseline mean, the purpose of setting this vertex is to accurately capture the asymmetric disruption caused by the weakening of kinetic energy in the left monitoring area; this coordinate point physically defines the left-side risk critical boundary of the target object's transformation from a healthy to a pathological state in the spatial neuromotor damage dimension; the coordinates of the right vertex are... The coordinates are set based on monitoring the motor nerve control ability of the right side of the target object's torso. Due to the imperfect symmetry of human anatomy, the reference center often has an inherent offset. By superimposing the first fluctuation standard deviation, pseudo-asymmetry caused by radar installation position or human anatomy deviation is eliminated. The x-coordinate of this vertex represents the maximum containment limit of physiological symmetry breaking in the right monitoring area, used to identify heterogeneous energy imbalance caused by right-side motor function decline.

[0110] In real time, it is determined whether the risk prediction point is located inside or on the boundary of the dynamic risk target area. Specifically, the horizontal and vertical coordinates of the risk prediction point are extracted. The absolute value of the difference between the horizontal coordinate of the risk prediction point and the first benchmark mean is divided by the product of the horizontal risk tolerance coefficient and the first standard deviation of fluctuation to obtain the first deviation. The absolute value of the difference between the vertical coordinate of the risk prediction point and the second benchmark mean is divided by the product of the vertical risk tolerance coefficient and the second standard deviation of fluctuation to obtain the second deviation. If the sum of the first and second deviations is less than or equal to 1, the risk prediction point is determined to be inside or on the boundary of the dynamic risk target area. If the sum of the first and second deviations is greater than 1, the risk prediction point is determined to be outside the dynamic risk target area. When the risk prediction point is outside the dynamic risk target area, the evolution slope G is calculated. The evolution slope refers to the instantaneous rate of change of the risk prediction point from the personalized health anchor point in the diagnostic evaluation plane. It is used to quantify the instantaneous acceleration of physiological state deterioration, thereby distinguishing between gradual physiological fluctuations and pathological evolution in the prodromal period of sudden stroke. The formula for calculating the evolution slope is: ,in, The Euclidean distance increment is obtained by calculating the Euclidean distance between the risk prediction point and the personalized health anchor point in the current monitoring period and the previous monitoring period, and then subtracting the Euclidean distance of the previous monitoring period from the Euclidean distance of the current monitoring period. This represents the total observation duration of a single frame of radio frequency echo signal. A graded risk warning is implemented based on the evolution slope. Specifically, a first slope threshold and a second slope threshold are set. The first slope threshold is set based on the Euclidean distance variation between each adjacent monitoring cycle within the historical healthy baseline period, and the standard deviation of this variation is calculated and defined as the baseline drift standard deviation. Since physiological parameters in a healthy state exhibit spontaneous slow drift, the first slope threshold aims to define the initial rate boundary of abnormal evolution, using a preset multiple of the baseline drift standard deviation as the first slope threshold. For example, the preset multiple is set to 2, meaning the first slope threshold equals 2 times the baseline drift standard deviation. The second slope threshold is set in conjunction with the pathological evolution dynamics of the hyperacute phase of clinical stroke. When the rate at which physiological characteristics deviate from the healthy baseline exceeds the regulatory limit of the body's compensatory mechanisms, it indicates that vascular occlusion or nerve damage is progressing rapidly. The second slope threshold aims to establish a physical boundary distinguishing between "slow degeneration" and "sudden onset." For example, the value is set to 5. When the evolution slope is less than the first slope threshold, the deviation rate of the risk prediction point is determined to be within the physiological compensation range of the target object, defined as low risk. At this time, high-frequency real-time monitoring is maintained, and silent data recording is performed without triggering any risk warning instructions. When the evolution slope is greater than or equal to the first slope threshold and less than the second slope threshold, the target object is determined to have non-specific physiological function decline or chronic pathological characteristic fluctuations, defined as medium risk. An abnormal physiological trend prompt is sent to the associated monitoring terminal. When the evolution slope is greater than or equal to the second slope threshold, the target object is determined to be in the high-risk prodromal period or early stage of stroke, defined as extremely high risk, and a loud alarm is issued to the associated monitoring terminal. The low risk, medium risk, and extremely high risk are combined into a risk warning set.

