A method and system for non-sensing monitoring of physiological indicators based on human body micro-motion sensor matrix

Abstract

The present invention discloses a method and system for non-invasive monitoring of physiological indicators based on a human micro-motion sensor matrix, the method comprising: step 1: performing all-round non-invasive collection of human chest and abdominal respiratory movements through a sensor matrix. Step 2: a data acquisition and processing module synchronously collects, performs noise reduction filtering and feature extraction on multi-sensor data to form a human micro-motion sensor matrix. Step 3: a respiratory event recognition module performs event recognition on the human micro-motion sensor matrix through a respiratory event classification model. Step 4: a user interaction module intuitively displays physiological indicators of respiratory frequency and respiratory depth, and provides auxiliary functions such as alarm. The present invention constructs a multi-sensor fusion sensor matrix to perform all-round non-invasive collection of human chest and abdominal respiratory movements, realizes real-time monitoring and display of key physiological indicators such as respiratory frequency and amplitude, and realizes high-precision, non-invasive monitoring of physiological indicators.

Description

Technical Field

[0001] The present invention belongs to the technical field of physiological monitoring, and in particular relates to a method and system for sensing physiological indicators based on a human body micro-motion sensing matrix. Background Art

[0002] With the continuous improvement of modern medical technology and health management needs, higher requirements are placed on the monitoring technology of human physiological indicators. Traditional physiological indicator monitoring methods, such as respiratory rate and respiratory amplitude, usually rely on contact sensors. Although these methods can provide relatively accurate physiological parameters, they have many limitations in practical applications, such as inconvenience in wearing, poor comfort, and irritation to the skin during long-term monitoring. In addition, contact sensors are prone to displacement during exercise, resulting in unstable data collection and affecting the accuracy of monitoring results.

[0003] In recent years, the development of non-contact sensing technology has provided a new way to monitor physiological indicators. This method not only improves the comfort and user experience of monitoring, but also enables long-term, real-time health monitoring and management without interfering with the daily life and work of the monitored person. Summary of the invention

[0004] In view of the defects existing in the above-mentioned prior art, the present invention provides a method for sensing physiological indicators based on a human body micro-motion sensor matrix, the method comprising:

[0005] Step 1: Perform all-round non-invasive collection of human chest and abdominal respiratory movements through a sensor matrix;

[0006] Step 2: The data acquisition and processing module performs synchronous acquisition, noise reduction filtering and feature extraction on multi-sensor data to form a human body micro-motion sensing matrix;

[0007] Step 3: The respiratory event recognition module performs event recognition on the human body micro-motion sensor matrix through the respiratory event classification model;

[0008] Step 4: The user interaction module intuitively displays the physiological indicators of breathing rate and breathing depth, and provides auxiliary functions such as alarm.

[0009] Among them, the sensor matrix integrates multiple sensors such as FMCW radar, pressure sensor, and acceleration sensor.

[0010] Among them, a clock frequency division and trigger control circuit is used to generate a unified sampling clock and a synchronous trigger signal.

[0011] Among them, the collected raw data is first preprocessed, including:

[0012] The radar signal uses an FIR bandpass filter. The pressure signal is corrected for baseline drift and filtered with a 60Hz power frequency notch. Wavelet transform is used for signal enhancement and smoothing. The acceleration data uses an adaptive threshold piecewise linear interpolation method to remove abnormal points, and then the Kalman filter algorithm is used for smoothing and noise reduction.

[0013] The characteristic parameters of the respiratory motion in the preprocessed sensor signals are extracted.

[0014] Among them, for radar signals, the breathing frequency is calculated based on Chirp-Z transform, and the breathing depth information is obtained through envelope extraction and energy integration; for pressure signals, the breathing frequency is obtained through peak and trough detection, and the breathing depth is estimated based on the cumulative distribution function of the pressure value; for acceleration signals, the autoregressive spectrum estimation is used to obtain the main breathing frequency.

[0015] Among them, feature extraction for FMCW radar signals includes:

[0016] Dominant frequency of the respiratory spectrum , the RMS value of the envelope curve representing the average level of breathing depth , and relative value index of tidal volume ;

[0017] The feature extraction of the fused pressure signal includes:

[0018] Average respiratory rate , respiratory rate standard deviation , average breathing depth , coefficient of variation of respiratory depth ;

[0019] Respiratory feature extraction for acceleration signals includes: respiratory frequency .

[0020] Difference judgment and calibration of common variables, including calculating the difference between the estimated values ​​of different sensors to determine whether the deviation exceeds the threshold. If so, weighted average of the estimated values ​​of the deviation is performed according to the reliability of each sensor to obtain a consistent estimate after calibration.

[0021] Wherein, the common variables are respiratory rate and respiratory depth;

[0022] Difference determination and calibration of non-shared variables, including linear scaling of the FMCW radar-based tidal volume index based on the radar breathing depth calibration scale factor;

[0023] Also, for the standard deviation of the respiratory rate and the coefficient of variation of the respiratory depth determined based on the pressure sensor, by calculating the difference between the standard deviation and the coefficient of variation of the corresponding variability index of the acceleration sensor and / or the FMCW radar, it is determined whether calibration is required. When the difference exceeds a threshold, a linear adjustment is performed according to the degree of difference to obtain a calibrated estimated value;

[0024] Wherein, the non-shared variables are tidal volume, respiratory rate standard deviation and respiratory depth variation coefficient;

[0025] Among them, the calibrated variable estimation values ​​are integrated into the human body micro-motion sensor matrix; each row of the human body micro-motion sensor matrix represents a time point, and each column represents a respiratory characteristic parameter. The human body micro-motion sensor matrix contains information on respiratory frequency, respiratory depth, tidal volume, respiratory frequency variability and respiratory depth variability.

[0026] Among them, a respiratory event classification model is constructed to classify the respiratory events belonging to the human body micro-motion sensor matrix and identify typical events, wherein the typical events at least include: normal breathing, rapid breathing, and apnea;

[0027] The present invention also discloses a physiological index non-sensing monitoring system based on a human body micro-motion sensor matrix. The system comprises a processor and a memory. The processor is used to run the computer code in the memory to execute the above method.

[0028] The present invention uses multi-sensor data to perform all-round non-invasive collection of human chest and abdominal respiratory movements, avoiding the discomfort and risks caused by traditional invasive collection methods, and improving the user experience. Through the fusion and synchronous processing of multi-sensor data, the accuracy and reliability of respiratory monitoring are significantly improved, and the respiratory status of the human body can be more accurately reflected. Through the respiratory event classification model and user interaction module, real-time monitoring and alarm functions of respiratory events are realized, abnormal respiratory events can be detected in time and alarms can be issued, improving the safety of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present disclosure will become readily understood. In the accompanying drawings, several embodiments of the present disclosure are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0030] Figure 1 is a flow chart showing a method for sensing physiological indicators based on a human body micro-motion sensing matrix according to an embodiment of the present invention;

[0031] Figure 2 is a schematic diagram showing a physiological index non-sensing monitoring system based on a human body micro-motion sensor matrix according to an embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings, and "multiple" generally includes at least two.

[0034] It should be understood that although the terms first, second, third, etc. may be used to describe ... in the embodiments of the present invention, these ... should not be limited to these terms. These terms are only used to distinguish .... For example, without departing from the scope of the embodiments of the present invention, the first ... may also be referred to as the second ..., and similarly, the second ... may also be referred to as the first ....

