Lightweight millimeter wave vital sign measurement method, system, equipment and medium

Abstract

The invention provides a lightweight millimeter wave vital sign measurement method, system, device and medium, and relates to the technical field of millimeter wave radar application, the method is characterized in that a human body reflection energy factor is generated by fusing an echo energy feature and a micro-motion energy feature, and the human body reflection energy factor is input into a lightweight deep learning network; the high-precision combined judgment of the personnel existence state and the human body distance is realized; after the existence of the personnel is confirmed, performing multi-stage filtering and unwrapping processing by combining the phase signal of the corresponding distance gate to obtain a smooth and continuous phase signal, so as to effectively support the stable extraction of subsequent vital sign parameters; furthermore, a self-adaptive harmonic spectrum is constructed based on the instantaneous breathing frequency and the human body distance, interference caused by breathing higher harmonics is accurately removed from heartbeat components, the accuracy of heart rate estimation is remarkably improved, meanwhile, invalid calculation in an unmanned state is avoided, and the measurement precision and the system efficiency are both considered.

Description

Technical Field

[0001] This invention relates to the field of millimeter-wave radar application technology, and in particular to a lightweight millimeter-wave vital sign measurement method, system, device and medium. Background Technology

[0002] With the development of millimeter-wave radar technology, it has demonstrated unique advantages in the field of vital sign monitoring, such as non-contact measurement and immunity to illumination and obstruction. However, existing technologies still face many challenges in practical applications: clutter signals in complex environments are diverse and uncertain, and traditional methods often rely on single signal features such as echo energy to determine the presence of a person, making it difficult to comprehensively reflect the differences between the human body's reflected signals and background clutter, resulting in low accuracy. Simultaneously, manually designed feature methods are difficult to adapt to complex environmental changes and cannot effectively capture the subtle differences between human body signals and clutter. Furthermore, during the extraction of vital sign parameters, high-order harmonics of respiratory signals are easily superimposed on heartbeat signals, and traditional filtering methods are insufficient to effectively eliminate such interference, leading to significant errors in heart rate measurement. Summary of the Invention

[0003] This invention aims to solve the technical problems existing in the above-mentioned related technologies, and proposes a lightweight millimeter-wave vital sign measurement method, system, device and medium. It can achieve high-precision identification of the presence status of personnel and distance estimation based on deep learning, and adaptively extract breathing and heartbeat parameters by combining prior knowledge of the presence of personnel, effectively suppressing clutter interference and respiratory harmonic aliasing, improving measurement accuracy and reducing computational resource consumption.

[0004] The solution to the technical problem of this invention is: This invention provides a lightweight millimeter-wave vital sign measurement method, comprising the following steps: The human body echo signal acquired by millimeter-wave radar is preprocessed to calculate the echo energy characteristics and micro-motion energy characteristics of each range gate; Based on the echo energy characteristics and the micro-motion energy characteristics, a human body reflection energy factor for each distance gate is generated by weighted fusion. The human body reflective energy factor is input into a pre-trained lightweight deep learning network, and the lightweight deep learning network outputs the human presence status identifier and the corresponding human body distance value. When the presence status of the person is indicated as present, the phase signal of the corresponding distance gate in the human body echo signal is processed based on the human body distance value to obtain a smooth and continuous phase signal. Based on the smooth and continuous phase signal, the respiratory component and the heartbeat component are extracted; The instantaneous respiratory rate is estimated based on the respiratory components, and an adaptive harmonic spectrum is constructed based on the instantaneous respiratory rate and the distance value to the human body. The adaptive harmonic spectrum is subtracted from the heartbeat component to eliminate the interference of higher harmonics of respiration, and the instantaneous heart rate is determined based on the heartbeat component after the interference is eliminated.

[0005] Furthermore, the preprocessing of the human body echo signal acquired by the millimeter-wave radar to calculate the echo energy characteristics and micro-motion energy characteristics of each range gate includes the following steps: Perform a Fast Fourier Transform on the echo signal of each distance gate in the slow time dimension to obtain a frequency domain complex echo sequence; Based on the frequency domain complex echo sequence, the mean echo energy and variance of the echo energy of the distance gate are calculated as the echo energy characteristics; The phase signal is extracted from the frequency domain complex echo sequence, and the phase signal is subjected to a fast Fourier transform. The maximum spectral peak value in the resulting spectrum is taken as the micro-motion energy characteristic of the range gate.

[0006] Furthermore, the lightweight deep learning network includes an input layer, two convolutional layers, two pooling layers, a fully connected layer, and an output layer connected in sequence. The lightweight deep learning network is obtained through the following training process: Collect millimeter-wave radar echo signal samples containing and without human targets in various complex environments; For each of the aforementioned samples, the human body reflection energy factor of each distance gate is calculated to form an input feature vector; Based on the actual scene of each sample, the corresponding personnel presence status identifier and the actual human distance value are labeled to form the output label; A training set is constructed based on the input feature vector and the output label, and the lightweight deep learning network is trained under supervision until convergence.

[0007] Further, the process of processing the phase signal corresponding to the distance gate in the human body echo signal to obtain a smooth and continuous phase signal includes the following steps: The phase signal is subjected to median filtering to remove instantaneous spikes caused by sudden interference; Based on the phase difference between adjacent radar frames, the filtered phase signal is subjected to phase dewinding processing to eliminate the jump caused by phase periodicity. The unwound phase signal is input into an extended Kalman filter, and the phase signal is dynamically smoothed by establishing state equations and observation equations to suppress low-frequency drift and random noise. Wavelet transform is performed on the phase signal after extended Kalman filtering, and the wavelet coefficients are processed using the soft thresholding method to suppress residual high-frequency noise components, thereby obtaining a smooth and continuous phase signal.

[0008] Further, the step of estimating the instantaneous respiratory rate based on the respiratory components and constructing an adaptive harmonic spectrum based on the instantaneous respiratory rate and the human body distance value includes: Autocorrelation calculation is performed on the respiratory components, and fast Fourier transform is performed on the autocorrelation results to obtain the respiratory power spectrum; The frequency corresponding to the maximum spectral peak is extracted from the respiratory power spectrum as the instantaneous respiratory rate; Based on the instantaneous respiratory rate, determine the position of its higher harmonic frequencies, and set the harmonic spectrum bandwidth with each higher harmonic as the center. Based on the human body distance value, the harmonic spectral power at each higher harmonic is weighted and adjusted, with a higher weight assigned to the harmonic spectral power for closer distances, thus constructing the adaptive harmonic spectrum.

[0009] Furthermore, when the personnel presence status flag is 0, the harmonic spectrum construction and subsequent heartbeat parameter extraction are skipped; when the personnel presence status flag is 1, a complete adaptive harmonic spectrum is generated to suppress respiratory harmonic interference in the heartbeat component.