[0111] The aforementioned early warning judgment module addresses the technical challenges of high false negative rates, inability to effectively distinguish between spontaneous physiological fluctuations and sudden pathological evolution in stroke early warning under a single vital sign monitoring model, and lack of dynamic early warning responses for different stages of disease progression. It achieves fully automated, graded stroke risk early warning under a cross-system collaborative analysis model involving the heart, lungs, and brain. Specifically, cross-modal temporal fusion eliminates dimensional differences in heterogeneous data and achieves semantic alignment, extracting the intrinsic correlations of multidimensional physiological features; the diagnostic evaluation plane constructs a high-dimensional risk carrier; the dynamic risk target area adaptively defines individualized physiological fluctuation thresholds using historical health baselines as anchors, resolving the contradiction between sensitivity and false alarm rates caused by a one-size-fits-all approach to early warning; and the evolution slope, by quantifying the instantaneous acceleration of physiological state deterioration, achieves precise capture of sudden pathological evolution and scientific triggering of graded early warning commands.

[0112] Example 2

[0113] This embodiment, based on Embodiment 1, provides a stroke risk diagnosis and early warning method based on millimeter-wave radar fusion, such as... Figure 5 As shown, it includes:

[0114] Step S10: Obtain the radio frequency echo signal characterizing the microscopic motion characteristics of the target object, perform complex feature reconstruction on the radio frequency echo signal to obtain the angular energy spectrum characterizing the spatial reflection intensity, perform peak detection and coordinate mapping on the angular energy spectrum to determine the spatial reflection anchor point, perform phase unwrapping and physical displacement mapping based on the spatial reflection anchor point to obtain the physiological displacement sequence, perform multi-scale digital filtering on the physiological displacement sequence to extract the physiological feature waveform.

[0115] Step S20: Construct a virtual anatomical projection surface based on spatial reflection anchor points, and call physiological feature waveforms within the virtual anatomical projection surface to perform physiological kinetic energy focusing mapping in the grid dimension to obtain a kinetic energy distribution heat map. Divide the bilateral monitoring areas within the kinetic energy distribution heat map, and perform two-dimensional spatial discrete summation calculation based on the bilateral monitoring areas to obtain the asymmetry index.

[0116] Step S30: Perform cross-modal temporal fusion based on physiological characteristic waveforms to obtain heterogeneous energy imbalance vectors. Construct a diagnostic evaluation plane based on asymmetric index to map the heterogeneous energy imbalance vectors to obtain risk prediction points. Calculate the evolution slope through the risk prediction points and perform hierarchical risk warning based on the evolution slope to obtain a risk warning set.

[0117] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0118] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A stroke risk diagnosis and early warning system based on millimeter-wave radar fusion, characterized in that, The system includes: Micro-feature decoupling module: used to acquire radio frequency echo signals characterizing the micro-motion characteristics of the target object, perform complex feature reconstruction on the radio frequency echo signals to obtain the angular energy spectrum characterizing the spatial reflection intensity, perform peak detection and coordinate mapping on the angular energy spectrum to determine the spatial reflection anchor point, perform phase unwrapping and physical displacement mapping based on the spatial reflection anchor point to obtain the physiological displacement sequence, and perform multi-scale digital filtering on the physiological displacement sequence to extract the physiological feature waveform; Spatial Heterogeneous Module: Used to construct a virtual anatomical projection surface based on spatial reflection anchor points, and to call physiological feature waveforms within the virtual anatomical projection surface to perform physiological kinetic energy focusing mapping in the grid dimension to obtain a kinetic energy distribution heat map. Bilateral monitoring areas are divided within the kinetic energy distribution heat map, and two-dimensional spatial discrete summation calculation is performed based on the bilateral monitoring areas to obtain the asymmetry index. Early warning judgment module: It is used to perform cross-modal time series fusion based on physiological feature waveforms to obtain heterogeneous energy imbalance vectors, and to construct a diagnostic evaluation plane based on asymmetric index to map the heterogeneous energy imbalance vectors to obtain risk prediction points. The evolution slope is calculated through the risk prediction points, and graded risk early warning is performed based on the evolution slope to obtain a risk early warning set. Bed frame; And sensor modules: The sensor modules include at least millimeter-wave radar sensors and weighing sensor arrays.

2. The stroke risk diagnosis and early warning system based on millimeter-wave radar fusion according to claim 1, characterized in that, The complex feature reconstruction includes: Multiple frequency-modulated continuous waves are continuously radiated into the monitored area containing the target object by a millimeter-wave radar sensor at a preset pulse period. The reflected signal is captured after being phase-modulated by the physiological micro-fluctuations of the thoracic cavity. The reflected signal is subjected to analog multiplication and low-pass filtering to extract the intermediate frequency signal and then quantized and sampled to obtain the radio frequency echo signal. Perform a fast Fourier transform on the radio frequency echo signal to generate a range spectrum, and define the ranging unit with the largest energy amplitude in the range spectrum as the target ranging unit; Based on the position of the radio frequency echo signal at the target ranging unit, a complex voltage value containing real and imaginary information is extracted, and the complex voltage values ​​are combined into a complex sampling sequence in time sequence; Calculate and sum the instantaneous power of each complex voltage value to obtain the total energy value of a single frame. Divide the total energy value of a single frame by the number of frequency-modulated continuous waves to convert it into average energy power. The average background energy of the monitored area under no-load conditions is obtained, and its reciprocal is taken as the environmental correction coefficient. The effective signal energy is obtained by multiplying the average energy power by the environmental correction coefficient.