[0035] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0036] As used herein, the words "if" and "if" may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0037] It should also be noted that the term "includes", "comprising" or any other variation thereof is intended to cover non-exclusive inclusion, so that a commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprising a ..." do not exclude the existence of other identical elements in the commodity or device including the elements.

[0038] In traditional methods, respiratory event recognition may rely on a single sensor, and the time and frequency of data collection from different sensors may be inconsistent. The present invention introduces a respiratory event classification model, which improves the accuracy and reliability of respiratory event recognition by fusing multi-sensor data.

[0039] The present invention proposes a method for non-invasive monitoring of physiological indicators based on a human body micro-motion sensor matrix. By constructing a sensor matrix of multi-sensor fusion, all-round non-invasive collection of human chest and abdominal respiratory movements is carried out. Through a data acquisition and processing module, a respiratory event recognition module and a user interaction module, real-time monitoring and display of key physiological indicators such as respiratory frequency and amplitude are realized, thereby achieving high-precision and non-invasive monitoring of physiological indicators.

[0040] like Figure 1 As shown, the workflow of a method for non-sensing monitoring of physiological indicators based on a human body micro-motion sensor matrix of the present invention is as follows:

[0041] Step 1: Use the sensor matrix to collect all-round non-invasive data on the respiratory movements of the chest and abdomen.

[0042] Step 2: The data acquisition and processing module performs synchronous acquisition, noise reduction filtering and feature extraction on multi-sensor data to form a human body micro-motion sensing matrix.

[0043] Step 3: The respiratory event recognition module performs event recognition on the human body micro-motion sensor matrix through the respiratory event classification model.

[0044] Step 4: The user interaction module intuitively displays the physiological indicators of breathing rate and breathing depth, and provides auxiliary functions such as alarm.

[0045] The present invention uses multi-sensor data to perform all-round non-invasive collection of human chest and abdominal respiratory movements, avoiding the discomfort and risks caused by traditional invasive collection methods, and improving the user experience. Through the fusion and synchronous processing of multi-sensor data, the accuracy and reliability of respiratory monitoring are significantly improved, and the respiratory status of the human body can be more accurately reflected. Through the respiratory event classification model and user interaction module, real-time monitoring and alarm functions of respiratory events are realized, abnormal respiratory events can be detected in time and alarms can be issued, improving the safety of users.

[0046] like Figure 2 As shown, the present invention provides a non-invasive respiratory monitoring system based on a human micro-motion sensor matrix, comprising a sensor matrix, a data acquisition and processing module, a respiratory event recognition module and a user interaction module.

[0047] Among them, the sensor matrix integrates multiple sensors such as FMCW radar, pressure sensor, and acceleration sensor.

[0048] In this embodiment, the sensor matrix consists of one 24 GHz FMCW radar, four piezoresistive pressure sensors, and one MEMS accelerometer.

[0049] The sensor matrix consists of a 24GHz FMCW radar, four piezoresistive flexible pressure sensors and a three-axis MEMS accelerometer, which are used to accurately and non-contactly collect the displacement, pressure and vibration signals caused by the respiratory movement of the human chest and abdomen.

[0050] Microwave radar technology uses the principle of electromagnetic wave reflection to monitor tiny displacements on the human body surface without contacting the human body, such as chest and abdominal micro-movements caused by breathing and heartbeats; acceleration sensors and pressure sensors can capture information about human body movements and pressure changes. By fusing and processing the data from multiple sensors, it is possible to achieve non-sensing monitoring of human physiological indicators.

[0051] Specifically, the FMCW radar is placed in the center of the headboard or the top of the sleeping chamber, keeping a vertical distance of 11.5 meters from the human body. The radar's transmitting and receiving antennas adopt a cross-polarization design, with a transmission power of less than 10mW and a working angle of 50°70°, which can detect micro-motion changes in respiratory displacement within the range of 0.1~10mm in the chest and abdomen of adults.

[0052] Four pressure sensors are arranged vertically inside the mattress, located on the left and right sides of the chest and abdomen. Each pressure sensor is made of conductive carbon slurry coated on a flexible piezoresistive material (such as PDMS) substrate, with a size of 1cm×1cm, a sensitivity of up to 0.1 Pa, and a range of 0~10kPa. It can take into account both sensitivity and measurement range to achieve accurate perception of breathing pressure. The sensor is connected to the data acquisition module through conductive fabric or ultra-fine elastic cable.

[0053] The accelerometer is fixed at the center of the bottom of the bed frame. It uses high-performance MEMS technology and is smaller than 5mm×5mm×2mm in size. It also has the characteristics of low noise (noise density less than 180μg / Hz), high sensitivity (up to 0.5mg / LSB), low power consumption (measurement current less than 500μA), etc. The three-axis range can be selected from ±2g / ±4g / ±8g / ±16g to meet different monitoring needs. By picking up tiny vibrations of the bed, macroscopic information of human respiratory movement can be obtained.

[0054] When the system is working, the FMCW radar, pressure and acceleration sensors synchronously collect original signals. The data collection and processing flow of the radar is as follows: the transmitting antenna sends out a linear frequency modulated continuous wave signal of a specific frequency, which is reflected after encountering the chest and abdomen of the human body, and the receiving antenna captures the echo signal; the radar obtains the intermediate frequency Beat signal through IQ demodulation, and then after amplification, filtering, ADC and other processing, it outputs the baseband signal related to the respiratory displacement and uploads it to the data acquisition module. The pressure sensor detects the changes in the chest and abdominal pressure distribution caused by respiratory movement in real time. When the pressure acts on the sensor, the resistance value of the sensor shows periodic changes. Through a simple resistance-to-voltage conversion and amplification circuit, a voltage signal reflecting the respiratory pressure rhythm can be obtained. The three-way acceleration analog output of the acceleration sensor is conditioned, amplified, and filtered, and then converted into a digital signal by ADC. The transformation obtains the synthetic acceleration, which reflects the vibration level of the bed caused by breathing.

[0055] Among them, the data acquisition and processing module is responsible for the synchronous acquisition and fusion processing of multi-sensor data.

[0056] The data acquisition and processing module is the core component of this system. It is responsible for the synchronous acquisition, preprocessing, feature extraction and multi-level fusion of multi-sensor data, integrating heterogeneous sensor data such as radar, pressure, acceleration, etc. into the human body micro-motion sensor matrix to provide high-quality data support for respiratory event recognition.

[0057] In terms of data synchronization acquisition, the system uses high-precision clock frequency division and trigger control circuits to generate a unified sampling clock and synchronous trigger signal. The FMCW radar samples the demodulated baseband signal into a digital signal through the ADC, and the pressure and acceleration sensors are also converted into digital signals after analog front-end conditioning. The sampling rate is set to 200Hz and the bit depth is 16bit, ensuring the real-time and synchronization of the data from each sensor.

[0058] The collected raw data are first preprocessed. The radar signal uses an FIR bandpass filter to filter out interference frequency components other than the respiratory frequency (0.10.8Hz), such as heartbeat (12Hz) and body movement (0.01~5Hz). After the pressure signal is corrected for baseline drift and filtered with a 60Hz power frequency notch, the wavelet transform is used for signal enhancement and smoothing. The acceleration data uses an adaptive threshold piecewise linear interpolation method to remove abnormal points, and then uses the Kalman filter algorithm for smoothing and noise reduction. The preprocessed radar, pressure, and acceleration signals all show obvious respiratory rhythm characteristics.