[0010] Further, the extraction of the respiratory and cardiac components based on the smooth and continuous phase signal includes: Only when the presence status of the person is indicated as present, a bandpass filter of 0.1 Hz to 0.5 Hz is applied to the smooth and continuous phase signal to obtain the respiratory component, and a bandpass filter of 0.8 Hz to 2.0 Hz is applied to obtain the heartbeat component.

[0011] On the other hand, this application provides a lightweight millimeter-wave vital sign measurement system, comprising: Millimeter-wave radar is used to collect human body echo signals; Memory, used to store computer programs; The processor is configured to support lightweight edge inference and is able to execute the computer program based on the human echo signal without cloud computing support to implement the aforementioned lightweight millimeter-wave vital sign measurement method.

[0012] On the other hand, this application provides an electronic device, including the aforementioned lightweight millimeter-wave vital signs measurement system.

[0013] On the other hand, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned lightweight millimeter-wave vital sign measurement method.

[0014] The beneficial effects of this invention are as follows: This application provides a lightweight millimeter-wave vital sign measurement method. This method generates a human body reflection energy factor by fusing echo energy characteristics and micro-motion energy characteristics, and inputs it into a lightweight deep learning network to achieve high-precision joint discrimination of the presence status of personnel and the distance to the human body. After confirming the presence of personnel, multi-level filtering and dewinding processing are performed on the phase signal of the corresponding distance gate to obtain a smooth and continuous phase signal, thereby effectively supporting the stable extraction of subsequent vital sign parameters. Furthermore, an adaptive harmonic spectrum is constructed based on instantaneous respiratory rate and human body distance, and interference caused by higher harmonics of respiration is accurately removed from the heartbeat component, significantly improving the accuracy of heart rate estimation, while avoiding invalid calculations in the absence of personnel, thus balancing measurement accuracy and system efficiency. This application also provides corresponding systems, devices, and media. The beneficial effects of the systems, devices, and media are the same as those of the above-described method, and will not be elaborated here.

[0015] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the description, claims and drawings. Attached Figure Description

[0016] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the technical solutions of the present invention, and do not constitute a limitation on the technical solutions of the present invention.

[0017] Figure 1 This is a flowchart of the lightweight millimeter-wave vital signs measurement method provided in this application; Figure 2 This is a structural diagram of the lightweight millimeter-wave vital signs measurement system provided in this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] The present application will be further described below with reference to the accompanying drawings and specific embodiments. The described embodiments should not be considered as limitations on the present application, and all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of the present application.

[0020] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0022] With the rapid development of intelligent sensing technology, non-contact vital sign monitoring has shown broad application prospects in fields such as healthcare, smart homes, and public safety. Millimeter-wave radar, due to its advantages such as strong penetration, immunity to lighting conditions, all-weather operation, and sensitivity to minute movements, is gradually becoming an important technical means for measuring vital sign parameters. Millimeter-wave radar can acquire key physiological signals such as respiration and heartbeat from a distance without being obstructed, enabling continuous monitoring without interfering with the normal activities of the monitored individual, thus possessing significant practical value and social significance.

[0023] In existing technologies, millimeter-wave radar vital sign monitoring primarily relies on time-frequency analysis or phase demodulation methods of echo signals. Typical approaches include performing Fast Fourier Transform (FFT) on slow-time-dimensional signals to extract respiratory and heart rate components, or estimating physiological parameters by inverting chest micro-movements through phase changes. However, these traditional methods face multiple challenges in complex real-world environments. First, in determining the presence of personnel, most systems rely solely on a single feature, such as echo energy intensity. This strategy is highly susceptible to failure in scenarios with strong background clutter or multipath interference, as clutter may have energy levels similar to those reflected by the human body, leading to misjudgments or missed detections. Furthermore, manually designed features struggle to fully characterize the essential differences between human micro-movement signals and environmental noise, lacking robustness and generalization ability.

[0024] Secondly, although deep learning has made significant progress in image recognition and speech processing in recent years, its application in millimeter-wave radar for vital sign monitoring is still in its early stages. Existing solutions rarely effectively integrate lightweight neural networks into embedded radar systems for human presence detection, and they haven't fully explored the ability of deep models to automatically learn discriminative features in complex clutter environments. This makes it difficult for the system to reliably distinguish between real human targets and false echoes when faced with dynamic interference sources such as furniture, pets, and fans, severely impacting the reliability of subsequent parameter extraction.

[0025] Third, in the extraction of vital signs parameters, respiratory signals typically contain significant high-order harmonic components, and their frequency range may overlap with that of heartbeat signals, especially in adults with high respiratory rates or low heart rates. Traditional bandpass filtering methods cannot effectively separate these two types of signals, leading to severe interference with heartbeat estimation and a significant increase in measurement error. Simultaneously, the raw phase signal is susceptible to system noise, phase entanglement, and environmental disturbances, exhibiting discontinuities or abrupt jumps. If used directly for parameter extraction without fine processing, the accuracy of the results will be further reduced.

[0026] More critically, most current technical solutions neglect the importance of fundamental prior information: "whether people are present." Regardless of whether anyone is in the environment, the system often continues to execute the entire signal processing flow, wasting computational resources and energy, and may even output false vital sign data when no one is present, causing false alarms. Even when people are confirmed to be present, there are few methods to incorporate the presence and distance information of people into harmonic modeling or filtering strategies, resulting in a lack of specificity and adaptability in the parameter extraction process.

[0027] To address the aforementioned issues, this application proposes a lightweight millimeter-wave vital sign measurement method. First, the system calculates and weights the echo energy characteristics and micro-motion energy characteristics of each distance gate to generate a human body reflection energy factor. This factor serves as the input to a lightweight deep learning network, enabling high-precision joint discrimination of the presence of a person and their distance. After confirming the presence of a person, the system performs median filtering, phase unwrapping, extended Kalman filtering, and wavelet soft thresholding on the phase signal of the corresponding distance gate to obtain a smooth and continuous phase signal. Subsequently, an adaptive harmonic spectrum is constructed by combining the instantaneous respiratory rate and the distance to the person, accurately removing interference caused by higher harmonics of respiration in the heartbeat component, and extracting a high-precision instantaneous heart rate. The entire process uses the presence of a person as a priori condition, activating subsequent parameter extraction only when a person is present, effectively avoiding invalid calculations and significantly improving system efficiency, measurement accuracy, and environmental adaptability. Furthermore, the lightweight network structure facilitates deployment on resource-constrained embedded platforms, balancing practicality and scalability.

[0028] First, the lightweight millimeter-wave vital signs measurement method provided in this application will be described in detail below with reference to the accompanying drawings.