3. The stroke risk diagnosis and early warning system based on millimeter-wave radar fusion according to claim 2, characterized in that, There are two ways to install the millimeter-wave radar sensor: the first is to fix the millimeter-wave radar sensor above the headboard, and the second is to embed the millimeter-wave radar sensor in the mattress of the bed.

4. The stroke risk diagnosis and early warning system based on millimeter-wave radar fusion according to claim 3, characterized in that, The method for obtaining the spatial reflection anchor point includes: The complex voltage values ​​of each independent channel of the millimeter-wave radar sensor at the target ranging unit position are obtained, and the complex voltage values ​​are sorted to construct a spatial measurement vector. Multiple candidate angles are preset, the theoretical phase offset of each independent channel at each candidate angle is calculated and the angle steering vector is constructed, and the reference vector set is composed of each angle steering vector. The spatial measurement vector and each angular steering vector are respectively subjected to inner product operation to obtain the spatial reflection intensity, and the corresponding candidate angles are combined to obtain the angular energy spectrum; The local extreme point with the largest value in the search angle energy spectrum is taken as the energy peak amplitude. The energy peak amplitude is mapped to obtain the target azimuth and target elevation angles. Based on the target ranging unit, the radial distance value is determined, and the lateral, longitudinal, and height components are calculated in the constructed radar local coordinate system in combination with the target azimuth and target elevation angles. A millimeter-wave radar sensor is used to perform rotation operations on the lateral, longitudinal, and height components to generate the rotated lateral, longitudinal, and absolute vertical height components. The three-dimensional physical coordinate points formed by these three components are defined as spatial reflection anchor points.

5. The stroke risk diagnosis and early warning system based on millimeter-wave radar fusion according to claim 4, characterized in that, The physiological characteristic waveforms include: The original phase value of each complex voltage value in the complex sampling sequence is calculated using the arctangent function, and the original phase sequence is obtained by arranging them in the time sequence of the pulse period. Calculate the phase difference between adjacent pulse periods in the original phase sequence, and perform numerical compensation operation based on the relationship between the phase difference and the preset transition threshold to eliminate phase entanglement and construct a continuous phase sequence; The phase values ​​in the continuous phase sequence are combined with the inherent wavelength of the radio frequency echo signal to convert them into instantaneous displacement characteristic values, and then combined in time sequence to obtain the physiological displacement sequence. Using a zero-phase shift digital bandpass filter, frequency-selective filtering is performed in parallel on the physiological displacement sequence within the preset respiratory frequency range and heart rate range to extract respiratory and heart rate characteristic waveforms. By combining the respiratory characteristic waveform with the heartbeat characteristic waveform, the physiological characteristic waveform is obtained.

6. The stroke risk diagnosis and early warning system based on millimeter-wave radar fusion according to claim 5, characterized in that, The kinetic energy distribution thermogram includes: Using the spatial reflection anchor point as the geometric reference center, a two-dimensional physical section is constructed and a rectangular region is extracted. The rectangular region is divided into multiple two-dimensional discrete grids with three-dimensional geometric coordinates. All two-dimensional discrete grids constitute a virtual anatomical projection surface. Calculate the physical spatial straight-line distance from the two-dimensional discrete grid to each independent channel, and determine the theoretical phase delay by combining the inherent wavelength of the radio frequency echo signal; By performing a reverse phase rotation on the complex voltage value using the theoretical phase delay, phase compensation is achieved, and the data is coherently accumulated to obtain a local time-domain reflection signal. Within the preset respiratory rate range and heart rate range, frequency-selective filtering is performed on the local time-domain reflectance signal to obtain the local physiological micro-motion component; The kinetic energy density points are calculated using local physiological micro-motion components. The kinetic energy density points corresponding to all two-dimensional discrete grids are combined according to their spatial topological positions to generate a kinetic energy distribution heat map.