[0059] For radar signals, a bandpass FIR filter is first designed, with a passband range of 0.10.8Hz, corresponding to the normal breathing frequency range of adults. The characteristics of FIR filters are linear phase, good stability, and the filter order and cutoff frequency can be flexibly set. A 64-order FIR filter is designed by Hamming window, with a passband ripple less than 0.1dB and a stopband attenuation greater than 40dB, which can effectively filter out non-respiratory frequency components such as heartbeat (12Hz) and body movement (0.015Hz), while retaining the respiratory signal to the maximum extent. FFT analysis of the filtered signal shows that the respiratory peak is located at 0.10.8Hz, and the heartbeat and body movement components are significantly suppressed.

[0060] In order to accurately extract human breathing frequency information from radar signals, a special digital filter needs to be designed to filter out other irrelevant frequency components, such as frequencies caused by heartbeat and body movement.

[0061] For normal adults, the respiratory rate is usually between 6 and 48 times per minute, which is 0.1 to 0.8 Hz when converted into Hertz. Therefore, the passband range of the filter is set to 0.1 to 0.8 Hz, that is, only signals within this frequency range are allowed to pass, while signals of other frequencies are blocked. Choose a filter type with good performance and easy implementation. The finite unit impulse response (FIR) filter is very suitable for this purpose because it is stable and reliable, and the parameters can be flexibly adjusted to achieve the best filtering effect. Using a 64-order FIR filter is equivalent to using 64 coefficients to perform weighted averaging on the signal to achieve an ideal frequency response curve.

[0062] In order to make the filter as flat as possible in the passband and as large as possible in the stopband, the Hamming window function is used to design the filter coefficients. This can control the fluctuation in the passband to within 0.1 dB, while increasing the stopband attenuation to more than 40 dB. The required breathing signal can pass through almost intact, while the unnecessary heartbeat and body movement signals are attenuated to less than one percent of the original.

[0063] When the radar signal is processed with the designed FIR filter, the respiratory frequency component will be highlighted, and other frequency components will be greatly weakened. In order to visually observe the filtering effect, a fast Fourier transform is performed on the filtered signal to convert the waveform in the time domain into a spectrum in the frequency domain. The respiratory frequency will appear in the form of a peak at 0.1 to 0.8 Hz, while the original heartbeat peak and body movement peak are greatly reduced or even disappear.

[0064] For the pressure signal, the moving average method is first used to estimate and deduct the baseline drift. A data segment of 110 seconds is taken, and the baseline is calculated using the weighted sliding mean. Then the baseline is subtracted from the original signal to achieve baseline drift correction. Next, in response to the 50Hz power frequency interference in my country, an IIR notch filter based on bilinear transformation is designed, with a center frequency of 60Hz, a bandwidth of 2Hz, and a stopband attenuation of more than 35dB, which effectively suppresses the power frequency interference. In order to further improve the signal quality, the wavelet transform is used to enhance and smooth the signal. The sym4 wavelet basis function is used to decompose the signal into 5 layers, and the soft threshold function is used on the 3rd to 5th layer wavelet coefficients for denoising. The threshold size is adaptively adjusted according to the coefficient variance, and then the enhanced pressure signal is reconstructed.

[0065] In order to obtain a more accurate and clearer pressure waveform from the signal collected by the pressure sensor, a series of preprocessing and enhancement operations are required on the original signal.

[0066] First, since the pressure sensor is easily affected by various environmental factors, the measured pressure value will drift or offset slowly. In order to correct this drift, a method called moving average is used, which is equivalent to sliding a fixed-size window on the signal continuously and calculating the average value of all data points in the window as the baseline estimate of the signal. 110 seconds is selected as the window length, which can better reflect the overall trend of the pressure signal. Subtracting the estimated baseline from the original signal, the drift-corrected pressure signal is obtained.

[0067] Because some regular interference noise is mixed into the pressure signal, the waveform has additional fluctuations. Through analysis, it is found that these interferences mainly come from the 60Hz mains. In order to remove this power frequency interference, a special digital filter is added, called a notch filter. Its function is to greatly attenuate the signal within a specific frequency range and allow signals of other frequencies to pass. Setting the center frequency of the notch filter to 60Hz and the attenuation bandwidth to 2Hz can effectively suppress power frequency interference by more than 35 decibels without affecting the pressure signal, which is equivalent to reducing the interference to 1 / 56 of the original.

[0068] In order to further improve the quality and clarity of the signal, the signal processing method of wavelet transform is used to enhance and smooth the pressure signal. Wavelet transform can be regarded as an intelligent signal decomposition and reconstruction tool. It can decompose the signal into basic components of different scales and directions. By processing these fragments, removing those components representing noise, and then reassembling the optimized fragments, an enhanced signal is obtained. Sym4 is selected as the wavelet basis function, and the pressure signal is decomposed into 5 layers. The soft threshold method is used to denoise the wavelet coefficients of the 3rd to 5th layers. The size of the threshold is adaptively adjusted according to the statistical characteristics of the coefficients of each layer, so as to achieve automatic noise suppression.

[0069] The soft threshold method is a commonly used wavelet coefficient processing method. Its basic principle is to regard those coefficients with smaller absolute values ​​as noise, reduce them or directly clear them, and retain those coefficients with larger absolute values.

[0070] The soft threshold method is considered as a process of shrinking coefficients. It compresses coefficients whose absolute values ​​are less than the threshold λ to make them closer to 0, and retains coefficients whose absolute values ​​are greater than or equal to λ, but also reduces them to a certain extent. This operation can be expressed by mathematical formula:

[0071] If the absolute value of the wavelet coefficient is less than the threshold λ, it is set to 0;

[0072] If the wavelet coefficient is greater than the threshold λ, then λ is subtracted from it;

[0073] If the wavelet coefficient is less than λ, add λ to it.

[0074] Through such soft threshold processing, those coefficients with smaller amplitudes that may represent noise are effectively suppressed, while those coefficients with larger amplitudes that represent real signals are retained and compressed to a certain extent.

[0075] The key to the soft threshold method is the selection of the threshold λ. If λ is too small, the denoising strength is insufficient and the noise cannot be effectively suppressed; if λ is too large, the useful signal components may also be removed, resulting in signal distortion. Therefore, it is necessary to adaptively adjust the threshold size according to the characteristics of the signal and the intensity of the noise.

[0076] The adaptive threshold selection method of the present invention is based on the statistical characteristics of the wavelet coefficients of each layer, such as standard deviation or variance, to determine the size of the threshold. For example, the standard deviation σ_i of the wavelet coefficients of the i-th layer can be taken and multiplied by a proportional coefficient k as the soft threshold of the layer:

[0077] .

[0078] Among them, the proportionality coefficient k can be determined by experience or cross-validation method, and is usually between 0.5 and 1. In this way, for those layers with large variance, the corresponding threshold will be larger and the denoising will be stronger; and for those layers with small variance, the threshold will be reduced accordingly to avoid excessive denoising.

[0079] Through the above three steps of processing, the pressure signal that originally had drift, power frequency interference, and hidden noise was corrected and purified step by step, and finally presented a stable, smooth, and clearly layered waveform, laying a high-quality data foundation for subsequent feature extraction and parameter analysis.