[0029] Reference Figure 1 The implementation process of the lightweight millimeter-wave vital sign measurement method provided in this application embodiment includes, but is not limited to, the following steps.

[0030] Step S110: Preprocess the human body echo signal acquired by the millimeter-wave radar, and calculate the echo energy characteristics and micro-motion energy characteristics of each range gate.

[0031] In step S110, the raw human echo signal acquired by the millimeter-wave radar undergoes preliminary processing to extract key features reflecting the target's characteristics. For each range gate, echo energy features and micro-motion energy features are calculated. The echo energy features are obtained by performing a Fast Fourier Transform on the signal in the slow time dimension and calculating its mean and variance, used to characterize the signal intensity stability within that range cell. The micro-motion energy features are obtained by performing a Fast Fourier Transform on the phase signal and extracting the maximum spectral peak value in the spectrum, used to capture the energy response corresponding to minute movements caused by human respiration or heartbeat. These two features together constitute the basic input for subsequent discrimination and analysis.

[0032] Step S120: Based on the echo energy characteristics and micro-motion energy characteristics, the human body reflection energy factor for each distance gate is generated by weighted fusion.

[0033] In step S120, the obtained echo energy features and micro-motion energy features are weighted and fused to generate a comprehensive index, namely the human body reflection energy factor. This factor is a linear combination of the normalized two features by setting weight coefficients, so that the final result can reflect both the overall energy level of the target echo and the dynamic characteristics generated by human body micro-motion. This fusion strategy effectively enhances the representation ability of human targets, while reducing the influence of static clutter or non-living object interference, providing more discriminative input features for subsequent deep learning networks.

[0034] Step S130: Input the human body reflective energy factor into a pre-trained lightweight deep learning network, and output the human presence status identifier and the corresponding human body distance value from the lightweight deep learning network.

[0035] In step S130, a pre-trained lightweight deep learning network is used to intelligently identify human body reflectance energy factors, thereby achieving accurate judgment of the presence status of personnel and estimation of the corresponding human body distance. This network takes the human body reflectance energy factors as input and outputs two key pieces of information after multiple layers of convolution, pooling, and fully connected operations: a personnel presence status identifier and a human body distance value. When the identifier indicates presence, it means that a valid human target has been detected in the current scene, and the output distance value indicates the specific distance gate position of the target. This process fully utilizes the ability of deep learning to automatically learn discrimination rules in complex feature spaces, significantly improving the accuracy and robustness of presence detection while suppressing background clutter.

[0036] Step S140: When the presence status of a person is marked as present, the phase signal of the corresponding distance gate in the human body echo signal is processed based on the human body distance value to obtain a smooth and continuous phase signal.

[0037] In step S140, after confirming the presence of a person, a series of refined processing steps are performed on the original phase signal of the distance gate where the person is located to obtain a high-quality continuous phase sequence. Since the original phase signal typically contains transient noise spikes, phase entanglement, and high-frequency interference, this application sequentially employs median filtering to eliminate outliers, utilizes inter-frame difference to achieve phase de-entanglement, and dynamically smooths the signal using extended Kalman filtering. Finally, a wavelet soft thresholding method is used to further suppress residual high-frequency components. After these processing steps, the phase signal becomes smooth and continuous, providing a reliable data foundation for subsequent high-precision extraction of vital sign parameters.

[0038] Step S150: Extract the respiratory component and the heartbeat component based on the smooth and continuous phase signal.

[0039] In step S150, physiological components corresponding to respiration and heartbeat are separated from the smooth continuous phase signal. Signal decomposition is achieved by applying a bandpass filter with a specific frequency band. Designed based on the frequency range of normal human physiological activities, this method can effectively filter out interference from irrelevant frequency bands while retaining the main physiological information, thus laying the foundation for independent estimation of their respective parameters.

[0040] Step S160: Estimate the instantaneous respiratory rate based on the respiratory components, and construct an adaptive harmonic spectrum based on the instantaneous respiratory rate and the distance value to the human body.

[0041] In step S160, the instantaneous respiratory frequency is accurately estimated based on the extracted respiratory components, and an adaptive harmonic spectrum model is constructed by combining this frequency with the known human body distance value. First, the respiratory frequency at the current moment is determined by performing autocorrelation analysis on the respiratory components and taking the maximum peak value of their power spectrum. Then, the positions of its higher harmonics are derived based on this frequency. On this basis, the power of the harmonic spectrum is weighted and adjusted using the human body distance value to form an adaptive harmonic spectrum that reflects both the harmonic structure and the propagation attenuation effect. This harmonic spectrum can more realistically simulate the harmonic interference pattern caused by breathing in actual signals.

[0042] Step S170: Subtract the adaptive harmonic spectrum from the heartbeat component to eliminate the interference of respiratory higher harmonics, and determine the instantaneous heart rate based on the heartbeat component after the interference is eliminated.

[0043] In step S170, interference components caused by higher harmonics of respiration are actively removed from the heartbeat component, thereby improving the accuracy of heart rate estimation. Specifically, the constructed adaptive harmonic spectrum is subtracted from the original heartbeat component to obtain a pure heartbeat signal after harmonic contamination. Then, a short-time Fourier transform is performed on the purified heartbeat signal, and the trajectory of frequency peak changes is tracked in the resulting time-frequency graph to finally determine the instantaneous heart rate at each moment. This targeted harmonic suppression mechanism effectively solves the measurement deviation problem caused by the aliasing of heartbeat and respiratory harmonics in traditional methods, significantly improving the reliability of heart rate parameters.

[0044] In some embodiments of this application, the human body echo signal acquired by millimeter-wave radar is preprocessed to calculate the echo energy characteristics and micro-motion energy characteristics of each range gate, including the following steps.

[0045] Step S210: Perform a Fast Fourier Transform on the echo signal of each distance gate in the slow time dimension to obtain a frequency domain complex echo sequence.

[0046] In step S210, a Fast Fourier Transform (FFT) is performed on the echo signal of each range gate acquired by the millimeter-wave radar in the slow time dimension, thereby converting the time-domain signal into a frequency-domain complex echo sequence. The slow time dimension corresponds to the changes between consecutive frames, reflecting the micro-motion characteristics of the target over time. Through this transformation, the minute periodic movements of the human body caused by breathing or heartbeat can be converted into specific frequency components in the frequency domain, providing a basic data representation for the subsequent extraction of energy features and micro-motion features.

[0047] Step S220: Based on the frequency domain complex echo sequence, calculate the mean echo energy and variance of the echo energy of the distance gate as echo energy characteristics.