7. The stroke risk diagnosis and early warning system based on millimeter-wave radar fusion according to claim 6, characterized in that, The asymmetric index includes: With the geometric reference center as the origin of the coordinate system, an anatomical sagittal projection line is drawn along the longitudinal direction in the radar local coordinate system, dividing the virtual anatomical projection plane into the left monitoring area and the right monitoring area. Extract the kinetic energy density points corresponding to each two-dimensional discrete grid in the left monitoring area and perform a two-dimensional spatial discrete summation operation to obtain the total value of micro-motion energy on the left. Extract the kinetic energy density points corresponding to each two-dimensional discrete grid in the right monitoring area and perform a two-dimensional spatial discrete summation operation to obtain the total value of micro-motion energy on the right. Acquire multiple frames of radio frequency echo signals of the target object during the historical health baseline period, calculate the ratio of the total left micromotion energy value to the total right micromotion energy value corresponding to each frame of radio frequency echo signal, and take the arithmetic mean of all ratios as the calibration factor. The asymmetry index is obtained by combining the calibration factor with the total energy values ​​of the left and right micromotions.

8. The stroke risk diagnosis and early warning system based on millimeter-wave radar fusion according to claim 7, characterized in that, The method for obtaining the risk prediction points includes: The effective peaks of the heartbeat characteristic waveform are obtained, the time difference between adjacent effective peaks is calculated, and the heartbeat interval sequence is obtained by arranging them in time sequence. The heart rate interval sequence is resampled and time-frequency converted to obtain the power spectral density function. Within the preset low-frequency band and high-frequency band, the power spectral density function is subjected to definite integral operation to extract the low-frequency power and high-frequency power respectively. The low-frequency power is divided by the high-frequency power to obtain the energy ratio. The heart rate variability index is obtained by combining the energy ratio and the short-term variability. Statistical calculations were performed on the heartbeat interval sequence to obtain the atrial fibrillation load characteristics; Statistical calculations were performed on the respiratory characteristic waveforms to obtain the cardiopulmonary physiological characteristics. Set up a feature integration sliding window, push the asymmetric exponent and heart rate variability index into the feature integration sliding window, and obtain the asymmetric exponent sequence and heart rate variability index sequence respectively; The arithmetic mean of the asymmetric exponential sequence and the heart rate variability index sequence within the feature integration sliding window is calculated separately. The arithmetic mean of the asymmetric exponential sequence and the heart rate variability index sequence is then mapped onto the diagnostic evaluation plane to obtain the risk prediction point.

9. The stroke risk diagnosis and early warning system based on millimeter-wave radar fusion according to claim 8, characterized in that, The risk warning set includes: Obtain and calculate the standard deviation and arithmetic mean of the asymmetric index sequence and the heart rate variability index sequence during the historical health baseline period, and establish the arithmetic mean of the asymmetric index sequence and the heart rate variability index sequence as a personalized health anchor point in the diagnostic evaluation plane; By combining the standard deviation of the asymmetric exponential sequence, the standard deviation of the heart rate variability index sequence, and the preset risk tolerance coefficient, a dynamic risk target zone with a personalized health anchor as the geometric center is constructed. When a risk prediction point moves away from the dynamic risk target area, the change in physical distance between the risk prediction point and the personalized health anchor point is calculated and defined as the Euclidean distance increment. Divide the Euclidean distance increment by the preset total observation duration to obtain the evolution slope; The evolution slope is compared with a preset first slope threshold and a second slope threshold to determine the risk level and trigger a corresponding warning. The risk levels are then combined to obtain a risk warning set.

10. A stroke risk diagnosis and early warning method based on millimeter-wave radar fusion, applied to the stroke risk diagnosis and early warning system based on millimeter-wave radar fusion as described in any one of claims 1-9, characterized in that, The method includes: The radio frequency echo signal characterizing the microscopic motion characteristics of the target object is acquired. Complex feature reconstruction is performed on the radio frequency echo signal to obtain the angular energy spectrum characterizing the spatial reflection intensity. Peak detection and coordinate mapping are performed on the angular energy spectrum to determine the spatial reflection anchor point. Phase unwrapping and physical displacement mapping are performed based on the spatial reflection anchor point to obtain the physiological displacement sequence. Multi-scale digital filtering is performed on the physiological displacement sequence to extract the physiological feature waveform. A virtual anatomical projection surface is constructed based on spatial reflection anchor points. Physiological feature waveforms are called within the virtual anatomical projection surface to perform physiological kinetic energy focusing mapping in the grid dimension, resulting in a kinetic energy distribution heatmap. Bilateral monitoring areas are divided within the kinetic energy distribution heatmap. Two-dimensional spatial discrete summation calculation is performed based on the bilateral monitoring areas to obtain the asymmetry index. Cross-modal temporal fusion is performed based on physiological characteristic waveforms to obtain heterogeneous energy imbalance vectors. A diagnostic evaluation plane is constructed based on asymmetric index to map the heterogeneous energy imbalance vectors to obtain risk prediction points. Evolution slopes are calculated through risk prediction points, and graded risk warnings are performed based on the evolution slopes to obtain a risk warning set.

Images