[0080] For acceleration signals, in order to remove outliers introduced by body motion, etc., Tukey's test method is first used to adaptively detect outliers and locate them in the acceleration sequence. Then, the acceleration sequence is segmented with outliers as the boundary, and cubic spline interpolation is performed on each segment to smooth the mutation at the outlier point. Finally, each segment is seamlessly spliced ​​to achieve the removal of outliers. On this basis, Kalman filtering is further used to smooth and reduce noise on the acceleration signal.

[0081] In the process of respiratory monitoring, the present invention uses an acceleration sensor to capture the ups and downs of the chest and abdomen of the human body. However, due to the influence of factors such as involuntary body movements and limb tremors, the collected acceleration signal is often mixed with some abnormal peaks or jumps, making the respiratory waveform uneven and discontinuous. In order to obtain a more stable and reliable acceleration signal, it is necessary to identify and remove abnormal points and noise.

[0082] First, we need to find the abnormal points in the acceleration signal. Abnormal points usually appear as maximum or minimum values ​​that are significantly different from the surrounding data. They are often caused by non-respiratory factors such as body movement. A statistical method called Tukey's test is used to automatically determine which points are abnormal points based on the overall distribution characteristics of the signal. This is like identifying "outliers" with abnormal behavior among a group of normal people.

[0083] After finding the abnormal points, they need to be processed to make the signal curve smooth and continuous at the abnormal points. The method used is to use the abnormal points as the dividing points and divide the acceleration signal into several segments. For each segment of data, a mathematical method called cubic spline interpolation is used to fit a smooth curve to remove the spikes and jumps at the abnormal points.

[0084] On the basis of removing abnormal points, the Kalman filter method is further used to smooth and reduce the noise of the acceleration signal as a whole. Kalman filtering is a dynamic and adaptive signal processing algorithm that makes full use of the physiological characteristics of respiratory movement and establishes a mathematical model that can reflect the changes in respiratory state. This model contains two key state variables: one is the speed of respiratory movement, and the other is the displacement of respiratory movement. At the same time, the model uses the data measured by the acceleration sensor as an observation to compare and correct the estimated state variables.

[0085] Kalman filtering predicts the breathing state at the next moment based on the current estimated breathing state. Then, this predicted value is compared with the actual measured acceleration value to calculate the predicted error. According to the size of the error, the parameters in the mathematical model are dynamically adjusted to make the estimated value closer to the true value. Through continuous prediction, comparison, and correction, Kalman filtering finally obtains a smooth, low-noise acceleration signal that accurately reflects the changing trend of respiratory movement.

[0086] Through the two steps of outlier removal and Kalman filtering, the acceleration signal that originally contained spikes, jumps, and random noise was purified and optimized, becoming more stable and coherent, showing the true face of respiratory movement. This laid a good data foundation for the subsequent extraction of key parameters such as respiratory rate and respiratory depth, and helped to achieve high-quality respiratory monitoring and health assessment.

[0087] After the above preprocessing steps, the non-respiratory components in radar, pressure and acceleration signals are effectively suppressed, and the respiratory rhythm characteristics become obvious. The time-frequency analysis of the signals before and after preprocessing shows that the energy of the processed signals in the respiratory frequency band of 0.1~0.8Hz is significantly enhanced, while the energy of the heartbeat, body movement and other frequency bands is significantly weakened.

[0088] On this basis, the time domain and frequency domain characteristic parameters closely related to respiratory movement are extracted from each sensor signal. For radar signals, the respiratory frequency is calculated based on Chirp-Z transform, and the respiratory depth information is obtained through envelope extraction and energy integration; for pressure signals, the respiratory frequency is obtained through peak and trough detection, and the respiratory depth is estimated based on the cumulative distribution function of pressure values; for acceleration signals, the main respiratory frequency is obtained by autoregressive spectrum estimation.

[0089] For radar signals, Chirp-Z transform (CZT) is used to accurately estimate the respiratory frequency. CZT is a flexible digital spectrum analysis technology that can calculate the spectrum at any frequency resolution in any frequency range. Through CZT, the radar signal spectrum is calculated at a high resolution of 0.001 Hz in the 0.1-0.8 Hz frequency band, and then the peak position of the power spectrum density is detected, which corresponds to the main respiratory frequency. At the same time, the radar signal is enveloped, and then the product of the RMS value of the envelope and the respiratory frequency is calculated, which can be used to characterize the displacement amplitude of each breath, that is, the depth of breathing. In addition, the relative value of the tidal volume can be obtained by integrating the envelope signal, which further characterizes the respiratory intensity.

[0090] FMCW, or frequency modulated continuous wave, is based on a linear frequency modulated pulse signal (chirp), which is a sine wave whose frequency increases over time. The basic principle of FMCW radar to detect the breathing frequency and depth of a person's sleep is to detect the phase change of the FMCW signal in a specific range bin caused by the tiny vibration of the target.

[0091] Specifically, FMCW radar transmits frequency modulated continuous waves to the human body, and the received echo frequency changes in the same pattern as the transmitted frequency, but there is a time difference. This tiny time difference can be used to calculate the target distance. The detection of breathing frequency and breathing depth is achieved by analyzing and processing the phase of the received signal.

[0092] FMCW radar can detect the breathing rate and depth of human body during sleep through the following steps:

[0093] Signal transmission: FMCW radar transmits frequency modulated continuous wave signals to the human body.

[0094] Echo reception: Receive the echo signal reflected from the human body.

[0095] Signal processing: Process the received echo signal, including amplification, filtering, mixing, etc., to extract the signal related to breathing.

[0096] Phase analysis: By performing phase analysis on the processed signal, the phase change caused by breathing can be obtained.

[0097] Frequency extraction: Perform spectral analysis on phase changes to extract the breathing frequency.

[0098] Amplitude measurement: measures the depth of breathing based on changes in breathing rate.

[0099] In order to accurately estimate the respiratory rate, the mathematical method of Chirp-Z transform (CZT) is used. CZT can calculate the spectrum of the signal with any desired accuracy within any specified frequency range.

[0100] According to the physiological characteristics of respiratory movement, the normal respiratory rate is usually between 6 and 24 times per minute, which is 0.1 to 0.4 Hz when converted into Hertz. In order to fully cover all possible situations, the analysis range of CZT is set to 0.1 to 0.8 Hz, which is enough to cover the respiratory rate changes of most people. At the same time, in order to be able to finely distinguish subtle frequency differences, the calculation accuracy of the spectrum is set to 0.001 Hz, which can distinguish respiratory rates that differ by only 0.001 Hz (0.06 times per minute).

[0101] After calculating the high-resolution spectrum of the radar signal through CZT, the position of the maximum value point in the spectrum is extracted. The frequency corresponding to it is the dominant frequency of the respiratory movement, that is, the average respiratory frequency.

[0102] In addition to the breathing frequency, we also want to know how deep each breath is, which reflects the amplitude of the breathing movement. To this end, the radar signal is envelope extracted to obtain an envelope curve that reflects the slow fluctuations of the breathing process. Then, the root mean square value of the envelope curve is calculated, which represents the average level of breathing depth.

[0103] And further calculate the amount of air exchanged in each breath, i.e. tidal volume. By integrating the respiratory envelope curve, a relative value index proportional to the tidal volume can be obtained.