[0048] In step S220, based on the aforementioned frequency domain complex echo sequence, the mean echo energy and variance of the echo energy of the signal within the range gate are calculated, and these two statistics are used as echo energy characteristics. The mean echo energy reflects the average reflection intensity of the range gate during the observation period, while the variance of the echo energy characterizes the degree of signal energy fluctuation over time. These two indicators together describe the stability and activity of the target echo at the energy level, helping to distinguish between static clutter and human body reflection signals with vital signs.

[0049] Step S230: Extract the phase signal from the frequency domain complex echo sequence, perform a fast Fourier transform on the phase signal, and take the maximum spectral peak in the obtained spectrum as the micro-motion energy feature of the distance gate.

[0050] In step S230, phase information is extracted from the frequency domain complex echo sequence, and the phase signal is subjected to a Fast Fourier Transform again. The maximum spectral peak in the transformed spectrum is then used as the micro-motion energy feature of the range gate. Since human respiration and heartbeat cause minute displacements in the chest cavity, resulting in periodic modulation on the radar echo phase, this maximum spectral peak effectively characterizes the energy concentration of such micro-motions. Therefore, this feature is specifically designed to capture dynamic information related to life activities, enhancing the ability to identify real human targets.

[0051] In some embodiments of this application, signal preprocessing and feature extraction are performed on the human body echo signals acquired by millimeter-wave radar. First, range-domain Fast Fourier Transform (FFT) processing is performed on the echo signals within each range gate to obtain spectral information for different range cells. The calculation formula for range-domain FFT is as follows: ; in, The imaginary unit, Indicates the first The first Chirp signal Intermediate frequency (IF) echo signal at each sampling point, This represents the number of points in the FFT, i.e., the number of points sampled within each chirp. This is a frequency index, corresponding to a specific range cell. This transformation can convert the time-domain signal to the frequency domain to identify the target's reflected energy at different distances. It is a complex radar echo at a specific distance and time (Chirp number) after range-dimensional FFT, used to characterize whether a human body and its micro-motion characteristics exist at that spatiotemporal location.

[0052] The mean echo energy of each distance gate is then calculated using the following formula: ; in The number of Chirp signals contained within a time window, representing the number of pulses in a continuous observation period, is the average value. It reflects the average intensity of the echo energy on that distance cell.

[0053] Further calculation of the echo energy variance is given by the following formula: ; in, The range gate is used to characterize the degree of energy fluctuation, which helps to distinguish between static clutter and dynamic targets.

[0054] Next, the phase signal within each distance gate is processed by FFT, and the formula is as follows: ; in, It is a complex signal phase angle, For FFT points, For frequency index, This indicates a gate at a specific distance. The phase signal at a given point is subjected to a Fourier Transform (FFT) in the slow time dimension (i.e., a sequence of consecutive chirp frames), and then indexed at the frequency. The spectral amplitude obtained at the location. The physical meaning is: the phase change caused by micro-movements of the human body (such as breathing and heartbeat) within this distance unit at a frequency... The energy intensity is used to analyze the phase variation over time, thereby capturing minute movements caused by human breathing or heartbeat. Final microkinetic energy characteristics The definition of is as follows: ; Micro-kinetic energy characteristics refer to the energy characteristics within a specified frequency range (i.e., The maximum amplitude of the phase FFT spectrum is taken as the microkinetic energy of the distance gate, which reflects the intensity of signal change caused by human micro-movement and can be used for subsequent vital sign detection.

[0055] In some embodiments of this application, a human body reflection energy factor for each distance gate is generated by weighted fusion based on echo energy characteristics and the micro-motion energy characteristics, thereby enhancing the ability to identify human targets. Specifically, the normalized echo energy characteristics are first calculated. Its formula is ; in, This is the average value of the echo energy within the gate at that distance, reflecting the overall signal strength; The variance of the echo energy characterizes the degree of energy fluctuation. The ratio of the two is used to eliminate the influence of signal strength differences at different distances, thereby achieving normalization of energy characteristics and improving robustness in complex backgrounds.

[0056] Subsequently, based on the normalized echo energy characteristics and microkinetic energy characteristics Calculate the human body's reflective energy factor, satisfying the following formula. ,in and Let be the weight coefficient, and satisfy... This is used to adjust the contribution ratio of normalized energy and micro-kinetic energy in the comprehensive judgment. Normalized energy mainly reflects the presence and stability of the target, while micro-kinetic energy reflects the micro-movement characteristics of the human body, such as phase changes caused by breathing or heartbeat. By weighted fusion of the two, static clutter and dynamic human targets can be effectively distinguished, improving the accuracy and anti-interference ability of personnel presence detection.

[0057] In some embodiments of this application, the lightweight deep learning network includes an input layer, two convolutional layers, two pooling layers, a fully connected layer, and an output layer connected in sequence. The lightweight deep learning network is obtained through the following training process.

[0058] Step S310: Collect millimeter-wave radar echo signal samples containing human targets and not containing human targets under various complex environments.

[0059] In step S310, millimeter-wave radar echo signal samples covering various complex environmental conditions are collected. These samples include scenes with real human targets as well as background clutter scenes without human presence. By collecting data under different environments such as indoors, outdoors, with furniture interference, and with dynamic non-human targets, the training data is ensured to have sufficient diversity and representativeness, thereby providing a comprehensive input foundation for subsequent model learning and improving the network's generalization ability and robustness in practical applications.

[0060] Step S320: For each sample, calculate the human body reflection energy factor of each distance gate to form an input feature vector.

[0061] In step S320, for each radar echo signal sample acquired, the human body reflection energy factor corresponding to each range gate is calculated according to a preset method. These factors are then organized into a vector form according to the range gate order, serving as the input feature of the deep learning network. This feature vector integrates echo energy and micro-motion energy information, effectively characterizing whether there are targets with vital signs within each range cell, providing structured and physically meaningful input data for network discrimination.

[0062] Step S330: Based on the actual scene of each sample, label the corresponding personnel presence status identifier and the actual human distance value to form the output label.

[0063] In step S330, based on the actual physical scene corresponding to each sample, the actual presence status identifier of the personnel and the corresponding human distance value are accurately labeled manually or through auxiliary means, forming an output label that corresponds one-to-one with the input feature vector. The personnel presence status identifier indicates whether the sample contains a valid human target, while the human distance value accurately reflects the target's position in the radar distance dimension. These labels constitute the ground truth information required for supervised learning and are the core basis for optimizing parameters during network training.

[0064] Step S340: Construct a training set based on the input feature vector and output label, and perform supervised training on the lightweight deep learning network until convergence.

[0065] In step S340, the input feature vectors are paired with the output labels to construct a complete supervised training dataset, which is then used to train the lightweight deep learning network end-to-end. During training, the network predicts the output through forward propagation, and then performs backpropagation based on the error between the prediction result and the true label, continuously adjusting the parameters of each layer until the loss function converges and the model performance stabilizes. This process enables the network to gradually learn to automatically extract discriminative features from human body reflectance energy factors, ultimately achieving a joint high-precision estimation of the presence state and distance values ​​of people.

[0066] In some embodiments of this application, step S140 involves processing the phase signal corresponding to the distance gate in the human body echo signal to obtain a smooth and continuous phase signal, including the following steps.

[0067] Step S410: Perform median filtering on the phase signal to remove instantaneous spikes caused by sudden interference.

[0068] In step S410, median filtering of the phase signal effectively removes sharp pulse noise caused by environmental interference or system transient anomalies. These transient spikes severely distort the local shape of the phase signal, affecting the accuracy of subsequent micro-motion feature extraction. Median filtering, by replacing the center point with the median value within a sliding window, can robustly suppress isolated outliers while preserving the main trend of the signal, providing a more reliable phase data foundation for subsequent processing.

[0069] Step S420: Based on the phase difference between adjacent radar frames, the filtered phase signal is subjected to phase dewinding processing to eliminate the jump caused by phase periodicity.

[0070] In step S420, phase dewinding is performed based on the phase difference between adjacent radar frames to address the inherent periodic ambiguity problem in millimeter-wave radar phase measurements. Since the phase value is confined to the range of negative π to positive π, phase changes caused by subtle human movements exceeding this range can result in jumps or reversals, disrupting signal continuity. By calculating and accumulating the phase difference between adjacent frames, a true and continuous phase evolution trajectory can be recovered, accurately reflecting minute displacements caused by breathing or heartbeats.

[0071] In step S430, the unwound phase signal is input into an extended Kalman filter. The phase signal is dynamically smoothed by establishing state equations and observation equations to suppress low-frequency drift and random noise.

[0072] In step S430, the unwound phase signal is input into an extended Kalman filter. By establishing a state equation describing the dynamic changes in phase and an observation equation reflecting the actual observed values, optimal estimation and smoothing of the phase signal are achieved. This step can simultaneously suppress low-frequency drift (such as that caused by temperature changes or radar platform fretting) and random Gaussian noise, improving the temporal continuity and stability of the phase signal while preserving physiological fretting characteristics such as breathing and heartbeat, thus laying the foundation for high-precision extraction of vital sign parameters.

[0073] In step S440, wavelet transform is performed on the phase signal after extended Kalman filtering, and the wavelet coefficients are processed using the soft thresholding method to suppress residual high-frequency noise components and obtain a smooth and continuous phase signal.

[0074] In step S440, wavelet transform is performed on the phase signal after extended Kalman filtering, and the wavelet coefficients are processed using a soft thresholding method to further remove residual high-frequency noise components. Wavelet transform can decompose the signal into multiple scales in the time and frequency domains, allowing noise and useful signals to exhibit different characteristics at different scales. The soft thresholding method effectively weakens the noise energy in high-frequency details by shrinking the wavelet coefficients, while avoiding physiological signal distortion caused by excessive smoothing, ultimately outputting a smooth continuous phase signal that retains key micro-motion information.

[0075] In some embodiments of this application, the phase signal corresponding to the distance gate in the human body echo signal is processed in multiple stages to obtain a smooth and continuous phase time series for subsequent extraction of vital signs parameters.

[0076] First, median filtering is used to remove transient spike interference, resulting in the filtered phase signal. The processing formula is as follows: ; in, The filter window size represents the current frame. Take from the surrounding area Phase values ​​of adjacent frames By taking the median, abnormal spikes caused by sudden noise or clutter are suppressed, while preserving the main trend of phase change.

[0077] Subsequently, phase unwrapping is achieved using the difference between adjacent frames, and the phase difference between adjacent frames is calculated using the following formula: ;when When this occurs, it indicates a phase transition has taken place, and unwinding correction should be performed at this time. This causes the phase difference to return to its normal value. Within the range, the true continuous phase change trajectory is recovered. The unwrapped phase is obtained through recursive updating. This enables continuous reconstruction of the phase signal.

[0078] Next, the unwound phase input extended Kalman filter is dynamically smoothed, and its state equation is: ,in The process noise represents the random perturbation of phase evolution over time; the observation equation is... ,in To observe noise and reflect errors in actual measurements, this filtering model effectively suppresses low-frequency drift and random noise, improving the stability of the phase signal. Finally, wavelet transform is applied to the filtered phase signal, and a soft thresholding method is used to suppress residual high-frequency components. The processing formula is as follows: ; in, These are wavelet coefficients. Using a set threshold, this method shrinks the wavelet coefficients to reduce noise while preserving the main signal features, ultimately obtaining a smooth, continuous phase signal with a high signal-to-noise ratio, providing a reliable basis for the accurate extraction of micro-motion parameters such as respiration and heartbeat.

[0079] In some embodiments of this application, the instantaneous respiratory rate is estimated based on the respiratory components, and an adaptive harmonic spectrum is constructed based on the instantaneous respiratory rate and the distance value to the human body, including the following steps.

[0080] Step S510: Perform autocorrelation calculation on the respiratory component and perform fast Fourier transform on the autocorrelation result to obtain the respiratory power spectrum.

[0081] In step S510, autocorrelation is performed on the respiratory component, and a fast Fourier transform is applied to the autocorrelation result. This enhances the periodicity of the respiratory signal and suppresses aperiodic interference. The autocorrelation operation highlights repetitive patterns in the signal, making the respiratory rhythm more prominent in the time domain. Subsequently, the autocorrelation result is converted to the frequency domain using a fast Fourier transform to generate a respiratory power spectrum, thus providing a clear spectral basis for accurately identifying the respiratory frequency.

[0082] Step S520: Extract the frequency corresponding to the maximum spectral peak from the respiratory power spectrum as the instantaneous respiratory rate.

[0083] In step S520, the frequency corresponding to the maximum spectral peak is extracted from the respiratory power spectrum as the instantaneous respiratory frequency. This is significant because it allows for real-time acquisition of the most prominent respiratory rhythm at the current moment in a data-driven manner. Since the respiratory signal typically manifests as the strongest energy peak in the frequency domain, the position of this peak directly reflects the actual respiratory rate. Therefore, selecting the frequency of the maximum spectral peak can effectively characterize the instantaneous respiratory state and provide an accurate fundamental frequency reference for subsequent harmonic analysis.

[0084] Step S530: Determine the position of the higher harmonic frequencies based on the instantaneous respiratory rate, and set the harmonic spectrum bandwidth with each higher harmonic as the center.