[0104] Among them, feature extraction of FMCW radar signals includes:

[0105] Step 1: Determine your breathing rate range:

[0106] The normal breathing rate range is 6 to 24 times per minute.

[0107] Convert to Hertz:

[0108] To fully cover all possible cases, set the analysis scope to:

[0109] Step 2: Calculate the high-resolution spectrum:

[0110] Use CZT (Chirp ZTransform) to calculate the high-resolution spectrum of the radar signal.

[0111] The spectrum calculation accuracy is set to: .

[0112] Step 3: Extract breathing rate:

[0113] Find the position of the maximum point in the spectrum, and the frequency corresponding to this position is the dominant frequency , which is the average respiratory rate.

[0114] Step 4: Envelope extraction:

[0115] Calculate the envelope of the radar signal to obtain an envelope curve reflecting the breathing process .

[0116] Step 5: Calculate breathing depth:

[0117] Calculate the root mean square value (RMS value) of the envelope curve, which represents the average level of breathing depth:

[0118] .

[0119] in, is the number of sampling points of the envelope curve, is the envelope curve at time point The value of .

[0120] Step 6: Calculate tidal volume (relative value):

[0121] Integrate the envelope curve to obtain a relative value index proportional to the tidal volume .

[0122] In actual calculation, the present invention uses the method of numerical integration: ,in, is the sampling interval.

[0123] All the parameters in the above process are defined as follows:

[0124] : Frequency

[0125] : Dominant frequency of the respiratory spectrum

[0126] : Envelope curve

[0127] : Root mean square value (RMS value) of the envelope curve

[0128] : Relative value index of tidal volume

[0129] : time

[0130] : Number of sampling points of the envelope curve

[0131] : Sampling interval

[0132] Among them, 4 pressure sensors are arranged vertically inside the mattress, located on the left and right sides of the chest and abdomen, respectively, to achieve accurate perception of breathing pressure. The sensors are connected to the data acquisition module through conductive fabric or ultra-fine elastic cables.

[0133] After preprocessing the original signals collected by the four pressure sensors, the preprocessed pressure signals of the four sensors are averaged to obtain a comprehensive pressure signal sequence. The fused signal more comprehensively reflects the respiratory pressure changes of the entire chest and abdomen.

[0134] For the fused pressure signal, the peak and trough detection algorithm with adaptive threshold is used to locate the starting point of inspiration and the end point of expiration of each respiratory cycle, and then the respiratory frequency is calculated based on the time difference, and the average frequency of each cycle is calculated. While obtaining the respiratory frequency, this method can also segment the complete respiratory cycle segment. For each respiratory segment, the cumulative distribution function (CDF) of the pressure value is calculated, and the pressure values ​​at the 10%, 50%, and 90% quantiles are extracted as characteristic parameters to characterize the dynamic range of pressure changes during the breathing process, thereby estimating the breathing depth.

[0135] Specifically, the pre-processed pressure signals of the four sensors are averaged to obtain a comprehensive pressure signal sequence. The fused signal more comprehensively reflects the respiratory pressure changes of the entire chest and abdomen.

[0136] The local mean and standard deviation are calculated using the sliding window method on the fused pressure signal. The window length can be set to 5 seconds, and the sliding step is 1 second. According to the local mean and standard deviation, the peak and trough detection threshold is adaptively set, such as the local mean ± 1.5 times the local standard deviation.

[0137] Then, peak and trough detection is performed. In each sliding window, all local maximum points exceeding the peak threshold are found as candidate peaks, and all local minimum points below the trough threshold are found as candidate troughs. The candidate peaks and troughs are paired in chronological order to obtain preliminary inhalation and exhalation cycle fragments.

[0138] Respiratory cycle screening: according to the physiological range of normal people, set the upper and lower limits of the respiratory cycle time (such as 110 seconds). Calculate the time span of each preliminary respiratory cycle segment, remove segments with too small or too large time span, and obtain the final respiratory cycle segment sequence.

[0139] Respiratory frequency calculation: For each respiratory cycle segment, the respiratory frequency of the cycle is calculated by the time difference between the start of inspiration and the end of exhalation. The average respiratory frequency of all cycles is taken to obtain the average respiratory frequency within the minute. At the same time, the standard deviation of the respiratory frequency is calculated to reflect the stability of the respiratory rhythm.

[0140] Respiratory depth estimation: For each respiratory cycle segment, calculate the cumulative distribution function (CDF) of its pressure value. Extract the pressure values ​​at the 10%, 50%, and 90% quantiles of the CDF curve as characteristic parameters to characterize the dynamic range of pressure changes in the respiratory cycle. The difference between the 90% quantile pressure and the 10% quantile pressure can be used as the depth estimation value of the respiratory cycle. Take the average of the depth estimation values ​​of all respiratory cycles to obtain the average respiratory depth within the minute. Calculate the standard deviation of the respiratory depth and then calculate the standard deviation of the respiratory depth to reflect the variability of the respiratory depth.

[0141] Among them, the feature extraction of the fused pressure signal includes:

[0142] Step 1: Sliding window to calculate local mean and standard deviation

[0143] Window length: Set to (seconds), can be 5 seconds.

[0144] Slide step: Set to (seconds), usually 1 second.

[0145] Calculate the local mean and standard deviation

[0146] For each sliding window :

[0147] Local mean : ;

[0148] in, Is the first The pressure value of each sampling point, is the number of sampling points in the window.

[0149] Local standard deviation : ;

[0150] Step 2: Peak and trough detection threshold

[0151] Peak Threshold:

[0152] Trough Threshold:

[0153] Step 3: Peak and trough detection

[0154] In each sliding window:

[0155] Candidate peaks: Find all peaks exceeding The local maximum point of .

[0156] Candidate troughs: Find all valleys below The local minimum point of .

[0157] Step 4: Breathing cycle screening

[0158] According to the normal physiological range of a normal person, set the upper and lower limits of the breathing cycle time (such as 110 seconds).

[0159] Respiratory cycle time span: is the time span of each preliminary respiratory cycle segment.

[0160] Filter: Remove time span Clips that are too small or too large are retained for 1 second Fragments of seconds.

[0161] Step 5: Respiratory rate calculation

[0162] Respiratory rate :

[0163] in, For the The time span of a respiratory cycle segment.

[0164] Average respiratory rate : ;

[0165] in, is the number of respiratory cycle segments in that minute.

[0166] Respiratory rate standard deviation : ;

[0167] Step 6: Breathing Depth Estimation

[0168] Cumulative Distribution Function (CDF): Calculate the cumulative distribution function of the pressure value for each respiratory cycle segment.

[0169] Quantile extraction:

[0170] The pressure value at the 10% quantile of the curve.

[0171] curve The pressure value at the quantile.

[0172] : CDF curve The pressure value at the quantile.

[0173] Respiratory depth estimation : .

[0174] Average breathing depth : .

[0175] Standard deviation of breathing depth : .

[0176] Coefficient of variation of respiratory depth : .