[0085] In step S530, based on the instantaneous respiratory rate f The purpose of determining the positions of its higher harmonic frequencies and setting the harmonic spectrum bandwidth centered on each higher harmonic is to accurately locate the components of the respiratory signal in the frequency domain that may interfere with heart rate measurement. Respiratory movements have non-sinusoidal characteristics, generating second, third, and other higher harmonics, whose frequency bands often overlap with the heart rate frequency band; by using nf By setting a reasonable bandwidth around the center (n=2,3,…), a frequency band model covering the actual harmonic energy distribution can be constructed, laying the foundation for subsequent accurate elimination of harmonic interference.

[0086] Step S540: Based on the distance value to the human body, the harmonic spectral power at each higher harmonic is weighted and adjusted. The closer the distance, the higher the harmonic spectral power weight is assigned, thus constructing an adaptive harmonic spectrum.

[0087] In step S540, the harmonic spectral power at each higher harmonic level is weighted and adjusted based on the distance to the human body. A closer distance assigns a higher weight to the harmonic spectral power, which aims to make the constructed harmonic spectrum more closely match the physical characteristics of the actual echo signal. Since the signal strength received by millimeter-wave radar attenuates with distance, the harmonic energy generated by the same physiological activity differs at different distances. Introducing distance information for adaptive weighting of harmonic power can improve the accuracy and robustness of the harmonic spectrum, thereby more effectively separating and removing respiratory harmonic components from the heartbeat component.

[0088] In some embodiments of this application, when the presence status indicator is 0, harmonic spectrum construction and subsequent heartbeat parameter extraction are skipped to avoid performing invalid signal processing operations in an unoccupied state. Since the millimeter-wave radar does not detect valid human reflection signals at this time, continuing harmonic spectrum construction and heartbeat parameter calculation is not only meaningless but also wastes system computing resources and may generate false vital sign outputs. By proactively terminating subsequent processes based on the presence status indicator, the system's operating efficiency and result reliability can be improved.

[0089] When the presence status indicator is 1, a complete adaptive harmonic spectrum is generated to suppress respiratory harmonic interference in the heartbeat component. Utilizing prior information confirming the presence of the person, targeted and high-precision vital sign extraction is performed. In this state, high-order harmonics generated by respiration may fall into the heartbeat frequency band, causing deviations in heart rate estimation. By constructing an adaptive harmonic spectrum that matches the current respiratory rate and distance from the person, these interfering components can be accurately modeled and subtracted, thereby effectively improving the accuracy of heartbeat parameter measurements.

[0090] In some embodiments of this application, respiratory and heart rate components are extracted based on a smooth, continuous phase signal. This includes applying a bandpass filter of 0.1 Hz to 0.5 Hz to obtain the respiratory component and a bandpass filter of 0.8 Hz to 2.0 Hz to obtain the heart rate component only when the presence status is indicated as present. The 0.1 Hz to 0.5 Hz frequency band covers the typical respiratory frequency range of a normal adult. Bandpass filtering can effectively separate low-frequency phase changes caused by chest cavity micro-movements, while avoiding false triggering of respiratory parameter calculations due to environmental noise or clutter in unoccupied situations, thereby improving the targeting of the system processing and the reliability of the results. The 0.8 Hz to 2.0 Hz frequency band corresponds to the normal heart rate range (approximately 48 to 120 beats / minute). Bandpass filtering can separate the weak high-frequency vibration components caused by heartbeats from the phase signal. Combined with prior information about the presence of personnel, ineffective filtering and misjudgment of noise can be prevented in targetless scenarios, ensuring the rationality and accuracy of heart rate parameter extraction.

[0091] In some embodiments of this application, a respiratory and heart rate parameter measurement method based on prior knowledge of the presence of a person is employed to achieve high-precision extraction of human vital signs. This method first utilizes a bandpass filter to process the phase signal... Separate the respiratory and cardiac components separately.

[0092] Specifically, the phase signal is filtered using a 0.1 to 0.5 Hz bandpass filter to obtain the respiratory component: ,in The transfer function of a bandpass filter with a frequency range of 0.1–0.5 Hz is used to preserve the low-frequency phase changes caused by human respiratory movements; * denotes convolution operation, indicating that the filtering process is the convolution of the signal in the time domain with the filter's impulse response. Subsequently, a bandpass filter of 0.8–2.0 Hz is used to extract the heartbeat component: ,in It is a bandpass filter transfer function with a frequency range of 0.8–2.0 Hz, corresponding to the heart rate frequency band of normal people, used to extract the mid-frequency phase fluctuations caused by heartbeats.

[0093] Next, regarding the respiratory component Autocorrelation calculations are performed to estimate the instantaneous respiratory rate; the autocorrelation function is defined as follows: ; in For time delay, This indicates the similarity of respiratory signals at different time delays. By performing a Fast Fourier Transform (FFT) on this autocorrelation result, the respiratory power spectrum can be obtained, and the frequency corresponding to the maximum spectral peak is the instantaneous respiratory rate. Based on this frequency, the higher harmonic frequencies of respiration can be further determined, i.e. ,in These harmonic components often overlap with the frequency band of heartbeat signals, which may cause interference. Therefore, they can be used to construct an adaptive harmonic spectrum to suppress interference.

[0094] In some embodiments of this application, an adaptive harmonic spectrum is constructed based on the presence status of personnel and the distance information of the human body to achieve precise suppression of respiratory harmonic interference in the heartbeat signal. Specifically, firstly, based on the estimated instantaneous respiratory rate... Determine its higher harmonic frequencies The harmonic spectrum bandwidth is set with each harmonic as the center, and its bandwidth range is... ,in The bandwidth coefficient is used to control the coverage width of the harmonic frequency band, ensuring effective capture of the energy distribution of respiratory harmonics. The innovation of this method lies in incorporating prior knowledge of the presence of personnel into the harmonic spectrum construction process: when the personnel presence status is 0, subsequent harmonic processing steps are skipped to avoid redundant calculations in an unoccupied state, saving system resources; when the personnel presence status is 1, the distance to the human body output by the deep learning network is used... The harmonic spectral power at each harmonic is dynamically adjusted. At this time, the... Power of first harmonic With distance They are inversely proportional and satisfy the following conditions. ,in The power coefficient reflects the physical characteristic that the radar echo intensity decreases with the square of the distance. This makes the harmonic spectrum power decrease with increasing distance, thus better matching the actual signal characteristics and solving the problem of inaccurate harmonic removal caused by ignoring the presence of people and distance factors in traditional methods.

[0095] Subsequently, in heart rate components Subtracting the adaptive harmonic spectrum from the middle yields the heartbeat component after harmonic interference removal: ,in Let be the Dirac function, representing that in An ideal narrowband filter is introduced to cancel harmonic components.

[0096] Finally, a short-time Fourier transform (STFT) is performed on the remaining signal, and the formula is as follows: ; in, It is an integral variable, representing a dummy variable on a continuous time axis, used to traverse the time domain of the signal; This is a window function used to localize time-domain signals, giving the spectral analysis time-varying characteristics. By tracking the changes in frequency peaks in the time-frequency plot, an estimate of the instantaneous heart rate over time can be obtained. This enables high-precision, low-interference extraction of heartbeat parameters.