[0177] All parameters in the above process are defined as follows:

[0178] : Window length (seconds)

[0179] : Sliding step length (seconds)

[0180] : No. The local mean of the sliding windows

[0181] : No. The local standard deviation of the sliding window

[0182] : Peak detection threshold

[0183] : Valley detection threshold

[0184] : Breathing cycle time span (seconds)

[0185] : No. Respiratory frequency per respiratory cycle segment

[0186] : Average respiratory rate

[0187] : Standard deviation of respiratory rate

[0188] : Coefficient of variation of respiratory rate

[0189] :CDF curve Pressure value at the quantile

[0190] : CDF curve Pressure value at the quantile

[0191] : CDF curve Pressure value at the quantile

[0192] : No. Depth estimate for respiratory cycle segments

[0193] : Average breathing depth

[0194] : Standard deviation of breathing depth

[0195] : Coefficient of variation of breathing depth

[0196] : The number of respiratory cycle segments in this minute.

[0197] The sliding window calculates the above breathing frequency and breathing depth characteristics minute by minute to obtain the continuous change trend of the breathing pattern parameters.

[0198] For the acceleration signal, the autoregressive (AR) model is used to estimate the power spectrum and obtain the respiratory frequency. The AR model parameters are estimated by the Burg algorithm, and then the power spectrum density is calculated by the AR parameters. The maximum value of PSD is calculated within the respiratory frequency band, and the corresponding frequency is the respiratory frequency.

[0199] Among them, the extraction of breathing features of acceleration signals includes:

[0200] Step 1: Burg algorithm estimates AR model parameters

[0201] The autoregressive (AR) model defines a The AR model of order can be expressed as:

[0202]

[0203] in:

[0204] is a time series signal

[0205] are the parameters of the AR model

[0206] It is white noise

[0207] The Burg algorithm is an effective method for estimating the parameters of the AR model. The main steps of the algorithm are as follows:

[0208] initialization: , ;

[0209] in:

[0210] is a time series signal.

[0211] is the initial forward prediction error.

[0212] is the initial backward prediction error.

[0213] and are the forward and backward prediction errors, respectively, and are iteratively calculated (for ):

[0214] Calculate the reflection coefficient : ,

[0215] Update AR parameters:

[0216] ,

[0217] ,

[0218] Update the forward and backward prediction errors:

[0219] ,

[0220] ,

[0221] Final AR parameters: .

[0222] All parameter definitions in the above process are as follows:

[0223] : Time series signal, indicating the The signal value at .

[0224] : Initial AR model parameters, set to 1.

[0225] : Initial AR model parameters, set to .

[0226] : Initial forward prediction error, equal to .

[0227] : Initial backward prediction error, equal to .

[0228] : The length of the time series.

[0229] : The order of the current iteration, .

[0230] : No. The reflection coefficient of the order.

[0231] : No. The forward prediction error at step .

[0232] : No. The backward prediction error at step .

[0233] No. AR model parameters at the time of order.

[0234] : The final AR model parameters are equal to The parameters of the order, .

[0235] Step 2: Calculate the Power Spectral Density (PSD)

[0236] Power spectral density of AR model The power spectral density of the model can be expressed as:

[0237]

[0238] in:

[0239] is the variance of white noise, are the parameters of the AR model, It's the frequency.

[0240] : Power spectral density, which represents the power distribution of the signal at different frequencies. : Variance of white noise. Represents the power of the part of the data that is not explained by the model. : Parameters of the AR model, specifically the coefficients estimated by the Burg algorithm, . : Frequency, usually measured in Hz during analysis. : Imaginary unit, satisfying . : The frequency response of the AR model, given by the AR parameters and frequency The complex sum of .

[0241] Step 3: Find the maximum value of PSD in the respiratory frequency band. The respiratory frequency band usually ranges from 0.1 Hz to 0.5 Hz.

[0242] In the respiratory frequency band, calculate the maximum value of PSD and its corresponding frequency: ,in, is the respiratory frequency, which indicates the frequency with the largest power spectral density in the respiratory frequency band.

[0243] Through the above steps and formulas, the Burg algorithm can be used to estimate the AR model parameters, and these parameters can be used to calculate the power spectral density, and finally the maximum value of PSD can be obtained within the respiratory frequency band, and the corresponding frequency is the respiratory frequency.

[0244] From the above process, it can be seen that feature extraction for FMCW radar signals includes:

[0245] Dominant frequency of the respiratory spectrum , the RMS value of the envelope curve representing the average level of breathing depth , and relative value index of tidal volume .

[0246] The feature extraction of the fused pressure signal includes:

[0247] Average respiratory rate , respiratory rate standard deviation , average breathing depth , coefficient of variation of respiratory depth .

[0248] Respiratory feature extraction for acceleration signals includes: respiratory frequency .

[0249] In non-invasive respiratory monitoring, a multi-sensor information fusion method is adopted to comprehensively utilize the respiratory features extracted by FMCW radar, pressure sensor and accelerometer to construct a sensor matrix that comprehensively reflects the micro-movements of the human body.

[0250] Among them, the common variables (respiratory rate and respiratory depth) are judged and calibrated for differences. By calculating the differences between the estimated values ​​of different sensors, it is judged whether there is a significant deviation. If the deviation is large, the deviated estimated values ​​are weighted averaged according to the reliability of each sensor to obtain a consistent estimate after calibration.

[0251] Among them, the non-common variables (tidal volume, respiratory rate standard deviation and respiratory depth variation coefficient) are also judged and calibrated. For the tidal volume index unique to the FMCW radar, linear scaling is performed according to the scale factor of the radar respiratory depth calibration. For the respiratory rate standard deviation and respiratory depth variation coefficient unique to the pressure sensor, the difference between them and the corresponding variability indicators of other sensors is calculated to determine whether calibration is required. If the difference is significant, it is linearly adjusted according to the degree of difference to obtain the calibrated estimate.

[0252] Finally, the calibrated common variable estimates and non-common variable estimates are fused into the human micro-motion sensor matrix. Each row of the human micro-motion sensor matrix represents a time point, and each column represents a respiratory characteristic parameter. Finally, a matrix form that comprehensively reflects human micro-motion is obtained, which contains information on respiratory rate, respiratory depth, tidal volume, respiratory rate variability, and respiratory depth variability.

[0253] Among them, for the common variables, calibration is performed in the following manner:

[0254] For respiratory rate calibration, the calibration process includes:

[0255] Step 1. Calculate respiratory rate:

[0256] The dominant frequency of the breathing spectrum of FMCW radar

[0257] Average respiratory rate of pressure signal

[0258] Respiration rate of acceleration signal

[0259] Step 2. Calculate the mean and standard deviation:

[0260]

[0261] Step 3. Calculate the difference: ;

[0262] Step 4. Determine whether calibration is required. If , calibration is required.

[0263] Step 5. Adjust the frequency estimate with the largest deviation to the weighted average of the frequency estimates of the other two sensors, with the weights determined according to their respective reliability (signal-to-noise ratio):

[0264] .

[0265] Among them, the parameters in the respiratory rate calibration process are defined as follows:

[0266] 1. Dominant frequency of respiratory spectrum :

[0267] The dominant frequency of the breathing rate measured by FMCW radar.

[0268] 2. Average respiratory rate :

[0269] Average respiratory rate measured from the pressure signal.

[0270] 3. Respiratory rate :

[0271] Respiratory rate measured from the acceleration signal.

[0272] 4. Mean respiratory rate :

[0273] The average of the respiratory rates measured by the three sensors.

[0274] 5. Respiratory rate standard deviation :

[0275] Standard deviation of respiratory rate measured by the three sensors.