[0097] In some embodiments of this application, a 24GHz millimeter-wave radar is used to collect human body echo signals. The radar's operating parameters are set as follows: sampling frequency 2MHz, bandwidth 250MHz, and 1024 sampling points per frame. The technical solution proposed in this application is achieved through the following steps.

[0098] Step 1: Perform range domain FFT processing on each range gate of the acquired echo signal, with N=256 FFT points. Calculate the mean and variance of the echo energy of each range gate. At the same time, perform FFT processing on the phase signal within each range gate (P=1024) and extract the maximum value of the spectral peak as the micro-motion energy.

[0099] Step 2: Based on the calculated mean, variance, and micro-motion energy of the echo energy, calculate the normalized energy and human body reflection energy factor according to the formula, where the weighting coefficients are... It is 0.6. It is 0.4.

[0100] Step 3: Construct a lightweight deep learning network consisting of two convolutional layers (using 3x3 kernels), two pooling layers (using 2x2 pooling windows), one fully connected layer, and an output layer. Human body reflectivity factors are input into the network for training and recognition, outputting a person's presence status identifier and distance value. During training, a large amount of labeled data allows the network to continuously learn and optimize, thus developing strong generalization capabilities and adapting to different environments and person states.

[0101] Step 4: Process the phase signal of the human body echo distance gate, with a median filter window size of k=5; in the extended Kalman filter, the process noise variance is set to 0.01, and the observation noise variance is set to 0.1; in the wavelet soft thresholding process, the db4 wavelet is selected, the decomposition level is 3, and the threshold... It is adaptively determined based on the noise level.

[0102] Step 5: Extract respiratory and heart rate parameters from the processed phase signal. A Butterworth filter of order 4 is used as the bandpass filter; bandwidth coefficient... The frequency is 0.08 Hz; the short-time Fourier transform uses a Hanning window with a window length of 256 and an overlap length of 128.

[0103] Through the above implementation methods, the presence status of personnel can be accurately determined, background noise can be effectively suppressed, the respiratory rate measurement error can be controlled within ±0.2 breaths / minute, and the heart rate measurement error can be controlled within ±2 breaths / minute.

[0104] Secondly, refer to Figure 2 This application provides a lightweight millimeter-wave vital signs measurement system, including a millimeter-wave radar, a memory, and a processor.

[0105] Millimeter-wave radar is used to collect human echo signals; memory is used to store computer programs; processor, coupled to memory, is configured to support lightweight edge inference, enabling it to execute computer programs to implement the aforementioned lightweight millimeter-wave vital signs measurement method without cloud computing support.

[0106] Specifically, millimeter-wave radar is used to collect human echo signals. Its function is to obtain raw radar data containing micro-motion information such as breathing and heartbeat by emitting millimeter waves and receiving signals reflected back from the human body surface. Because millimeter waves have the characteristics of being non-contact, having strong ability to penetrate clothing, and being sensitive to minute movements, this radar can continuously sense dynamic characteristics related to human vital signs without relying on cameras or wearable devices, providing basic input for subsequent signal processing.

[0107] The memory is used to store computer programs, and its function is to store all the instruction code and parameter configurations required to implement the lightweight millimeter-wave vital sign measurement method described in this application. These programs include modules such as signal preprocessing, feature extraction, deep learning models, phase processing algorithms, and vital sign parameter calculation, ensuring that the stored content can be directly called by the processor locally without relying on external storage or network transmission, thereby ensuring the independence and real-time performance of the system operation.

[0108] The processor is configured to support lightweight edge inference, enabling it to execute a memory-stored computer program based on human echo signals acquired by millimeter-wave radar, without cloud computing support, to implement the aforementioned lightweight millimeter-wave vital sign measurement method. By deploying the entire measurement process on a local hardware platform, it achieves low-latency, high-privacy, and highly reliable vital sign monitoring. Employing lightweight deep learning networks and optimized signal processing algorithms, the processor can efficiently complete tasks such as human presence detection, distance estimation, phase calculation, and respiratory and heart rate parameter extraction on resource-constrained embedded devices without uploading data to the cloud. This makes it suitable for applications with stringent real-time and offline capabilities, such as medical monitoring, smart homes, and security.

[0109] Furthermore, embodiments of this application provide an electronic device, including the aforementioned lightweight millimeter-wave vital signs measurement system.

[0110] In addition, a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the aforementioned lightweight millimeter-wave vital signs measurement method.

[0111] In summary, the lightweight millimeter-wave vital sign measurement method, system, device, and medium provided in this application have the following technical effects.

[0112] This application's embodiments utilize a lightweight deep learning network to identify human body reflection energy factors that integrate normalized energy and micro-motion energy. This network autonomously learns the essential differences between human body echo signals and background clutter, significantly improving the accuracy of personnel presence determination while simultaneously achieving high-precision distance measurement and clutter suppression. Furthermore, by incorporating personnel presence as key prior knowledge into the parameter extraction process, filtering and subsequent processing of respiratory and heartbeat components are only performed upon confirmation of personnel presence. This effectively avoids invalid calculations and spurious outputs, improving system robustness and resource utilization efficiency.

[0113] To address the susceptibility of millimeter-wave phase signals to noise, phase entanglement, and high-frequency interference, this application employs a multi-level processing strategy combining median filtering, inter-frame differential de-entanglement, extended Kalman filtering, and wavelet soft thresholding to recover a smooth and continuous phase signal, laying the foundation for high-precision vital sign extraction. Furthermore, an adaptive harmonic spectrum is constructed based on instantaneous respiratory rate and distance to the human body, accurately modeling and subtracting the interference of higher respiratory harmonics on the heart rate band, controlling the respiratory rate measurement error to within ±0.2 breaths / minute and the heart rate error to within ±2 breaths / minute. The entire solution is lightweight and can run independently on edge processors without cloud support, meeting the requirements of embedded scenarios for low power consumption, low latency, and high privacy, facilitating large-scale deployment in medical, home, and security fields.

[0114] It should be noted that in all specific embodiments of this application, all data processing activities related to user identity or personal characteristics, such as user information, user behavior data, historical data, and location information, will be conducted in accordance with the principles of legality, legitimacy, and necessity. All data collection, use, storage, and processing will be subject to compliance with applicable national and regional laws, regulations, and industry standards, and informed consent from users will be obtained in a clear and explicit manner before processing. For the processing of sensitive personal information, separate consent from users will be obtained through prominent means such as pop-up prompts and independent confirmation pages. If any processing conflicts with laws and regulations, the laws and regulations will prevail, and necessary data processing will only be carried out within the scope permitted by laws and regulations, ensuring that all data-based applications, analyses, and technical implementations are conducted within the scope permitted by laws and regulations.

[0115] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this application are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0116] Furthermore, although this application is described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding this application. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of ordinary skill of an engineer. Therefore, those skilled in the art can implement the application set forth in the claims using ordinary skill. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of this application, which is determined by the full scope of the appended claims and their equivalents.

[0117] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several programs to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0118] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequential list of executable programs for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, a program execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can retrieve and execute a program from or in conjunction with such a program execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit a program for use by or in conjunction with a program execution system, apparatus, or device.

[0119] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Additionally, computer-readable media can even be paper or other suitable media on which programs can be printed, for example, by optically scanning the paper or other media, then editing, interpreting, or, if necessary, processing it in a suitable manner to obtain the program electronically, and then storing it in computer memory.

[0120] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable program execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0121] In the foregoing description of this specification, the reference to terms such as "one embodiment / implementation," "another embodiment / implementation," or "certain embodiments / implementations," etc., indicates that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in an embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0122] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0123] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A lightweight millimeter-wave vital sign measurement method, characterized in that, Includes the following steps: The human body echo signal acquired by millimeter-wave radar is preprocessed to calculate the echo energy characteristics and micro-motion energy characteristics of each range gate; Based on the echo energy characteristics and the micro-motion energy characteristics, a human body reflection energy factor for each distance gate is generated by weighted fusion. The human body reflective energy factor is input into a pre-trained lightweight deep learning network, and the lightweight deep learning network outputs the human presence status identifier and the corresponding human body distance value. When the presence status of the person is indicated as present, the phase signal of the corresponding distance gate in the human body echo signal is processed based on the human body distance value to obtain a smooth and continuous phase signal. Based on the smooth and continuous phase signal, the respiratory component and the heartbeat component are extracted; The instantaneous respiratory rate is estimated based on the respiratory components, and an adaptive harmonic spectrum is constructed based on the instantaneous respiratory rate and the distance value to the human body. The adaptive harmonic spectrum is subtracted from the heartbeat component to eliminate the interference of higher harmonics of respiration, and the instantaneous heart rate is determined based on the heartbeat component after the interference is eliminated.

2. The lightweight millimeter-wave vital sign measurement method according to claim 1, characterized in that, The preprocessing of the human body echo signal acquired by millimeter-wave radar to calculate the echo energy characteristics and micro-motion energy characteristics of each range gate includes the following steps: Perform a Fast Fourier Transform on the echo signal of each distance gate in the slow time dimension to obtain a frequency domain complex echo sequence; Based on the frequency domain complex echo sequence, the mean echo energy and variance of the echo energy of the distance gate are calculated as the echo energy characteristics; The phase signal is extracted from the frequency domain complex echo sequence, and the phase signal is subjected to a fast Fourier transform. The maximum spectral peak value in the resulting spectrum is taken as the micro-motion energy characteristic of the range gate.

3. The lightweight millimeter-wave vital sign measurement method according to claim 1, characterized in that, The lightweight deep learning network comprises an input layer, two convolutional layers, two pooling layers, a fully connected layer, and an output layer connected in sequence. The lightweight deep learning network is obtained through the following training process: Collect millimeter-wave radar echo signal samples containing and without human targets in various complex environments; For each of the aforementioned samples, the human body reflection energy factor of each distance gate is calculated to form an input feature vector; Based on the actual scene of each sample, the corresponding personnel presence status identifier and the actual human distance value are labeled to form the output label; A training set is constructed based on the input feature vector and the output label, and the lightweight deep learning network is trained under supervision until convergence.

4. The lightweight millimeter-wave vital sign parameter measurement method according to claim 1, characterized in that, The process of processing the phase signal corresponding to the distance gate in the human body echo signal to obtain a smooth and continuous phase signal includes the following steps: The phase signal is subjected to median filtering to remove instantaneous spikes caused by sudden interference; Based on the phase difference between adjacent radar frames, the filtered phase signal is subjected to phase dewinding processing to eliminate the jump caused by phase periodicity. The unwound phase signal is input into an extended Kalman filter, and the phase signal is dynamically smoothed by establishing state equations and observation equations to suppress low-frequency drift and random noise. Wavelet transform is performed on the phase signal after extended Kalman filtering, and the wavelet coefficients are processed using the soft thresholding method to suppress residual high-frequency noise components, thereby obtaining a smooth and continuous phase signal.

5. The lightweight millimeter-wave vital sign parameter measurement method according to claim 1, characterized in that, The step of estimating the instantaneous respiratory rate based on the respiratory components and constructing an adaptive harmonic spectrum based on the instantaneous respiratory rate and the human body distance value includes: Autocorrelation calculation is performed on the respiratory components, and fast Fourier transform is performed on the autocorrelation results to obtain the respiratory power spectrum; The frequency corresponding to the maximum spectral peak is extracted from the respiratory power spectrum as the instantaneous respiratory rate; Based on the instantaneous respiratory rate, determine the position of its higher harmonic frequencies, and set the harmonic spectrum bandwidth with each higher harmonic as the center; Based on the human body distance value, the harmonic spectral power at each higher harmonic is weighted and adjusted, with a higher weight assigned to the harmonic spectral power for closer distances, thus constructing the adaptive harmonic spectrum.

6. The lightweight millimeter-wave vital sign parameter measurement method according to claim 5, characterized in that, When the presence status of the person is 0, the harmonic spectrum construction and subsequent heartbeat parameter extraction are skipped; when the presence status of the person is 1, a complete adaptive harmonic spectrum is generated to suppress respiratory harmonic interference in the heartbeat component.

7. The lightweight millimeter-wave vital sign measurement method according to claim 1, characterized in that, The extraction of respiratory and cardiac components based on the smooth and continuous phase signal includes: Only when the presence status of the person is indicated as present, a bandpass filter of 0.1 Hz to 0.5 Hz is applied to the smooth and continuous phase signal to obtain the respiratory component, and a bandpass filter of 0.8 Hz to 2.0 Hz is applied to obtain the heartbeat component.

8. A lightweight millimeter-wave vital sign measurement system, characterized in that, include: Millimeter-wave radar is used to collect human body echo signals; Memory, used to store computer programs; The processor is configured to support lightweight edge inference and is capable of executing the computer program based on the human echo signal without cloud computing support to implement the lightweight millimeter-wave vital signs measurement method as described in any one of claims 1 to 7.

9. An electronic device, characterized in that, Including the lightweight millimeter-wave vital signs measurement system as described in claim 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the lightweight millimeter-wave vital signs measurement method as described in any one of claims 1 to 7.

Images