[0276] 6. Respiratory rate difference :

[0277] The absolute difference between the respiratory rate of a single sensor and the mean.

[0278] 7. Calibrated breathing rate :

[0279] Calibrated respiratory rate estimate.

[0280] Among them, the calibration of breathing depth includes:

[0281] Steps: 1. Calculate breathing depth: RMS value of FMCW radar , average breathing depth of pressure signal .

[0282] Step 2. Normalization:

[0283]

[0284]

[0285] Step 3. Calculate the difference: ;

[0286] Step 4. Determine whether calibration is required: If , calibration is required.

[0287] Step 5. Take the weighted average of the breathing depth estimates from the two sensors: .

[0288] The parameters in the calibration process of breathing depth are defined as follows:

[0289] 1. RMS value :

[0290] The RMS value of breathing depth measured by FMCW radar.

[0291] 2. Average breathing depth :

[0292] Average breathing depth measured from the pressure signal.

[0293] 3. Normalized RMS value :

[0294] The RMS value of the FMCW radar breathing depth after normalization.

[0295] 4. Normalized average breathing depth :

[0296] The normalized pressure signal represents the average breathing depth.

[0297] 5. Breathing depth difference :

[0298] The absolute difference between the normalized RMS value and the normalized mean breathing depth.

[0299] 6. Calibrated breathing depth :

[0300] Calibrated estimate of breathing depth.

[0301] Difference determination and calibration for non-shared variables, including:

[0302] For the tidal volume calibration determined by FMCW radar, the following process is involved:

[0303] Steps: 1. Relative value indicator :

[0304] Independently estimated by FMCW radar, reflecting changes in breathing depth.

[0305] Step 2. Calibrate the scale factor based on the radar breathing depth ,right To perform linear scaling:

[0306] .

[0307] Among them, the parameters in the tidal volume calibration process are defined as follows:

[0308] 1. Tidal volume :

[0309] A relative value indicator reflecting changes in breathing depth estimated by FMCW radar.

[0310] 2. Calibrated tidal volume :

[0311] Tidal volume adjusted according to the scale factor calibrated to the radar breathing depth.

[0312] The standard deviation calibration of the respiratory rate determined by the pressure sensor includes the following steps:

[0313] Steps: 1. Calculate the standard deviation:

[0314] Standard deviation of frequency estimates for FMCW radar , the standard deviation of the accelerometer frequency estimate ;

[0315] Step 2. Calculate the mean and standard deviation:

[0316] ,

[0317] ,

[0318] Step 3. Calculate the difference: ;

[0319] Step 4. Determine whether calibration is required: If , calibration is required.

[0320] Step 5. Calibration method: .

[0321] Among them, the parameters in the respiratory rate standard deviation calibration process are defined as follows:

[0322] 1. Frequency standard deviation :

[0323] The standard deviation of the respiratory rate estimate measured by the pressure sensor.

[0324] 2. Radar frequency standard deviation :

[0325] Standard deviation of breathing rate estimates measured by FMCW radar.

[0326] 3. Accelerometer frequency standard deviation :

[0327] The standard deviation of the respiratory rate estimate measured by the accelerometer.

[0328] 4. Frequency Standard Deviation Mean :

[0329] Average of the standard deviations of the frequencies measured by the three sensors.

[0330] 5. Frequency Standard Deviation Standard Deviation :

[0331] The standard deviation of the frequency standard deviations measured by the three sensors.

[0332] 6. Frequency standard deviation difference :

[0333] The absolute difference between the standard deviation of the frequency of a single sensor and the mean.

[0334] 7. Frequency standard deviation after calibration calibrated:

[0335] The calibrated estimate of the frequency standard deviation.

[0336] The calibration of the coefficient of variation of breathing depth determined by the pressure sensor includes the following process:

[0337] Steps: 1. Calculate the coefficient of variation: The coefficient of variation of the RMS value of the FMCW radar ;

[0338] Step 2. Calculate the mean and standard deviation:

[0339]

[0340] Step 3. Calculate the difference: ;

[0341] Step 4. Determine whether calibration is required. If , calibration is required.

[0342] Step 5. According to The ratio of the threshold value will Linear Adjustment:

[0343] Among them, the parameters in the calibration process of the coefficient of variation of breathing depth are defined as follows:

[0344] 1. Coefficient of variation of breathing depth :

[0345] Coefficient of variation of breathing depth measured by pressure sensor.

[0346] 2. Radar RMS coefficient of variation :

[0347] Coefficient of variation of the RMS value of breathing depth measured by FMCW radar.

[0348] 3. Accelerometer Amplitude Coefficient of Variation :

[0349] Coefficient of variation of estimates of breathing depth measured by accelerometer.

[0350] 4. Mean coefficient of variation :

[0351] The average of the coefficient of variation measured by the three sensors.

[0352] 5. Coefficient of variation standard deviation :

[0353] Standard deviation of the coefficient of variation measured by the three sensors.

[0354] 6. Difference in coefficient of variation :

[0355] The absolute difference between the coefficient of variation of a single sensor and the mean.

[0356] 7. Coefficient of variation after calibration :

[0357] Calibrated coefficient of variation estimates.

[0358] After completing the difference judgment and calibration of shared variables and unshared variables, a human body micro-motion sensing matrix that comprehensively reflects human body micro-motion is constructed.

[0359] Sensor Matrix Each row of represents a time point (e.g., every 10 seconds), and each column represents a respiratory characteristic parameter. The matrix elements are the estimated values ​​after calibration.

[0360] Among them, every 10 seconds, a human body micro-motion sensing matrix that comprehensively reflects human body micro-motion can be constructed based on a 1-minute window covering the collected data of all sensors in the time period one minute before the current time point.

[0361] For example, in time point, sensor matrix No. A row can be represented as:

[0362]

[0363] in:

[0364] After calibration Estimated respiratory rate at each time point

[0365] After calibration Estimated breathing depth at each time point

[0366] After calibration Relative value of tidal volume at each time point

[0367] After calibration Estimated standard deviation of respiratory rate at each time point

[0368] After calibration Estimated coefficient of variation of respiratory depth at each time point

[0369] Through the above fusion, the radar, pressure and acceleration data fully interact and complement each other, overcoming the limitations of a single sensor and greatly improving the accuracy and robustness of respiratory motion restoration. The fused respiratory signal not only accurately reflects the respiratory rhythm characteristics such as respiratory frequency and depth, but also contains the dynamic evolution information of the respiratory pattern, providing more comprehensive and reliable data support for subsequent respiratory event analysis and anomaly detection.

[0370] The respiratory event recognition module builds a respiratory event classification model, classifies the respiratory events belonging to the human micro-motion sensor matrix, and identifies three typical events: normal breathing, rapid breathing, and apnea.

[0371] In order to realize the respiratory event recognition module, a machine learning method can be used to build a respiratory event classification model. The model takes the sensor matrix as input, and by learning the characteristic patterns of different respiratory events, it classifies the input respiratory data and identifies three typical events: normal breathing, rapid breathing, and apnea.

[0372] A labeled respiratory event dataset needs to be prepared. The dataset should contain enough samples, each of which corresponds to a sensor matrix within a time window and has a corresponding respiratory event label (normal breathing, rapid breathing, or apnea). The labeled dataset is divided into a training set and a test set. The training set is used to train the classification model, and the test set is used to evaluate the performance of the model. You can choose a suitable machine learning algorithm, such as support vector machine (SVM) or convolutional neural network (CNN), to build a classification model.

[0373] In the training phase, the sensor matrix in the training set is used as input, and the corresponding respiratory event label is used as the target output. The model is trained using the selected machine learning algorithm. During the training process, the model automatically learns the characteristic patterns of different respiratory events in the sensor matrix, such as the changing rules of respiratory frequency, respiratory depth, respiratory rhythm, etc.

[0374] After training is completed, the performance of the model is evaluated using the test set. The sensor matrix in the test set is input into the trained classification model. The model will classify each input sensor matrix according to the learned feature pattern and output the corresponding respiratory event category. The model's prediction results can be compared with the actual respiratory event labels to calculate the model's accuracy, precision, recall and other evaluation indicators to quantify the model's performance.

[0375] If the performance of the model meets the requirements, it can be integrated into the respiratory event recognition module.

[0376] In actual application, when new sensor matrix data arrives, a new human micro-motion matrix is ​​constructed every 10 seconds and input into the trained classification model. The model will automatically identify the respiratory event category to which the sensor matrix belongs and output the corresponding results (normal breathing, rapid breathing or apnea).

[0377] The user interaction module provides respiratory monitoring information display and human-computer interaction functions, including key indicator display, alarm and other functions. The user interaction module displays the trend changes of key indicators such as respiratory rate and depth-shallowness ratio in real time through radar charts, line charts, etc., and automatically triggers an alarm prompt when a apnea event is detected that lasts for more than 10 seconds (one detection cycle is a apnea event).

[0378] The present invention uses multi-sensor data to perform all-round non-invasive collection of human chest and abdominal respiratory movements, avoiding the discomfort and risks caused by traditional invasive collection methods, and improving the user experience. Through the fusion and synchronous processing of multi-sensor data, the accuracy and reliability of respiratory monitoring are significantly improved, and the respiratory status of the human body can be more accurately reflected. Through the respiratory event classification model and user interaction module, real-time monitoring and alarm functions of respiratory events are realized, abnormal respiratory events can be detected in time and alarms can be issued, improving the safety of users.

[0379] It should be noted that the computer-readable medium mentioned above in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. Computer readable signal media may also be any computer readable medium other than computer readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0380] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0381] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0382] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0383] The units involved in the embodiments described in the present disclosure may be implemented by software or hardware, wherein the name of a unit does not, in some cases, limit the unit itself.

[0384] The above introduces the preferred embodiments of the present invention, which is intended to make the spirit of the present invention clearer and easier to understand, but is not intended to limit the present invention. All modifications, substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection outlined by the claims attached to the present invention.

Claims

1. A method for non-sensing monitoring of physiological indicators based on a human body micro-motion sensor matrix, the method comprising: Step 1: Use the sensor matrix to collect the respiratory movements of the chest and abdomen of the human body in an all-round and non-invasive manner: Step 2: The data acquisition and processing module synchronously collects, de-noises, filters and extracts features from multi-sensor data to form a human body micro-motion sensing matrix; Step 3: The respiratory event recognition module performs event recognition on the human body micro-motion sensor matrix through the respiratory event classification model; Step 4: The user interaction module intuitively displays the physiological indicators of respiratory rate and respiratory depth, and provides auxiliary functions of alarm; Among them, the sensor matrix integrates multiple sensors such as FMCW radar, pressure sensor, and acceleration sensor; Among them, feature extraction for FMCW radar signals includes: The dominant frequency f of the respiratory spectrum peak , represents the RMS value of the envelope curve of the average level of breathing depth A rms , and the relative value index of tidal volume V tidal ; The feature extraction of the fused pressure signal includes: Average respiratory rate Respiratory rate standard deviation σ f , average breathing depth Coefficient of variation of respiratory depth CV D ; The respiratory feature extraction of acceleration signal includes: respiratory frequency f resp ; Among them, the common variables are respiratory rate and respiratory depth; the non-common variables are tidal volume, respiratory rate standard deviation and respiratory depth coefficient of variation; Among them, the difference judgment and calibration of the common variables include calculating the difference between the estimated values ​​of different sensors to determine whether the deviation exceeds the threshold. If so, the estimated values ​​of the deviation are weighted averaged according to the reliability of each sensor to obtain the consistent estimate after calibration; Difference judgment and calibration of non-shared variables, including linear scaling of the tidal volume index determined by the FMCW radar according to the scale factor calibrated by the radar breathing depth; Also, for the standard deviation of the respiratory rate and the coefficient of variation of the respiratory depth determined based on the pressure sensor, by calculating the difference between the standard deviation and the coefficient of variation of the corresponding variability index of the acceleration sensor and / or the FMCW radar, it is determined whether calibration is required. When the difference exceeds a threshold, a linear adjustment is performed according to the degree of difference to obtain a calibrated estimated value; After completing the difference judgment and calibration of shared variables and unshared variables, a human body micro-motion sensing matrix that comprehensively reflects human body micro-motion is constructed, including: The calibrated variable estimation values ​​are integrated into a human body micro-motion sensor matrix; each row of the human body micro-motion sensor matrix represents a time point, and each column represents a respiratory characteristic parameter, and the human body micro-motion sensor matrix includes information on respiratory frequency, respiratory depth, tidal volume, respiratory frequency variability, and respiratory depth variability; Among them, every 10 seconds, based on a 1-minute window covering the collected data of all sensors in the time period from the current time point to one minute before, a human body micro-motion sensing matrix reflecting human body micro-motion is constructed; Among them, a respiratory event classification model is constructed to classify the respiratory events belonging to the human body micro-motion sensor matrix and identify typical events, wherein the typical events at least include: normal breathing, rapid breathing, and apnea.

2. The method for non-sensing monitoring of physiological indicators based on human body micro-motion sensor matrix as claimed in claim 1, characterized in that: Clock frequency division and trigger control circuit are used to generate unified sampling clock and synchronous trigger signal.

3. The method for monitoring physiological indicators without sensing based on human body micro-motion sensing matrix as claimed in claim 1, characterized in that: The collected raw data is first preprocessed, including: The radar signal uses an FIR bandpass filter. The pressure signal is corrected for baseline drift and filtered with a 60Hz power frequency notch. Wavelet transform is used for signal enhancement and smoothing. The acceleration data uses an adaptive threshold piecewise linear interpolation method to remove abnormal points, and then the Kalman filter algorithm is used for smoothing and noise reduction.

4. The method for non-sensing monitoring of physiological indicators based on human body micro-motion sensor matrix as claimed in claim 3, characterized in that: Extract characteristic parameters of respiratory motion from preprocessed sensor signals.

5. The method for non-sensing monitoring of physiological indicators based on human body micro-motion sensor matrix as claimed in claim 3, characterized in that: For FMCW radar signals, the breathing frequency is calculated based on the Chirp-Z transform, and the breathing depth information is obtained through envelope extraction and energy integration; for pressure signals, the breathing frequency is obtained through peak and trough detection, and the breathing depth is estimated based on the cumulative distribution function of the pressure value; for acceleration signals, the autoregressive spectrum estimation is used to obtain the main breathing frequency.

6. A physiological index non-sensing monitoring system based on a human body micro-motion sensor matrix, the system comprising a processor and a memory, the processor being used to run a computer code in the memory to execute a method as claimed in any one of claims 1 to 5.

Images